Model-Based Spindle Bearing Monitoring Using Vibration Sensors and Artificial Neural Networks

To ensure the longevity of the bearings of a motor spindle, it is advantageous to know the precise loads on the bearings during operation. Since sensor-based monitoring involves a great deal of effort due to the limited space available, and simulating the bearing load is not real-time capable, we investigated how the bearing loads can be estimated using machine learning methods. To estimate the bearing load, a co-simulation was first set up that generates large amounts of training data based on measured cutting forces and spindle vibration velocities. Measured and simulated quantities are then used to train artificial neural networks. The best-performing neural networks can estimate the surface pressure between rolling elements and bearing rings with an error of less than 2%. This deviation refers to the contact stress that was calculated for comparison with the simulation results.