A machine learning approach to predictive maintenance: Remaining useful life and motor fault analysis

Abstract

Rotary motors are integral to various modern technological domains, playing a crucial role in areas such as manufacturing and medical equipment. Consistency in motor performance is vital in these domains, as any downtime can lead to substantial time and financial losses. The advent of Predictive Maintenance (PdM) has provided a means to mitigate this challenge. This paper presents a comprehensive framework designed to predict the specific fault type occurring within a given motor and determine its remaining useful life (RUL). Utilizing Industry 4.0 applications, the proposed framework harnesses real-time vibration and motor current signature analysis (MCSA) data, feeding it into Machine Learning (ML) classification and regression models. These models promptly alert maintenance personnel of potential motor faults. To validate the effectiveness of the proposed framework, experimental verification was conducted using a one-horsepower (HP) motor, in which faults were systematically introduced at specified time intervals. The experimental results affirm the efficacy of the proposed framework in accurately classifying various fault conditions and determining the RUL of the motor. Consequently, the framework enhances the PdM capabilities for motors deployed in practical settings.