Problem Set 1

UCSI University Department of Electrical & Electronics Engineering EE406 Control Systems Introduction to Control Systems Problem Set
1 Due : In-class, 10 May 2012
Question 1
Many closed-loop and open-loop control systems can be found in homes. List several examples and provide explanations and justifications for your classifications.
Question 2
One of the important applications of control system is in the air traffic control systems, where engineers are developing an air traffic control system and collision avoidance system by using the Global Positioning System (GPS) navigation satellites. The GPS allows each aircraft to know its position in the air space landing corridor very precisely. Sketch a block diagram depicting how an air traffic controller might use the GPS for aircraft collision avoidance. Provide also necessary justification and explanations.
Question 3
Another interesting application of control system is in the development of an intermittent automobile windshield wiper in which it operates based on the concept of adjusting its wiping cycle according to the intensity of the rain. Sketch a suitable block diagram for a windshield wiper control system.
Note
(1) Related articles are attached. (2) Plagiarized work will result in zero marks.
AFRL-VA-WP-TP-2002-317 AUTOMATIC AIR COLLISION AVOIDANCE SYSTEM

Y. Ikeda B. Nguyen A. Barileld B. Sundqvist S. Jones
JUNE 2002
Approved for public release; distribution is unlimited.
2002 SICE This work is copyrighted. The United States has for itself and others acting on its behalf an unlimited, paid-up, nonexclusive, irrevocable worldwide license. Any other form of use is subject to copyright restrictions.
AIR VEHICLES DIRECTORATE AIR FORCE RESEARCH LABORATORY AIR FORCE MATERIEL COMMAND WRIGHT-PATTERSON AIR FORCE BASE, OH 45433-7542
î^ms6 m
REPORT DOCUMENTATION PAGE
Form Approved 0MB No. 0704-018
The public reporting burden for this collection of information is estimated to average 1 hour per response, inchiding (he time for reviewing instructions, seardiing existing data sources, searching eristing data sources gathering and mamtammg the data needed, and completing and reviewing the collectfan of information. Send comments regarding this burden estimate or any other aspect of this colleetbn of information including ' suggestions for reducing this burden, to Department of Defense, Washington Headquarters Services, Directorate for Information Operatbns and Reports (0704-0188), 1215 Jefferson Davis Highway, Suite 1204 Arlmgton, VA 22202-4302. Respondents should be aware that notwithstanding any other provision of law, no person shall be subject to any penalty for Ming to comply with a collection of information if it does not display a currently valid 0MB control number. PLEASE DO NOT RETURN YOUR FORM TO THE ABOVE ADDRESS.
1, REPORT DATE (DD-MM-YY) Jxme 2002

4. TITLE AND SUBTITLE
2. REPORT TYPE
3. DATES COVERED fFrom-To)
Journal Article Preprint

5a, CONTRACT NUMBER
AUTOMATIC AIR COLLISION AVOIDANCE SYSTEM
F33615-01-2-3103 5b. GRANT NUMBER 5c. PROGRAM ELEMENT NUMBER 6321IF
6, AUTHOR(S)
5cl. PROJECT NUMBER
Y, Bceda (The Boeing Company) B. Nguyen and A. Barfield (AFRL/VACC) B. Sundqvist (Saab AB) S. Jones (Lockheed Martin Aeronautics)
7. PERFORMING ORGANIZATION NAME(S) AND ADDRESS(ES)
486U
5e. TASK NUMBER
17
5f. WORK UNIT NUMBER
00
8. PERFORMING ORGANIZATION REPORT NUMBER
Lockheed Martin Aeronautics P.O. Box 748 Fort Worth, TX 7610]
The Boemg Company AFRL/VACC
Saab AB
10. SPONSORING/MONITORING AGENCY ACRONYM(S)
9. SPONSORING/MONrrORING AGENCY NAME(S) AND ADDRESS(ES)
Air Vehicles Directorate Air Force Research Laboratory Air Force Materiel Command Wright-Patterson Air Force Base, OH 45433-7542
12. DISTRIBUTION/AVAILABILITY STATEMENT
AFRLAÂCC
11. SPONSORING/MONITORING AGENCY REPORT NUMBER(S)
AFRL-VA-WP-TP-2002-317 Approved for public release; distribution is unlimited.

13. SUPPLEMENTARY NOTES
To be presented in the Society for Instrument and Control Engineers (SICE) Annual Conference, August 5,2002. O 2002 SICE. This work is copyrighted. The United States has for itself and others acting on ite behalf an unlimited, paid-up, nonexclusive, irrevocable worldwide license. Any other form of use is subject to copyright restrictions. The other contract on this report is F33615-01-2-3101 for The Boeing Company.
14. ABSTRACT (Maximum 200 IVords)
This paper presents an algorithm for an Automatic Air Collision Avoidance System under development by the U.S. Ah Force and its Swedish counterpart, Forsvaret Matcrielverk (FMV). The algorithm uses optimal coordmated escape maneuvers to avoid mid-air colhsion, while satisfying the imposed system requirements. In addition, the algorithm can simultaneously accommodate multiple aircraft in a collision course by activating the coordinated escape nmneuvers. On the other hand, the algorithm has logic to allow close formation flight and rejoin without activating the escape maneuver. The algorithm is designed to operate safely against failure and GPS/data link dropout.
IS. SUBJECT TERMS
Air Colhsion Avoidance, Automatic System, Coordinated Escape Maneuver, Nuisance Free, Failure Safe
16. SECURITY CLASSIFICATION OF; a. REPORT b. ABSTRACT c. THIS PAGE 17. LIMITATION 18. NUMBER OF 19a, NAME OF RESPONSIBLE PERSON (Monitor) OF ABSTRACT: PAGES
Unclassified
Unclassified
Unclassified
SAR
12
Ba Nguyen
19b. TELEPHONE NUMBER (Include Area Code)
(937)255-1135
Standird Form 298 (Rev. 8-98) Prescribed by ANSI Sl4 Z39-18
SICE02-0631
Automatic Air Collision Avoidance System

Y. Ikeda', B. Nguyen^ A. Barfield^ B. Sundqvist^ S. Jones*
1 The Boeing Company, P.O. Box 516, St. Louis, MO 63166-0516, U.S.A. 2 Air Force Research Laboratory, AFRL/VACC, WPAFB, OH 45433, U.S.A. 3 Saab AB, Linkoping, Sweden 4 Lockheed Martin Aeronautics, P.O. Box 748, Fort Worth, TX 76101, U.S.A. Abstract: This paper presents an algorithm for an Automatic Air Collision Avoidance System under development by the U.S. Air Force and its Swedish counter part, Forsvaret Materielverk (FMV). The algorithm uses optimal coordinated escape maneuvers to avoid mid-air collision, while satisfying the imposed system requirements. In addition, the algorithm can simultaneously accommodate multiple aircraft in a collision course by activating the coordinated escape maneuvers. On the other hand, the algorithm has logic to allow close formation flight and rejoin without activating the escape maneuver. The algorithm is designed to operate safely against failure and GPS / data link dropout. Keywords: Air Collision Avoidance, Automatic System, Coordinated Escape Maneuver, Nuisance Free, Failure Safe Introduction: The United States Air Force (USAF) Safety Center has reported that mid-air collisions are the leading cause of fighter aircraft losses. Mid-air collisions pose a threat to aviation safety whether it is air-to-air combat training or a formation rejoin mission. The Swedish AF recently had an incident where the flight lead was almost hit by his wingman during air-to-air combat training. Tomorrow's USAF will use uninhabited air vehicles (UAVs) for a number of missions. High-risk missions are ideal candidates for these vehicles. However, for them to realize their full potential and become an integral part of USAF airspace operations, they must be safe to operate in the same airspace with manned aircraft. For manned aircraft, several air colUsion avoidance systems are in use that alert pilots to potential collisions at distances of several miles. All of these systems require pilot action to avoid the collision. These manual collision avoidance systems often create nuisance events that interfere with the pilot's ability to perform the mission. Pilots quickly become tired of nuisance alerts and turn the systems off. Currently, the United State Air Force Research Laboratory, The Boeing Company, Lockheed Martin, FMV, and Saab are jointly developing the Automatic Air Collision Avoidance System (Auto ACAS). The program is divided into two phases; a conceptual study phase (phase 1), and a system design and flight test phase (phase 2). Boeing, Saab and Lockheed Martin completed phase 1 in May, 2001. During this phase, Saab and Boeing developed two different algorithms. The architecture of the Boeing algorithm was presented in Reference 1. The system design and flight test phase started in August 2001 and will be completed in August 2003. Saab and Boeing are jointly developing a generic algorithm as a result of the study phase. Lockheed Martin will tailor the algorithm and integrate it into an F-16 for flight test in July 2003. The following are the basic requirements for the Auto ACAS. The system will: 1) provide a last resort emergency maneuver to avoid collision with another air vehicle(s) - (It is expected to initiate an escape maneuver less than L5sec prior to the minimum time-to-evade), 2) provide a predictable response similar to the response a pilot would command in order to avoid a collision, 3) protect against the unforeseen events that cause collisions, 4) accommodate failure conditions with acceptable results, 5) not interfere with the pilot's tasks except to prevent aircraft loss, 6) work with GPS or data link loss, 7) be fully verified, validated, and tested with redimdant elements to insure safe vehicle operation, 8) force a UAV to maneuver to avoid a collision with a manned aircraft before the manned vehicle must maneuver.
ASC- 0 2
13
In what follows, the architecture of the Auto ACAS algorithm is first discussed. Then discussion on how the Auto ACAS algorithm is designed to meet the requirements follows. Simulation results showing the performance of the algorithm are also presented. Finally, the ACAS algorithm development status is summarized in the concluding remarks. 1. Algorithm Architecture Figure 1 illustrates the architecture of the algorithm. The algorithm operates with three basic assumptions. First, because all aircraft maneuver to avoid collision, all aircraft must operate with data provided at the same absolute time. For data received over a data link, this means that the data latency must be considered, and dead-reckoning correction must be applied to bring all data to the same absolute time. Second, the computations in each aircraft processor must result in the same escape maneuver, so the algorithm in each aircraft must operate on the same data. Third, the escape angle computed by each aircraft must not be allowed to vary by a large amount as the time to perform the escape maneuver approaches, since the escape solution for all aircraft must be stable.
Initialization For the first time through the algorithm, the first escape maneuver is initialized using the Accurate Aircraft Response Model (AARM) for a zerobank escape, and two other escapes are computed at a preselected maximum spread angle using the Simplified Aircraft Response Model (SARM). Data Link Data In Brings in data for "In-Network" intruders. Out-Network Brings in data for "Out-Network" intruders such as radar, sensor, or third-party data. Coarse Filter Selects up to three intruders (aircraft other than your own) that should be considered the biggest threats to the host (your ovra aircraft). Selected aircraft are evaluated by the remainder of the algorithm. Dead Reckon Adjusts intruder and host data so that all aircraft (intruders and host) are evaluated at the same absolute time. Positions are corrected using constant velocities and known data latencies.
n=>
Intruder Data In Data Link Out-Network
C=>
Coarse Filter
'==0
^
Dead Reckon
^=>
Correct Flight Path
Escape Angle Selection
t=>
AARM
Collision Detection
^=>
Roll Logic Host Data Out
SARM
C=>
Data Link
==>
Figure 1 - Algorithm Architecture
Correct Flight Path Corrects the flight path for intruders if two aircraft in a collision coarse differ in maneuverability. If a tanker and a fighter are on a collision coarse, for instance, then the fighter (more maneuverable than a tanker) maneuvers first to get out of the tanker's flight path. In this process, the tanker's flight path is corrected such that it flies straight along the current velocity vector. Flight correction is also made when the host is uninhabited and intruders are inhabited. Due to system requirement No. 8 discussed above, the inhabited aircraft flies a straight path along the current velocity vector and the uninhabited aircraft gets out of the way. Escape Angle Selection Compares all possible combinations of each intruder's three escape maneuvers (single straight-line flight for out-network intruders) with the host's three escape maneuvers. Escape maneuvers are provided by the projected aircraft position at 0, 3, and 6 seconds away fi-om the current time using the fly out model. Escape maneuvers are compared along a quadratic fit to the three time points, and the minimum separation distance (MSD) and minimum separation time are computed. Computes the reciprocals of each MSD, and computes the sum of the reciprocals for each multiple-intruder case. Selects the preferred host escape maneuver that corresponds to the minimum of the reciprocal sums. Returns the angle of the preferred host escape maneuver as the Escape Angle. AARM Accurate Aircraft Response Model is a detailed model of the host aircraft. Degrees-of-freedom depend on accuracy desired. .Calculates the Escape Maneuver along the Escape Angle provided using aircraft state information. Output is position at 0, 3, and 6 seconds, and velocity at 0 and 6 seconds. Collision Detection Computes the Minimum Safety Separation Distance (MSSD) based on host and intruder AARM data plus a radius of uncertainty in current and fiiture positions. The time at minimum separation fi-om Escape Angle Selection is used as a first approximation. The MSSD and time at MSSD are returned. If MSSD is zero, an escape maneuver command is sent to the host flight controls. If the host and closest mtruder are converging, computes the
time remaining to escape maneuver activation by dividing MSSD by closure speed. Roll Logic Determines how much the escape angle solution can vary fi-om fi-ame to flame by limiting the spread between the computed AARM and two SARMs. The logic compares data fi-om older fi-ames with the last frame to determine where to recenter the escape angle search for the current computational frame. The spread of the SARMs is increased if the solution is changing rapidly or decreased if the solution is stable. However, changes in the spread of the SARMs are limited to small amounts as the time until maneuver execution decreases in order to keep the overall solution stable for all aircraft. SARM Simplified Aircraft Response Model is a simplified 4-degree-of-freedom model of an aircraft with specific available g and available roll rate supplied by the host aircraft. Calculates the Escape Maneuver along the Escape Angles at the spread provided by Roll Logic. The model uses aircraft state information from the host. Output is position at 0, 3, and 6 seconds. Data Link Data Out Sends host data to "In-Network" intruders. 2. Algorithm Description In the previous section, components that constitute the algorithm were described. In this section, the logic within the algorithm that enables air colhsion avoidance is described. 2.1 Optimal Coordinated Escape Angle In order to select'the best'sescape maneuver against multiple intniders, the algorithm first calculates the estimated minimum separation distance between the host and each intruder. This is done by curve fitting both the intruder and host flight path with quadratic curves as a fimction of time, and by finding the analytic solution for the minimum separation distance between two curves (see Figure 2 and 3 for a single intruder case). The quadratic curve fit uses three extracted position points (position at time 0, 3, and 6 seconds) from the host and intruder's 6 second long projected flight paths. The 6 second long flight paths for host and intruder are calculated by 4 degree-of-freedom equations of motion.
Host
6 sec
6 sec
Intruder Figure 2. 6 seconds long flight envelope confining flight paths associated with the 3 possible maneuvers
Host
aht'+bbt+Ch
Intruder Figure 3. Quadratic Curve Fit using 0, 3, 6 seconds position along a flight path.
Since each approximated flight path is represented by a second order polynomial fiinction in time of the form a.t'' + b.t + c (1)
the necessary condition for optimality applied to a norm of relative position d/dt ||(ai-ah)t^+(brbh)t+crCh|| = 0 (2)
yields time instance Tmin at which the minimum separation occurs. The advantage of using a
quadratic curve fit is that Tmin is obtained by solving a cubic polynomial equation whose analytic solution is readily available. With this Tmin so determined, it is an easy exercise to find the minimum separation distance. To find the best combination of escape maneuvers between the host and intruders, the minimum separation distance for each possible combination of escape maneuvers is first calculated. Here, each aircraft has three possible choices of escape maneuver in the optimization process. The first one is roll with a bank angle selected in the previous iteration together with simultaneous pull with the maximum allowable normal acceleration (Nz) command. The second and the third ones are rolls with plus and minus delta bank angle around the first one, respectively, together with the simultaneous pull with the maximum allowable Nz command. Thus, there are nine possible combinations of escape maneuvers between a host and an intruder. In addition, this process is repeated for each intruder. Therefore, three (the maximum number of intruders) times nine (number of possible combination of escape maneuver between a host and an intruder) minimum separation distances are calculated. Among all the possible maneuver combinations, the one which gives rise to the minimum reciprocal sum of minimum separation distances is selected as the optimal combination of the escape maneuvers. The reason for minimizing the reciprocal sum of minimum separation distances as opposed to maximizing the sum of minimum separation distances is to penalize the combinations of the escape maneuvers that provide the worst separation distances. 2,2 Escape Maneuver Activation and Termination The escape maneuver determined via the optimization process is activated when the predicted minimum separation distance becomes smaller than the prescribed minimum safety distance. To increase the accuracy of this prediction, the distance between the host and the intruder is recalculated based on the flight path which is curve fitted by cubic spline using the position and velocity information at 0, 3, and 6 seconds. The escape maneuver is terminated when the separation distance starts increasing, i.e., when the minimum separation has been reached. This satisfies system requirement No. 1 discussed earlier.
3. Simulation Results To evaluate the logic in the algorithm and make assessment of the performance, the results from non-linear simulation in different collision scenarios are discussed. Basic collision scenarios such as head-on, trail, and beam (90 degree aspect angle) are used to test the performance of the algorithm. A PC-based desktop flight simulator called D-Six (a product of Bihrle Applied Research, Inc) is used to perform simulation. D-Six executes the Auto ACAS algorithm using non-linear 6 DOF equations of motion. Head-on Collision Scenario: To evaluate a case of high closure rate, a head-on collision scenario was simulated. Host and an intruder both fly at Mach 0.65, 7 km altitude in a head-on collision course, resulting in 430 m/sec closure rate. Figure 4 shows the resulting optimal combination of escape maneuvers (far left), escape maneuver activation status (middle), and the separation distance between the host and intruder (far right). In the far left plots, bank angle and ACAS roll command are shown with a blue curve and a red curve, respectively. It can be seen that both aircraft turned 1 radian or about 60 degree right to avoid collision.. In the middle plots, the status of escape maneuver activation is shown where 1 means active and 0 means nonactive. The plots show that it took a little over 2 seconds to complete the escape maneuver. In the far right plots, separation distance (blue) is shown against the prescribed minimum safety distance (red). The plots for both the host and intruder show that the minimum separation achieved is 80m against the minimum safety separation of 100m, resulted in 20 % penetration. Trail Collision Scenario: To evaluate a case of low closure rate, a trail collision scenario was simulated. The host flies at Mach 0.7, 7 km altitude, while the intruder flies at Mach 0.65, 7 km altitude, resulted in 30m/sec closure rate. The far left plots in Figure 5 shows that the host and intruder chose opposite direction for roll of about 60 degree. The middle plots show that time to complete the escape maneuver is less than a second, much less than that for the high closure rate case. The far right plots show that the minimum separation achieved is about 100m, almost no penetration into the safety bubble.
jtSfeOi.
--ptll 3.00 2.00 1.00 0.00 ^ -1.00 .2.00 .3.00 0 12 Tkne 3 4 12 rTie 3 4 6 7 2 Tttx 9 4 6
IZ
fj^-(a)
140 120 too
p-Slxl.T.
3.00 200 1.00 O.0O -1.00 .2.00 .3.00 0 1 2
.-Dmselect
/
20 0 3 Tim 4 6 7 2 3 4 0 12 ThK 3 4 6 7
(b) Figure 4. Head-on Collision Scenario - (a) Intruder, (b) Host

