FUZZY FAULT DETECTION SCHEME FOR SYNCHRONOUS GENERATOR

Academic Open Internet Journal

www.acadjournal.com

Volume 14, 2005

A FUZZY BASED FAULT DETECTION SCHEME FOR SYNCHRONOUS GENERATOR

V. Duraisamy¹, N. Devarajan², P.S.Vinoth Kumar³, S.N.Sivanandam⁴,D.Somasundareswari⁵

1 Assistant Professor,

Department of Electrical and Electronics Engineering,

Kumaraguru College of Technology, Coimbatore - 641 006,India.

Email:vduraisamy@yahoo.com

2 Assistant Professor, 3 P.G. Scholar,

Department of Electrical and Electronics Engineering,

Government College of Tech., Coimbatore-641 013,India.

4 Professor and Head,

Department of Computer Science and Engineering,

PSG College of Technology, Coimbatore -641 004,India.

5 Senior Llecturer,Kumaraguru College of Tech,Ciombartore

Abstract

A fuzzy system based inter-turn fault detection scheme for synchronous generator is proposed. The negative sequence components of voltages and currents are used as fault indicators for inter-turn fault detection. The laboratory synchronous generator is used as a test machine to obtain the training data for fuzzy fault detector. The negative sequence voltage and current are obtained from the line voltages and currents and used as inputs for fuzzy fault detector (FFD). The effectiveness of fuzzy fault detector has been analyzed through simulation studies with triangular, trapezoidal and Gaussian membership functions and results are compared. It is observed from the results that the performance of FFD with triangular membership function is better than trapezoidal and Gaussian membership functions.

Key Words: Inter turn fault, negative sequence component, fuzzy fault detector, membership functions

I. INTRODUCTION

Synchronous generators are considered to be the most important and expensive part in an electrical power system. Earlier detection of abnormalities in generators will help to minimize the maintenance expenses. The synchronous generator is exposed to variety of operating conditions. The operating conditions coupled with aging of insulation lead to weakening of insulation. This process will lead inter-turn fault in the winding. There is no experimental data to differentiate between the inter-turn faults and insulation weakness. It is probable that the transition between the two states is not instantaneous. Therefore, early detection of inter-turn faults would eliminate subsequent damage to adjacent coils and stator core, reducing repair cost and generator outage time.

Instrumentation currently exists to indicate the presence of a shorted turn, which can be performed off-line. Recently researches have concentrated on online monitoring of electrical machines [1]-[3]. Modeling of synchronous machines with shorted turns is the first step in the design of turn fault detection systems [4], [5]. Models exhibit complex relationship between parameters and exact fault signature extraction is very difficult. The utility of machine model is restricted because it is even theoretically impossible to include all no idealities of the machine. A twin signal sensing method is used for the detection of incipient faults in the windings of turbine-generator. The environmental factors such as temperature of machine and operational factors like speed and excitation current can affect the fault signature [6]. The fault detection based on the monitoring of spectrum vibration has been reported in [7]. The accurate sensing devices are required to monitor the vibration spectrum for fault signature and they are expensive

Many applications of intelligent systems techniques for fault diagnosis have been proposed in the technical literature. The authors have presented expert systems in [8], which require a set of patterns for construction of a knowledge base. However, expert systems perform satisfactorily only for specific situations that have been previously considered for the development of the knowledge base. The major drawback of expert system is the difficulty to deal with new or corrupted patterns. Alternatively, some methods based on artificial neural networks (ANN) have been proposed in [9] and [10]. The ANN-based methods overcome some drawbacks of expert systems approaches. But this method requires large amount of data to be stored for the given machine to avoid false detection.

In this paper, a fuzzy logic based technique is proposed to detect the inter turn faults. Fuzzy logic based fault detection is user friendly and requires less amount of data to detect the faults. This paper also presents a comparative study on the effectiveness of fuzzy fault detection scheme with triangular, gaussian and trapezoidal membership functions.

II. NEGATIVE SEQUENCE COMPONENT BASED FAULT DETECTION SCHEME

A. Problem definition

The stator winding turn fault introduces the unbalance in the line currents. The winding unbalance injects negative sequence current and voltage in the machine. The negative sequence voltage and current can be measured from machine and used as fault indicator. The negative sequence voltage and current is calculated from the line voltages and currents using (1) and (2).

