Recurrent Neural Network (RNN) Based Bearing Fault Classification of Induction Motor Employed in Home Water Pump System

Abstract

In home appliances, the water pump is used to supply the water from a room to the other rooms. Defects of the water
pump are not distributing the water, inner stator winding short circuit, and bearing failure. In this paper, bearing
fault detection of induction motor (IM) used in home water pump system is developed by using recurrent neural
network (RNN) method. It is difficult to detect fault bearing of IM using a mathematical model. So that, a recurrent
neural network (RNN) method is applied to solves this problem. These bearing faults classifications are based on IM
stator current waveform. Bearing fault types are all normal (AN), front fault (FF), rear fault (RF), and all fault (AF).
While, the detection process consist of three step. They are taking bearing fault data, features extraction, and RNN
fault detection. The bearing fault data is taken from the stator currents of IM by using soundcard oscilloscope
software. Second step is features extraction process to obtain more bearing fault signs. In this step, stator currents
of IM is converted from time domain into frequency domain by using Fast Fourier Transform (FFT). Last stage is
RNN model to clasify the bearing fault of IM. The effectiveness of proposed RNN method is clarified by using four
bearing fault types.