Contact Defect Detection of Gas-Insulated Line Via Thermal–Vibration Feature Fusion and Deep Neural Network Technique

By using thermal infrared images, vibration waveforms, and deep learning, this work offers a way for identifying contact failures. Infrared thermal imaging and vibration waveform datasets were created by data enhancement and the Gramian Angular Field (GAF) approach on a 1100 kV GIL prototype capsule with normal and contact faults under varying currents. Infrared image and vibration waveform high dimensional features are extracted by dual-branch convolutional neural network (CNN) architecture, followed by feature embedding and parameter sharing to achieve feature fusion of heat -vibration information, and fully connected (Fc) neural network via double constraint loss function to classify contact defects. Results show that purposed technique could out - performed the single heat/vibration method and the decision- level fusion method, reaching accuracy rates of 93.75%, precision ratios of 93.97%, and recall rates of 93.50%.