Anfis Based Soft-starting And Speed Control Of Ac Voltage Controller Fed Induction Motor

ANFIS Based Soft-Starting and Speed Control of AC Voltage Controller Fed Induction Motor Mohammad Ayyub, Member, IEEE

Abstract—An intelligent ac voltage controller is proposed for the control of induction motor. It controls the motor speed by adjusting the firing angles of the thyristors. Adaptive Network fuzzy Inference System (ANFIS) based controller was designed for open-loop sensor less control. The results obtained were satisfactory and promising. In addition to simplicity, stability, and high accuracy such controller gives soft starting. It is suitable to control induction motor as a soft starter and speed adjustment in compressors, blowers, fans, pumps, and many other applications. Index Terms—AC voltage controller, adaptive neuro fuzzy inference system (ANFIS), firing angle, soft starter, thyristors.

T

I. INTRODUCTION

application of AC voltage controllers are increasingly used for control of small induction motor fed fans, blowers, compressors, centrifugal pumps etc. [1-2]. Although the performance of above drives are satisfactory but their starting is accomplished with high current and losses. The electrical and mechanical stresses on the motor are more. AC voltage controllers are used in these applications where soft starting is also achieved. The static soft starters has following advantages over conventional starters: • Smooth acceleration, which reduces mechanical stress on the drive system. • Reduced stress on electrical supply due to reduced starting current, hence elimination of voltage dip and brown out conditions [3]. • Energy saving at light load conditions. A reduced voltage can be applied if the load torque requirement is met at reduced flux. Copper and iron losses will also be reduced [4]. HE

Soft starting can be achieved by simple change in the control logic of the static AC voltage controller. Thus single controller can be used for speed control, starting and stopping with minimum electrical and mechanical stresses on the drive system. Appropriate delay angle (α) has to be applied, which gives appropriate reduced voltage corresponding to the required operating condition of speed at a given torque. M. Ayyub is with Electrical Engineering Department, Z. H. C. O. Engineering & Technology, A. M. U. Aligarh, U. P., India (e-mail [email protected]).

0-7803-9525-5/06/$20.00 ©2006 IEEE.

However, adjusting the voltage to achieve an operating condition for given speed and torque is a difficult job. Also, it is not easy to find the exact value of α [5]. Closed loop control method of voltage regulators that require speed sensor is reported in [2]. Sastry et al [6] have proposed an optimal soft starting method without speed sensor, but it requires sensing of voltages across the thyristors. Gastli et al [5] have discussed an ANN-based soft starting of voltage-controller fed induction motor drive. Adaptive neuro fuzzy inference system (ANFIS) is an intelligent technique and is proved to be efficient in problems like classification, modeling, and control of complex systems [7[-[9]. ANFIS has advantages over conventional fuzzy systems as it provides standard method of interpreting system information into specific fine-tuned rules and membership functions through training process [9][11]. Therefore, this paper aims at applying ANFIS for selecting the thyristors’ firing angles of a voltage-controller feeding the induction motor. The proposed controller operates in open loop without any requirement of speed or voltage sensing. High accuracy was achieved with the proposed controller and is simple. II. AC VOLTAGE CONTROLLER The configuration of the drive is shown in Fig. 1, in which a three-phase voltage controller fed induction motor is shown. The firing sequence for the six thyristors was the standard symmetrical firing [2] as shown in Fig 2. It should be noted that to have a current flowing into the motor at least two thyristors must conduct. The reference for delay angle (α) is the zero crossing of the supply voltage for the respective phase. The three-phase induction motor was modeled with orthogonal axes representation fixed with stator and the devices were modeled according to their operating characteristics. The simulation model was developed in Matlab-Simulink environment. This model was found to give satisfactory performances. The data for training the adaptive neuro fuzzy inference system (ANFIS) was obtained by having several runs of the simulation and getting the firing angle, speed, and torque pair from each run. Complete system to obtain training data with constant load torque is shown in Fig. 3. Fig. 4 shows the speed-torque characteristics obtained from the simulation runs that give steady-state speed for constant load torque with different values of delay angle (α).

1

A

2.5

α = (0 - 30 0)

4 B

Torque (p.u.)

Induction Motor

6 5

C

40

2

3

50 1.5

60 70

1

2

Fig.1. AC voltage-controller fed Induction Motor.

