Anda di halaman 1dari 6

WIRELESS SENSOR NETWORKS IN ELECTRICAL MANUFACTURING

Purdue University
Abstract: Sensors along with processing and communication electronics are becoming small enough to be tightly integrated into a wide variety of systems from biological and environmental to manufacturing. Along with this miniaturization, wireless sensor networks have recently become an area of great interest to researchers and industry. They offer the promise of monitoring, data collection, and control of systems with unprecedented scale and spatial granularity. It is widely believed that wireless sensor network system design, installation, and maintenance are simplified due to the absence of physical wires. As with any emerging technology, pitfalls are as significant as the benefits. Today the performance of these networks is highly application dependent, with many different, yet tightly coupled challenges. This paper provides an overview and description of what wireless sensor networks are and are not. Challenges with their design and deployment in electrical manufacturing environments are also presented. The current wireless sensor network state of the art and obstacles to overcome are reviewed. Finally, a speculative view of future trends is offered.

Jeffrey J. Evans

pursued by materials scientists, engineers, and industry. Sensor networks are now actively addressing the latter. For a sensor network to be most effective in applications monitoring environmental or other parametric characteristics, it must become an integral part of the environment while not disturbing it. For a sensor network monitoring critical parameters of a production line the sensors must ideally become part of the production line (like a piece of smart dust [7]). Applications monitoring a dynamic process should become part of that process without disturbing it. Assuming the sensor technology exists, enormous constraints are still placed on the sensor network in terms of processing, basic communication, and data management.

Key Words: Sensor networks; Wireless Systems I. INTRODUCTION


The area of wireless sensor networks has been gaining wide spread attention at a rate complimentary with that of the reduction of device size. Device size reduction combined with improvements in wireless communications has sparked interest in deploying distributed networks of sensors (i.e. sensor networks) in a wide variety of application scenarios. Application areas include object tracking systems, environmental monitoring and in some cases, like buildings and manufacturing systems, control. Some monitoring applications continue to be addressed using manual or semi-automated techniques. One reason for this is due to complexity constraints imposed on the use of certain sensors. Some sensors require significant amounts of energy to operate, special implementation techniques, or both. Challenges of sensor robustness, calibration, and packaging are another. This is further complicated with packaging the computing and communications electronics with sufficient space and energy efficiency. The former constraint continues to be

Information gathered from the sensors, whether processed or not (another sensor network design decision), must somehow propagate from each sensor to the location where the bulk of the processing and decision-making will take place. Depending on the communication network topology a sensor "node" may serve a dual purpose, as both a sensor and a data "router". Furthermore, if the sensors are providing feedback for the purpose of process control, latency variations induced by processing and communication can have adverse affects on process (and

ultimately product) stability.

Aspects of communications influence the performance of any sensor network. Wired networks are topologically less flexible and exhibit failure modes due to mechanical wear. Wireless systems offer topological flexibility, but can be plagued by electro-magnetic interference and data integrity/security concerns. The communications medium is shared and available to anyone who wishes to listen. Therefore the designer is faced with balancing data integrity and security with the cost associated with extra processing and communications overhead. These overheads ultimately combine to influence the energy budget of each node and the system, since access to unlimited power becomes more complicated and difficult.
The remainder of this document is presented as follows. An overview of sensor networks and motivations for manufacturing applications is presented in section II. Section III presents sensor network enabling technologies as well as threats and pitfalls to achieving desired outcomes by virtue of their use (and misuse). Finally, a summary and conclusions are offered in section IV.

0-7803-91 45-4/05/$20.0020051EEE

460

II. SENSOR NETWORK MOTIVATION


A. Sensor Network Overview One of the primary descriptions often used with sensor networks is that its purpose is to enable the solving of a large problem using a collection of simple, autonomous "'sensor nodes" that work collectively to provide data processing and propagation for the solving of the much larger problem. The idea is that many sensors can work together in a robust (fault tolerant) fashion to provide data (and in some cases control) of a much larger system.
A typical sensor network is illustrated in Figure 1. The network consists of many sensing "nodes" and a communications network such that the data flow is normally from the end point (sensor) to (or through) an aggregation point or points. From here the data may continue on to a "head-end", sometimes called a data sink node or cluster head. Here data may be further processed or forwarded to a data collection point or application task manager node. This node is often in the form of a computer workstation and may be remotely located.

