Automated Space Charge Detection Inside XLPE HVDC Cable Insulation Employing Customized Convolutional Neural Network

Space charge accumulations taking place inside the high-voltage direct current (HVDC) polymeric insulation causes significant distortion in electrical field distribution i.e., homo-charges increases the field value past uniform measure and the hetero-charges decrease it. Effect of such unbalanced electrical fields reflect in severe insulation degradation, resulting into reduced service life of the cable and ultimately breakdown if not detected early. Hence, space charge identification is extremely crucial to maintain safe and reliable operations of HVDC cables as well as to increase their service lifespan. Keeping this in mind, primary purpose of the proposed method is to fabricate an indigenous deep learning framework for automated space charge detection and categorization inside HVDC cable insulations. To this end, space charge accumulations inside cross-linked polyethylene (XLPE) insulation were measured and the acquired red-green-blue (RGB)-colored space charge image data were subsequently augmented in size and dimensionally pre-processed. The automated space charge recognition task was accomplished utilizing a light-weight convolutional neural network (CNN) model, whose architecture was custom-designed for the particular space charge classification problem in hand, keeping in mind the objective to perceive optimum accuracy engaging the least computation time. Comparative survey against different classical CNN models reveals persuasive potential of the proposed customized CNN-based methodology for automated space charge detection and categorization purpose, which substantiate its presumable application in real-time HVDC insulation diagnostic scenarios