Multi-view generative synergistic optimization agent for few-shot machinery fault diagnosis under limited labeled data

Abstract

Mechanical fault diagnosis is crucial for maintaining the stability of industrial systems. However, the limited labeled fault data in real-world applications severely restricts diagnostic accuracy and generalization. To address this problem, this paper proposes a deep reinforcement learning–enhanced multi-view generative synergistic optimization agent, termed MGSOA, for machinery fault diagnosis under data-scarce conditions. First, a deep data sampling augmented generative adversarial network is designed to generate high-quality fault samples and mitigate data bias in few-shot scenarios. Furthermore, a multi-view feature extraction with domain adaptation architecture is constructed to adaptively extract multi-scale and multi-view features, enhancing feature representation under few-shot and mixed operating conditions. Finally, a novel priority-embedded reward-centering deep deterministic policy gradient algorithm is developed to improve the accuracy and robustness of the agent. Experimental results demonstrate that the proposed method outperforms other state-of-the-art methods in few-shot machinery fault diagnosis.