Using multisensor data fusion for allocating systemic faults of the hydraulic control subsystem of a rubber mixing machine

The rise of machine and deep learning algorithms in predictive maintenance has led to the influx of multidimensional data analysis studies. However, although studies dedicate towards increasing the accuracy of regression and classification models, many commit to resolving the issues by addressing a single fault mechanism, neglecting the latent degradation of other fault mechanisms. In this paper, we dedicate our efforts in understanding multiple and systemic faults through multidimensional data analysis. Using Knowledge Graph via Network Analysis we allocate markers of fault mechanisms that are used as features for fault classification. The features are extracted from discretised hydraulic power signal, hydraulic fluid physical and chemical data, and system response data. Using feature extraction we were able to observe latent degradation mechanisms that are used for multi label classification using machine learning algorithms. The results obtained show that neural network had highest, i.e., 85% accuracy (AUC = 0.88) among classification algorithms in allocating systemic faults within the hydraulic power system.