>Slx1.7
.*-DftiSstecl 3.0O 2.00 1.00 O.00 .1.00 -2.00 J.0O 140 130 icn
'
i>^'^
I
'
1
1
4 Tkne ."flrwwSnri
4 Time .- TkTiÛn.'>nA -t^ r4iW^^n
(a)
a-sixi.7
3.00 2.x 1.00 000 .100 .2.00 L^.-- Z^
1
T
1111
.11
40 20
(b) Figure 5. Trail Collision Scenario - (a) Intruder, (b) Host 4. Conclusions Simulation results to date indicate that the Automatic Air Collision Avoidance System presented in this paper meets the basic requirements. In particular, it seems promising that the system can provide mid-air collision protection for fighter aircraft during combat training and formation flight without interfering with the pilot's mission. There is more logic developed and implemented in the system than is presented in this paper such as logic to accommodate time delay, data dropout, failure mode, close formation flight and so on. Discussion on these issues is deferred to a future paper due to limited space. The system will be flight tested in July, 2003. Reference [1] Ba Nguyen, Arthur Barfield, Yutaka Ikeda, Cari-Olof Carlsson, "Preliminary Simulation Predictions of Nuisance Criteria for an Automatic Air Collision Avoidance System", The 9* St. Petersburg International Conference on Integrated Navigation Systems, May 2001.
L>
KUCHAR AND DRUMM
The Traffic Alert and Collision Avoidance System

James K. Kuchar and Ann C. Drumm
n The Traffic Alert and Collision Avoidance System (TCAS) has had extraordinary success in reducing the risk of mid-air collisions. Now mandated on all large transport aircraft, TCAS has been in operation for more than a decade and has prevented several catastrophic accidents. TCAS is a unique decision support system in the sense that it has been widely deployed (on more than 25,000 aircraft worldwide) and is continuously exposed to a high-tempo, complex air traffic system. TCAS is the product of carefully balancing and integrating sensor characteristics, tracker and aircraft dynamics, maneuver coordination, operational constraints, and human factors in time-critical situations. Missed or late threat detections can lead to collisions, and false alarms may cause pilots to lose trust in the system and ignore alerts, underscoring the need for a robust system design. Building on prior experience, Lincoln Laboratory recently examined potential improvements to the TCAS algorithms and monitored TCAS activity in the Boston area. Now the Laboratory is pursuing new collision avoidance technologies for unmanned aircraft.
The Trafc Alert and Collision Avoidance System
collision between aircraft is one of the most sudden and catastrophic transportation accidents imaginable. These tragic events are rarely survivablehundreds of people may die as the two aircraft are destroyed. In response to this threat, Lincoln Laboratory has been pursuing surveillance and alerting system technologies to protect aircraft operations both on the ground and in the air. Recent developments in the Runway Status Lights Program, for example, greatly reduce airport-surface collision risk due to runway incursions [1]. In the air, other systems have been developed and are currently in use to prevent midair collisions. This article focuses on the widely fielded, crucial technology called the Traffic Alert and Collision Avoidance System (TCAS). In the context of integrated sensing and decision support, TCAS illustrates the particular challenge of developing effective decision aids for use in emergency situations involving extreme time pressure. Despite the terrifying prospect of a mid-air collision, aviation travel is incredibly safe. A person who flew con-
tinuously on a jet transport aircraft in todays environment could expect to survive more than 11,000 years of travel before becoming the victim of a mid-air collision. This accomplishment has only recently been realized. As shown in Figure 1, the number of hours flown annually by jet transport aircraft has more than quadrupled since 1970, but the rate of mid-air collisions over that period of time has dropped by an order of magnitude. The result is that today we can expect one mid-air collision every 100 million flight hours. Such an exceptional safety level was achieved through advances in air traffic surveillance technology and relentless attention to improving operational procedures. But as the September 2006 mid-air collision between a Boeing 737 and an Embraer Legacy 600 business jet over the Amazon jungle in Brazil demonstrates, maintaining safety is an ever present challenge. This challenge has been eased, but not eliminated, with the development and deployment of TCAS. TCAS is one component of a multi-layered defense against mid-air collisions. The structure of airspace and
VOLUME 16, NUMBER 2, 2007 LINCOLN LABORATORY JOURNAL
277
KUCHAR AND DRUMM
Mid-air collision rate (per million flight hours)

0.10 0.08 0.06
Worldwide annual jet transport flight hours (millions)

40 30 20
highlights the fact that TCAS does not operate in a vacuum and any technological progress needs to mesh into a continuously operating environment.
History
Interest in development of a collision avoidance system dates back to at least 0.04 the mid-1950s, when a mid-air collision 10 occurred between two U.S. air carrier air0.02 craft over the Grand Canyon. For several 0 0 decades thereafter, a variety of approach1970 1975 1980 1985 1990 1995 2000 2004 es to collision avoidance were explored, Year until 1974, when the Federal Aviation FIGURE 1. Worldwide annual flight hours and mid-air collision rate (collision Administration (FAA) narrowed its forate based on a ten-year moving average). Flight hour data are from Boeing. cus to the Beacon Collision Avoidance System (BCAS), a transponder-based operational procedures provide the first, strategic layer airborne system. In 1978, a second mid-air collision of protection. Traffic flows are organized along airways occurred near San Diego between an air carrier and a at segregated altitudes to aid air traffic controllers (ATC) general-aviation aircraft, leading to the expansion of in managing aircraft and predicting potential conflicts the BCAS effort; in 1981, the name was changed to the well before problems arise. Aircraft are normally kept Traffic Alert and Collision Avoidance System (TCAS). three to five miles apart laterally or 1000 ft vertically, A third mid-air collision in 1986 near Cerritos, Calito provide sufficient safety margins. Air traffic control fornia, prompted Congress in 1987 to pass legislation ensures that separation minima are not violated by isrequiring the FAA to implement an airborne collision suing tactical commands (including altitude restrictions avoidance system by the end of 1992. The mandate apand heading change vectors) to the pilots in response to plied to all large (more than 30 passenger seats) turbinenearby traffic. Should these nominal traffic separation powered aircraft in the United States. A subsequent law processes fail, the TCAS system aids pilots in visually extended the original deadline by one year to the end of acquiring potential threats and, if necessary, provides 1993. The first commercial TCAS systems began flying last-minute collision avoidance guidance directly to the in 1990. flight crew. Monitoring and safety assessments led to a series of It is obviously imperative that TCAS alert the flight changes resulting in an international version of TCAS crew early enough that evasive action can be taken. But referred to as Version 7, or the Airborne Collision it is also important that TCAS not alert unnecessarily. Avoidance System (ACAS). Starting in January 2003, Collision avoidance alerts represent high-stress, timethe International Civil Aviation Organization mandated critical interruptions to normal flight operations. These the use of ACAS worldwide for all turbine-powered airinterruptions, in addition to distracting the aircrafts craft with passenger capacity of more than 30 or with crew, may lead to unnecessary maneuvering that dismaximum take-off weight exceeding 15,000 kg. In Janrupts the efficient flow of traffic and may over time also uary 2005, that mandate was extended to cover aircraft cause pilots to distrust the automation. with more than 19 passenger seats or maximum take-off This article outlines some of the challenges in achievweight of more than 5700 kg. Today, more than 25,000 ing this balance. A critical aspect is the need to accuaircraft worldwide are equipped with TCAS. rately model sensors, system dynamics, and human inLincoln Laboratorys involvement in BCAS/TCAS volvement in the collision avoidance process. The wide dates back to 1974, when the FAA tasked the Laboradeployment of TCAS provides a wonderful opportunity tory to develop the surveillance subsystem and MITRE to collect feedback on performance and to understand Corp. to develop the collision avoidance algorithms, the environment in which the system operates. It also also known as the threat logic. Lincoln Laboratorys sur278
LINCOLN LABORATORY JOURNAL VOLUME 16, NUMBER 2, 2007
KUCHAR AND DRUMM
Intruder aircraft
Interrogation
Reply TCAS
Range, bearing, altitude
Surveillance
Trajectory extrapolation Time to collision Threat detection
Maneuver templates
Response selection
Coordination
Traffic display
Resolution advisory display
Threat resolution
Other information sources (ATC, visual acquisition)
Pilot
Flight controls
FIGURE 2. TCAS relies on a combination of surveillance sensors to collect data on the state of intruder aircraft
and a set of algorithms that determine the best maneuver that the pilot should make to avoid a mid-air collision.
veillance activities continued throughout the next three decades; significant development took place during the BCAS-to-TCAS transition and during the design of TCAS Version 7 [2]. Lincoln Laboratory was involved in two additional TCAS activities besides surveillance development. In the mid-1970s the Laboratory, using first a Lincoln Laboratorydeveloped prototype Mode S sensor and then FAA production Mode S sensors, began TCAS-related monitoring of aircraft in the Boston airspace. Early monitoring focused on identifying transmitted data errors that would impact the performance of a collision avoidance system, such as garbled aircraft-reported altitude. Later monitoring focused on assessing the appropriateness of collision avoidance advisories and the impact of these advisories on airspace operation. In the mid-1990s, the Laboratory undertook a third area of activityassessing the threat logic. Because of the growing complexity of the threat logic, Lincoln Laboratory and the FAA William J. Hughes Technical Center began developing simulation and analysis tools to perform specific types of threat-logic assessment. This work was a precursor to the much more complex Lincoln Laboratory simulation tool that we describe later.
How TCAS Works

TCAS processes are organized into several elements, as shown in Figure 2. First, surveillance sensors collect state information about the intruder aircraft (e.g., its relative position and velocity) and pass the information to a set of algorithms to determine whether a collision threat exists. If a threat is identified, a second set of threat-resolution algorithms determines an appropriate response. If the intruder aircraft also has TCAS, the response is coordinated through a data link to ensure that each aircraft maneuvers in a compatible direction. Collision avoidance maneuvers generated and displayed by TCAS are treated as advisories to flight crews, who then take manual control of the aircraft and maneuver accordingly. Pilots are trained to follow TCAS advisories unless doing so would jeopardize safety. The following sections provide more detail on the methods used to perform surveillance, threat detection, and threat resolution.
Surveillance
Surveillance of the air traffic environment is based on air-to-air interrogations broadcast once per second from antennae on the TCAS aircraft using the same frequenVOLUME 16, NUMBER 2, 2007 LINCOLN LABORATORY JOURNAL
279
KUCHAR AND DRUMM
cy (1030 MHz) and waveform as ground-based air traffic control sensors [3]. Transponders on nearby intruder aircraft receive these interrogations and send replies at 1090 MHz. Two types of transponders are currently in use: Mode S transponders, which have a unique 24-bit identifier, or Mode S address, and older Air Traffic Control Radar Beacon System (ATCRBS) transponders, which do not have unique addressing capability. To track ATCRBS intruders, TCAS transmits ATCRBS-only all-call interrogations once per second; all ATCRBS aircraft in a region around the TCAS aircraft reply. In contrast, Mode Sequipped intruders are tracked with a selective interrogation once per second directed at that specific intruder; only that one aircraft replies. Selective interrogation reduces the likelihood of garbled or overlapping replies, and also reduces frequency congestion at 1030/1090 MHz. Replies from most ATCRBS and all Mode S transponders contain the intruders current altitude above sea level. TCAS computes slant range on the basis of the round-trip time of the signal and estimates the bearing to the intruder by using a four-element directional antenna. Alpha-beta and non-linear filters are used to update range, bearing, and altitude estimates as well as to estimate range rate and relative-altitude rate. Mode S transponders also provide additional data-link capabili-
ties. All aircraft with TCAS are equipped with Mode S transponders so that this data link can coordinate collision avoidance maneuvers. One of the most difficult challenges in the development of TCAS is balancing the surveillance requirements of TCAS and air traffic control ground sensorsin particular, managing their shared use of the 1030/1090 MHz frequencies. As the density of TCASequipped aircraft grows, transponders in an airspace are interrogated by more and more TCAS units. As a result, transponders now devote more of their time to responding to TCAS and less of their time responding to ground interrogations. Because of concerns about frequency congestion, TCAS uses interference-limiting algorithms to reduce competition between TCAS and ground sensors. Each second, TCAS determines the number and distribution of other TCAS units in its vicinity. With that information, TCAS can reduce its maximum transmit power (i.e., reduce its surveillance range)limiting the impact on the victim transponders and, in turn, on the ground sensors. National and international requirements in this area are quite strict. Interference limiting is intended to ensure that for any given transponder, no more than 2% of its available time is consumed in communications with all nearby TCAS units. Because TCAS requires a minimum surveillance range to provide adequate collision avoidance protection, however, a limit is imposed on how much the TCAS transmit power can be reduced. As a result, it is possible for a transponder to exceed the 2% utilization figure in highdensity airspace. Transponder utilization due to TCAS has been the focus of worldwide monitoring, and monitoring results continue to motivate the development of innovative TCAS surveillance techniques. FIGURE 3. TCAS is an advisory systemi.e., it tells the pilot what to do to avoid Many such techniques were develcollision but does not take control of the aircraft. Here, a TCAS traffic display oped for Version 7, including using (left) indicates that a threatening intruder aircraft is approximately 14 nautical Mode S interrogation schemes that miles ahead and to the right of the TCAS aircraft; this aircraft is 100 ft above the are different for distant, non-threatTCAS aircraft and is descending. A second, non-threatening intruder aircraft loening intruders than for potential cated about 22 nautical miles ahead is flying level 2500 ft above the TCAS aircraft. threats, and transmitting sequences The Resolution Advisory (RA) display (right) indicates that the aircraft is flying of variable-power ATCRBS interrolevel (i.e., vertical speed of 0). TCAS is instructing the pilot to climb at a rate of gations to reduce garble, or overlap, 1500 to 2000 ft/min, as shown by the green arc. The red arc on the display indicates vertical rates that do not comply with the RA. among concentrations of ATCRBS
280
KUCHAR AND DRUMM
intruders. In addition, standards are nearing completion for TCAS Hybrid Surveillance. This is a new technique that allows TCAS to make use of passive (Automatic Dependent SurveillanceBroadcast, or ADS-B) transmissions, thereby reducing TCAS interrogation rates. Two other issues affect the ability of TCAS to track intruders. First, some older transponders do not report altitude information when interrogated. TCAS can not generate collision avoidance commands against these threats. (Large aircraft, aircraft flying in the vicinity of large airports, and aircraft flying above 10,000 ft are required to be equipped with altitude-reporting transponders.) Second, aircraft without a functioning transponder cannot be detected or tracked by TCAS at all. Some small aircraft, such as gliders or ultralights, may not carry any electronic equipment or transponders. Pilots therefore must take the responsibility to see and avoid such traffic.
as a preparatory cue in case maneuvering becomes required. If the situation worsens, a resolution advisory (RA) warning is issued 15 to 35 seconds before collision (again depending on altitude). The RA includes an aural command such as climb, climb and a graphical display of the target vertical rate for the aircraft. A pilot receiving an RA should disengage the autopilot and manually control the aircraft to achieve the recommended vertical rate. Figure 3 shows both the traffic and RA displays.
Threat Resolution
Once the criteria for issuing an RA have been met, TCASs threat-resolution algorithms determine what maneuver is appropriate to avoid a collision. First, the algorithm decides the vertical sense of the maneuverthat is, whether the aircraft needs to climb or to descend. Second, the system figures the strength of the RAthat is, how rapidly the plane needs to change its altitude. TCAS works only in the vertical direction; it does not select turning maneuvers, because bearing accuracy is generally not sufficient to determine whether a turn to the left or right is appropriate. Figure 4 shows a simplification of the sense-selection process. In general, two maneuver templates are examined: one based on a climb, and one based on a descent. Each template assumes a 5 sec delay before a response begins, followed by a 0.25 g vertical acceleration until reaching a target vertical rate of 1500 ft/min. In the meantime, the intruder aircraft is assumed to continue in a straight line at its current vertical rate. The TCAS algorithm selects the maneuver sense providing the largest separation at the predicted closest point of approach. In the situation shown in Figure 4, TCAS would on the basis of these criteria advise the aircraft to descend. If the intruder is also TCAS equipped, the sense of the RA is coordinated through the Mode S data link to ensure that both aircraft do not select the same vertical sense. Should both aircraft simultaneously select the same sensesay, both select a climb RAthe aircraft with the lower numerical-valued Mode S address has priority and will continue to display its climb RA. The aircraft with the higher Mode S address will then reverse its sense and display a descend RA. Once the sense has been selected, the strength of the RA maneuver is determined by using additional maneuver templates (Figure 5). Each template again assumes a 5 sec delay, followed by a 0.25 g acceleration to reach the target vertical rate. TCAS selects the template that
Threat Detection and Display

