CSA-SVM method for internal cavitation defects detection and its application of district heating pipes

The goal of this paper is to develop an ultrasonic detection device that can be mounted on an underwater snake vehicle (USV) for underwater district heating pipe (DHP) detection in the future. Ultrasonic detection technology (UDT) is the detection means used, and the cavitation defects in polyurethane (PUR) layer of DHPs are the object being detected. Due to the large thickness of PUR layer and the complex interface information of multi-layer structure, detecting defects of DHPs quantitatively is a difficult task. To address this issue, this paper proposes an approach that combines feature extraction and crow search algorithm (CSA) optimized support vector machine (SVM). Firstly, the main parameters and detection method of UDT are designed after investigation. Secondly, defective signals are pre-processed by signal processing to extract the features form three domains. Finally, four different classifiers are used to identify cavitation defects based on the feature-set. When compared among optimized random forest (RF), k-nearest neighbor (KNN), and ordinary SVM, the experimental results show that CSA-SVM had the highest accuracy in defect size prediction, and the validation-experiment verifies the practicability and feasibility of the CSA-SVM classifier. All experiments illustrate that the issue could be well solved by our method.