LSGAN-Transformer life prediction method for rolling bearings under few samples

Aiming at the problem that it is difficult to obtain a large amount of data for bearings with complex working conditions, which leads to the inability to accurately predict their life, a rolling bearing life prediction method based on few samples, LSGAN-Transformer, is proposed. A dropout layer is added to the LSGAN generator to avoid the overfitting phenomenon that often occurs during few-sample training. The normalization of each layer in the traditional Transformer model is moved forward to the input of the decoder and encoder submodules before the residual network, forming a direct gradient path from input to output, avoiding the problem of excessive expected gradient near the output layer that often occurs in the traditional Transformer network. Verification on the PHM2012 dataset and the XJTY-SY dataset shows that the MAE and RMSE of the proposed method are greatly improved; compared with other common prediction models, the MAE and RMSE of the proposed method are improved by 30.61 % and 35.93 % respectively.