Sensor networks exhibit several unique and challenging characteristics. As outlined in [6], small physical size and low power consumption are primary attributes of the sensing "nodes". The small physical size made possible by improved hardware design must be complimented with efficient software to ensure reasonable processing, storage, and communication overhead. Robust operation is also required as operation will be largely unattended. Application reliability can be enhanced by tolerance of individual sensor node failure, as long as sufficient numbers and density of nodes are present. Some of the characteristics of sensor networks and other systems (ad-hoc networks or single-sensor systems) are:
* The number of sensor (nodes) in a sensor network may be significantly larger (by orders of magnitude) than ad-hoc or traditional single-sensor systems. * Individual sensors/nodes may be prone to failures. * The communication "topology" of a sensor network frequently changes. * Sensor nodes often communicate using a broadcast paradigm. Ad-hoc and single-sensor systems more often use point-to-point communications. * It is assumed that sensor nodes are limited in terms of power availability, computational capacity and memory. * Depending on the application sensor nodes may or may not have a form of global identification. Global identification comes at the price of increased overhead.

Figure 1. Sensor Network


Many factors determine the complexity of the sensor network. Power management requirements may place restrictions on communication speed, modulation technique, and medium access mechanisms. Applications unconcerned with power management will often use industry standard communications mechanisms and the Internet to move data to the user. An urban traffic monitoring system using video monitors as the sensors is an example of such a system.

Figure 2. Sensor (Actuator) Node Major Components

461

Sensor (or actuator) nodes (Figure 2) have a lesser degree of sophistication than most other computer systems. Some sensor network nodes approach the size of a U.S. quarter [31. The CPU-memory-communication interfaces are considerably more primitive than their workstation-based cousins. Communications is often handled using low frequency, low bit rate radios to conserve power, rendering them unusable in traditional 802.11 x wireless LAN systems. Space, power constraints, and limited onchip configurability requires support of concurrencyintensive management of embedded microprocessors.

the manufacturer or other interested user (or mischievous eavesdropper).


On the other hand, there are applications in this category that definitely qualify as sensor networks. Consider the trucking industry. For some time now major trucking companies and component manufacturers (engine and power train in particular) have used the sensor network concept to track the movement of the fleet. Moreover, engines can be re-tuned on the fly as they operate at different altitudes. This kind of monitoring and control adhere nicely to the sensor network concepts of continually changing communication topology, as the trucks are frequently moving, and global identification. As each truck may have a unique identifier or "address", the address cannot be used to establish location, like a zip or area code can.

Sensor networks provide concurrent operation in that there is communication (data movement) along with some amount of on-the-fly processing. In general, sensor networks do not operate on a command, sense/process, then respond/communicate sequence. Some applications, particularly those that are severely constrained may actually benefit from this approach.

Scalability and fault tolerance are key design factors in sensor networks. In [4], a graph-theoretic optimization to determine the minimal sensor set (MSS) and pseudominimal sensor set (PMSS) is developed. These sets are defined as the minimum set of sensors that guarantee the system's desired observability property. In other words, it is the minimum set that still provides the desired function in the presence of failures. In this context the reliability (probability of no failures) is modeled using a Poisson distribution

JR, (t)= e-i(1)


where X, is the sensor node failure arrival rate. The idea here is that it is assumed that some set of sensors is observing essentially the same phenomenon at all times. As the number of sensor nodes scales into the hundreds, thousands, or even beyond, the reliability, cost, and data coordination considerations can become overwhelming. This problem generally occurs with very large systems monitoring very large geographic areas but can also be found wherever a high density of sensors is required. Quality control monitoring of very complex assemblies can require literally thousands of sensors.