80 0.5

Q1

0

90 100 0.4

0.2

0

0.8

0.6

1

Speed (p.u.)

α

Fig. 4. Speed-torque characteristics for different α.

Q2

III. ADAPTIVE NEURO FUZZY INFERENCE SYSTEM

Q3

The adaptive neuro fuzzy inference system (ANFIS) is a fuzzy system and used in classification, modeling and control problems [9]. It is based on Takagi and Sugeno’s fuzzy if-then rules representation [7]-[9][11], which is different from commonly used fuzzy logic controllers [11][12]. The consequent part of the rule is a function of input variables. The system considered in this paper has two inputs, viz. the load torque (T) and the desired motor speed (ωm), and the firing angle (α) is the only output. The inference mechanism of ANFIS is mathematically expressed by the set of rules. These rules are generated through the experience of operating the system, which may be feedback from the plant operator, design engineer, or the expert. The kth rule is generally expressed in the form (If premise THEN consequence) and is given by:

Q4 π+α Q5

Q6 π/3

2π/3

π

4π/3

5π/3

2π

Fig. 2. Firing sequences of thyristors.

Ia

rm s

s ignal

alpha a

g

g

g a ti n g

+ -

v

a

Tm

A

A

b B

B

C

C

3 -p h so u rce

0 .5 m

A

a

B

b

C

c

IM p u

c

wm

A C Re g u l a to r

m

Fig. 3. Matlab/Simulink model to generate training data.

It is to be noted that for α ≤ 30 the characteristics are similar. This may be because of the firing angle a being smaller than the load impedance angle ( α ≤ φ ) and continuous conduction of current takes place. Thus changing α below 300 will not change the applied voltage across the motor. The system parameter considered for the simulation studies are [5][10]: 1/3 hp, 4 pole, 50 Hz 220 V, squirrel-cage three-phase induction motor. The motor parameters in per unit with rated voltage taken as base voltage and base power taken as 375 watts are as given below: 0

Rs = 0.0566 p.u. Rr = 0.1252 p.u. Ls = Lr = 1.0318 p.u. M = 0.969 p.u. Moment of inertia (J) = 1.5 p.u.

k

k

If (T is A i ) and (ωm is B j ) Then (α is fk) k

where, A i and B

k j

are the ith and jth fuzzy term sets of

representing T and ωm respectively. The consequent function fk is represented as f k = pk T + qk ωm + rk . The parameters pk, qk, and rk are adjustable. These parameters were tuned in training phase. The proposed work considered seven fuzzy subsets for each variable and resulting 49 rules cover the whole input space. Bell-shaped membership function was considered for each fuzzy subset each having three parameters. Each fuzzy subset (say Ai) is defined by a membership function µ A i (x ) as in (1).

µ A ( x ; a i , bi , c i ) = i

1 ⎛ x − ci 1 + ⎜⎜ ⎝ ai

⎞ ⎟⎟ ⎠

2 bi

(1)

where (ai, bi, ci) is the parameter of antecedent fuzzy set Ai of the ith membership function. These parameters control the shape of the bell-shaped membership function. The ranges of these parameters are chosen to accommodate most of the standard shapes used in fuzzy system.

The architecture of the ANFIS considered in this paper is shown in Fig. 5, which has five layers. Per unit values are considered for input. Layer 1 is the fuzzification layer where the values of membership for each input are calculated. Layer 2 is the rule inference layer where the firing strength of each rule (wk) is calculated using product inference [9]. These firing strengths are normalized in layer 3. The consequent functions (fk) are calculated in layer 4 and layer 5 calculates the weighted sum consequent functions and output from this layer is a, which can be expressed as: N

α = ∑ϖ k f k

(2)

region. More rules are required to classify large training data present in this region.

Premise parameters Consequent parameters Signal

TABLE I ANFIS hybrid learning Forward pass

Backward pass

Fixed

Gradient descent

LS-estimator

Fixed

Node output

Error signal

1

k =1

where α is the firing angle, N is the number of rules, and ϖ k is the normalized firing strength of kth rule. An overview of the ANFIS considered in this paper is given in Fig. 5. It can be observed from the architecture that for fixed values of premise parameters the overall output is a linear combination of consequent parameters. There will be 147 consequent parameters (49 rules x 3 parameters = 147 parameters). Thus forward path will calculate till layer 4 with fixed premise parameters and the consequent parameters can then be identified by least-square method. In backward path the error signal is considered and premise parameters (2 inputs x 7 fuzzy term sets x 3 parameters = 42 parameters) are updated by gradient decent method. This hybrid-learning algorithm is summarized in Table I, which tunes 189 parameters in all (147 linear parameters + 42 nonlinear parameters = 189 parameters). T

N N

...

