Leveraging Auxiliary Task Relevance for Enhanced Bearing Fault Diagnosis through Curriculum Meta-learning

TL;DR

The paper addresses fault diagnosis under data scarcity and variable operating conditions by proposing Related Task Aware Curriculum Meta-learning (RT-ACM). RT-ACM extends MAML with a related-task relevance weighting, autoencoder-derived task similarity, and an easy-first curriculum guided by a teacher recommender, followed by target-task fine-tuning with layer freezing. Across two real-world bearing datasets, RT-ACM outperforms DL and several meta-learning baselines, demonstrating faster convergence and higher accuracy in few-shot settings. The work offers a practical approach to robust fault diagnosis in smart manufacturing with limited labeled data and diverse working conditions.

Abstract

The accurate diagnosis of machine breakdowns is crucial for maintaining operational safety in smart manufacturing. Despite the promise shown by deep learning in automating fault identification, the scarcity of labeled training data, particularly for equipment failure instances, poses a significant challenge. This limitation hampers the development of robust classification models. Existing methods like model-agnostic meta-learning (MAML) do not adequately address variable working conditions, affecting knowledge transfer. To address these challenges, a Related Task Aware Curriculum Meta-learning (RT-ACM) enhanced fault diagnosis framework is proposed in this paper, inspired by human cognitive learning processes. RT-ACM improves training by considering the relevance of auxiliary sensor working conditions, adhering to the principle of ``paying more attention to more relevant knowledge", and focusing on ``easier first, harder later" curriculum sampling. This approach aids the meta-learner in achieving a superior convergence state. Extensive experiments on two real-world datasets demonstrate the superiority of RT-ACM framework.

Figures (7)

• Figure 1: Learning process of MAML, which optimizes for a representation $\theta$ that can quickly adapt to new tasks.
• Figure 2: The overall framework of our proposed RT-ACM.
• Figure 3: Comparison of (a) 3-Way and (b) 6-Way 5-Shot Performance on $T_1$.
• Figure 4: Accuracy in the (a) outer loop and (b) model performance on $T^1$.
• Figure 5: The visualization of classification from four models on (10-way 5-shot).