FPGA Implementation of a Bearing Fault Classification System Based on an Envelope Analysis and Artificial Neural Network

Bearings are one of the most widely used components of rotary machines. To keep these bearings running in the best condition, several techniques for the early diagnosis of faults are applied to enable continuous monitoring of their condition and avoid unexpected faults that may cause damage to humans and materials. Several works have focused on the development of such technologies, including those that apply artificial intelligence, in the classification and diagnosis of faults. This work reports on a multi-layer perceptron (MLP) to classify the conditions of faulty bearings, using the envelope analysis method to extract the faulty features of the bearings. The proposed architecture is implemented on a field programmable gate array (FPGA) board, where the Digilent Zybo Z7-20 platform with a Zynq-7000 FPGA circuit from Xilinx was selected as the target. The Case Western Reserve University (CWRU) dataset, which is considered the standard reference for testing bearing fault classifications, is used to evaluate the performances. The results of the implemented embedded system are first compared to those obtained through MATLAB simulations and then to those obtained from the literature. These practical results provide an average accuracy of 95 and 89% for the fault-type identification and fault-severity identification, respectively.