TCASs complex threat-detection algorithms begin by classifying intruders into one of four discrete levels [4]. To project an aircrafts position into the future, the system performs a simple linear extrapolation based on the aircrafts estimated current velocity. The algorithm then uses several key metrics to decide whether an intruder is a threat, including the estimated vertical and slantrange separations between aircraft. Another parameter, called tau, represents the time until the closest point of approach between aircraft. A display in the cockpit depicts nearby aircraft, indicating their range, bearing, and relative altitude; an arrow indicates whether the intruder is climbing or descending. Such traffic display information aids the pilot when attempting to visually acquire traffic out the windscreen. Distant, non-threatening aircraft appear as hollow diamond icons. Should the intruder close within certain lateral and vertical limits, the icon changes to a solid diamond, alerting the flight crew that traffic is proximate but is not yet a threat. If a collision is predicted to occur within the next 20 to 48 seconds (depending on altitude), TCAS issues a traffic advisory (TA) in the cockpit. This advisory comes in the form of a spoken message, traffic, traffic. The traffic icon also changes into a solid yellow circle. The TA alerts the pilot to the potential threat so that the pilot can search visually for the intruder and communicate with ATC about the situation. A TA also serves
281
KUCHAR AND DRUMM
Own TCAS
mb cli f lt o su Re
Descend sense selected CPA
FIGURE 4. The TCAS algorithm selects the
Re su lt o fd es ce nd
maneuver sensethat is, whether to climb or descendthat provides the largest separation at the predicted closest point of approach (CPA). In the scenario shown here, the correct maneuver would be to descend.
requires the smallest vertical-rate change that achieves at least a certain minimum separation. In the example shown in Figure 5, the TCAS aircraft is currently descending at a rate of 1000 ft/min when an RA is issued. Five maneuver templates are examined, with each template corresponding to a different target vertical rate. The minimum-strength maneuver that would provide the required vertical separation of at least 400 ft would be to reduce the descent rate to 500 ft/min; the pilot would receive an aural message stating that instruction. Descent rates exceeding 500 ft/min would appear in red on the RA display. Note that in Figure 5 if the intruder were 100 ft higher, then the selected RA would instead be dont descend. If the intruder were another 100 ft higher still, the selected RA would be climb. Due to TCASs 1 Hz update rate and filtering lags, its estimates may lag the actual situation during periods of sudden acceleration. This lag may in turn lead to an inappropriate RA sense or strength. To help alleviate this problem, TCAS refrains from issuing an RA if there are large uncertainties about the intruders track.
TCAS also includes algorithms that monitor the evolution of the encounter and, if necessary, issue a modified RA. The strength of an RA can be increasedfor example, changing from dont descend to climb (target rate of 1500 ft/min) to increase climb (target rate of 2500 ft/min). Under certain conditions, if it becomes clear that the situation is continuing to degrade, TCAS can even reverse the sense of the RA, from climb to descend, or vice versa. Coordination of this reversal with a TCAS-equipped intruder aircraft will also be performed through the Mode S data link. Sense reversal is especially challenging because only a few seconds may remain before collision. Any latencies involved in pilot and aircraft response could result in an out-of-phase response that further reduces separation.
Performance Assessment
The main functions of TCAS are to identify a potential collision threat, communicate the detected threat to the pilot, and assist in the resolution of the threat by recommending an avoidance maneuver. As an alerting
Re su lt of cl im b
Vertical separation (feet) at closest point of approach

FIGURE 5. Once TCAS determines whether
Own TCAS
Result of dont descend
600 500 400 300 200 100 Required minimum separation
Resu Re lt of su limit lt o d f li 500 ft/mi escent n mi td es ce nt 100 0f t/m in
to advise an aircraft to climb or to descend, it calculates the speed at which the plane must maneuver to avoid collision. TCAS selects the template that requires the smallest change in vertical rate that achieves the required separation.
in m ft/ 00 20 nt ce es it d im fl lt o su Re
282
LINCOLN LABORATORY JOURNAL
VOLUME 16, NUMBER 2, 2007
KUCHAR AND DRUMM
system, TCAS operates quietly in the background most of the time. When the algorithms determine that action is needed, TCAS interrupts the flight crew to bring the threat to their attention. This interruption may be vitally important if the pilots are not aware of the threat. In some situations, however, aircraft may operate safely close together; in those cases, the TCAS alerts are more of a nuisance than a help. An example is during an approach to closely spaced parallel runways. In good visibility conditions, pilots can be given the authority to maintain separation from parallel traffic by monitoring nearby aircraft visually through the windscreen. TCAS, however, does not know that visual separation is being used and may issue a TA or an RA, thus introducing a distraction on the flight deck when pilots should be especially focused on performing their approach procedures. TCAS does inhibit issuing RAs when an aircraft is less than 1000 ft above the ground, both to reduce nuisances at low altitude and to help ensure that any TCAS advisories do not conflict with potential terrain hazards. TCAS operates in a complex, dynamic environment. Each decision maker (Air Traffic Control, pilots, TCAS itself) uses different information sources and operates under different constraints and with different goals. TCAS may have more accurate range or altitude information about an intruder than flight crews or ATC do. But TCAS cannot observe all the factors affecting a traffic encounter, such as the location of hazardous weather, terrain, aircraft without transponders, or ATC instructionsa major reason that TCAS is certified to operate only as an advisory system. Pilots are ultimately
responsible for deciding on the correct course of action, weighing TCAS alerts with the other information available to them. TCAS is extremely successful in providing a last-resort safety net, and does not necessarily need to operate perfectly to be effective. Still, it is important to identify situations where TCAS may have difficultyand, if possible, modify the logic to better handle such circumstances.
Lessons from a Disaster

On the night of 1 July 2002, a Boeing B-757 operated by the cargo carrier DHL collided with a Russian Tu-154 passenger jet at 34,940 ft over the small town of berlingen, Germany (Figure 6). The accident destroyed both aircraft and killed all 71 crew members and passengers aboard the two planes. What was especially troubling about this accident is that both aircraft were equipped with TCAS. As with most aviation accidents, a string of events occurred leading up to the collision. First, the nominal separation standards between aircraft were lost through a combination of problems and errors at the air traffic control facility monitoring the aircraft. As a result, the two aircraft were on a collision course much closer together than is normal while cruising at 36,000 ft. Figure 7 schematically summarizes the event. Fortythree seconds before the collision, ATC instructed the Russian aircraft to descend because of the traffic conflict. Before the controller finished his verbal instruction, however, TCAS on the Russian aircraft issued an RA advising the pilot to climb. A coordinated descend
DHL B-757
Russian Tu-154
FIGURE 6. A mid-air collision over berlingen, Germany, in 2002 killed 71 peopleeven though both planes were equipped with
TCAS. (Photograph source: German accident investigation report.)

283
KUCHAR AND DRUMM
ATC instruction to descend Russian Tu-154 Climb, climb
AS TC
y tor ec j a tr
DHL B-757 Descend, descend
Ac tua l tr aje cto ry
FIGURE 7. The berlingen mid-air collision occurred after the Russian pilot decided to heed the air
traffic control instruction to descend rather than the TCAS advisory to climb.
RA was issued on the DHL aircraft at the same time. The DHL pilots followed their RA and began to descend; the Russian flight crew followed the ATC instruction and also descended. Shortly thereafter the RAs on each aircraft were strengthened to increase climb on the Russian aircraft and increase descent on the DHL aircraft. About 35 seconds after the TCAS RAs were issued, the aircraft collided. One of the immediate causes for the accident, as described in the German accident report, was the fact that the Russian flight crew chose to follow the ATC clearance to descend rather than follow the TCAS RA to climb [5]. The Russians choice to maneuver opposite to the RA defeated the coordination logic in TCAS. An advisory system like TCAS cannot prevent an accident if the pilots dont follow the systems advice. The DHL crew, however, did follow the TCAS RA and yet they still collided. The question thus arises: why didnt TCAS reverse the sense of the RAs when the situation continued to degrade? Had it done so, the Russian aircraft would have received a descend RA, which presumably it would have followed, since the crew had already decided to descend in response to the ATC clearance. The DHL aircraft would have received a climb RA, which it likewise would have presumably followed, since its crew had obeyed the original RA. This is not to say that a reversal is always a good idea, however. In many encounters, a reversal would reduce separation and increase the risk of a collision. Because of sensor limitations and filtering lags, it turns out to be quite difficult to trigger reversals when they are needed while avoiding them when they are not needed. A closer examination of the reversal logic revealed several areas in which earlier design assumptions proved inadequate in situations when one aircraft maneuvers
284
opposite to its RA. In order for an RA reversal to be issued, the Version 7 threat logic requires four basic conditions to be satisfied; these conditions are illustrated in Figure 8. First, a reversal will be triggered only by the aircraft with prioritythat is, the aircraft with the lower Mode S address. If the aircraft has a higher Mode S address than the intruder, the RA sense will be reversed only when directed to do so by the priority aircraft through the data link. Second, the maneuver templates projecting the situation into the future need to predict that insufficient separation between aircraft will occur unless a sense reversal is issued. Third, a maneuver template projecting the response to a reversed-sense RA needs to predict adequate separation between aircraft. Fourth, the two aircraft in danger of colliding must be separated by at least 100 ft vertically. (This last condition is intended to prevent reversals from occurring just as aircraft cross in altitude.) A closer look at the berlingen accident, as shown in Figure 9, reveals why TCAS did not issue an RA reversal. Responsibility for triggering the reversal rested with the Russian aircraft, which had a lower Mode S address. The Russian aircraft was operating under an active climb RA. The climb-RA maneuver template predicted adequate separation between aircraft, at least until the final few seconds; therefore, TCAS did not issue an RA reversal. Since the Russian aircraft was not actually following the climb maneuver, of course, the templates predictions were invalid. What is startling, however, is that even if the DHL aircraft had the lower Mode S address (and therefore priority), the planes still probably would have collided. In the hypothetical case in which the DHL aircraft had priority, three of the four conditions required to trigger a reversal, as shown in Figure 8, would have held:
KUCHAR AND DRUMM
4. Must currently be separated by >100 ft
3. Reversed RA is adequate Descend, descend
1. Has priority
2. Current RA is not adequate FIGURE 8. In order for TCAS to reverse its maneuver instructione.g., from descend to climbfour con-
ditions must hold. (1) The reversal can be triggered only by the aircraft with priority. (2) The maneuver template must predict that insufficient separation between aircraft will occur if the present RA is followed. (3) A maneuver template must predict that a reversed RA will result in adequate separation between aircraft. (4) The two aircraft in danger of colliding must be separated by at least 100 ft vertically.
the DHL aircraft would have had priority; the DHL aircrafts descend RA would have shown that a collision was still predicted; and the projection of a reversal-climb RA would have predicted adequate separation. However, both aircraft remained within 100 ft vertically of each other throughout the encounter, and so this fourth criterion for permitting a reversal still would not have been met. To reduce the risk of this type of collision, researchers funded by the European Organization for the Safety of Air Navigation, or Eurocontrol, have proposed a change to the TCAS threat logic. Eurocontrols proposal aims to improve reversal performance in encounters in which both aircraft become involved in a so-called vertical chase, as occurred at berlingen. The proposal includes two major components. First, when using maRussian Tu-154 Climb, climb TCAS trajectory
neuver templates, TCAS would no longer assume that the TCAS aircraft would follow its RA. Instead, TCAS would check the recent vertical motion of the aircraft; if this motion is not compatible with the RA that had been issued, then TCAS would revert to models using the aircrafts current vertical rate instead of its predicted motion in response to the RA. Second, the proposal would eliminate the 100 ft separation requirement, allowing TCAS to reverse sense in vertical-chase situations. The combination of these changes would have produced RA reversals in the berlingen accidentno matter which aircraft had priority. Starting in 2004, the FAA funded Lincoln Laboratory to answer two fundamental questions: how often do RA reversal problems occur in U.S. airspace, and how effective would the European change proposal be?
DHL B-757 Descend, descend
Russian aircraft Descend, descend now
DHL aircraft Climb, climb now
FIGURE 9. The berlingen accident might have been averted if TCAS had issued an RA reversal as shown. Respon-
sibility for triggering the reversal rested with the Russian aircraft, which had priority and which was operating under a climb RA. But until the final few seconds, the climb RA maneuver template predicted adequate separation between aircraft; therefore, TCAS did not issue an RA reversal. Since the Russian aircraft was not actually following the climb maneuver, but rather the air traffic control instruction to descend, the templates predictions were tragically invalid.
285
KUCHAR AND DRUMM
TCAS Monitoring
Following the berlingen accident, researchers set about monitoring the European airspace to estimate how common this type of situation was. A total of ten events, including the berlingen accident, were positively identified in which one aircraft flew opposite to its RA, a reversal did not occur, and either a collision or near miss occurred. Eurocontrol estimated on the basis of the number of flight hours examined that these types of situations occur more than fifty times per year in European skies, and that a mid-air collision in Europe due to this problem might be expected once every four years. In recent years, several countriesthe United States, Britain, France, Germany, and Japanhave been monitoring TCAS to find out if the systems advisories are appropriate and to understand the impact that these TCAS advisories have on airspace operation. All U.S. monitoring/analysis has been performed at Lincoln Laboratory, using an FAA production Mode S sensor located in Lexington, Massachusetts. Following the berlingen accident, the FAA tasked Lincoln Laboratory to begin monitoring for occurrences of the type of situation described above. To accomplish this, we pass sensor data through a series of software tools (Figure 10) to examine the details of TCAS events occurring in the Boston area airspace.
Mode S sensor
Statistics e.g., RA number, type location, version
Procedures for transmitting TCAS RA information to Mode S ground sensors are a part of the basic Mode S and TCAS designs. Whenever TCAS issues an RA to an aircraft within the coverage area of a Mode S sensor, the aircrafts transponder automatically informs this ground sensor that information is available for read-out. On each radar sweep over the duration of the RA, the sensor requests the aircrafts RA Report. This report contains the Mode S address of the TCAS aircraft, the type of RA, and (for TCAS Version 7) an identification of the intruder triggering the RA. Correlation of RA Reports with the TCAS aircraft surveillance data is performed via the aircrafts unique Mode S address, which is present in both the RA reports and the aircraft surveillance data, as well as via time stamps, which are applied by the sensor and show the time of receipt for all communication and surveillance data. All datacommunication and surveillanceare recorded for later playback and analysis. As shown in Figure 10, Lincoln Laboratory performs four types of processing: statistics; pilot response (compliance with RAs); filtering for specific events; and playback. We discuss the first three types in the following sections. The fourth, the playback feature, allows detailed review of the TCAS logic performance. The 5 sec radar surveillance position reports are converted to 1 sec inputs for TCAS. The data can then be played through one of several different versions of the TCAS
Comm data Surveillance data Data recording computer