I_sn = (I_a+ α²I_b + αI_c) (1)

V_sn = (V_ab+ α²V_bc + αV_ca) (2)

Where α = -0.5+j0.866

I_sn= Negative sequence current

V_sn = Negative sequence voltage

B. Proposed scheme

The schematic of the proposed scheme is shown in figure.1. The negative sequence component of voltage and current is used as fault signature for stator winding inter-turn faults. The sequence components of voltages and currents are measured from the machine being monitored. They are used as inputs to fuzzy fault detector (FFD). The FFD can be trained with suitable membership functions and training samples for effective fault detection. The training data for the FFD is obtained by conducting the experiment on the generator for different fault conditions.

To obtain the training samples for FFD, a 400V, 5KVA, 1500 rpm is used as a test machine. Figure 2 shows the experiment set up for fault simulation. The generator is connected to load through variable rheostats and switch across the rheostats connected separately in each phase. The unbalance in the winding can be created including the resistance keeping the switches opened. This creates similar effect as inter turn faults. The percentage of inter turn short depends on the value resistance included. The experiment is conducted for different values of resistance for constant load.

Figure 2 Experimental set-up for fault simulation

Table 1 shows the line voltages and currents of synchronous generator at constant load for different fault conditions. The sequence component of current and voltage are calculated and used as training data for FFD. Fault detection process is the mapping between the inputs and fault code, which represents the fault condition. The simulation procedure is explained in section III.

Table1. Training Data

Fault condition	Fault code	Line voltage V_ab(v)	Line voltage V_bc(v)	Line voltage V_ca(v)	V_sn (v)	Line current I_{a (A)}	Line current I_{a (A)}	Line current I_{a (A)}	I_sn _(A)
No fault	0	370	372	370	0.022	2.9	2.9	2.9	0
1 turn short (phase A)	1	375	370	370	1.92	2.0	2.7	2.85	0.260
2 turn short (phase A)	2	372	370	370	0.334	2.3	2.8	2.8	1.110
1 turn short (phase B)	3	370	375	372	1.44	2.6	1.9	2.7	0.230
2 turn short (phase B)	4	368	370	370	-0.65	2.7	2.3	2.8	1.691
1 turn short (phase C)	5	370	370	377.5	2.5	2.6	2.9	1.9	1.607
2 turn short (phase C)	6	367.5	370	372.5	1.44	2.7	2.8	2.4	0.119

III. SIMULATION OF FUZZY FAULT DETECTOR

The madmani’s model is used for simulation. The fuzzy fault detector is constructed with two inputs and one output. The V_sn and I_sn are the inputs to the fault detector and fault code is used as output. This fault detection process is the mapping from V_sn and I_sn to fault code, which represents the fault condition. The simulation is carried out using commercially available package.

The defuzzification scheme used is centroid method. The fuzzy rules are obtained from experimental values and expert knowledge. The fuzzy rules used for simulation are given in table 2. The input variables are classified into three membership functions and they are given as low, medium and high. The output variables are classified into seven membership functions, which will indicate the fault condition. The V_sn range from is from 0 to 2A, and I_sn ranges from 0 to 2V. The fault code ranges from 0 to 6. The FFD is trained with triangular, trapezoidal and Gaussian membership functions and are shown in figures (figure 3 – figure 5).

Table 2 Fuzzy rules

V_sn I_sn	L	M	H
L	0	6	NULL
M	NULL	3	1
H	4	2	5

L: Low M: Medium H: High

IV. results and discussion

The FFD is tested with triangular, Gaussian and trapezoidal membership functions. Since there is no hard criterion to qualify the best membership function for the proposed scheme, great efforts have been incorporated to fetch the suitable membership function. The simulation results are compared in table 3. The FFD is tested for V_sn=1.92 V, I_sn= 2A and testing results are shown in figures (figure 6 – figure 8). From the simulation results, it is found that the accuracy of FFD with triangular membership functions better than other popularly used membership functions.

