Anda di halaman 1dari 6

Fuzzy Application Library/Technical

Applications/AC Induction Motor


Fuzzy Logic Enhanced Control of an AC Induction Motor with a DSP

S. Beierke; Texas Instruments, Germany, sbei@msg.ti.com -- C. von Altrock; INFORM


GmbH, Aachen, Germany, cva@inform-ac.com -- R. Hemmerlein, Inform Software Corp.,
Chicago, U.S.A., rod@informusa.com

Citation Reference: This paper was published at the Fifth IEEE International Conference
on Fuzzy Systems, held in New Orleans September 1996. The Fuzzy Logic Application
Note series is published by Inform Software Corporation on its Internet server to promote
the use of fuzzy logic technologies in applications.

Fuzzy logic is a new and innovative technology being used to enhance control engineering
solutions. It allows complex system design directly from engineering experience and
experimental results, thus quickly rendering efficient solutions. In a joint application project,
Texas Instruments and Inform Software have used fuzzy logic to improve AC induction
motor control. The results were intriguing: control performance has been improved while
design effort has been significantly reduced.

1. Introduction

Market analysis shows that 90% of all industrial motor applications use AC induction type
motors. The reasons for this are high robustness, reliability, low cost, and high efficiency.
The drawback of using an AC induction type motor is its difficult controllability, which is due
to a strong nonlinear behavior stemming from magnetic saturation effects and a strong
temperature dependency of electrical motor parameters. For example, the rotor time
constant of an induction motor can change up to 70% over the temperature range of the
motor. These factors make mathematical modeling of motor control systems difficult. In
real applications, only simplified models are used. The commonly used control methods
are:

• voltage/frequency control (U/f)


• stator current flux control (Is/f2)
• field oriented control

Of these approaches, the field-oriented control method has become the de-facto standard
for speed and position control of AC induction motors. It delivers the best dynamic behavior
and a high robustness under sudden momentum changes. Alas, the optimization and
parameterization of a field oriented controller is laborious and must be performed
specifically for each motor. Also, due to the strong dependency of the motor's parameters,
a controller optimized for one temperature may not perform well if the temperature
changes.

To avoid the undesirable characteristics of the field oriented control approach, the
companies Texas Instruments and Inform Software have developed new alternative control
methods, and compared them with the field oriented control approach. The alternative
methods involved two types of flux controllers enhanced by fuzzy logic and NeuroFuzzy
techniques, respectively. The goal was to use fuzzy logic to improve the dynamic behavior
of the flux control approach such that the robust behavior of the flux controller and the
desirable dynamic properties of the field oriented controller are achieved simultaneously.
For a practical introduction to fuzzy logic and NeuroFuzzy technologies, see [1].

Figure 1: Demonstration of the Test


Motor at the Embedded Systems
Conference, San Jose 1995 (large)

2. Field Oriented Control Method

Figure 2 shows the principle of field oriented control. It allows for control of the AC
induction motor in the same way a separately exited DC motor is controlled. The flux model
computes the "phase shift" between rotor flux field and stator field from the stator currents
iu and iv, and the rotor angle position n. The field oriented variables of the two independent
controller units are subsequently computed by the transformation of the stator currents
using this "phase shift".

The actual control model consists of two components of cascaded standard PI controllers.
The upper component comprises outer magnetizing current (imR) controller and inner isd
current controller. The lower component comprises a speed controller and momentum
controller. The input of the speed controller is computed as the difference between set
speed nref and filtered measured speed n.