Pilot response Opposite Proper
Playback Playback radar data through selected TCAS logic version 6.04A, 7, CP112E Filtering Filter for close encounters, reversals
Analysis
Encounter plots
Partial
RA reports
Manually examine plots for safety issues
FIGURE 10. A Lincoln Laboratory facility monitors TCAS operations in the United States to find out if the systems
advisories are appropriate and to understand the impact that these TCAS advisories have on airspace operation.
286
KUCHAR AND DRUMM
logic, allowing comparison of the performance of difreversal will not occur when necessary, such as at berferent TCAS versions, or examination of the effect of lingen. Examination of pilot response is therefore a key a proposed logic change. The playback can be stepped component of the Lincoln Laboratory monitoring. through the encounter in 1 sec increments and allows As an example, Figure 12 shows pilot response to viewing of key TCAS logic parameters at each step. climb RAs during the eight-month period from June Figure 11 shows the location of RAs as recorded by 2005 through January 2006. For a climb RA, TCAS exthe Lincoln Laboratory monitoring program from June pects the pilot to begin to maneuver within 5 sec and 2005 through January 2006. Over this time period, to achieve a 1500 ft/min vertical rate. If the aircraft is monitoring took place for approximately 190 days, and already climbing faster than 1500 ft/min, the RA is roughly 200,000 Mode S flight hours were observed instead maintain climb, and the pilot is expected to within the sensors 60-nautical-mile coverage area. We continue at the current rate. The delay in pilot response observed a total of 1725 RA events, corresponding to was estimated as the time required for the vertical rate to an average of 9 RAs per day, or about one RA every 116 change by at least 400 ft/min. flight hours. Examination of the data shown in Figure 12 indiThis RA rate is typically an order of magnitude larger cates that only 13% of pilot responses met the assumpthan that in European terminal airspace. The higher tion used by TCAS: pilot responses within 5 seconds RA rate in Boston is thought to be due, at least in part, and achieving a 1500 ft/min vertical rate. In 63% of to U.S. air traffic control use of visual-separation prothe cases, the pilots maneuvered in the proper direction cedures when visual meteorological conditions (VMC) but were not as aggressive or prompt as TCAS assumed. prevail, increasing the number of encounters in which Pilots maneuvered in the opposite direction to the RA aircraft pass each other safely even though they are within 24% of the cases. Some of these opposite responses in TCAS RA thresholds. In particular, Figure 11 shows a are believed to be due to visual acquisition of the innumber of RAs along the parallel approaches to runways truder aircraft and the pilots decision that following the 4L and 4R at Bostons Logan Airport. In VMC, aircraft RA was not necessary. Although such a decision may may be vectored onto final approaches to these two runways as the pilots accept responsibility of maintaining safe separation from the parallel traffic. In some cases, TCAS may still alert because of the close proximity of aircraft. Many of the RAs elsewhere in the airspace are due to the large density of small general-aviation aircraft in the United States that may operate more closely to other air traffic in VMC than is typical in Europe. When air traffic control allows use of visual-separation procedures, pilots may ignore an RA, or move contrary to the RA, because the intruder aircraft is in FIGURE 11. The location of RA events as recorded by the Lincoln Laboratory monitorsight. Pilot non-compliance to ing program for the time period June 2005 through January 2006. Over this time perian RA may not necessarily comod, monitoring took place for approximately 190 days, and roughly 200,000 Mode S flight promise safety in a particular enhours were observed within the sensors 60 nmi coverage area. We observed a total of counter. It can, however, lead to 1725 RA events, corresponding to an average of 9 RAs per day, or approximately one RA every 116 flight hours. a degrading situation in which a
287
KUCHAR AND DRUMM
5000 4000
Vertical rate during RA (ft/min)
3000 2000 1000 0 1000 2000 3000 4000 5000 5000 4000 3000 2000 1000 0 1000 2000 3000 4000 5000
Pilot response delay

< 5 sec 510 sec > 10 sec
Vertical rate prior to RA (ft/min)

FIGURE 12. Pilot compliance with climb RAs. Each circle corresponds to one
RA event for a TCAS aircraft. The green-shaded region indicates responses that achieved the intended vertical rate; the yellow segment indicates responses that were in the correct direction but did not achieve the intended vertical rate; the red shading indicates responses that were in the wrong direction.
be reasonable in certain cases, the fact remains that maneuvering opposite to the TCAS RA invites exactly the kind of vertical chase that happened at berlingen. The European change proposal would provide an additional safety net should such a vertical chase occur and aircraft continue to move on a collision course.
Detecting Reversal Problems

A key part of Lincoln Laboratory monitoring efforts since mid-2004 has been to find instances of RA reversal problems in the Boston airspace. Automated tools extract encounters in which a TCAS RA occurs and separation between aircraft is small. These encounters are then examined to find those in which the two aircraft moved in the same vertical direction after the start of the RA event. The Lincoln Laboratory monitoring program detected two RA reversal problem events. Figure 13 diagrams one of these encounters, in which a TCAS-equipped aircraft came within 0.3 nmi horizontally and 100 ft vertically of another aircraft that did not have TCAS. As the aircraft neared each other, TCAS issued a descend RA, which the pilot followed. At about the same time,
288
however, the non-TCAS aircraft also began a descent, presumably on the basis of visual identification of the TCAS aircraft and an attempt to avoid a collision. TCAS then issued an increase descent RA, which the pilot again followed. At the same time, the non-TCAS aircraft also increased its descent rate. TCAS maintained the descend sense even after the non-TCAS aircraft had dropped below the TCAS aircraft. TCAS did eventually issue a reversal, instructing the aircraft to climb. However, this reversal occurred essentially as the two aircraft passed each other and was too late to be of benefit. This encounter, which took place in April 2005, is of the same general type as that in the berlingen accident in that an RA reversal should have been issued earlier. The findings of the Lincoln monitoring program indicate that RA reversal problems occur in U.S. airspace at a rate comparable to that in Europe.
Assessing Safety
Safety analysis of TCAS is based in part on a comprehensive, statistically valid set of data describing TCAS performance across a wide range of encounter situations. Specific problem situations also need to be identified
KUCHAR AND DRUMM
Intruder (without TCAS)

2500
TCAS aircraft
2400
D D
Altitude (ft)
2300
Descend RA
D
2200
Reversal: climb RA
C
2100
2000 20 15
Descend RA
10 5 0 5
Climb RA
10 15 20 25
Time from closest point of approach (sec)

FIGURE 13. In this encounter, a TCAS-equipped aircraft came within 0.3 nmi horizontally and
100 ft vertically of another aircraft that did not have TCAS. The RA reversal came too late.
and judged as to their criticality and likelihood. Extensive flight testing is required to support modeling sensor performance, automation, human interaction with TCAS advisories, and flight characteristics. However, flight tests alone cannot provide enough data to make a complete system assessment. Thus a combination of modeling based on flight experience and fast-time simulation of many encounters is needed. A key performance metric is the reduction in collision risk achieved by equipping with TCAS. This risk is expressed in terms of Near Mid-Air Collision (NMAC) events, defined to occur when separation between two aircraft is less than 100 ft vertically and 500 ft horizontally. The probability of Near Mid-Air Collision is P (NMAC). The ratio of P (NMAC) when TCAS is used to P (NMAC) without TCAS is commonly referred to as the risk ratio. Changes in TCAS algorithms, such as those included in the European change proposal, can be evaluated by examining their effect on risk ratio. Assessment of safety requires more than simply the application of a single analytical model. Several tools must be brought to bear, each focusing on a different aspect of the overall system. In particular, the collision risk problem can be partitioned into two regimes: an outer loop that encompasses system failures and events that lead up to a critical close-encounter event, and an inner
loop that covers the second-by-second details of an encounter in a dynamic analysis, given the conditions that were defined in the outer-loop regime (Figure 14). A fault tree is typically used to model the outer-loop system failures or events that in turn define the environment for a fast-time inner-loop simulation of a close encounter. For example, the probability that a transponder will fail to provide altitude information can be estimated in the fault tree, and P (NMAC) for that type of encounter can be computed in a detailed fast-time simulation. Results are then combined in the fault tree with corresponding performance data and probabilities for other conditions, leading to a global estimate of system safety. Researchers can perform sensitivity studies by modifying event probabilities in the fault tree and observe their impact on overall risk, without requiring new fast-time simulations. The outer-loop regime defines what conditions apply to the set of close encounters that are dynamically simulated in the inner loop. Outer-loop conditions include airspace environment (e.g., low altitude, high altitude, U.S. airspace, European airspace); encounter characteristics (e.g., speeds, geometry of encounter), intruder aircraft equipage (e.g., transponder-equipped, TCASequipped); system component failures; pilot response to TCAS RAs (e.g., failure to respond, normal response);
289
KUCHAR AND DRUMM
environmental conditions; and finally, Outer loop Encounter ATC involvement in resolving the model close encounter. The outer-loop modeling requires System Encounter a valid model of the types of close failures Inner loop conditions Results encounters that may occur. This soFast-time called encounter model specifies a simulation Pilot-response number of variables that are selected model randomly in every fast-time simulaInner-loop variations in tion run. Key variables include the Aircraft encounter geometry equipage and sensor noise geometry of the encounter, aircraft speeds, and vertical accelerations. The Outer-loop encounter modeling process begins variations in encounter conditions by collecting thousands of hours of air traffic radar data and using a set of filters to extract from these data any FIGURE 14. The aircraft mid-air collision risk problem has two parts: an outer loop close encounters between aircraft. The that encompasses system failures and events that lead up to a close encounter, characteristics of each close encoun- and an inner loop that covers the second-by-second details of an encounter in a ter are then compiled into a statistical dynamic analysis. distribution describing the likelihoods that various conditions are present. When generating trajectories to determine miss distances and to compute encounter scenarios, separate software randomly selects P (NMAC). To reduce computation time, batch simuparameter values from these distributions, computes the lation runs are performed with Lincoln Laboratorys initial conditions for the simulation, and stores the reLLGrid parallel computing facility [6]. LLGrid enables sults in an input file. simulation of one million encounter situations in apThe inner-loop dynamic simulation takes the status proximately 3.5 hours, allowing enough flexibility to of system components and the environment and cominteractively investigate changes to TCAS or other colputes P (NMAC) over a representative range of encounlision avoidance systems. ter situations. Each encounter scenario is executed once The simulation includes flight-certified TCAS code without TCAS and once with TCAS. Additional runs obtained from a TCAS vendor. The logic in the simulamay be performed to compare the performance of diftion is thus identical to that in actual aircraft, providferent TCAS algorithms. These runs, using identical iniing high fidelity and an ability to replicate the full range tial conditions, facilitate making direct estimates of the of logic behavior. Information from the TCAS logic is safety provided by TCAS. passed to a pilot-response model (to respond to RAs), to a visual-acquisition model (triggering improved pilot Lincoln Laboratorys Safety Assessment Tool visual-search efficiency) and to the other aircrafts TCAS Lincoln Laboratory recently designed and implemented unit (if equipped) to handle maneuver coordination. (using The MathWorks MATLAB, Simulink, and Real A visual-acquisition model estimates the probability Time Workshop software packages) a fast-time Monte that a pilot will see the other aircraft through the windCarlo simulation capability called the Collision Avoidscreen. This model relies on a technique developed for ance System Safety Assessment Tool. This system takes accident investigations, safety analyses, and regulatory encounter model data as an input and simulates threeprocesses [7]. The models basis is that visual acquisition dimensional aircraft motion. The simulation includes is limited by target search time over a given volume of several integrated sub-models, as shown in Figure 15. space. In the model, the probability of visually acquiring These sub-models include TCAS logic, a visual-acquisia threat during one time step is given by tion model, a pilot-response model, and a vehicle dyA = 2 , namics model. A sensor noise model is also included. r A performance analysis module examines the aircraft
290
KUCHAR AND DRUMM
where b is a constant, A is the visual area presented by the target, and r is the range to the target. If the aircraft are on a collision course, r decreases with time, so the acquisition probability increases smoothly until the point of closest approach. The value of A may vary as an aircraft changes orientation. The value of depends on visibility, contrast, the number of pilots searching, and whether those pilots have been cued by an ATC or TCAS traffic advisory. Values for have been validated in flight experiments. The visual-acquisition model estimates the probability of a pilot visually detecting another aircraft by a certain time and thus helps identify encounters that might be avoided by visual acquisition. Aircraft motion normally follows a scripted set of maneuvers as specified by the encounter model. These maneuvers can include vertical or lateral acceleration such as a level-off or turn, plus changes in airspeed. If a TCAS RA is issued, the motion transitions to a new set of control behaviors as defined by a pilot-response model. We use different models to explore a variety of possible pilot behaviors, including pilots who respond exactly as TCAS assumes as well as pilots who respond slowly, move more aggressively than expected, maneuver in the opposite direction as the RA calls for, or make no maneuver at all.
European ChangeFor the Better?

Between 2004 and 2006, the FAA and Eurocontrol conducted an international study to assess the performance of the European change proposal [8]. Figure 16 shows data from Lincoln Laboratory simulations, as
Airspace encounter model Radar data
part of this study, that demonstrate how the European change proposal would affect the measured vertical miss distance between aircraft in encounters similar to berlingenthat is, when both aircraft are equipped with TCAS but when one aircraft does not follow its RA. Clearly, the European proposal would in a vast majority of cases affected by the proposal92%result in an increase in vertical separation. A full 22% of the affected cases are considered saves; that is, a near mid-air collision would have occurred with the current version of TCAS but would not occur if the proposed change were to be implemented. In only 2% of the affected cases would the situation be reversed, with the proposed change resulting in a near miss, while the current TCAS would not. Lincoln Laboratory simulations were also run with an encounter model representing European airspace. In encounters involving two TCAS-equipped aircraft in which both pilots respond appropriately to their RAs, TCAS provides a risk ratio of approximately 0.02. That is, if pilots obey the RA, the use of TCAS reduces the risk of mid-air collision by about a factor of 50. If instead one pilot does not respond to the RA, the risk ratio rises by an order of magnitude, to 0.23. Figure 17 summarizes the overall impact of TCAS according to the estimated number of years between mid-air collisions over Europe. These estimates, which were based on the Lincoln Laboratory simulation studies, consider two factors: the likelihood with which pilots follow their RAs, and the type of TCAS logic being used. With no TCAS at all, one mid-air collision
Simulation
TCAS
Pilotresponse model
Aircraft dynamic model
Metrics Vertical separation Risk with TCAS Risk without TCAS
Filter
Encounter model distributions Random samples
TCAS
Pilotresponse model
Aircraft dynamic model
Years between collisions
FIGURE 15. A set of software modules implemented at Lincoln Laboratory simulates three-dimensional aircraft mo-
tion based on data from encounter models. The fast-time simulator runs on the LLGrid parallel computing facility.
291
KUCHAR AND DRUMM
complex collision avoidance system at a desired level of performance even as the types of aircraft and procedures for 1200 air traffic management change. Triggering an alert requires the auto1000 mation to determine on the basis of its internal models that intervention is re800 quired. This decision may conflict with the human operators mental model. 600 Such tension is good when the human needs to be alerted to a problem that 400 requires attention, but can be undesirable if the human has access to information that disagrees with the need for 200 an alert or for action. Along these lines, areas of potential improvement in col0 lision avoidance system design include 0 200 400 600 800 1000 1200 1400 enhanced surveillance information and Separation with current TCAS (ft) more sophisticated modeling and deciFIGURE 16. Each data point represents one encounter that was simulated twice: sion-making algorithms. once with the existing TCAS logic and once assuming adoption of the European TCAS can be only as good as the change proposal. Encounters in which the proposed change does not affect the data that it works fromparticularly vertical miss distance lie on the diagonal line. Points above the line represent encounters where the proposed change would increase separation; points below the estimates of an intruder aircrafts the line indicate encounters where the change would reduce separation. position, velocity, and acceleration. Because position measurements are could be expected over Europe approximately every updated only once a second, rapid changes in aircraft three years. With deployment of TCAS Version 7 (blue trajectory cannot be detected or tracked immediately. curve), the years between collisions depend heavily Acceleration information obtained directly from the on how often crews conform to RAs. Safety increases flight management system on the TCAS aircraft could sharply as more pilots follow their RAs. Introducing the European change proposal should improve safety even 20 further (green curve). Clearly, aviation safety can be best enhanced through a combination of measures: upgrading the TCAS algorithms while also improving pilot 15 Current TCAS training and procedures to increase RA conformance. The FAA is considering the economic and safety tradeoffs involved in mandating an upgrade to all TCAS 10 Change proposal units worldwide. Because of this analysis, it is possible that within the next year the FAA will issue directives requiring aircraft operators to implement new algorithms 5 incorporating the European change proposal.
1400
Separation with change proposal (ft)
Years between collisions
Better Sensing and Algorithms

