关键词: 皮带机托辊故障诊断, 噪声干扰, Jeffrey 散度, 多通道残差去噪对抗自编码器

Abstract: The idler is a critical rotating component of belt conveyors, responsible for bearing the weight of the conveyor belt and ensuring its stable operation. Its operation is often accompanied by complex noise and signal instability, posing significant challenges for fault diagnosis. To effectively suppress noise interference in acoustic signals and improve diagnostic accuracy, this proposes an idler fault diagnosis framework based on an enhanced denoising adversarial autoencoder. The framework incorporates a multi-channel residual denoising adversarial autoencoder based on Jeffrey divergence, which not only denoises the signals but also enhances feature reconstruction capability. An independently designed fault classifier is then employed to accurately identify the reconstructed signals. The proposed method is evaluated under varying noise levels. Experimental results demonstrate that the method achieves excellent diagnostic performance in both noise environments, with an average recognition accuracy exceeding 99%, significantly outperforming existing mainstream deep learning models. The constructed residual structure exhibits stronger robustness against noise compared to traditional residual networks.

Key words: belt conveyor idler fault diagnosis, noise interference, Jeffrey divergence, multi-channel residual denoising adversarial autoencoder