Intelligent defect diagnosis of GIS basin insulator via multi-source ultrasonic fusion

To take advantage of ultrasonic-based non-destructive testing (NDT) and data-driven intelligent defect diagnosis, the current study proposes a feature tensor classifier based on multi-source ultrasonic fusion, to enhance the defect diagnosis adaptability and reliability for gas-insulated switchgear (GIS) basin insulator. First, multi-source ultrasonic signals are acquired by finite element modelling (FEM), describing the healthy states of the GIS basin insulator completely. Second, time of flight (Tof)-featured tensors are expressed by wavelet transform (WT), and used to create the datasets. Third, a deep learning-based feature tensor classifier is proposed, and concerned training, validation, and testing processes are carried out. Finally, the effectiveness of feature tensor extraction is evaluated, and the anti-noise performance of the Tof-featured tensor classifier is verified. The main contributions indicate that the Tof-featured tensor classifier can realise excellent diagnosis performance, the average accuracy is, respectively, 90.53%, 99.75%, and 100% in training, validation, and testing sets, while the signal tensor classifier has poor performance. In addition, three other noised datasets are applied, and the result shows that the anti-noise performance of the Tof-featured tensor classifier is feasible, when SNR is greater than 1 dB.