Agriculture is an application space that has much in common with manufacturing operations. Recent research in soil characterization sensor technology has made interesting progress. A primary motivation for soil characterization is pH balancing, where pH is the measure of acidity. Traditional methods of soil mapping include taking random samples, approximately one per hectare (about 2.5 acres). These samples are then taken to a laboratory for analysis. The results are then used to determine how much "lime" to apply (for the purpose of optimizing pH, and hence field production yield), and where to apply it. The range of pH is from 0 to 14, with 7.0 being "neutral".
It has been documented that soil pH can vary from below 5.0 to nearly 8.0 in a single field in the Midwestern United States [7]. More significant variability has been noted in Australia, where pH can vary by as much as 100% in a square meter [1]. Sensor networks could contribute to better-optimized and consistent soil chemistry by precise application of chemicals, hence better utilization and yield from the land with less waste.

B. Applications in Manufacturing A primary motivation for using distributed sensing technologies (sensor networks) in manufacturing is to monitor and optimize quality control. Like the agricultural system just described product quality and yield are critically important. Electrical systems and their subassemblies are increasingly complex in shape, function and required performance. This inevitably increases the probability of system failure. The absence of sufficient measurement and control capability complicates the problem of ensuring reliability.

Remote sensing is not the same as a sensor network. Remote sensing of operational parameters of a product already sold and deployed does not really constitute a sensor network, yet the term is often used. The normal operation of the device does not include the sensor, at least not in real-time. In these cases a product or device logs data about its operation, waiting to be queried from

462

Diagnosis in assembly systems requires fault localization to elements of the assembly mechanism itself. This generally occurs based on a sensor measurement of a partially or fully assembled part. In large parts such as the sheet metal exoskeleton of an automobile, there are 150250 different fixtures required to constrain 800 distinct oriented surfaces that comprise the assembly [2][5]. Therefore it is desirable to perform fault diagnosis in assembly as close to the source as possible. This may require positioning of sensors in dimensionally demanding or environmentally hazardous locations. Interconnecting these sensors may be difficult if not impossible, making sensor network technology a potential alternative.
The use of sensor networks in electrical manufacturing suggests yet another unique set of sensor network application scenarios. Monitoring dimensions of an assembly or controlling the coating thickness on a winding are just a few examples. While there may be similarities in terms of "what" sensors are used hence the data collected, there is no "one size fits all" set of tools that can account for characteristics unique to a given manufacturing problem. This reinforces the notion that sensor networks are significantly application dependent.
When monitoring and controlling a manufacturing process it is important to "sample" the parameter of interest often enough to ensure reconstruction of the original signal. Perhaps more importantly it is critical that the acquired data reach its destination sufficiently fast to make it possible for a decision to be made and corrective actions to be taken if necessary. A decision may be of the form of pass/fail and therefore may be made locally (at the point of assembly). Alternatively, the raw data may be forwarded on to another (possibly remote) location for analysis and future corrective action (perhaps in another area of production, or even inventory control).

communications at higher (GHz) frequencies. Advanced modulation and spectrum management techniques include Orthogonal Frequency Division Multiplexing (OFDM) and spread-spectrum techniques such as Code Division Multiple Access (CDMA), now being used more frequently in cell phones. These techniques tend to enhance the medium's robustness to interference and offer a certain amount of security while the bits are in flight.
Internet communications allow for globally remote access and control of a sensor network. Data can be collected, then forwarded by a "gateway", a device that can speak the wireless language of the sensors as well as the wireless (or wired) language used by standardized Local Area Network (LAN) and Internet protocols.

Advanced routing algorithms in wireless sensor networks facilitate data movement from the sensor end points through the network to the gateway or head end. These routing algorithms consider combinations of parameters to determine the shortest path to the destination. Parameters include energy consumption, radio link strength and quality, failed nodes, and the ability of a node (it might be another sensor node) to forward the data. Recall that each sensor node might be called upon to serve a dual purpose yet each node has finite resources (power, memory, etc.).
B. Challenges