Table 3 Membership Functions: Comparison

Expected Fault Condition Code	Triangular		Trapezoidal		Gaussian
Expected Fault Condition Code	FFD Output	Error	FFD Output	Error	FFD Output	Error
0	0.54	0.54	0.54	.54	0.68	0.68
1	0.928	7.75	0.928	7.75	1.32	24.24
2	2	0	2	0	2.32	13.78
3	2.895	3.62	2.835	5.82	2.684	11.73
4	4.21	4.98	4.22	5.21	4.43	9.7
5	5.1	1.96	5.1	1.96	5.32	6.01
6	5.98	.33	5.98	.33	5.86	2.38
Average Error	2.72		3.08		9.78

(a) Membership for I_sn(b) Membership for Vsn (c) Membership for fault code

Figure 3 Triangular membership functions

(a) Membership for I_sn(b) Membership for Vsn (c) Membership for fault code

Figure 4 Gaussian membership functions

(a) Membership for I_sn(b) Membership for Vsn (c) Membership for fault code

Figure 5 Trapezoidal membership functions

Figure 6 Testing of FFD with Triangular membership functions

Figure 7 Testing of FFD with Trapezoidal membership functions

Figure 8 Testing of FFD with Gaussian membership functions

V. conclusion

The fuzzy fault detector have been proposed to monitor the inter turn fault in the synchronous generator based on monitoring negative sequence components of line voltages and currents. The experiment is conducted on a laboratory machine to obtain the training data for the proposed FFD. The accuracy of the proposed scheme is tested with triangular, trapezoidal and Gaussian membership functions. From the results it is observed that FFD with triangular membership function is more efficient for fault detection application. This scheme does not require machine parameters for fault detection. Hence, the proposed scheme can be applied to other types of electrical machines.

ACKNOWLEDGEMENT

The authors thank Dr.K.K.Padmanabhan Principal, Kumaraguru College of Technology, Coimbatore for his guidance and support. The authors also thank Dr. S. Arumugam, Principal, Government College of Technology, and Coimbatore for providing research facilities.

REFERENCES

[1]. G.B.Kliman, W.T.Premerlati, R.A.Roegl and D.Hoeweler, “A new approach to on line fault detection in AC motors,” IEEE Industry applications society Annual meeting, San Diego, CA, pp 687-693, March, 2000.

[2]. J. Sottile and J. K. Kohler, “An on-line method to detect incipient failure of turn insulation in random-wound motors,” Proceedings of IEEE Winter Power Meeting, Columbus, OH, paper 93 WM 021-6 EC, 1993

[3]. J. Penman, H. G. Sedding, B. A. Lloid, and W. T. Fink, “Detection and location of inter-turn short circuits in the stator windings of operating motors,” IEEE Trans. Energy Conversion, vol. 9, pp. 652-658, 1994.

[4]. H Jiang, R Aggarwal, “A new approach to synchronous generator internal fault simulation using combined winding function theory and direct phase quantities,” Ninth International Conference on Electrical Machines and Drives, Conf. Publication No. 468, 1999.

[5]. Gojko M. Joksimovic, Jim Penman, “The Detection of Inter-Turn Short Circuits in the Stator Windings of Operating Motors,” IEEE Trans. on Industrial Electronics, vol. 47, no. 5, October, 2000.

[6]. Amol S. Kulkarni, Mohamed A. El-Sharkawi, “Development of a Technique for On-Line Detection of Shorts in Field Windings of Turbine-Generator Rotors: Circuit Design and Testing,” IEEE Trans. on Energy Conversion, Vol. 15, no. 1. March 2000

[7]. Tommy W.S. Chow, and Hong-Zhou Tan,“HOS-based nonparametric and parametric methodologies for machine fault detection,” IEEE Trans on Industrial Electronics, vol. 47, no. 5, October 2000.

[8]. Y. M. Park, G.W Kim, “A logic based expert system for fault diagnosis of power systems,” IEEE Trans. Power Systems, vol. 12, pp. 363-369, Dec. 1994.

[9]. A.I.Megahed & O.P.Malik, “An Artificial Neural Network Based Digital Differential Protection Scheme for Synchronous Generator stator Winding Protection” IEEE Trans on Power del, Vol. 14, no. 1, pp. 86-93, Jan 1999.

[10]. Hatem A. darwish, Adbel-Maxoud I.Taalab, “Development and Implementation of an ANN-Based Fault Diagnosis Scheme for Generator Winding Protection,” IEEE Trans. on Power Delivery, vol. 16, no. 2, April 2001.

Technical College - Bourgas,