Federated Learning-Based Bearing Fault Classification

Abstract

Motors play an important role in converting various energy sources into mechanical workforce. In this context, there should be a motor in every area where movement is required. Electric motors are preferred because they have advantages over other types of motors due to minimum energy consumption and minimum fault rate. The use of electric motors covers a wide range of applications, from large-scale industrial systems to agriculture, automotive, medical, smart home systems, white goods and small household appliances. Although it has low failure rates, there is a possibility of failure. These malfunctions can cause serious financial losses for businesses using electric motors. In this study, a federated learning(FL) based method is proposed to minimize the losses caused by faults. Models created as a result of training with data sets from different businesses through federated learning are combined on a server. The updated model created by combining the models is sent back to the businesses, and the businesses conduct their subsequent training with this updated model. With the federated learning method, data sets are not shared outside the businesses and training is carried out with more diverse data sets. In this way, the performance rate can be increased. As a result of the training performed with the proposed method and The Case Western Reserve University (CWRU) dataset, a accuracy rate of 99.06% was achieved. In this way, motor faults can be detected early, possible errors can be prevented and losses will be minimized. ©2024 IEEE.