Real-Time Localization Method of Large Pressure Vessel Leaks Based on Improved CNN and STCA of Elastic Wavefield

Abstract

In this paper, a real-time leak source localization method based on convolutional neural network (CNN) of elastic wavefield images and spatio-temporal correlation analysis (STCA) is developed for the pressure vessel leakage. This method uses a single sensor array coupled to the wall to collect the elastic wave data excited by the leak source. Besides, the distance $R$ and the direction $\theta $ between the leak source and the sensor array are calculated based on CNN and STCA respectively, to finally obtain the location ( $R$ , $\theta$ ) of the leak source. In this paper, the digital twin model of the experimental platform is established, the training set is obtained by the finite element simulation, and the CNN model applied to the elastic wavefield images is studied and constructed. The experimental results show that the maximum locating error is 1.46 cm and the average locating error is about 0.56 cm within the range of a 1 m2 experimental plate based on the method proposed in this paper.

Keywords

• Elastic wave
• leakage
• location
• sensor array
• deep learning