Fault Size Estimation of Ball Bearings: A Machine Learning Approach for Noisy Data

Accurate bearing condition monitoring, including fault size estimation has the potential of identifying the conditions that causes bearing failure. However, noisy data can exacerbate data analysis and may lead to wrong bearing condition assumptions. This study shows that a convolutional neural network (CNN) trained with synthetically generated data from 3D multi-body simulations of a roller bearing with different faults can estimate the fault sizes from noisy measurements of defective ball bearings.