Abstract
During long-term operation in complex environments, the pressure pipeline systems are prone to damage and faults, and serious safety accidents may occur without real-time condition monitoring. Moreover, in traditional non-contact monitoring approaches, acoustic signals are widely employed for condition monitoring for pressure pipelines, which are easily contaminated by background noise and provide unsatisfactory accuracy. As a tool for quantifying uncertainty and complexity, signal entropy is applied to detect abnormal conditions. Based on the characteristics of entropy and acoustic signals, an improved entropy-based condition monitoring method is proposed for pressure pipelines through acoustic denoising. Specifically, this improved entropy-based noise reduction model is proposed to reduce the noise of monitoring acoustic signals through adversarial training. Based on the denoising of acoustic signals, an abnormal sound detection method is proposed to realize condition monitoring for pressure pipelines. In addition, the experimental platform is built to test the effectiveness and reliability of the proposed method. The results indicate that the quality of signal denoising can reach over 3 dB, while the accuracy of condition monitoring is about 92% for different conditions. Finally, the superiority of the proposed method is verified by comparing it with other methods.