III. ENABLERS AND CHALLENGES


A. Enabling Technologies

As enabling technologies become available there is a desire to take advantage of them. Smaller silicon devices tend to allow higher clock frequencies, so more processing work can be performed in less time. Smaller RF components imply higher carrier frequencies. Faster wireless transmission can reduce the time to move data through the network. As enticing as this sounds, there are ramifications to these enablers that must be considered when designing, deploying, and maintaining a sensor network.
Power consumption is the primary design consideration, and it enters into the design of all other aspects of the sensor network. In [9] for example, A/D conversions are estimated to consume 1nJ/sample and processing has a cost of lpJ/instruction. Considering battery and capacitor storage of IJ/mm3 and 10mJ/mm3 respectively, energy consumption must be minimized in order to ensure reasonable sensor node life. It is obviously impractical to change batteries on thousands of sensors every few days or weeks. Future trends in energy harvesting may go a long way to addressing this issue.
Advanced wireless communications techniques come at the price of complexity and energy consumption. Complexity tends to add physical hardware (and in some

Sensor network components (sensors, microprocessors, RF transceivers, etc.) are taking advantage of Micro Electro-Mechanical Systems (MEMS) technology. Studies have demonstrated devices that can monitor and communicate whose size is on the order of 1OOmm3 [6][9]. It turns out that the largest contributor to the size of the node is the power supply, which in this case is a

hearing-aid battery.

Wireless communications technology has steadily progressed over the last decade. New materials and manufacturing techniques allow for cost effective

463

cases software) components. While hardware adds to size software consumes valuable memory and processing resources.

Simplifying RF communications increases the probability of degraded system performance due to RF interference. Even sophisticated RF communication techniques can suffer from this. The result is re-transmission or data loss, both of which have negative impacts on the power budget. Even in applications where power is unlimited the effects of RF interference can result in lost (or at least untimely) data.
Data integrity and security in wireless sensor networks is an important consideration. This is especially true in systems where something (like a piece of machinery) is controlled as well as monitored. Providing more secure over a shared wireless medium means more processing overhead to encrypt and decrypt the data. Small microprocessors may not be capable of performing these operations in reasonable time. Reference [8] provides a good overview of security in wireless sensor networks. Many of the concepts are equally appropriate in wired sensor networks as well. Sensor network security tends to follow security practices in general computer networks. Data is encrypted, possibly more than once, before it is transmitted on the medium. On the receiver side, the data must be decrypted before it can be used. The function to do this generally requires both sides to retain a private "key", known only by the sender, receiver, and possibly a trusted third party. One of the significant challenges being faced by developers of large sensor networks is how to manage and distribute large numbers of keys where the send/receive pair frequently changes.

Node capture can have a similar "signature" to data manipulation. The intruder essentially captures the node and forces it to act as he desires by perhaps installing new software on the node. The system must then be capable of detecting this erroneous behavior and taking corrective action. This could be by ignoring the data or taking the node out of service, if either of these choices is available.

Finally, denial of service has the same effect as data loss. An intruder can "flood" the network with so much traffic that a node or set of nodes cannot gain access to the communications medium. This has the same effect as generating sufficient RF interference or unplugging a cable or cables on a wired network. Wired networks unfortunately can also suffer from this type of attack.
Two more areas requiring sophisticated communication protocols are admission control and authentication. Admission control addresses the question of who is allowed to use this network? Authentication attempts to ensure that those using the network are indeed who they say they are. To accommodate these issues at the sensor node requires still more overhead, reducing the amount of useful work the sensor node is able to accomplish.

Security protocols generally allow for the periodic changing of keys. This has the effect of making it even more difficult for an intruder to decrypt data. In manufacturing settings where send/receive node pairs do not often change the problem of getting the keys to all the nodes and periodically changing them still exists.
The most significant threats to sensor network security in a manufacturing environment are message (data) manipulation, node capture, and denial of service. Message or data manipulation has the effect of providing false information to the system. Sensor data provides the decision-making machine with misleading information, potentially causing an ill-formed action. False actuator data controlling a critical aspect of a machine can have harmful ramifications.

Finally, consider a manufacturing scenario where a fairly large part requires 100 nodes to provide sufficient observability of the process. If a part is made every three seconds, and each sensor takes three samples per part, then there is the potential for nearly 10 million pieces of data every 24 hours. The required bit rate of the link is dependent on whether the sensor node performs additional processing (like filtering) and how quickly the entire sample set from the sensors must reach the head-end. One consideration not mentioned here is that of redundancy. The requirement of 100 nodes could easily become an implementation of 200 nodes or more, directly affecting the volume of data being processed.