N

. . .

... . . .

N N

∑ N

...

N

Normalization layer

Rule inference layer

Fuzification layer

0 0

0.1

0.2

0.3

0.5 0.4 (a)Speed ( )

0.6

0.7

0.8

0.9

1

1 0.8 0.6 0.4 0.2 0 1.5 1 (b)Torque ( after ) training. Fig. 6. Final membership functions 0

0.5

2

2.5

IV. SIMULATION RESULTS The ANFIS model and motor model were implemented with Simulink Power System Blockset and Simulink Toolbox as in Fig. 7. The load considered was a pump or a fan load having following torque-speed characteristics:

TL = kω m2

Nm

(3)

where k = 1.8 (N m s2/rad2) is a constant The input was the reference speed ( ω ref ) and simulation

N

N

Input layer

α

ωm

Output layer

ωm

. . . . . .

Consequent layer

T

0.5

Fig. 5. Structure of ANFIS ac voltage controller.

The initialization for ANFIS training was done by grid partition [9]. The training data obtained from off line simulation of the motor with different torques and firing angles and the corresponding steady state speed was recorded for each set. The training was done for 150 epochs, after which the sum of squared error was reached to 0.006, which is within acceptable limit. The final membership functions after the training is shown in Fig. 6. It can be noted that the membership functions crowded near the rated value. This may be due to overlapping and bunching of the curves in this

studies for motor starting, braking and speed control were carried. The reference torque (Tref) was generated from the reference speed and the load model (2). These two references ( ω ref and Tref) were the input to the ANFIS controller. The output from the controller is the firing delay angle α. The triggering logic and gate pulse generator block (it is indicated as Pulses block in the diagram) generated the required gating patterns for the thyristors as in Fig. 7. The simulation speed was done for speed control with reference speed was changed from 0 to 0.45 p.u. At 6 seconds it was changed to 0.7 p.u. and at 8.5 seconds it was set to 0.5 p.u. It is obvious from the result depicted in Fig. 8 that the actual speed effectively tracks the reference speed and there is no oscillation in the response. The ac voltage controller described above can be used for soft starting, in addition to speed control. Soft starting can be accomplished by manipulating the speed reference signal in such a way that the stress on the motor is minimized. A ramp

Ia

w a

T

T

w

w_ref

load model

ANFIS controller a

g

C

3-ph source

wm

v

m

Te

a

k*u^2

A b

B

signal + -

g

Pulse A

rms

Tm

i -

+

m

A

a

B

b

B

C

C

c

c

IM

AC voltage regulator

Fig. 7. Simulation model with ANFIS controller.

0 .8 0 .6 R e f. S p e e d

A c tu a l S p e e d

0 .4 0 .2 0

(a)

Fig. 8. Response for speed change.

signal with appropriate slope was used to start the motor from zero to the final speed. Fig. 9 shows the variation of speed, current and torque with respect to time during the start-up of the motor from zero to the final speed. Studies with direct-online starting were also conducted to compare the performances. Fig. 10 shows the results for this study. It can be noticed that for direct-on-line starting the current attains more than four-and-half times the rated current before settling down to its steady-state value. The developed electric torque also reaches the value as high as four-and-half times the rated torque. We can also note that the torque and current oscillate for some time. Because of the oscillations in current the electric supply system is subjected to stress and may result in voltage dips. On the other hand, the oscillations in torque decreases the life of gearboxes, motor bearing, motor shaft, belt etc. During soft starting, Fig. 9, the magnitude of current remains around its steady-state value and it results in the developed electric torque sufficient enough to overcome the load torque. Fig. 9(a) shows the variation of actual speed of the motor. It follows the reference speed and smoothly settles to the final speed where it exactly matches with the reference. Because of smooth rise in speed the motor current and developed electric torque (Te) have smooth variations and has no oscillations. It can be noted that Te has a peak value slightly above one-and-half times the rated torque, which is much reduces a value compared to its value with direct-online starting.

