Induction motors fault diagnosis using a stacked sparse auto-encoder deep neural network

In this article, deep neural network and stacked sparse auto-encoder deep neural network performances in fault diagnosis are compared. Methods are employed experimentally for the detection and isolation of an induction motor's condition (healthy, bearing outer race fault, stator winding short circuit, and rotor broken bar) in the presence of unbalanced power supply and pump dry running disturbances. Pre-processing and de-noising is performed on three-phase electrical current signals using fast Fourier transform and independence component analysis algorithm, respectively. Experimental results show that sparse auto-encoder deep neural network method has outperformed and diagnosed the aforementioned faults in the presence of disturbances with a highly reliable accuracy rate of 90.65%.