Tracking Evolution of Stator-based Fault in Induction Machines using the Growing Curvilinear Component Analysis Neural Network

Stator-based faults are one of the most common faults among induction motors (IMs). The conventional approach to IM control and protection employs current sensors installed on the motor. Recently, most studies have focused on fault detection by means of stator current. This paper presents an application of the Growing Curvilinear Component Analysis (GCCA) neural network aided by the Extended Park Vector Approach (EPVA) for the purpose of transforming the three-phase current signals. The GCCA is a growing neural based technique specifically designed to detect and follow changes in the input distribution, e.g. stator faults. In particular, the GCCA has proven its capability of correctly identifying and tracking stator inter-turn fault in IMs. To this purpose, the three-phase stator currents have been acquired from IMs, which start at healthy operating state and, evolve to different fault severities (up to 10%) under different loading conditions. Data has been transformed using the EPVA and pre-processed to extract statistical time domain features. To calibrate the GCCA neural network, a topological manifold analysis has been carried out to study the input features. The efficacy of the proposed method has been verified experimentally using IM with l.lkW rating and has potential for IMs with different manufacturing conditions.