INTRODUCTION TO PREDICTIVE MAINTENANCE BASED MACHINE LEARNING WITH RISK AND UNCERTAINCIES: A STATE OF THE ART

Uncertain and risky environment surrounding industries makes equipment maintenance planning and scheduling very complex. Standard maintenance approaches such as corrective maintenance and preventive maintenance have been used for long time. More recently, predictive maintenance based on machine learning (PdM–ML) is taking more place within scientific and industrial community. It is preventive maintenance activities coupled with equipment monitoring condition. However, the implementation of such strategy is more complex and requires more financial and technological means than traditional approaches. Therefore, fully understanding the deployment aspects of the said maintenance strategy and the resulting benefits are prerequisites for business wishing to get in. This is even much more true for metallurgical companies where archaic and sophisticated characters of equipment coupled with highly typical skills of working personnel makes the situation more complex. This work aims to lay the foundations that will support the development of PdM–ML. The said approach will be tested in a metallurgy company to validate its robustness and applicability. For this, a literature review is carried out to identify, define and understand the subtleties surrounding (PdM–ML). The concepts of uncertainty and risk are also addressed. A new literature review methodology based on the systematic literature review (SLR) is presented and used to extract the relevant papers. The study reveals that, for the last then decade, the main fields of application of PdM–ML are manufacturing, electric-power, structure reliability. Also, the main goal observed for PdM–ML are failure diagnosis, failure prediction, failure detection, fault analysis and remaining useful life (RUL) prediction.