Monitoring the Thermal Aging of Rubber Bearings Using Virtual Temperature Sensors

Recurring thermal loads in a vehicle can lead to the failure of rubber bearings due to thermal aging within the expected vehicle lifetime. The disadvantages of a preventive or reactive maintenance strategy are high warranty costs and low customer satisfaction, respectively. This work proposes a predictive maintenance system, which monitors the thermal aging of rubber bearings and indicates their timely replacement. Since no real temperature sensors are installed at rubber bearings in production vehicles, virtual temperature sensors are used to monitor component temperatures during customer operation. As a virtual sensor, a feedforward neural network is trained on measurement data in order to learn to predict the component temperatures of several rubber bearings in a combustion engine vehicle based on existing vehicle signals. The neural network achieves an average mean absolute error of 1.78 K and a coefficient of determination of 0.95 over all components after hyperparameter tuning. The remaining useful life of the rubber bearing can be estimated based on the predicted temperature load collective. By integrating this predictive thermal aging system into the vehicle, the disadvantages of a preventive or reactive maintenance strategy can be avoided. Potential limitations for the implementation in production vehicles may be data availability for the model training or the reliability of model predictions under real-world conditions.