Conditional monitoring and remaining useful life prediction of electric motor in predictive maintenance: review

Predictive maintenance of an electric drive as part of technological equipment reduces the probability of unplanned production downtime and minimizes repair costs by continuously monitoring the condition and predicting the remaining useful life of the electric drive. Based on the analysis of foreign sources and ISO standards, the article presents aspects of predictive maintenance systems: the stages of their construction and the methods used at each stage are considered. The emphasis is placed on the most time-consuming and combining various methods of stages - diagnosis of the condition of the electric drive and faults prediction. The methods of determining the technical condition of the electric drive are highlighted, their advantages and disadvantages are considered, as well as the types of faults that can be diagnosed using each of the methods. The most effective and widely used diagnostic methods turned out to be methods based on the analysis of vibration and electrical signals, since they allow detecting a wide range of faults. In addition, they are versatile, as they are suitable for determining the technical condition of almost all types of electric motors. A comparative analysis of machine learning methods used to predict faults and remaining useful life of an electric drive is carried out. Such machine learning methods as the Random Forest method, Long short-term memoryneural networks, the support vector machine method, and the k-nearest neighbor method are noted. The analysis showed that the choice of an algorithm depends on many factors and cannot be based on universal approaches.