Incorporating bearing flexibility into a digital twin for supervised machine learning-based ball bearing diagnostics

Vibration analysis remains the most used technique for the maintenance of rotating machines, despite requiring their availability. However, with the emergence of Industry 4.0, digital twins provide the opportunity to generate signals of several operating modes. These signals can be used to diagnose or predict the state of the machine being monitored. The digital twins rely on working on hypotheses in the conception of the numerical model. Rolling bearings are one of the most critical components in rotating machines, due to their high requirements for machine dynamics. In this study, the flexibility of the bearings are integrated into the numerical model of a rolling bearings test bench. To achieve this, a discrete element model and a finite element model is created, which communicate with each other to translate the structure's dynamics. To convert the numerical model into a digital, it is recalibrated on the basis of the signals measured on the test bench, by minimizing an objective function. Then, to assess the reliability of the data generated by the digital twin, a diagnostic by classification is performed. The data generated by the digital twin are used to train a MSVM classification algorithm. Predictions are made on the experimental data initially acquired, describing the same operating modes as the generated data. The results show an improvement in the classification accuracy with the model integrating flexibility compared to the model existing in the literature.