A Structured Approach to Machine Learning for Condition Monitoring: A Case Study

This chapter details the application of a machine learning condition monitoring tool to an industrial case study. The process follows the content of the corresponding tutorial chapter and is a step-by-step example of the setup of a monitoring kit in a packaging machine. The case study is particularly interesting since it is focused on Independent Carts System. This consists of a closed path made up of modular linear motors having a straight or curved shape and controls a fleet of carts independently. The application, not so common nowadays, proves the feasibility of the proposed condition monitoring approach in a non-trivial case, with scanty literature on it. The target is the diagnostics of ball bearings present in the wheels of the carts in order to reduce downtime due to the breakage of these components and to maximize their life cycle cutting down spare part costs. This chapter details the phase of feature extraction, the Machine Learning methods used, the results and the metrics for measuring them. Considerations will be made in particular on the acceptability/interpretability of the results and the industrial significance of the metrics.