Machine learning modeling on acoustic emission leak detection of metal-sealed pressure vessel

The metal-sealed pressure vessels are often used to provide a sealed pressure environment for storage and transportation of gas or liquid. Once the leakage occurs, it will cause serious financial loss and even casualties. Therefore, it is very necessary to monitor the operation state of the metal-sealed pressure vessel in time. This paper describes a real-time detecting method for metal-sealed pressure vessels, which is based on a machine learning (ML) framework and acoustic emission (AE) signal. The result of leakage detection based on AE is susceptible to different factors, such as environmental noise, the material of detected pressure vessels, and the position of the AE sensor, these factors may lead to a poor detection result. To address these drawbacks, an effective ML model, the decision tree (DT) based on AE signals is presented. At the core of this methodology, a DT model was trained on over 20 thousand AE signal records for classifying the leak sections. The experimental result illustrated that the accuracy of the proposed model to detect leakage of the metal-sealed pressure vessel is 98.3%, F1
score
is 0.98, which has significant advantages compared with the traditional method. Instead of tedious feature extraction, ML methods have the advantage of being simple and real-time.