A Data-Driven Methodology to Assess Raw Materials Impact on Manufacturing Systems Breakdowns

Abstract

Data-driven Prognostics and Health Management (PHM) become a crucial layer in the realm of predictive maintenance (PM), particularly for metal-forming industries. In fact, noncompliant material characteristics affect badly the manufacturing tools leading to high machine breakdown frequency and poor quality products. To cope with this situation, a new methodology for breakdown prediction is proposed. In detail, the methodology starts by implementing an Extract, Transform, Load (ETL) process to create a new dataset from heterogeneous sources. Then, a feature selection method is used for dimensionality reduction and keeps only useful information. After that, a Machine Learning (ML) model predicts system breakdown occurrences using the selected features. Finally, thanks to these steps above, an auto-labeling algorithm to evaluate the severity impact of the material data is proposed and makes the originality of this paper. The developed methodology is applied to a real dataset of a French company, SCODER, that shows and points out promising perspectives in PM.