As the European change proposal illustrates, improvements are still possible and being investigatedas are completely new technologies for collision avoidance. A key issue is how to best apply sensors, displays, automation, procedures, and controls to enable operating a
292
No TCAS 0 0 0.2 0.4 0.6 0.8 1.0
Rate of conformance to TCAS

FIGURE 17. For a given rate of pilot conformance to TCAS
instruction, the risk of collision would be lower under the proposed European change than with the current system.
KUCHAR AND DRUMM
be used to provide lead information to the dynamic models that TCAS uses to determine whether an alert is needed or whether an RA is being followed. Access to additional intent information from the intruder aircraft, such as the altitude at which it intends to level off, would also greatly improve TCASs ability to determine what type of RA should be issued, or if one should be issued at all. Improvements are also possible for the trajectory model templates on which TCAS bases its predictions. TCAS uses a two-stage decision-making process. First, the system projects current traffic into the future on the assumption that both aircraft continue in straight lines with no acceleration. The rationale behind using this nominal template is that alerts are issued only when they are necessary to avoid a collision. The accuracy of trajectory prediction generally degrades into the future, so some cutoff or maximum look-ahead time is typically required to avoid nuisance alerts. That uncertainty precludes TCAS from making accurate collision avoidance decisions more than 30 to 40 seconds into the future. If the first-stage modeling predicts insufficient separation between aircraft, TCAS then selects one of several avoidance templates and instructs the pilot to make this recommended maneuver. The safety of the two-stage system is ensured by tuning the alerting parameters so that, on balance, first-stage alerts are issued early enough that ample avoidance trajectories remain. Encounter models and Monte Carlo simulations are integral to this evaluation and tuning process. A single integrated decision-making stage could further improve performance. A system that simultaneously examined the nominal and avoidance templates would issue alerts only when they were necessary and likely to be successful. Alert timing could then be tightened to ensure that alerts would be issued only when both templates showed that it was appropriate to do so.
TCAS for Unmanned Aircraft

The increasing demand for unmanned aerial vehicles (UAV) adds a new wrinkle to the task of collision avoidance. UAVs are being developed to serve a variety of roles, including border patrol, sea-lane monitoring, vehicle tracking, environmental observation, cargo delivery, and military surveillance. Many of these missions require UAVs to coexist with civilian aircraft. Like piloted aircraft, UAVs are required to see and avoid other air traffic.
TCAS was not designed to be a sole means for the see-and-avoid directive, nor was it originally intended for use on UAVs. TCAS presumes the existence of conventional separation processes, including air traffic control and visual separation, and the surveillance, display, and algorithm designs of TCAS were developed and validated for aircraft with onboard pilots. Four issues in particular are of special concern. First, TCAS can detect only transponder-equipped aircraft, which means that UAVs would be blind to small aircraft such as gliders, hot-air balloons, or ultralights they might encounter. Second, maneuvering is not permitted on the basis of the TCAS traffic display or TAs because of limited bearing accuracy and vertical rate information. Third, remote-control latencies in reacting to RAs may result in maneuvers that induce collisions. Finally, it may be difficult for a UAVs pilot to detect anomalous situations such as altitude-reporting errors or intruders that are maneuvering in a manner incompatible with the RA. Initial studies by Lincoln Laboratory have provided estimates on how command and control latency affects TCAS performance on a particular UAV, the U.S. Air Forces RQ-4A Global Hawk [9, 10]. Figure 18 shows how risk ratio increases with increasing latency when a pilot is responding to a TCAS RA on Global Hawk. Data are shown for five altitude bands representing different mixes of aircraft and encounter characteristics. To arrive at these results, we adjusted the encounter models to account for the unique flight characteristics of Global Hawk, but no system failures were considered. Nor did we consider issues related to the inability of Global Hawk to visually acquire threats. As Figure 18 shows, risk ratios increase sharply if latencies exceed 10 to 15 sec, especially at low altitudes. Performance at lower altitudes is a serious concern because that is where TCAS may be needed the most: there is a higher potential for encountering aircraft not being managed by ATC. Large latencies reduce TCAS performance to the point where it adds no value. And if latencies grow too large, TCAS may actually induce more risk than exists without TCAS. In some cases, delayed RA reversals can be generated that are out of phase with aircraft motion, potentially causing a collision. We are pursuing further study into the performance of TCAS on UAVs to address these concerns. We are also investigating the potential for autonomous response to RAs, which might significantly improve safety. Another important factor is the potential for additional sensor
293
KUCHAR AND DRUMM
tion of more than a hundred years of experience with manned air2.5 craft, such a foundation does not yet exist for unmanned aircraft. 2.0 The prospect of an aircraft autonomously maneuvering to avoid 1.5 500013,500 ft a passenger jet carrying hundreds No TCAS on Global Hawk of people requires intense scrutiny. 1.0 13,50021,500 ft Collision avoidance systems will be a focal point of this concern. 0.5 21,50029,500 ft At the same time, new tech29,50041,500 ft 0 nologies are arriving that promise 5 0 15 20 10 to improve the ease with which RA response latency (sec) collisions can be detected and FIGURE 18. Impact of response latency on TCAS performance in an unmanned aerial avoided. These technologies make vehicle (UAV) called Global Hawk. As response latency increases, so does a value use not only of enhanced data link called risk ratio, which is defined as the probability of a collision with TCAS divided capabilities to provide information by the probability of a collision without TCAS. The higher the risk ratio, the less effecon the intentions of an intruding tive is TCAS. aircraft, but also additional modalities such as visual, infrared, technologies to aid in detecting and avoiding air trafor radar sensors. The optical environment, including fic that is not transponder equipped. Systems based on lighting, haze, clouds, and background clutter, repreelectro-optical sensors or radar may fill this role, though sents aspects that did not need to be considered with how these systems and their collision avoidance algoradio-frequency-based TCAS but that are now critically rithms interact in the existing air traffic environment important. Fusing TCAS information with these other needs to be investigated. Such study would require the surveillance sources represents an opportunity and a development of new encounter models for aircraft that challenge. Extensive flight testing, modeling, and simudo not have transponders as well as environmental modlation need to be conducted to fully explore design isels needed to evaluate electro-optical sensor performance sues with these new technologies. across a range of conditions. The real challenge lies in integrating new collision avoidance technologies with the existing systems and Future Challenges procedures. The berlingen accident demonstrated the TCAS represents a clear success story in aviation safety. catastrophic outcome that can result from dissonance Its successful design was achieved through detailed conbetween two different decision makers in a time-critisideration of sensor characteristics and the coupled dycal situation: namely, an air traffic controllers decision namic interactions among pilots, air traffic controllers, to request a descent and TCASs Resolution Advisory and aircraft. The result is a fine balance that provides to climb. While this specific problem is being solved by sufficient time to take action and that minimizes alert improving pilot training to comply with RAs and refinrates. As the berlingen accident shows, however, safety ing the TCAS algorithms, related problems are likely cannot be taken for granted, and areas of improvement to surface as unmanned aircraft and enhanced collision will always exist in systems that rely on integrating huavoidance technologies mix. Ensuring compatible opmans and automation for information processing and eration also extends well beyond TCAS or aviation to decision making. many integrated sensing and decision support system The future still holds several vital issues to be overapplications. Lincoln Laboratorys experience with sencome. Remotely piloted or autonomous unmanned airsor fusion, decision support, and systems prototyping craft introduce a novel element into an already complex will greatly facilitate the path forward in these areas, and environment. Although the high degree of confidence we are continually exploring new areas of complex sysin todays aviation system has been built on a foundatem design.
3.0
< 5000 ft
Normalized risk ratio
294
KUCHAR AND DRUMM
Acknowledgments
TCASs development spans several decades and is the product of contributions from many individuals. We would particularly like to acknowledge Thierry Arino, Ken Carpenter, Kathy Ciaramella, Roger Elstun, Stephen George, John Law, Paul Fontaine, Stuart Searight, Andrew Zeitlin, Thomas Billingsley, Barbara Chludzinski, Garrett Harris, Val Heinz, Katherine Sinclair, and David Spencer. This work was funded by the Federal Aviation Administration and by the U.S. Air Force.
REFERENCES
1. J.R. Eggert, B.R. Howes, M.P. Kuffner, H. Wilhelmsen, and D.J. Bernays, Operational Evaluation of Runway Status Lights, Linc. Lab. J. 16 (1), 2006, pp. 123146. 2. W.H. Harman, TCAS: A System for Preventing Midair Collisions, Linc. Lab. J. 2 (3), 1989, pp. 437458. 3. V.A. Orlando, The Mode S Beacon Radar System, Linc. Lab. J. 2 (3), 1989, pp. 345362. 4. Minimum Operational Performance Standards for Traffic Alert and Collision Avoidance System II (TCAS II) Airborne Equipment, RTCA/DO-185A, Washington, D.C., 16 Dec. 1997. 5. Investigation Report AX001-1-2/02, Bundesstelle fr Flugunfalluntersuchung (BFU, German Federal Bureau of Aircraft Accident Investigation), Braunschweig, Germany, May 2004. 6. N.T. Bliss, R. Bond, J. Kepner, H. Kim, and A. Reuther, Interactive Grid Computing at Lincoln Laboratory, Linc. Lab. J. 16 (1), 2006, pp. 165216. 7. J.W. Andrews, Air-to-Air Visual Acquisition Handbook, ATC-151, Lincoln Laboratory, Lexington, Mass., 27 Nov. 1991, DTIC #ADA-243807. 8. Safety Analysis of Proposed Changes to TCAS RA Reversal Logic, RTCA DO-298, Washington, D.C., 8 Nov. 2005. 9. J. Kuchar, Update on the Analysis of ACAS Performance on Global Hawk, Aeronautical Surveillance Panel, International Civil Aviation Organization (ICAO), SCRSP WG A/WP A10-04, Montreal, 15 May 2006. 10. T.B. Billingsley, Safety Analysis of TCAS on Global Hawk Using Airspace Encounter Models, S.M. Thesis, MIT, Cambridge, Mass., June 2006.
295
KUCHAR AND DRUMM
james k. kuchar is an assistant leader in the Surveillance Systems group, where he oversees surveillance and safety programs spanning civil and homeland defense mission areas. From 1995 to 2003, he was an MIT faculty member in the Department of Aeronautics and Astronautics. While at MIT, he focused on decision support system research with topics including TCAS, parallel runway approaches, terrain avoidance displays, and automobile warning technologies. He has S.B, S.M., and Ph.D. degrees in aeronautics and astronautics from MIT, and he has authored more than sixty journal articles and refereed conference papers. He has been a private pilot since 1995, and believes he has not yet triggered a TCAS alert.
ann c. Drumm is a staff member in the Surveillance Systems group. She received a B.A. degree in mathematics and computer science from Miami University (Ohio) in 1969. She joined Lincoln Laboratory in 1969, and became a member of the Air Traffic Control division when it was formed in 1970. She was part of the team that developed the software for the FAAs first prototype Mode S sensor. She developed the protocols for TCAS air-to-air coordination, a process used worldwide to ensure that TCAS aircraft encountering one another select compatible maneuvers. She has served as technical lead for the Lincoln Laboratory FAA programs in Mode S Data Link, Automatic Dependent Surveillance-Broadcast (ADS-B), and TCAS. Her current work involves airspace monitoring to assess the performance of TCAS in the United States and investigation of TCAS on unmanned aerial vehicles.
296
innovation for a standard intermittent windshield wiper is situation which often arises during a car race. The driver of a race e concept of a smart wiper that automatically turns itself car has many continuously onerous responsibilities that require on or off and adjusts its wiping cycle according to the intensity critical decisions and precise execution to win arace. Many races of rain. This article describes the development of a smart auto- have been won by cars that finished only fractions of seconds matic windshield wiper system using a fuzzy logic-based con- ahead of others. Concentration becomes so crucial and time so ductive rain sensor scheme. Fuzzy logic is used to compensate precious that a race car driver does not wish to spend any effort nonlinearities in the sensor and implement intuitive reasoning on menial yet important tasks such as controlling the windshield for automatic operation of the wiper. An M68HCll-based mi- wiper [l-51 in case of a rainy condition during a race. Consecrocontroller was developed to implement and test the smart quently, most drivers simply leave the wipers turned on through wiper concept for a car. The proposed approach described in this the race, whether it is raining heavily or drizzling. In such situations, a driver might aparticle can-be extended io preciate having a smart autoother applications related to rain detection and system ac- Cheok and Kobayashi are with the Department of Electrical and matic intermittent wiper that Systems Engineering, School of Engineering and Computer Science, turns itself on or off and adjusts tuation. Oakland University, Rochester, MI 48309-4401, telephone: 810- its speed and delay interval acIntroduction 370-2232,fax: 810-370-4261, email: cheok@ oakland.edu. Scaccia cording to rain conditions durTo appreciate the motiva- and Scaccia are with Technoresearch Inc., I766 Greenwich, Troy, ing a race. The research tion of this project in this arti- MI 48098, telephone: 810-370-0042, fax: 810-370-0042, email: described herein was initiated cle, consider the following scaccia@oakland.edu. and sponsored by the Alfa Ro-
28
0272- 1708/96/$05.000 1996IEEE
IEEE Control Systems
Physical process Rain or Drizzle
!
I
Sensors & Actuators
!
I
Electronic Conditioning
!
I
Computer Controller
+!7
Windshield
I
+zk
Washer
Motor
Fig. 1. Schematics of a smart wiper with fuzzy logic controller
meo Race Team to investigate the development of such a wiper, Aside from racing, a smart wiper could be also beneficial for commercial automobiles. An automatic wiper would eliminate momentary distractions ofhaving to adjust the wiper control. The advantage is obvious since such smart automation would allow a driver to concentrate on driving or to race in varying rainy conditions. In this article, we describe the concept and design considerations for a smart wiper, and discuss the potential of multi-sensor fusion for detecting rain conditions (second section). Next, we focus on how a fuzzy logic controller [6-101 can be used with a conductive rain sensor to provide a smart wiper control scheme. We then present an outline of the approach we used to develop and implement the real-time stand-alone fuzzy logic hardware (Motorola microcontroller [ l 11) and software (FIDE [12]). Finally, we demonstrate the test results of the fuzzy logic controller for the smart wiper.
Concept for a Smart Wiper

