Fault detection and diagnosis of rubber mixing machine using machine and deep learning: an energy-based maintenance approach

In this study, we offer an Energy-Based Maintenance (EBM) solution as an alternative to Predictive Maintenance (PdM) practices relying on discretised hydraulic power signals. The study offers new insights into fault detection and diagnosis by incorporating machine and deep learning algorithms for resolving multiclass classification problems of systematic faults in a rubber mixing machine process. By incorporating the Recursive Feature Elimination Feature Selection (RFE-FS) approach, we have successfully extracted the most relevant feature from various machine learning algorithms and analysed features using GGM (Gaussian Graphical Models). The obtained findings uncover latent degradation of a hydraulic system, specifically the degradation of a hydraulic signal at the axial piston pump outlet that led to> 7% degradation in hydraulic power output.