Mlamsan-based cross-case fault diagnosis of rotor-bearing systems
In response to the low accuracy in fault diagnosis caused by data distribution deviations due to variable operating conditions of mechanical equipment, this paper constructs a multi-scale linear attention multi-source subdomain adaptation network (MLAMSAN) that integrates the multi-scale linear attention (MLA). It can achieve fault diagnosis under cross operating conditions through subdomain feature alignments. This method uses the local maximum mean difference that can align subdomains as the metric function, integrates convolutional neural networks and MLA with lower computational complexities as the feature extractor, and constructs the MLAMSAN model for cross condition fault diagnosis of rotor-bearing systems. By comparing different convolutional layer effects, training batch sizes, and learning rates on the network model, parameters that optimize the model performance are selected. Through experimental verification on both the self-constructed dataset and the publicly available dataset, the MLAMSAN model can realize the diagnosis of single and compound faults of rotor-bearing systems under cross-working conditions, and exhibits the superior diagnostic performance and the strong generalization ability.