The general concept of a smart wiper, including use of multisensing devices, is briefly discussed as part of the preliminary investigation carried out for this project. Functional Configuration of a Smart Wiper A general schematic block diagram for a smart wiper control system is shown in Fig. 1. The innovative components in the configuration are the rain sensors and fuzzy logic decision scheme for the wiper motor hardware. Desirable Features for a Smart Wiper The desirable automatic features of a smart wiper are as follows: Turn itself on as needed when it drizzles, rains, or snows. Adjust its intermittent delay interval according to the condition of drizzle, rain, or snow. Apply washer fluid when needed. Turn itself off when not needed. As a result of these requirements, a smart wiper will need new components to supplement its standard components.
New Components of a Smart Wiper A selection of potential new components, including user interface sensor, decision and actuation hardware, is discussed below: Control Panel. The control panel of a smart wiper will have a three-position switch for selecting Automatic I Manual I Off modes. The automatic mode will turn on the wiper, decide the intermittent interval, and select a Low or High speed for the wiper motor. It will also activate the windshield washer pump as needed. Sensors. Different sensor methods, including the use of reflective, audio, and conductive sensors, were considered for detecting rain conditions and quantifying its intensity. They are briefly described below: a. Reflectivity sensor. A reflectivity sensor is a photo diodephoto transistor unit where the transistor is used to detect the amount of reflected light emitted by the LED. Optimum location: The photo transmitter-receiver unit can be mounted on the windshield from within the automobile to monitor changes in the reflectivity of the windshield as rain droplets fall on it. The reflective sensor, for instance, requires a function generatodtransmitter circuit to trigger the photo diode and a photo transistor receiver circuit for capturing the reflected light intensity. b. Audio sensor. An electret microphone could be used to listen to the rain drops as they fall on a metallic panel such as the roof of a vehicle. However, there is considerable ambient noise encountered in this approach, and cancellation of ambient noise will be required. A frequency selective filter for the microphone will be needed to remove ambient audio noise and pass through only useful splattering sound of rain as useful information. c. Conductivity sensor. A conductive sensor is an approximately one-inch-square plate coated with a simple pair of conductors whose terminal resistance changes with ambient dampness or wetness. This is actually a very simple idea that works very well. As will be presented in more detail in this article, the signal from the conductive sensor needs to be electronically decomposed into DC and AC components via low-pass and high-pass filters. Note that it might be possible to evaporate the rain as well as melt icelsnow, if high current was designed to sufficiently heat the conductive sensor to an appropriate high temperature. Sensor Fusion. These three sensors emulate the basic inputs to visual, auditory, and tactile sensory capab driver. The information from these sensors can be fused together to complement each other and deduce the conditions of drizzle, rain, or snow. Decision can then be made to control the wiper and washer accordingly. Control Decision. As demonstrated in this article, fuzzy logic [6-lo] is a good approach for implementing a knowledge-based logic to control the wiper system. It is also useful for handling system uncertainties and nonlinearities, without the use of a sophisticated mathematical model. Sensor fusion and control decisions can be carried out using intuitive reasoning via fuzzy logic. Another powerful tool is the method of neural networks, which is not considered here but will be investigated in the future. Microcontroller HardwardSoftware. The real-time processing and fusion require the hardware and software of a microcontroller. In this project, we found that the combination of an M68HC11EB microcontroller board hardware and a PC-based
December 1996
29
FIDE (Fuzzy Inference Development Environment) software provide a convenient method for converting the fuzzy logic concept into a stand-alone application. Standard Components of a Wiper The existing components of a O"f C12 0 . Q l u , C.IL"I. 50k standard wiper are briefly outlined beRIQ low: Output Signal Conditioning. A microprocessor is used to control the external analog power electronic circuitry that drive the windshield wiper motor and washer fluid motor. Wiper Motor and Mechanism. There are numerous types of wiper motor mechanisms made by various manufacturers. Almost all wiper assemblies are equipped with special cams for indicating the home position for the wiper mechanisms. Motor and Pump' Fig. 2 Electronics hardware components for smart wipel: Again, - assem- a variety of motorlpump blies can be found from various manufacturers. human observation-action-decision process in operating a wiper. As mentioned in the Introduction, the scope of the project was Other Issues to develop an automatic windshield wiper for a race car. Many other important issues to be considered include: Mounting and Packaging. One of the main issues in develHardware Configuration and Implementation opment of a stand-alone application system is the packaging and The implemented hardware components for the smart wiper mounting of the sensors and controller, which involves brackets, for this project is shown in Fig. 2 and outlined below: wiring, and locations. The search for optimum setting requires Control Panel. A Control Panel was designed with a threecareful considerations and innovative design. Cost Consideration. The additional cost of sensors, signal position switch for selecting Automatic I Manual I Off modes and conditioners, and microcontrollers will be added to that of a a Washer fluid push button for the wiper. standard wiper system. a. Automatic mode: In the Automatic mode the smart controlReliability Consideration. Reliability of the smart wiper ler will turn on the wiper as needed. The smart controller will sensor andlor sensor fusion technique must be evaluated through decide the intermittent interval and select a "Low" or "High" extensive and thorough testing under various extreme weather speed for the wiper motor. Note: Any time the automatic mode conditions. is selected the smart controller will turn on the washer motor for For research purposes, one should attempt to investigate all four cycles and turn it off at the fifth cycle. This will tell the potential desirable components for developing a smart wiper. drivers that helshe has selected the automatic mode. Under project resource constraints of time and cost, however, one b. Manual mode: In the Manual mode, a driver can select the might single out only the minimum number of components for intermittent delay interval for low-speed wiper action, constant development as well as production consideration. It has been low-speed action, and constant high-speed action. Note: For the estimated that a mass-produced smart wiper unit might cost a few U S . dollars for minimal component specification to several dollars for maximal specifications.
l"P"l
Automatic Wiper With Conductive Rain Sensor
Fuzzy Logic Scheme

In this project, the above three types of sensors were tested and investigated for possible use in detecting rain condition as well as sensor fusion. Because of the project constraints, however, we have selected to investigate only the approach using a conductive rain sensor and fuzzy logic inference for controlling the wiper. Hence, the scope is more focused and limited than the broader capabilities described above. The concept of multi-sensor fusion and more advanced control decision will be conducted in a future extension. Fuzzy logic is employed to imitate the
Fig. 3. Motor, gear box switches, conductive sensor and microcontroller hardware for smart wiper:
30
race car version, the Manual mode will only operate on the motor at a constant high speed. c. Off mode: In the Off mode the controller will deactivate and park the wiper. d. Washer fluid: A push button is used to activate the washer fluid motor and also operate the wiper motor for a duration of four cycles. Conductive Rain Sensor Characteristics. Two important characteristics of the conductive rain sensor were observed. The dc characteristic of sensor conductivity varies accordingly with the average wetness on the sensor plate, and the ac characteristic changes when fresh drops of water fall on the sensor. These characteristics can be exploited to detect the wetness of rain and droplet frequency. Input/ Output Signal Conditioning. A amplifier circuit was designed to extract the dc level component of the conductive sensor characteristics; the output of this filter would indicate the average ambient wetness of the rain condition. Next a band-pass
Knowledge Representation
filter with a cut-in comer frequency of 0.36 Hz and cut-off comer frequency of 16 Hz was used to extract the ac level characteristics from the sensor by removing the ambient wetness and undesirable high frequency noise. Output signal conditioning FET driver circuits for the wiper and washer are also shown in Fig. 2. Microcontroller. The M68HC 1 1EB Evaluation Board (EB) was used as a target system for implementing and debugging the control scheme for the wiper. The EB is supplied with monitor/ debugger firmware (BUFFALO), including a one-line assemblerldisassembler. An RS-232-compatible terminal or a host computer can be used to download object code and to enter run-time commands. The wire-wrap area on the EB is available for implementing additional circuitry; in this case, we needed signal conditioning and device driver circuits. In the FIDE implementation, the fuzzy logic callable functions resides in the EPROM of the microcontroller, and the I/O programs reside in the EEPROM. Packaging and Mounting. Fig. 3 shows the hardware of the Smart Wiper and the packaging and the packaging box. The PC laptop computer was used for programming development only.
Fuzzy Logic Scheme In this section, we will only present the fuzzy logic scheme used for controlling the speed of the wiper motor. Control of the washer motor/pump for the wiper was not a requirement at the time of the project, and was to be operated manually. The fuzzy Measured Decision Inputs Outcome logic scheme [6-101 for the wiper is the Mamdani-type scheme as shown in Fig. 4. The measured inputs for the logic are comprised of two variables: Rain: The dc output component from the sensor signal Fig. 4. Fuzzy logic decision process. conditioning unit; it reflects the wetness condition of the rain.. Intensity: The ac output component of Table 1. General Linguistic Rules for Controlling a Wiper the sensor signal conditioning unit; it reflects the droplet frequency of the rain. IF it is not raining or drizzling, THEN the wiper should be set to Off. The decision outcomes are output variables defined as IF it is drizzling lightly, THEN use a long delay interval setting on the wiper. Speed: Determines the wiping speed of the IF it is drizzling heavily, THEN use a short delay interval setting on the wiper. wiper motor. Interval: Determines the intermittent delay IF it is raining lightly, THEN use the continuous low speed setting in the wiper. interval for the wiper. IF it is raining heavily, THEN use the continuous high speed setting on the wiper. Knowledge Representation Rule Base. A basic reason, ing used by a driver in adjusting the control knobs for a wiper in an automobile may be Antecedent Consequence expressed linguistically as shown in Table 1. THEN set Speed to Zero and Interval to Long. Rule 1: IF Rain is Dry and Intensity is Light, This set of linguistic IFRule 2: IF Rain is Dry and Intensity is Medium, THEN set Speed to Low and Interval to Long, THEN rules can be readily understood by a human. But it is THEN set Speed to Low and Interval to Short Rule 3: IF Rain is Dry and Intensity is Heavy, highly ambiguous for comRule 4: IF Rain is Drizzle and Intensity is Light, THEN set Speed to Low and Interval to Long. puter operation since it does not stipulate any numerical Rule 5: IF Rain is Drizzle andlntensity is Medium, THEN set Speed to Low and Interval to Short values or algorithm. A means Rule 6: IF Rain is Drizzle and Intensity is Heavy, THEN set Speed to Low and Interval to Zero. for translating these linguistic rules into computer codes is to THEN set Sueed to Low and interval to Short. Rule 7: IF Rain is Wet and lntensitv is Light, use fuzzy logic and fuzzy set theory [6-101. The above deciRule 8: IF Rain is Wet and Intensity is Medium, THEN set Speed to Low and Interval to Zero. sions can be more systematiTHEN set Sped to High and Interval to Zero. Rule 9: IF Rain is Wet and Intensity is Heavy,
IF-THEN Fuzzy Logic Rules
Fuzzy Membership Functions
~
1
1
1
1
December 2996
31
pSlow(Speea)
pFast(Speed)
sets orfum labels ((Dry,Drizzle, Wet)or (Light,Medium, Heavy)). The degrees of associations are simply given by the values p ~ v y ( R o, ) pDDrizzle(R0) , pwet(R0) pLght(T0) , pMedium(1O) , pHeavy(I0) .
0.5
' 0
20
40 60 Rain
80
4 6 Speed
Intensity
Interval
Fig. 5 Membershipfunctions offuzzy variables.
rewriting the above fuzzy rules as a set of fuzzy logic rules (see Table 2). In setting up the fuzzy logic rules, we have defined Rain and Intensity as the Input Fuzzy Variables; Speed and Interval as the Output Fuzzy Variables. Each of these fuzzy variables is associated with descriptors or labels such as Dry, Drizzle, Wet, Light, Medium, Heavy, Stop, Low,High, Zero, Short, and Long, which are the Fuzzy Sets. Fig. 4 shows the Membership Functions that psriSpeed) = A i * pstop(Speed) define the degree or extent to which the fuzzy variables belong psz(Speed) = A2 * pLow(Speed) to the fuzzy sets. Data Base. The membership functions for the input variables 3 * p.Low(Speed) ps3(Speed) = A Rain and Intensity are defined as ps.i(Speed) = A4 * pLow(Speed) ( 4 ) pDry(Ruin)= Trapezoid(0,0, 10,50) p s s ( S p e 4 = A5 * pLow(Speed) pDrizzle(Rain) = Trupezoid(l0, 50, 50, 90) pss(Spee4 = A 6 * pLow(Spee4 pwet(Rain)= Trapezoid(50,90, 100, 100) (1) ps7(Speed) = AI * pLow(Speed) pLight(Zntensity) = Trapezoid(0,0, 0, 100) pss(Spee4 = A8 * pLow(Spee4 pMedium(IntensityY) = Trapezoid(0, 100, 100,200) ps9(Speed) = A9 * pHigh(Speed) p,yeavy(lntensity) = Trapezoid( 100, 200, 250, 250) and and the output variables Speed and Interval as pzl(IntewuZ)= A1 * pbng(lntewaZ) pstop(Speed)= Trapezoid(0,0,0.8, 1 ) = A2 * pLong(lntervaZ) piz(lnterva1) pLow(Speed) = Trapezoid(4,4.5,5,5.5) p13(ZntervaZ) = A3 * pshor&tewal) pHigh(Speed)= Trapezoid(9,9.5, 10, 10) (2) p14(Zntewal)= A4 * pbng(Zntewal) pzero(Interval)= Trapezoid(0,0, 0, 6 ) p15(Interval)= A5 * pshort(Intewal) pshort(Interval)= Trapezoid(0,6, 12, 18) (5) pLons(Interval)= Trapezoid(l2, 18, 20, 20) p16(Interval) = A 6 * pzero(lntewal) where Trapezoid denotes a unit height trapezoidal function pn(lnterva1) = AI * pShort(lnterval) whose break points on the abscissa are defined by the four p~(Intewu1) = A8 * pze,(1ntewal) arguments. Fig. 5 illustrates the membership functions described p19(Intewal) = A9 * pzero(lnterval) by Equations ( 1 ) and (2).Note that the ranges (the universes of discourse) for the input and output variable are scaled and tuned Equations (4) and (5) represent the contributions of the individual according to hardware specifications. These membership func- fuzzy logic rules shown in Table 2. tions were determined by trial-and-success optimization through Defuzzification extensive experimental tests (next section). Defuzzification is a stage where the individual contributions Fuzzification are combined to produce an overall outcome for the fuzzy logic The input variables,Rain (wetness) and Intensity (density), will decision. The mechanism used for the wiper logic is as follows. receive numerical readings from the sensor and signal conditioning Using the method of Maximum on Sum, the overall implied units, say, Rain = Ro and Intensity =lo. The process of fnzzification Wiper Speed is given by the Peakspeed, which corresponds to is to assign a degree (a.k.a. an extent), of how closely a measured the peak or maximum of the aggregated or summed individual input value (Rain = Ro and Intensity = Io) is associated to thefum implied speed function
Inference Fuzzy logic inference is a process where the strength, implication, and contribution of each of the fuzzy logic rules is evaluated. The inference mechanism chosen for the wiper logic is as follows. Applying the product method [ l o ] ,the strengths (A:'s) of the antecedents in the nine rules are given by A1 = PDry(R0) * pLight(I0) * FMedium(I0) A2 = PD.O.~(RO) A3 = p~ry(R01 * pHeavy(I0) A4 = pDrizzLe(R0) * /J-Light(Io) (3) A 5 = pDLDrizzle(R0) * pMedium(10) A6 = pDrizzLe(R0)* pfhavy(10) A 7 pWet(R0) * p~ight(10) A 8 = pWet(R0) * pil/ledium(IO) A9 = p~et(R01 * pHeavy(I0) If we also apply product method to the consequence, the implied individual contributions (in the form of membership functions) of the speed and interval rules are given by
32
10
8
6
4 2
0 0
100
300 -0 20
15
10
5 0 0
100
300 -0
Translate linguistic productiodlogic rules into a fuzzy logic inference system. Convert the fuzzy logic into assembly code and download it to the 68HC11 target system using FIDE. Execute fuzzy logic control to test the outcome using integrated hardwarehoftware smart wiper system. Fine-tune the membership functions and/or alter the logic. Once the fuzzy code is assembled in the PC, it can be embedded into a main microcontroller unit (MCU) program. FIDE employs a simple code insertion technique shown in Fig. 7. The first block of code is the pre-processing block that reads the analog to digital representation of conditioned external sensor signals. The second block is the fuzzy logic routine that reads the acquired data, evaluate the membership functions, execute the fuzzy logic operations to determine the individual strengths and outputs of the rules, and compute the maximum location and center of gravity of the aggregated outputs. The third block sends the results to the output routine, which drives the actuators. For our case, the fuzzy logic outputs the values for Speed and Interval. The Speed determines wiping action (High-Slow-Stop) for the wiper. A pulse modulation routine was used to interpret the Interval and generate the intermittent time delay between wiping actions. Testing and Optimization Test Vehicle and Facility. To evaluate the smart wiper system, the fuzzy logic microcontroller was hooked up to the wiper motor of a passenger car in a vehicle research laboratory at the School of Engineering and Computer Science, Oakland University, Rochester, MI. To facilitate testing, a portable shower head fixture was constructed to simulate rain and drizzle conditions. Simulated light, medium, and heavy rain conditions were used to testing and calibrate the microcontroller decisions for the smart wiper. Optimization. To modify the fuzzy logic code in the microcontroller, it is necessary to erase the previously transferred code, then transfer the new code to the microcontroller embedded EPROM. Although this process can be a little cumbersome and time-consuming, it is a convenient way for developing a standalone, real-time microcontroller-based fuzzy logic controller. Optimization of the smart wiper performance was carried out via extensive testing in this manner.
Fig. 6. Fuzzy logic decision map.

