Data Augmentation Method for the Temperature Rise of Multi-disc Wet Clutch Based on the Joint Drive of Multiple Data Streams

In response to the incomplete coverage of operating conditions and the high acquisition costs of critical operating condition data in the sliding friction temperature rise test data set of multi-disc wet clutch, this paper proposes an improved TimeGAN data augmentation method based on numerical simulation and test data fusion. Firstly, a improved TimeGAN architecture based on multi-layer Transformer and a data reconstruction method (TTimeGAN) are proposed to enhance the model generation quality. Secondly, a multi-data stream training strategy based on the temperature rise model of friction element is proposed, and a coefficient of condition transformation is constructed through mechanism constraints to improve the domain generalization ability of the model for the operating conditions. Finally, a multi-module joint training process under the constraints of three loss functions is established. The comparison results show that the proposed method is used to improve the value range index of the generated data by 15.2% to 16.2%, and increase the time series index by more than 87.8%. The results of the ablation experiments show that the joint training of multiple data streams can increase the quality of generated data under new conditions by more than 14.2%. The average performance improvement brought by data reconstruction and multi-layer Transformer is approximately 48.6%. The quality of the generated data has a significant negative correlation with the length of the generated sequence.