Lightweight Acoustic Anomaly Detection for Converter Valves

Accurate and lightweight anomaly detection

An unsupervised anomaly detection method based on lightweight transformer generative adversarial network achieves high accuracy and low resource consumption on converter valve acoustic data, addressing sample imbalance and computational resource constraints.

Tang, Mingzhu · Chen, Yan · Hu, Haijun · Wang, Zhiyong · Fuqiang, Xiong · Yang, Wei · Chen, Zhiwen

ISA Transactions 2025

Specifications

million parameters
{'zh': '0.506', 'en': '0.506'} GFLOPs

Advantages

AUC of 0.9806, significantly outperforming baseline methods

p-AUCs of 0.9295, 0.9122, and 0.9559 in low false-positive regions [0,0.1], [0,0.2], and [0,0.3], respectively