0
pSpeedSum(Speed) =
i=l
Using the method of Center of Gravity,the overall implied Wiper Interval is given by
Interval,,
pIidInterval
i=l
(7)
In summary, Equations (1) through (7) define the steps for evaluating the fuzzy logic in Table 2. The decision maps for the fuzzy logic are shown in Fig. 6.
Microcontroller Realization of Fuzzy Wiper Logic

FIDE Hardware and Software Development The Fuzzy Inference Development Environment (FIDE) is a powerful PC-based software tool that allows users to easily edit, simulate, debug, and tune the membership functions and rules of a fuzzy logic application. To use the FIDE, a user must understand the physical system and its control requirements. The fuzzy logic controller was developed in stages as follows. Identify the inputs and outputs for the controller. Express experience and expertise in the form of rules, and define a set of preliminary membership functions based on best guesses for the fuzzy variables in the antecedents and consequences of the rules.
and light rain condition. The subplots in the figure show (l)thedc(rain wetness) component from conductivesensor signal
( 2 ) the ac (rain in-
Other MCU routines
pqfj
Fuzzy code Other MCU routines
tensity) components
December 1996
33
[3] Y.-K. Chin, A. Kade, J. Kowalik, andD. Graham, Electronic Windshield Wiper System 11: Control and Sensitivity Study, Int. J. of Vehicle Design, vol. 12, no. 2, 1991, pp. 183.196.
35
40
4C signal
45
50
[4] Electric Wiper Motor Specijcations, Sprague Devices Inc., Michigan City, IN.
[ 5 ]Windshield WiperSystems, United Technology Automotive-Motor tems Inc., Columbus, MS.
Sys-
f
01
15
20
25
30
35
40
45
50
Wiper Motor Delay lntelval 20

15
0 ; 10
[6] D. Warbutton, How to Design Fuzzy Logic Controllers, Machine Design, Nov. 26, 1992, pp. 92-93. [7] R.R. Yager and L.A. Zadeh (eds.), An Introduction to Fuzzy Logic Applications in Intelligent Systems, Kluwer Academic Publishers, 1993.
01
15
20
25
30
35
40
45
50
Wiper Mobr Speed
1 [9] D. Driankov, H. Hellendoom, and M. Reinfrank, An Introduction to Fuzzy

Control, Springer-Verlag, 1993.
I 1
[8] L.-X. Wang, Adaptive Fuuy Systems and Control: Design and Stability Analysis, Prentice-Hall, 1994.
1 ;
2 ;
25
3b Time in seconds
3;
4b
45
50
[lo] T. Terano, K. Asai, and M. Sugeno, Fuzzy Systems Theory and Its Applications, Academic Press, 1992. [ l l ] M68HC11 Manual, Motorola, Inc., 1995. [12] FIDE Manual, Aptronix Inc., 1995. Ka C. Cheok is a professor in the Department of Electrical and Systems Engineering at Oakland University, Rochester, MI. He received his B.S.E.E. in 1977 from the University of Malaya, and his M.S. in electrical and computer engineering in 1979 and his Ph.D. in systems engineering in 1982 from Oakland University. Cheok is active in the research areas of computercontrolled systems involving artificial intelligence and computer visualization. His contributions include the development of heuristics search, fuzzy logic and neural network techniques for guidance of autonomous robotic vehicles, and control of active suspension, automotive traction and tracking systems for the automotive and defense industries. He has published more than 90 technical joumal and conference articles. He received a state of Michigan-sponsored Teaching Excellence Award from Oakland University in 1990.
Light Rain Heavy Rain Medium Rain No Rain Condition Condition Condition Condition Fig. 8. Experimental test results from actual fuzzy logic-based smart
automatic windshield wiper
(3) the command for the wiper delay interval from the FL controller (4) the command for the wiper motor speed from the FL controller As can be seen, the performance of the smart wiper resulting from these outcome was quite satisfactory. The wiper motor is off when there is no rain. Then in a fuzzy sense, it operates at full speed with zero delay interval when the rain is heavy, at low speed with short delay intervals when the rain is medium, and low speed with long delay interval when the rain is light.
Conclusions
Performance of a control system is an ultimate issue in the controller design. The design of a fuzzy logic controller uses intuitive reasoning and does not necessarily require detailed mathematical modeling of the plant dynamics to be controlled. Unavailability of a good plant dynamic model makes it difficult to use computer simulation to accurately predict and tune the performance of the fuzzy control system. In such situations, a way to evaluate the performance is to build a physical hardware simulator, implement fuzzy logic using a target controller board, test its behaviors, and tune the system until the desired output is achieved, as we have done in this project. This project successfully demonstrates how fuzzy logic can be used to implement a smart wiper system that uses a conductive rain sensor. FIDE was used as a convenient systematic tool to implement the fuzzy logic for automatic operation of a wiper. Future extension of the effort would be to conduct more extensive research and development with multi-sensor fusion.
K. Kobayashi received his B S in electrical engineering from Hosei University, Japan, in 1989, and his M.S. and Ph.D. in systems engineenng from Hosei University in 1991 and 1994, respectively From 1994 to 1996, he was a research associate at Oakland University He is currently a lecturer at Hosei University His research area includes vehicle traction control, vehicle absolute speed measurment using fuzzy logic, and sensor fusion for automotive applications
Sandro Scacciais the president and CEO of TechnoRes S E E and B S C E from search Inc He received h ~ B Oakland University in 1988 and his M S E in 1993 He is a member of the SAE and IEEE.
References
[l] B.S. Hsu and S.F. Ling, Windshield Wiper System Design, Int. J. of Vehicle Design, vol. 11, no. 1, 1990, pp. 63-78.
[21 Y-K. Chin, A. Kade, J. Kowalik, andD. Graham, Electronic Windshield Wiper System I: Modeling and Validation, Int. J. of Vehicle Design, vol. 12, no. 2, 1991, pp. 175-182.
Giamberto Scaccia is a senior electronic engineer at Magneti Mare111 U.S.A. He received the B.S.E.E. and B.S C E from Oakland University in 1988 and his M.S.E. from Wayne State University in 1994. He is a member of the SAE and IEEE
34
Intelligent Rain Sensing and Fuzzy Wiper Control Algorithm for Vision-based Smart Windshield Wiper System
Joonwoo Son Seon Bong Lee Department of Mechatronics, Daegu Gyeongbuk Institute Science & Technology, Deagu 700-742, Korea * Man Ho Kim Suk Lee Department of Mechanical Engineering, Pusan National University, Busan 609-735, Korea ** Kyung Chang Lee Department of Instrumentation Engineering, Pukyong National University, Busan 609-735, Korea
Key Words :
Smart Wiper System, Intelligent Vehicle, Rain Sensing, Wiper Control, Vision Sensor, Fuzzy Control Abstract
Windshield wipers play a key role in assuring the drivers safety during precipitation. The traditional wiper systems, however, requires drivers constant attention in adjusting the wiper speed and the intermittent wiper interval because the amount of precipitation on the windshield constantly varies according to time and vehicles speed. Because the manual adjustment of the wiper distracts driver's attention, which may be a direct cause of traffic accidents, many companies have developed automatic wiper systems using some optical sensors with various levels of success. This paper presents the development of vision-based smart windshield wiper system that can automatically adjust its speed and intermittent interval according to the amount of water drops on the windshield. The system employs various image processing algorithms to detect water drops and fuzzy logic to determine the speed and the interval of the wiper. transportation system (A. Broggi, 2000; L. Figueiredo et al., 2001). Among the vehicle parts that influence the driver's safety and convenience, a windshield wiper is an important part that allows a driver to collect visual information during precipitation. The traditional wiper systems, however, requires drivers constant attention in adjusting the wiper speed and the intermittent wiper interval because the amount of precipitation on the windshield constantly varies according to time and vehicles speed. Because the manual adjustment of the wiper distracts driver's attention, which may be a direct cause of traffic accidents, many companies have developed automatic wiper systems using some optical sensors with various levels of success. Many smart wiper systems try to regulate the wipers speed and intermittent interval automatically according to the amount of rain or snow (K. C. Cheok et al., 1996; H. Kato et al., 1990). The key element of these systems is the sensor to measure the amount of water on the windshield. In most systems, an optical sensor is used for this purpose (K. C. Cheok et al., 1996). This type of sensors uses the fact that the refraction angle and the amount of reflection of the light are different when the 1
1.
Introduction
Intelligent vehicles offer the potential to enhance safety and convenience significantly for both drivers and passengers. As a component of the intelligent transportation system (ITS), the intelligent vehicle uses various intelligent sensing and control algorithms to assess the vehicles environment and assist the driver with safe driving. These algorithms include the driver assistance system that partially controls the vehicle for the drivers convenience, and the collision warning system that provides emergency information to allow the driver to avoid a collision. Because of their ability to enhance the safety and convenience, intelligent vehicles become a crucial research area for the intelligent
* Corresponding Author
E-mail : slee@pnu.edu TEL : +82-51-510-2320; FAX : +82-51-514-0685 Department of Mechanical Engineering, Pusan National University, Busan 609-735, Korea. (Manuscript Received February 11, 2006; Revised April 30, 2006)
** Co-corresponding Author
windshield is wet. Many sensors use several LEDs to emit the light into the windshield at a small incident angle. The light travels through the glass with several reflections at the glass-air border. Opposite to the LEDs, there are several photo sensors to detect the arriving light after the reflections. When the windshield has water on its surface, the refraction angle and the amount of reflected light are changed, resulting in the different photo sensor output. This difference is used to determine the amount of water on the windshield. The optical rain sensors have some disadvantages even though they are widely used. One of disadvantages is the sensitivity to external light. That is why many systems have a dark filter on the windshield surface where the sensor is attached in order to block the external light. However, many systems still activate the wiper when the car comes out of tunnels or underground parking lot. Another shortfall, maybe a major one is that the sensing area is a relatively small portion of windshield. This makes the wiper system operate only with limited information. The wiper system may fail to activate when there are some raindrops on the drivers line of sight, but not on the sensing area. For solving these problems, this paper presents a vision-based smart wiper system that can measure a relatively wider windshield area than the conventional system. Especially, the vision sensor has advantage that can measure not only the amount of rain but also the distribution of raindrops. This information can allow the system more intelligently without disturbing drivers. In this paper, we present the concept and key algorithms for a vision- based smart wiper system. More specifically, this paper shows an effective rain sensing algorithm that can characterize the rainfall during the day and night. In addition, this paper introduces an intelligent wiper control algorithm based on fuzzy logic that can regulate the wiper speed and interval. Finally, experimental results under the laboratory settings are presented to evaluate the efficacy of the proposed system. This paper consists of six sections including this introduction. Section 2 introduces the basic structure of a vision-based smart wiper system, and Section 3 presents the rain sensing algorithm. The intelligent wiper control
vision sensor windshield rain LEDs fuzzy controller
system using fuzzy logic is described in Section 4, and the experimental results are given in Section 5. Finally, the conclusions are presented in Section 6.
2.
Structure of vision-based smart wiper system
In many cases, the sensing area of an optical smart wiper system is very small; the area is at most 10 mm wide and 30 mm long. This makes the whole system operate on very limited information. When it starts to rain, the wiper begins to move only after some raindrops fall on to the sensing area. Moreover, the wiper may not start at all when water is splashed by other cars on to the area other than the sensing area. In order to avoid these problems, we developed a smart wiper system using a vision sensor as shown in Fig. 1. The smart wiper system consists of a wiper motor, a wiper switch, a fuzzy controller, a vision sensor, and light emitting diodes (LEDs). Here, the wiper switch is used for activating the smart wiper system. The fuzzy controller performs two functions: the rain sensing algorithm that calculates the rain intensity and rain distribution from the image captured by the vision sensor, and the fuzzy wiper control algorithm that computes the wiper speed and wiper interval based on rain intensity and distribution. Finally, the vision sensor is installed on the ceiling of the vehicle to capture the windshield image, and the LEDs are installed in order to provide some lighting during the nighttime. To guarantee the performance of the vision-based smart wiper system, the external background outside of the vehicle must be ignored, and only the raindrops on the windshield can be extracted from the image captured by the vision sensor. For this purpose, we have focused
wiper motor
wiper switch
Fig. 1 Schematic diagram of vision-based smart wiper system 2
Fig. 2 Raindrop image with low depth of field
the vision sensor on the windshield with low depth of field by opening the aperture to the maximum. The captured images as shown in Fig. 2 show that the external background outside of the windshield is blurred while the raindrops on the windshield are distinct. Finally, the vision sensor can be affected by an external light source such as sunlight in the daytime, street lamps, tail lamps, and headlamps in the nighttime. Part of disturbances such as sunlight and street lamps can be avoided by installing the vision sensor tilted downward as shown in Fig. 1. However, other disturbances such as tail lamps and headlamps cannot be eliminated. Therefore, an appropriate image processing algorithm is necessary to eliminate the effect of these external light sources. For this purpose, we have an additional image processing step for the nighttime.
are applied, the algorithm computes rain intensity and distribution that are used to determine the wiper speed and interval. The rain intensity represents the amount of rain on the windshield while the rain distribution indicates how widely the raindrops are distributed on the windshield. 3.1 Daytime image processing algorithm In order to detect raindrops on the windshield, the algorithm finds the boundaries of raindrops by using an edge detection technique. Among many techniques such as Sobel, Prewitt, Roberts, Laplacian, and Laplacian of Gaussian (R. C. Gonzalez et al., 1992), the Sobel mask is selected because its edge detection capability is neither too bad to detect raindrops nor too good so that the edge of background is detected. Fig. 4(a) shows an original image of raindrops while Fig. 4(b) shows the image after the Sobel mask is applied. In the figure, we can see that the boundaries of raindrops are clear without any noticeable edges from the background. The image in Fig. 4(b) still has the boundaries of the background, though not very visible, that are represented gray pixels. This is because the background image is blurred due to the low depth of field. In order to remove these background boundaries and shorten the processing time, a single threshold T is applied as shown in Eq. (1) where and are the old and new values for the pixel, respectively (R. Crane, 1997). In this equation, the value for T is selected as 125 after some trials and errors..
3.
Rain sensing algorithm of smart wiper system
Fig. 3 shows the rain sensing algorithm for the visionbased smart wiper system. The algorithm takes a captured image as its input and produces the state of precipitation described by two variables, i.e., rain intensity and rain distribution as its outputs. In order to obtain the outputs, the algorithm employs a series of image processing techniques. In the figure, after the wiper switch is set to auto mode, the vision sensor acquires an image of the windshield. Then, depending on whether the headlamps are on or off, the algorithm decides whether the additional nighttime processing is required. After appropriate image processing techniques
high (255) if f ( x, y ) T F ( x, y ) = otherwise low (0)
(1)
wiper switch set to auto image capture nighttime ? daytime nighttime image filtering
Fig. 4(c) shows the image after thresholding where we can see that the gray background boundaries are completely gone. The next step is to whiten the pixels inside the
edge detection binary image processing dilation processing characterization of precipitation
(a) original image
(b) image after Sobel mask
(c) image after threshold (d) image after dilation

Fig. 3 Flowchart of rain sensing algorithm 3 Fig. 4 Result of image processing for the daytime
boundary. For this purpose, we use the dilation operation (R. C. Gonzalez et al., 1992) as an approximation. In general, the dilation makes an object enlarged because the operation extends the outermost pixels of the boundary. Therefore, the raindrops will be slightly larger and some of raindrops still may have black pixels inside the boundary. In this work, a 33 dilation mask, whose pixel values are all zeros, is used for faster computational speed (D.H. Jang 1999). Fig. 4(d) shows the final image after the dilation operation. Comparing Fig. 4(c) with Fig. 4(d), we can verify that most raindrop boundaries are filled with white pixels. 3.2 Nighttime image processing algorithm As mentioned earlier, the nighttime images tend to contain more disturbances such as street lamps, tail lamps, and headlamps. When the daytime processing algorithm is applied to the nighttime image, the effect of disturbances is still remaining in the final image. This indicates that the nighttime processing algorithm should be different from the daytime algorithm. For this purpose, we focus on the fact that sharp features, in general, have relatively high frequency elements, while blurred features have relatively low frequency elements. That is, the raindrops under the internal LED light are distinct and have high frequency elements whereas the external light sources are blurred by the low depth of field and have low frequency elements. Therefore, disturbances by the external light source can be removed by eliminating low frequency elements from the image. For removing low frequency elements, we choose the Butterworth high pass filter that can be expressed as follows:
The wiper operation algorithm requires some input values derived from the processed image. Therefore, we need a step, i.e., characterization of precipitation, to represent the processed image by some values. In order to derive these values we rely on the notion that the factors that hinder the drivers view are the amount of water and the number of raindrops on the windshield. First, when the windshield is covered by the significant amount of water, the driver will have difficulty in finding lanes and obstacles. However, the amount of water is not the only factor because the drivers view is considerably blocked when numerous but very small raindrops are covering the windshield. Therefore, we define two variables: rain intensity and rain distribution. The rain intensity is defined as the number of pixels representing raindrops (its value is high) in the processed image (256256 pixel). The rain distribution is computed as follows: 1) divide the overall area of 256256 pixel image into 1616 pixel unit areas, 2) if there is at least one pixel representing a raindrop in the 1616 pixel unit area, assign high value to all pixels of the 1616 unit area, 3) the rain distribution is defined as the ratio of the number of pixels with high value to the number of overall pixels (65,536). Fig. 6 shows an example how the rain intensity and rain distribution are calculated. The large square represents the 256256 captured image while the small square represents the enlarged 1616 unit area. Fig. 6(a) represents the case for drizzle or mist. Suppose that raindrops are of the size of one pixel and they are evenly
H (u , v ) =
1 1 + [ D0 / D (u , v)] 2 n
(2)
where D(u,v) is the distance from the origin to (u,v) on the frequency plane. In this work, the cutoff frequency is selected to be 32, and n is selected to be 16 via some trials and errors. Fig. 5(a) shows a nighttime image containing a tail lamp. Fig. 5(b) shows the image where the tail lamp is effectively removed after the high pass filter is applied to the original image. 3.3 Characterization of prediction
(a) mist
(a) original image
(b) image after high pass filtering
(b) a single large raindrop

