Weak Fault Diagnosis Method Based on SEGAN and KL Divergence for Industrial Equipment

Abstract

With the complexity and large-scale of modern control systems, more and more weak faults threaten the safe operation of system equipment. Therefore, finding an effective method for diagnosing weak fault signals has become an urgent problem to be solved. However, the weak fault amplitude is small and the detection is difficult. Therefore, this paper proposes a semi-supervised weak fault symptom enhancement and diagnosis method Signal Enhancement Generative Adversarial Networks(SEGAN) to realize early warning and diagnosis of industrial equipment under complex conditions. The semi- supervised fault symptom enhancement architecture of Signal Enhancement Generative Networks(SEGAN) is constructed, and the weak fault symptom is enhanced by anti-encoding. For the enhanced signal, the fault feature identification method based on KL divergence is proposed to effectively realize timely and rapid fault diagnosis. In this paper, the effectiveness of the proposed method is verified by the natural vibration and percussive vibration data under the fault state of the power tower. Experiments show that the fault diagnosis of the natural vibration signal containing weak fault signs can be realized.