ANN-based detection of broken coils of small generator stator with two parallel branches in phase

This paper presents a fault detection system based on multilayer feedforward artificial neural network (ANN) for detection of one or several broken parallel coils of a small generator. This fault type cannot visibly influence on machinepsilas normal performance and since the performance of a faulty machine with broken coils is similar to that of a healthy one with unbalance load, therefore detection of this kind of fault at the beginning is important. Normal and faulty machines are simulated using the parameters obtained by finite element analysis (FEA). A massive harmonic content analysis of field and stator phase currents are performed and ultimately harmonic content of field current from 1^st to 7^th order (as parameters with maximum variations) are chosen as training signals for a feedforward ANN. Feedforward neural network is trained using the back propagation algorithm. This ANN-based fault detection system identifies the broken coil faults of the stator windings with proper convergence to desired target values.

I. Introduction

Currently, power electric industry and consumer demand is developing increasingly. Power plant installation and operating problems increase trend toward embedded generation.