Parameter-Based Transfer Learning for Bearing Fault Diagnosis Using Small Samples Under Variable Working Conditions

While measured vibration signals are widely accepted as an indicator of machine health conditions, whether debilitating or ideal, it remains challenging to effectively use them for real-time condition monitoring. The difficulty associated with data acquisition from faulty equipment due to hazard-related concerns further exacerbates the problem. Massive simulation data generated from a rotor-bearing dynamic model is utilized to build a machine learning model using a convolutional neural network (CNN). Several transfer learning strategies are utilized in conjunction with the small samples to facilitate real-world diagnosis. Two case studies are used to substantiate the model’s reliability. In one case, three out of four strategies show a perfect diagnosis, as opposed to the other case, where two out of four strategies achieve the same level of accuracy. The study reveals that positive transfer strategies are superior when developing a diagnosis framework using relatively small samples in both cross-domain and cross-working conditions.