Multiple production lines directly influence the amount of data being processed and moved. Design considerations here include isolation of network bandwidth, aggregation points for data collection, admission and authentication controls, and centralizing the overall system, which could be useful in comparing and contrasting production lines.
IV. SUMMARY AND CONCLUSIONS

This paper has offered an overview of the present state-ofthe-art in (wireless) sensor network technology. Sensor networks promise to revolutionize how we monitor and control environments. This comes as a result of reduced

464

device size and power consumption combined with improved wireless communications capabilities. Wireless sensor networks also allow remote monitoring and control by taking advantage of technologies already found with Internet communications.
The use of sensor networks in electrical manufacturing environments is motivated by increasing demands on monitoring and controlling critical assembly parameters. This needs to be done in spatially intolerant or hostile environments. While many manufacturing operations can benefit from the application of sensor networks they cannot be designated as a "silver bullet" just yet. Sensor network deployment is significantly application dependent, evading a "one size fits all" or "plug and play" solution. It is one thing to gather large amounts of data while it is quite another to determine how the data is useful.
It was previously noted that the small size, low energy consumption, and wireless communication is a good combination for cost effective, yet potentially disposable sensor nodes. The idea here is when a node or nodes fail, simply throw them away and replace them.

4. G. Hoblos, M. Staroswiecki, A. Aitouche, "Optimal Design of Fault Tolerant Sensor Networks", Control Applications, 2000, Proceedings of the 2000 EEE International Conference on 25-27 Sept. 2000 Page(s) 467 - 472

5. A. Khan, D. Ceglarek, J. Ni, 1998 "Sensor Location Optimization for Fault Diagnosis in Multi-Fixture Assembly Systems," Trans. of ASME, Journal of Manufacturing Science and Engineering, Vol. 120, No. 4, pp. 781-792, also in 1996 ASME International Mechanical Engineering Congress and Exposition, MED-4, pp. 473-484, Atlanta, GA.

6. J. M. Kahn, R. H. Katz, and K. S. J. Pister, "Next Century Challenges: Mobile Networking for "Smart Dust". In International Conference on Mobile Computing and Networking (MOBICOM), pages271-278, 1999.
7. J. Lowenberg-Deboer and A. Hallman, "Value of pH Soil Sensor information. In Proceedings of the 5th International Conference on Precision Agriculture, July 2000.
8. A. Perrig, J. Stankovic, and D. Wagner, "Security in Wireless Sensor Networks", Communications of the ACM, Vol. 47, No. 6, pp. 53-57, June 2004

The wireless communications medium is a wide-open, shared medium. Techniques continue to improve robustness and security, making monitoring or interference less of a problem, whether it is caused accidentally or by malicious intent. REFERENCES
1. V. I. Adamchuk, M. T. Morgan, and J. LowenbergDeboer, A model of Agro-Economic Analysis of Soil pH Mapping. Precision Agriculture, Vol. 5, pp. 111-

9. B. Warneke, B. Atwood, and K. Pister, "Smart Dust Mote Forerunners", In Proceedings of the 14t EEE International Conference on Micro Electro Mechanical Systems (MEMS), January, 2001.

129, 2004.

2. Y. Ding, P. Kim, D. Ceglarek, J. Jin, 2003, "Optimal Sensor Distribution for Variation Diagnosis in Multistation Assembly Processes," IEEE Transactions on Robotics and Automation, Vol. 19(4), pp. 543-556.

3. J. Hill, R. Szewczyk, A. Woo, S. Hollar, D. Culler, and K. Pister, "System Architecture Directions for Networked Sensors", In Proceedings of the Ninth International Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS2000), pages 93-104, November 2000.

Jeffrey J. Evans holds a BS in Electrical Engineering Technology from Purdue University, an MS in Computer Science and is a Ph.D. candidate in Computer Science at the Illinois Institute of Technology. He worked with embedded systems in industry for twenty years prior to joining the faculty at Purdue. He is an Assistant Professor in the Electrical and Computer Engineering Department where he teaches computer architecture and networking courses. His research interests include network performance of large-scale systems, such as computer clusters and sensor networks.

465

Anda mungkin juga menyukai