Polymer gear contact fatigue reliability evaluation with small data set based on machine learning

Abstract Polymer gears have shown potential in power transmission by their comprehensive mechanical properties. One of the significant concerns with expanding their applications is the deficiency of reliability evaluation methods considering small data set circumstances. This work conducts a fair number of polyoxymethylene (POM) gear durability tests with adjustable loading and lubrication conditions via a gear durability test rig. A novel machine learning-based reliability model is developed to evaluate contact fatigue reliability for the POM gears with such a data set. Results reveal that the model predicts reasonable POM gear contact fatigue curves of reliability–stress–number of cycles with 2.0% relative error and 18.8% reduction of test specimens compared with the large sample data case. In contrast to grease lubrication, the oil-lubricated POM gear contact fatigue strength improves by 10.4% from 52.1 to 57.6 MPa.