Remaining Useful Life prediction and challenges: A literature review on the use of Machine Learning Methods

Approaches such as Cyber-Physical Systems (CPS), Internet of Things (IoT), Internet of Services (IoS), and Data Analytics have built a new paradigm called Industry 4.0. It has improved manufacturing efficiency and helped industries to face economic, social, and environmental challenges successfully. Condition-Based Maintenance (CBM) performs machines and components' maintenance routines based on their needs, and Prognostics and Health Management (PHM) monitors components' wear evolution using indicators. PHM is a proactive way of implementing CBM by predicting the Remaining Useful Life (RUL), one of the most important indicators to detect a component's failure before it effectively occurs. RUL can be predicted by historical data or direct data extraction by adopting model-based, data-driven, or hybrid methodologies. Model-based methods are challenging, expensive, and time-consuming to develop in complex equipment due to the need for a lot of prior system knowledge. Data-driven methods have primarily used Machine Learning (ML) approaches. They require little historical data, are less complex and expensive, and are more applicable, providing a trade-off between complexity, cost, precision, and applicability. However, despite the increased use of data-driven methods, several studies have pointed out different challenges to RUL prediction. Some works have proposed solutions from individuals and unconnected work structures to overcome these challenges, and there is still a lack of an explicit framework for general process analysis. Moreover, none of them have correlated the different challenges with each micro-step of the RUL prediction process. This work describes the structures, systems and components approached, and datasets used. Next, it proposes a compact framework for the RUL prediction process. Also, it maps the main challenges of this process and the advantages and drawbacks of the most relevant ML methods. Further, it discusses the operational datasets, the accuracy concern, the use of file log systems in the RUL prediction, and approaches the ML Interpretability (MLI) issue. Finally, it concludes with some future research directions.