一种动态稀疏化的域增量轴承故障诊断方法OA

Industrial scenarios with continuous dynamic changes in rotational speed and load maybe lead to the failure of already trained models,thus degrading the performance of bearing fault diagnosis.To allow the model to learn new domain knowledge while effectively retaining the old domain knowledge that has already been learned,a dynamic sparsification method for domain incremental bearing fault diagnosis is proposed.Firstly,the L1-norm-based dynamic screening module for representative samples is utilized to select representative samples that contain important knowledge in different domains,and protect previously learned knowledge in different domains while learning new knowledge.Secondly,a sparsity rate parameter,which can be set autonomously according to the storage requirements,is introduced during the screening process to adjust the number of representative samples to be stored to save memory space.Finally,catastrophic forgetting is avoided by using the importance loss function to impose constraints on important parameters in the old domain.The proposed method is used to experiments on two rotating machinery datasets,and the results show that the proposed method achieves the highest average diagnostic accuracies of 100％and 99.06％at the same and different rotational speeds(loads),respectively.The memory capability of the model is also superior to traditional fault diagnosis methods.