Machine Learning Techniques for Predictive Maintenance of Industrial Machinery

Rotating machines such as motors, pumps, gearboxes andbearings are at the heart of almost every industrial process. Whenone of them fails without warning, the result is lost production,possible safety hazards and expensive emergency repair. Theshift toward Industry 4.0 and cyber-physical production systems[1] has made it cheap to instrument machines with sensors and tocollect data continuously, which creates an opportunity to predictfailures before they occur.Traditional condition-based maintenance relies on fixedthresholds or on the judgement of experienced engineers [4].Machine learning offers a data-driven alternative that can learncomplex relationships between sensor signals and machine healthwithout an explicit physical model. This paper gives a structuredoverview of how ML is applied to PdM. Section II comparesmaintenance strategies, Section III covers data acquisition andfeature extraction, Section IV reviews ML techniques, Section Vpresents a generic framework, Section VI discusses evaluationmetrics, and Section VII lists challenges and future directions