5

0

T im e (s e c .)

10

15

Reference and actual speeds

5

4

3

I (rms)

2 1

0 0

5

10

Time (sec.)

15

(b) Line current (rms value)

2 1 .5 D e v e lo p e d T o r q u e

1

L o a d T o rq u e

0 .5 0

0

5

T im e ( s e c . )

(c) Developed electric torque and load torque. Fig. 9. Soft-starting with ramp reference.

10

15

voltage starting of squirrel-cage induction motors”, IEEE Trans. Ind. Appl. vol. IA-20, no. 1, pp. 46-55, Jan./Feb. 1984.

5 4 .5 4

[4]

3 .5 3

[5]

2 .5 2

T o rq u e

[6]

1 .5

Speed

1 0 .5

[7]

0 0

1

2

T im e (s e c .)

3

4

5

[8]

(a) Variation of speed and developed electric torque.

[9]

5 4 .5

[10]

4 3 .5

[11]

3 2 .5

I (rm s )

2

[12]

Tl

1 .5 1

N. Mohan,” Improvement in energy efficiency of induction motors by means of voltage control”, IEEE Power App. Syst., vol. PAS-99, pp. 146-1471, Jul./Aug. 1980. A. Gastli and M. M. Ahmed,” ANN-based soft starting of voltagecontrolled-fed IM drive system”, IEEE Trans. Energy Convers., vol. 20, no. 3, pp. 497-503, Sept. 2005. V. V. Sastry, M. R. Prasad, and T. V. Sivakumar,” Optimal soft starting of voltage-controller-fed IM drive based on voltage across thyristors”, IEEE Trans. On Power Electronics, vol. 12, no. 6, pp. 1041-1059, Nov. 1997. T. Takagi and M. Sugeno,” Fuzzy identification of systems and its applications to modeling and control”, IEEE Trans. on System, Man and Cybernetics, Vol. , no. , pp. 116-132, 1985. M. Sugeno, editor,” Industrial Applications of fuzzy control”, Elsevier Science Pub. Co., 1985. J. S. Jang, C. T. Sun, and E. Mizutani,” Neuro-Fuzzy and Soft Computing – A computational approach to learning and machine intelligence”, Prentice-Hall, Inc., Upper Saddle River, New Jersey, 1997. T. A. Lipo,” The analysis of induction motor with voltage controlled by symmetrically triggered thyristors”, IEEE Trans. on PAS, vol. 90, pp. 515-525, March 1971. N. Gulley and J. S. R. Jang,” Fuzzy Logic Toolbox – a reference guide”, The MATHWORKs Inc, 1995. M. Ayyub,” Application of genetic algorithm for optimal design of fuzzy logic controller”, in Proc. Of AICTE sponsored Natl. Conf. on Control, Communication and Information Systems, at Goa Government Engg. College, Goa during Jan. 23-24th, 2004.

0 .5 0

0

0 .5

1

1 .5

2

2 .5 3 T im e ( s e c .)

3 .5

4

4 .5

5

(b) Load torque and motor current. Fig. 10. Direct-on-line starting.

V. ACKNOWLEDGMENT The author gratefully acknowledges the suggestions from F. Azeem and A. Iqbal that resulted in improvements in this work. VI. REFERENCES [1] [2] [3]

P. C. Sen,” Power Electronics”, Tata McGraw Hill, 1987. S. B. Dewan, G. R. Slemon, and A. Straughen,” Power Semiconductor Drives”, Wiley Interscience, 1984. F. M. Bruce, R. J. Craefe, A. Lutz, and M. D. Panlener,” Reduced

VII. BIOGRAPHIES Mohammad Ayyub (M’1995) was born in Chilha in the state of Jharkhand, on January 15, 1961. He graduated from Institute of Technology, B. H. U. Varanasi, U. P. in 1984 and completed his M. Tech. in Power Apparatus and Systems from I. I. T. Delhi. He joined Electrical Engineering Department; A. M. U. Aligarh, U. P., India in 1985 and is now Associate Professor in the Department. His field of interest includes intelligent control of machines and processes, electric drives and SVC for power transmission system. Ayyub is a life member of ISTE (India), System Society of India and member of Industrial Engineering and Computational Intelligent Societies of IEEE, USA. He received best paper award and best paper presentation award at several national conferences.