Classification of Insulation Defects in High Voltage Equipment Using Computer Vision

局部放电（PD）分析是用于检测电气绝缘中的早期故障的电气方法。PD活动导致介电材料的逐渐退化（老化），介电材料受到高电应力，并最终导致故障。通过绘制放电的表观电荷、相位角和重复率与工频正弦曲线的关系，可以产生相位分辨局部放电（PRPD）图案，其可以用于识别高压设备中存在的缺陷的类型。目前，PRPD模式的视觉解释只能由具有专业知识的工程师执行。由于需要不断收集和分析大量数据，因此实施起来非常耗时和昂贵。 该项目建议开发一种系统，该系统可以通过使用计算机视觉分析PRPD模式来自动检测和分类绝缘缺陷。与其他人工智能技术不同，计算机视觉更类似于人类解释PRPD模式的方法，因此保留了PRPD分析的基本原则，这些原则与直接从视觉图像中提取信息有关。该项目将涉及实验工作，从各种绝缘缺陷中产生标准化的PRPD图案图像，开发用于处理的仪器，以及设计，训练和验证用于自动缺陷检测，分类和严重性报告的计算机视觉和机器学习模型。 PRPD分析的自动化可以使高压设备的状态诊断更快、更准确。此外，它将使该技术更容易实施，更具成本效益，这反过来将使更广泛的高压资产的监测成为可能。及时可靠的诊断可以为维护或更换决策提供信息，同时提供额外的好处，包括防止设备的灾难性故障，提高人员的安全性，并最大限度地减少因计划外停机造成的损失。项目产出可用于实现行业实践和资产管理方法的现代化，从而实现更可持续的能源基础设施。

Partial Discharge (PD) analysis is an electrical method for detecting incipient faults in electrical insulation. PD activity causes progressive degradation (ageing) of dielectric materials that are subjected to high electrical stress and can eventually lead to failure. By mapping the apparent charge, phase angle and repetition rate of the discharges with relation to the power frequency sinusoid, phase-resolved partial discharge (PRPD) patterns can be produced which can be used to identify the type of defect present in high voltage equipment. Presently, visual interpretation of PRPD patterns can only be performed by engineers with specialist knowledge. It is very time consuming and expensive to implement due to the large volume of data that needs to be continuously collected and analysed. The project proposes the development of a system that can automate the detection and classification of insulation defects by using computer vision to analyse the PRPD patterns. Unlike other artificial intelligence techniques, computer vision is much more akin to the human approach of interpreting PRPD patterns, hence retaining the fundamental principles of PRPD analysis which relate to the extraction of information directly from visual images. The project will involve experimental work to produce standardised PRPD pattern images from a variety of insulation defects, the development of instrumentation for their processing, and the design, training and validation of computer vision and machine learning models for automated defect detection, classification, and severity reporting. Automation of PRPD analysis can make condition diagnosis of high voltage equipment faster and more accurate. Furthermore, it will make the technique easier and more cost-effective to implement, which in turn will enable the monitoring of a wider range of high voltage assets. Timely and reliable diagnosis can inform maintenance or replacement decisions while providing additional benefits including prevention of catastrophic failure of equipment, increased safety for personnel, and minimisation of losses due to unplanned downtime. The project outputs can be used to modernise industry practices and asset management methodologies leading to more sustainable energy infrastructure.