Intelligent Fault Diagnosis Across-Datasets Based on Second-Level Sequencing Meta-Learning for Small Samples

Mechanical equipment functioning in intricate surroundings is prone to malfunctions, which can lead to accidents and significant financial losses. A key component of machinery health management, fault diagnosis creates a connection between equipment state and health data monitoring. This paper presents a second-level sequencing meta-learning approach to tackle the constraints of insufficient fault data and cross-dataset issues, which might impair the accuracy of intelligent fault detection models under normal operating situations. By utilizing the Model-Agnostic Meta-Learning (MAML) core, this technique effectively addresses the problem of limited sample size. Second-level sequencing is implemented for cross-dataset fault diagnostics. The experimental findings, using Paderborn University bearing datasets and University of Connecticut gear datasets, demonstrate the superiority of the suggested Second-Level Sequencing Meta-Learning (SSML) model. SSML demonstrates superior performance compared to other models, with a 95.1% accuracy rate for bearing datasets and a 97.0% accuracy rate for gear datasets. This makes it very useful for diagnosing faults in complicated situations with limited data samples and across different datasets. The importance of sequencing in improving model stability and attaining high accuracy across datasets is emphasized in the study.