Bearing remaining useful life estimation under varying speed

This workshop presents the results of our research project on  “A Digital Twin for Health Monitoring and Predictive Maintenance”. In this project we focused on methodology for bearing remaining useful life estimation under varying speed. Two novel bearing prognostics approaches based on vibration measurements will be presented , including an experimental validation.

• The first approach exploits a bearing phenomenological model to generate training data which are used in a machine learning algorithm. This allows to avoid having to train the algorithms based on degradation data, which are typically not available.
• The second approach does not need any models but requires one degradation dataset and it is based on feature extraction and fusion.

For both approaches, the estimation of the end of life clearly outperforms the state of practice, which relies on design rules. This allows to reduce machine downtime and optimize the maintenance plan.