To optimize the field oriented control model, all controllers must be parameterized and
optimized individually. In this application project, the method of optimized amplitude
adaptation was used to tune the current controller, and the method of the symmetrical
optimum was used for the velocity controller. Implementation effort for the field oriented
controller was 3 person months, including parameterization and design of the flux model.
The computation time for the inner current controllers, the flux model, and the coordinate
transformation is 100µs on a TMS320C31-40MHz digital signal processor. The resulting
performance is shown in Figure 6. When switching the set speed from -1000 rmp to +1000
rpm, the new set speed is reached within only 0.25 seconds without any overshoot.
However, this excellent performance is not always available. When the motor heats up the
control performance drops significantly, and a motor with slightly different characteristics
will achieve only mediocre results utilizing the same controller.
Figure 2: Field-Oriented Control of AC
Induction Motors (large)

3. Fuzzy Flux Control Method

The conventional flux control model has been enhanced by fuzzy logic in two steps. In the
first step, the non-linear relation between slip frequency and stator current was described
by a fuzzy logic system (Fuzzy Block #1). Figure 3 shows the principle of the resulting
fuzzy flux controller. The control model consists of three inner control loops and one outer
control loop. The inner control loops control the three stator phase currents using standard
PI controllers. The outer control loop determines the slip frequency n2, also using a
standard PI controller. The slip frequency is the input to Fuzzy Block #1, which outputs the
set value of the stator current. The primary objective for Fuzzy Block #1 is to keep the
magnetizing current constant in all operating modes. The magnetizing current is a non-
linear function of the slip frequency, the rotor time constant, the rotor leakage factor, and a
non-constant offset current.

Figure 3: Principle of a Fuzzy Flux


Controller (large)

The stator frequency n1 is the sum of the measured rotor frequency n and the slip
frequency n2. The reference position is determined by integration of the stator frequency n1.
Modulated by sin/cos, the reference position is multiplied with the set value of the stator
current, and split back into a 3 phase system of the stator current set values.

The rules of the fuzzy block were not manually designed, but rather generated from
existing sample data by the NeuroFuzzy add-on module of the fuzzyTECH design
software. NeuroFuzzy utilizes neural network techniques to automatically generate rule
bases and membership functions from sample data. The benefit of the NeuroFuzzy
approach over the neural net approach is that the result of NeuroFuzzy training is a
transparent fuzzy logic system that can be explicitly optimized and verified. In contrast, the
result of a neural net training is a rather non-transparent black box [1].

4. Comparison with Field Oriented Control

Figure 6 shows the performance of the fuzzy flux controller in comparison with the field
oriented controller. The overshoot performance is almost as good as that provided by the
field oriented control, however, it takes the fuzzy flux controller almost twice as long to
reach the new set speed (curve Fuzzy_1). On the other hand, parameterization and
optimization of the fuzzy flux controller only required 4 person days. The computation time
for the entire controller is 150µs on the TMS320C31-40MHz digital signal processor.
To improve the performance of the fuzzy flux controller, in a second step, the standard PI
controller for the outer control loop was replaced by a fuzzy PI controller (Fuzzy Block #2 in
Figure 4). This fuzzy PI controller does not use the proportional (P) and integral (I)
component of the error signal, but rather the differential (D) and proportional (P)
component then integrates the output. This type of fuzzy PI controller has been used very
successfully in a number of recent applications, especially in the area of speed and
temperature control [1]. In contrast to the standard PI controller, the fuzzy PI controller
implements a highly non-linear transfer characteristic. The sub-window in the lower left part
of Figure 5 shows the transfer characteristics for the fuzzy PI controller implemented in this
application.

Figure 4: Enhanced Fuzzy Flux


Controller (large)

The enhanced fuzzy flux controller reveals a much improved dynamic performance. Figure
6 shows the performance of the enhanced fuzzy flux controller (curve Fuzzy_2) in
comparison to the fuzzy flux controller with only one fuzzy block. The enhanced fuzzy flux
controller reaches the new set speed as fast as the field oriented controller and shows no
overshoot at all. Initial analysis has also shown that the enhanced fuzzy flux controller is
significantly more robust with regards to variances in motor parameters than the field
oriented controller.

The good performance attained in this case hinges on the non-linear behavior of the fuzzy
PI controller. In contrast to the conventional linear PI controller, the non-linearity of the
fuzzy PI controller produces stronger control action for a large speed error, and a smoother
control action for a small speed error. This also results a higher robustness of the
enhanced fuzzy flux controller against parameter changes. The implementation of the
second fuzzy block with the fuzzy flux controller only required an additional day for the
fuzzy logic system itself, and two additional days for the optimization of the total system.
Hence, the total development effort for the enhanced fuzzy flux controller was 7 person
days in comparison to 3 person month for the field oriented controller. The computation
time for the entire controller is 200µs on the used TMS320C31-40MHz digital signal
processor.

