Small sample bearing fault diagnosis based on combined prediction model.

Rolling bearings are an important part of problems that often occur in rotating machinery. Their fault conditions are complex and difficult to diagnose. Aiming at the problem that small sample learning has a large difference between the true feature value and the target feature value and the generalization ability is weak, this paper proposes a small sample learning model that combines semi-supervised variational autoencoder and LightGBM classification model. The Bayesian optimization and improvement algorithm based on Gaussian process is used to optimize the LightGBM hyperparameters, thus effectively solving the defects such as unstable performance, weak ability of extracting features, and overfitting in small sample learning. Comparative verification on the bearing experimental data set released by the Western Reserve University in the United States shows that the method has better diagnostic accuracy when facing small sample data space. [ABSTRACT FROM AUTHOR]