Fig. 6 Example of rainfall condition by wetness and distribution
Fig. 5 Result of high pass filtering for the nighttime image
distributed so that each 1616 unit area contains four raindrops as shown in the small square in Fig. 6(a). Then, the rain intensity is the number of raindrop pixels which equals 1024 (4 in each unit area 16 16 unit areas in the processed image). The rain distribution is 100% because all pixels of a unit area should have high value due to four pixels in the unit area, and all unit areas have high value due to the even distribution. In contrast, Fig. 6(b) represents the case of a single big raindrop at the center of the image. In this case, the raindrop is as large as four unit areas that contain 1024 pixels. Therefore, the rain intensity is 1024. The rain distribution is 1.56% because only four unit areas have high value out of 256 unit areas (4/256100).
Fig. 8 shows the membership functions of the fuzzy input and output linguistic variables. The linguistic variables for the rain intensity are defined as Dry, Drizzle, and Rain while those for the rain distribution are defined as Narrow, Medium, and Wide as shown in Fig. 8(a). In the figure, the membership functions of rain intensity are different depending on whether it is daytime or nighttime. This is to compensate the boundary detection capability in the nighttime because almost all the boundaries of raindrops are detected in the daytime while the algorithm tends to miss some raindrops in the nighttime. The linguistic variables for the wiper interval are defined as Zero, Short, and Long as shown in Fig. 8(b). Here, we
D ry D rizzle R ain
4.
Fuzzy wiper control algorithm
In this paper, a fuzzy control algorithm is used to regulate the wiper speed and wiper interval using the calculated rain intensity and distribution after the image processing. The major rationale to use fuzzy logic is that the fuzzy logic allows mimicking the human drivers decision making with relative ease and that the outcomes of the logic can be tuned or customized for individual drivers by adjusting membership functions (H. C. Lee et al., 2003; S. Y. Foo, 2000). Fig. 7 shows the structure of the fuzzy wiper control algorithm of the smart wiper system. In the figure, the fuzzy controller consists of three parts: fuzzifier, inference engine, and defuzzifier. The fuzzifier converts the rain intensity and distribution into linguistic values. The inference engine creates the fuzzy outputs using fuzzy control rules generated from expert experiences. Finally, the defuzzifier calculates the wiper speed and wiper interval from the inferred results. To acquire the fuzzy control rules, we summarize the wiper control rules based on the general behavior of human drivers as shown in Table 1 (K. C. Cheok et al., 1996). That is, if it is a light drizzle or several raindrops are on the windshield, the wiper should be set to low speed with a long interval. However, if it is raining heavily or raindrops are all over the windshield, the wiper should be set to high speed with zero interval. The rules in Table 1 are converted into more structured rules considering the input variables of rain intensity and distribution as shown in Table 2.
Fuzzy controller rain Windshield Vision sensor Image processing Fuzzifier distribution Inference engine speed Wiper DC motor interval Defuzzifier Knowledge base
(pixel) 3000 2000 7000 8000 4000 5000 R ain Intensity M edium 14000 (daytim e) 11000 (nighttim e)
N arrow
W ide
10
30 40 R ain D istribution
60
(% )
(a) membership functions of fuzzy inputs

Zero Short Long
500
1000
2000 Interval Low
2500 3000 (ms)
Zero
High
4 Speed
9 10
(b) membership functions of fuzzy outputs

Fig. 8 Membership functions of fuzzy input and output variable
Fig. 7 Schematic diagram of fuzzy controller
Table 1. General linguistic rule for wiper control
Antecedent IF it is not raining or drizzling IF it is drizzling lightly IF it is drizzling heavily IF it is raining lightly IF it is raining heavily
Consequence THEN the wiper should be set to Off. THEN use a long delay interval setting on the wiper. THEN use a short delay interval setting on the wiper. THEN use the continuous low speed setting on the wiper. THEN use the continuous high speed setting on the wiper with a frame grabber (Euresys Picolo board) and a CCD camera (Pulnixs TM-200). The PC is also connected to an interface circuit that can drive the wiper motor. The cameras aperture is adjusted to open as wide as possible (f=1.4). 5.1 Experimental results for daytime operation In order to emulate the drizzle in the daytime, water is sprayed on to the windshield with the laboratory lights turned on. Fig. 10 shows the images throughout the processing, i.e., Sobel mask, thresholding, and dilation as explained earlier. The final image contains 2,183 white pixels (rain intensity of 2, 183) and has 30.5% of 1616 unit areas with at least one raindrops (rain distribution of 30.5%). Using these two values, we can calculate that the wiper interval is 2,450msec, and that the wiper speed is low from the fuzzy wiper control algorithm. Fig. 11(a) shows the captured image of the windshield when a large amount of water is dropped to emulate the rain in the daytime. From the final image, the rain intensity is 15,831 pixels, and the rain distribution is 83.2%. This
Interval
Table 2. Fuzzy control rules for wiper control
(a) speed Rain Intensity Dry Distribution Narrow Medium Wide (b) Interval Rain Intensity Distribution Narrow Medium Wide Zero Low Low Drizzle Low Low Low Rain Low Low High
Dry Long Long Short
Drizzle Long Short Zero
Rain Short Zero Zero
define that the wiper interval varies from zero to 3,000 msec. The linguistic variables for the wiper speed are defined as Zero, Low, and High as shown in Fig. 8(b). For faster execution of the fuzzy logic controller, the Mamdanis min-max inference method is used. For defuzzification, the MoM (Mean-of-Maximum) method is used for the wiper speed, and the CoA (Center-ofArea) method is used for the wiper interval (C. C. Lee, 1990; C. C. Lee, 1990). The reason for using the MoM method is to obtain the output that changes discretely instead of varying continuously. This is because the wiper motors operate at two different speeds, i.e., high and low instead of continuously varying speed. In this paper, we have the wiper stop if the output value is smaller than one, run at high speed if the output value is larger than 9, and run at low speed otherwise. The output maps for speed and interval are shown in Fig. 9.
Rain Intensity Rain Distribution
(a) wiper interval

Interval
5.
Performance evaluation of vision-based smart wiper system
In order to evaluate the efficacy of the vision-based smart wiper, we have set up an experimental test bed in a laboratory environment. The test bed has a windshield and a wiper removed from a car. The image processing and fuzzy logic algorithms are implemented on a PC 6
Rain Distribution
Rain Intensity
(a) wiper speed

Fig 9. Output maps for wiper interval and speed
(a) original image
(b) image after Sobel mask
(a) original image
(b) image after high pass filter

Fig. 10 Result of image processing for drizzle in the daytime gives us 350msec for wiper interval and high for wiper speed. Comparing these results, we can see that the fuzzy logic behaves as intended although the maximum wiper interval of three seconds may be too small for real applications.
(c) image after Sobel mask (d) image after threshold dilation
Fig. 12 Result of image processing for drizzle in the nighttime raindrops are represented by white pixels. From the final image, we can calculate that the rain intensity is 3,352 pixels, and that the rain distribution is 44.5%. These input values result in the wiper interval of 1,840msec and the low wiper speed. Fig. 13 shows the results of an experiment for nighttime rain with an external light disturbance. In the experiment, several LEDs are used as the external light source as shown in Fig. 13(a). After performing image processing procedures for nighttime images, most of disturbance is removed as shown in Fig. 13(d). In this case, the rain intensity is 8,449 pixels and the rain distribution is 73.0%. These results in the wiper interval of 350msec and high wiper speed. In real application, if the wiper interval is too short, e.g., 500msec or less, the wiper should operate continuously. The experimental evaluation shows that the proposed
Experimental result for nighttime operation 5.2 The nighttime environment is emulated in the laboratory by conducting the experiment after nightfall with the lights turned off. Fig. 12 shows the images obtained during an experiment mimicking a drizzle in the nighttime. The capture image shown in Fig. 12(a) shows some background objects due to other external light. After applying the Butterworth high pass filter to the original image, we obtain an image as shown in Fig. 12(b) where raindrops are too dark to see. After edge detection, thresholding, and dilation, however, we can obtain the final image shown in Fig. 12(d) where
(a) original image (b) image after high pass filter (a) original image (b) image after Sobel mask

Fig. 11 Result of image processing for rain in the daytime
(c) image after Sobel mask (d) image after threshold dilation
Fig. 13 Result of image processing for rain in the nighttime
vision-baser smart wiper system can be used instead of existing optics-based systems. The proposed system may allow the developers to avoid existing patents and make the system more intelligent not to irritate the drivers due to improper operations. For practical automotive application, the algorithms should be implemented on some type of embedded system such as an electronic control unit (ECU) that has some moderate computation speed because the algorithms need to be run from time to time, say once in a few seconds. In addition, the wiper system may share camera and image capture board in the future with other vision-based smart systems such as lane tracking system or collision warning system.
Acknowledgment
This work was supported by fund of Daegu Gyeongbuk Institute Science & Technology in Ministry of Science & Technology
Reference
(1) A. Broggi, 2000, Intelligent vehicle applications worldwide, IEEE Intelligent Systems, Vol. 15, No. 1, pp. 78-81. (2) L. Figueiredo, I. Jesus, J.A.T. Machado, J.R. Rerreira, and J.L. Martines de Carvalho, 2001, Towards the development of intelligent transportation systems, 2001 IEEE Intelligent Transportation Systems Conference Proceedings, pp. 1206-1211. (3) K. C. Cheok, K. Kobayashi, S. Scaccia, and G. Scaccia, 1996, A fuzzy logic-based smart automatic windshield wiper, IEEE Control Systems, vol. 16, no. 6, pp. 28-34. (4) H. Kato and T. Matsuki, 1990, Raindrop sensor using electric double layers, Sensors and Actuators B: Chemical, Vol. 1, No. 1-6, pp. 308-311. (5) R. C. Gonzalez and R.E. Woods, 1992, Digital image processing, Addison-Wesley Publishing Company. (6) R. Crane, 1997, Simplified approach to image processing, Prentice-Hall. (7) D. H. Jang, 1999, Implementation of digital image process using Visual C++, PC Advances. (8) H. C. Lee, M. Tomizuka, 2003, Adaptive vehicle traction force control for intelligent vehicle highway systems (IVHSs), IEEE Transactions on Industrial Electronics, Vol. 50, No. 1, pp. 37-47. (9) S. Y. Foo, 2000, A fuzzy logic approach to fire detection in aircraft dry bays and engine compartments, IEEE Transactions on Industrial Electronics, Vol 47, No. 5, pp. 1161-1171. (10) C. C. Lee, 1990, Fuzzy Logic in Control Systems: Fuzzy Logic Controller Part I, IEEE Transactions on Systems, Man, and Cybernetics, Vol. 20, No. 2, pp. 404-418. (11) C. C. Lee, 1990, Fuzzy Logic in Control Systems: Fuzzy Logic Controller Part II, IEEE Transactions on Systems, Man, and Cybernetics, Vol. 20, No. 2, pp. 419-435.
6.
Summary and conclusion
In this paper, we proposed a vision-based smart wiper system to improve the safety and convenience of drivers. The system employs a series of image processing steps to remove background objects and identify raindrops on the windshield. From the processed image, rain intensity and distribution are calculated as two inputs to the fuzzy logic algorithm selecting wiper speed and interval. In order to demonstrate the efficacy of the proposed system, the system was evaluated through a series of experiments under laboratory settings. The conclusions derived from the research are as follows. It is demonstrated that the vision-based wiper system is technically viable by combining appropriate image processing steps. This vision-based approach provides far more information that the existing optics-based systems. The optical sensors sensing area may be too small to detect the sparsely distributed raindrops and the sensor cannot detect the raindrops directly on the line of sight of driver. In contrast, the vision sensor can cover an area large enough to detect sparsely distributed raindrops and raindrops on the drivers line of sight. Rain intensity and distribution may be suitable variables to characterize a rainfall. The amount of water used to be the only variable to characterize a rainfall in existing systems. However, we have added another variable, rain distribution to represent how raindrops are distributed on the windshield. By using these to variables and a set of simple fuzzy rules, we were able to determine the wiper speed and intermittent interval. In spite of the above conclusions, the vision-based wiper system is not still economically viable due to the extra cost for camera, frame capture circuitry, and controller. We hope that the advantages of vision-based wiper system will outweigh the extra cost and complexity in the near future when intelligent vehicles begin to appear. When the proposed system begins to make sense economically, the proposed algorithms should be tested in the field and the logics should be fine-tuned by observing drivers behavior.

Problem Set 1

Diunggah oleh

Informasi Dokumen

Deskripsi Asli:

Hak Cipta

Format Tersedia

Bagikan dokumen Ini

Bagikan atau Tanam Dokumen

Opsi Berbagi

Apakah menurut Anda dokumen ini bermanfaat?

Apakah konten ini tidak pantas?

Hak Cipta:

Format Tersedia

Problem Set 1

Diunggah oleh

Hak Cipta:

Format Tersedia

UCSI University Department of Electrical & Electronics Engineering EE406 Control Systems Introduction to Control Systems Problem Set

1 Due : In-class, 10 May 2012

AFRL-VA-WP-TP-2002-317 AUTOMATIC AIR COLLISION AVOIDANCE SYSTEM

Approved for public release; distribution is unlimited.

REPORT DOCUMENTATION PAGE

Form Approved 0MB No. 0704-018

1, REPORT DATE (DD-MM-YY) Jxme 2002

3. DATES COVERED fFrom-To)

Journal Article Preprint

AUTOMATIC AIR COLLISION AVOIDANCE SYSTEM

F33615-01-2-3103 5b. GRANT NUMBER 5c. PROGRAM ELEMENT NUMBER 6321IF

5cl. PROJECT NUMBER

Lockheed Martin Aeronautics P.O. Box 748 Fort Worth, TX 7610]

The Boemg Company AFRL/VACC

9. SPONSORING/MONrrORING AGENCY NAME(S) AND ADDRESS(ES)

AFRL-VA-WP-TP-2002-317 Approved for public release; distribution is unlimited.

IS. SUBJECT TERMS

19b. TELEPHONE NUMBER (Include Area Code)

Automatic Air Collision Avoidance System

Intruder Data In Data Link Out-Network

Correct Flight Path

Escape Angle Selection

Roll Logic Host Data Out

Figure 1 - Algorithm Architecture

(b) Figure 4. Head-on Collision Scenario - (a) Intruder, (b) Host

4 Time .- TkTi^Un.'>nA -t^ r4iW^^n

KUCHAR AND DRUMM

The Traffic Alert and Collision Avoidance System

The Trafc Alert and Collision Avoidance System

KUCHAR AND DRUMM

The Trafc Alert and Collision Avoidance System

Mid-air collision rate (per million flight hours)

Worldwide annual jet transport flight hours (millions)

KUCHAR AND DRUMM

The Trafc Alert and Collision Avoidance System

Range, bearing, altitude

Trajectory extrapolation Time to collision Threat detection

Resolution advisory display

Other information sources (ATC, visual acquisition)

How TCAS Works

KUCHAR AND DRUMM

The Trafc Alert and Collision Avoidance System

KUCHAR AND DRUMM

The Trafc Alert and Collision Avoidance System

Threat Detection and Display

KUCHAR AND DRUMM

The Trafc Alert and Collision Avoidance System

FIGURE 4. The TCAS algorithm selects the

Vertical separation (feet) at closest point of approach

Result of dont descend

600 500 400 300 200 100 Required minimum separation

Resu Re lt of su limit lt o d f li 500 ft/mi escent n mi td es ce nt 100 0f t/m in

LINCOLN LABORATORY JOURNAL

VOLUME 16, NUMBER 2, 2007

KUCHAR AND DRUMM

The Trafc Alert and Collision Avoidance System

Lessons from a Disaster

TCAS. (Photograph source: German accident investigation report.)

VOLUME 16, NUMBER 2, 2007

KUCHAR AND DRUMM

The Trafc Alert and Collision Avoidance System

ATC instruction to descend Russian Tu-154 Climb, climb

Ac tua l tr aje cto ry