5. System Simulation Using Matlab/Simulink and fuzzyTECH

The initial design of the system was implemented in a software simulation. The
fuzzyTECH® fuzzy-system development software was used together with the
Matlab™/Simulink™ control-system simulation software. fuzzyTECH allows using fuzzy
blocks in Simulink's control diagrams [4]. This tool combination allows for the design of
simulations combining conventional and fuzzy logic control engineering technologies in the
same software environment. Figure 5 shows the development of the fuzzy blocks with
fuzzyTECH/Simulink. The differential equation model used for the simulation of the AC
induction motor is discussed in [3].
Figure 5: Simulation of the Enhanced
Fuzzy Flux Controller Using the
Software Products fuzzyTECH and
Matlab/Simulink (large)

6. Fuzzy Logic on Digital Signal Processors

Because of the increasing number of successful of applications of fuzzy logic in both


control engineering and signal processing, DSP market leader Texas Instruments was
looking for a software partner to implement fuzzy logic on DSP. In 1992, a formal
partnership was formed with Inform Software Corp., a company specializing in fuzzy logic.
One product of the partnership was the design of dedicated versions of fuzzyTECH that
allow the implementation of fuzzy logic systems on standard TI-DSPs. The primary
objective was to reach an acceptable computing performance level for fuzzy logic on
DSPs, a quality previously unknown to software implementations of fuzzy logic. Using the
fuzzyTECH assembly kernel for 16 bit resolution, 2.98 million fuzzy rules per second can
be computed on the TMS230C52 (25ns instruction cycle DSP), including fuzzification and
defuzzification. For comparison: the most recent dedicated fuzzy processor of VLSI
(VY86C500/20) only computes 0.87 million fuzzy rules per second (not including
fuzzification and defuzzification) with just 12 bit resolution (VLSI data sheet). While the
referenced DSP only costs a few dollars in large quantities, the fuzzy processor is quoted
at $75 each. This comparison shows that in most applications, the use of dedicated fuzzy
processors is not necessary.

7. Result Discussion

The application project discussed in this paper shows that even in areas where traditional
control engineering already offers comprehensive solutions, fuzzy logic can deliver
substantial benefits. The authors were astounded by the dramatically shorter design time
for the fuzzy approach which delivered similar performance and higher robustness than the
traditional approach. While the traditional approach required three (3) people months by a
known expert in the area, the enhanced fuzzy flux controller only took seven (7) people
days to implement. The fuzzyTECH assembly kernel for DSPs developed by Texas
Instruments and Inform Software Corp. allows for the integration of fuzzy logic systems
together with conventional algorithms on the same chip, even when control loop times of a
fraction of a millisecond are required. Texas Instruments and Inform Software Corp. now
work on further enhancements of the fuzzy flux controller. The companies are currently
striving for even better dynamic performance by adding a fuzzy air-gap flux observer to the
system.
Figure 6: Set Speed Response
Comparison of the two Fuzzy Flux
Control Approaches with the Field
Oriented Control Approach (large)

8. Literature

[1] v. Altrock, C.: "Fuzzy Logic and NeuroFuzzy Applications Explained"; Prentice Hall;
ISBN 0-13-368465-2; 1995.
[2] S. Beierke, R. Konigbauer, B. Krause, and C. v. Altrock: "Fuzzy Logic Enhanced Control
of AC Motor Using DSP"; Embedded Systems Conference California; 1995.
[3] D. Naunin, S. Beierke, and P. Heidrich: "Transputer Control of Asynchronous Servo
Drives"; EPE Florence; 1991.
[4] "fuzzyTECH User's Manual"; INFORM Software Corp., 2001 Midwest Rd., Oak Brook,
IL60521, U.S.A.; 1995.
[5] "TMS320C3X User's Guide; Texas Instruments; 1994.

Anda mungkin juga menyukai