Non-Destructive Testing and Defect Identification of High-Voltage Cable Insulators Based on Fiber-Coupled Terahertz Time-Domain Pulsed System

Objective With the rapid development of high-voltage transmission technology, greater demands are placed on the performance and reliability of key power equipment. High-voltage cable insulators (typically referring to insulators used in overhead transmission lines) serve as core components of power transmission systems, providing electrical isolation between conductors at different potentials or between conductors and grounded structures, while also offering mechanical support. Therefore, it is essential to test insulators during production, installation, and operation to avoid potential failures after long-term service and to ensure the continuous and stable operation of power systems. However, traditional testing methods for high-voltage cable insulators have several limitations. To address this, terahertz technology has been introduced as a complementary solution. In this study, a fiber-coupled terahertz time-domain pulse system was developed for non-destructive testing of high-voltage cable insulators. The proposed system can effectively characterize various preset defects in ceramic insulators, demonstrating good imaging resolution and defect identification capability. Furthermore, to achieve intelligent defect recognition, a residual neural network classification model fusing time-domain and frequency-domain features was proposed. The classification results indicate that the model can accurately identify and classify different defect types. Methods In this study, a fiber-coupled terahertz time-domain spectroscopy (THz-TDS) system was designed, with an effective operating frequency range of 0.1–1.5 THz. The system employed a high-speed delay line device driven by a voice coil motor to achieve rapid scanning of terahertz time-domain signals, and a six-fold folded optical path design was implemented to extend the scanning range to 600 ps. High-voltage cable insulators made of aluminosilicate ceramics, typically coated with an anti-pollution flashover layer, were selected as the research objects. First, the optical parameters of coated samples in the terahertz frequency band were extracted using the time-of-flight method. Second, by taking the reflection signal of flat materials or metallic mirrors with known optical parameters as a reference and comparing it with the reflection signal of the samples, the refractive index and absorption coefficient of ceramic materials without metallic substrates were calculated. To further evaluate the application potential of the system in non-destructive testing, high-voltage cable insulator simulation samples with preset defects were fabricated. These samples were tested and analyzed using the developed terahertz system for defect detection and imaging. Finally, to realize online monitoring and intelligent diagnosis of potential defects during production and use, improve detection efficiency, and effectively prevent safety accidents caused by insulation performance degradation, a residual neural network classification model based on the fusion of time-domain and frequency-domain features was proposed. Through the deep fusion of dual-domain features, effective intelligent classification of typical insulator defects was achieved. Results and Discussions According to the time-of-flight method, the average refractive index of the coating in the terahertz domain was determined to be 1.51. After processing and imaging, the contours of several different defects were clearly displayed. The analysis of depth information at different positions is shown in Fig. 8, where the signals corresponding to different defect types exhibited distinctive features that could be effectively analyzed and displayed by the terahertz system. Based on these results, it is evident that the fiber-coupled terahertz time-domain pulse system demonstrates strong detection capability for high-voltage cable insulators and their various defect characteristics. During the detection and imaging of high-voltage cable insulators, if potential defects can be monitored online during their production and operation stages, detection efficiency can be significantly improved. A residual neural network classification model fusing time-domain and frequency-domain features was established. A dataset consisting of six defect categories was prepared and used for model training. The model achieved an accuracy of 97.98%, which represented a notable improvement compared with both the dual-channel defect recognition algorithm based on a Transformer using time-frequency features and the residual neural network classification model based solely on time-domain features. To more intuitively demonstrate the classification accuracy of the model, three-dimensional reconstructions of the test set data were performed. Conclusions The acquisition of the three-dimensional structural distribution of high-voltage cable insulators is of great significance for condition monitoring and fault diagnosis of power equipment and holds broad industrial application prospects. In this work, a terahertz three-dimensional imaging system for non-destructive testing of high-voltage cable insulators was designed and implemented based on a fiber-coupled terahertz time-domain pulse mechanism. By introducing two parameter extraction methods, the key optical parameters of insulator materials were successfully measured. Experiments were carried out on actual high-voltage cable insulator samples, and their terahertz three-dimensional images were successfully reconstructed, which verified the effectiveness of the developed system in defect identification and imaging. Furthermore, a residual neural network classification model fusing time-domain and frequency-domain features was proposed for intelligent recognition of insulator defect types. The classification accuracy reached 97.98%, representing a significant improvement in detection performance compared with traditional single-feature extraction methods.