A Hybrid Modelling And Evidence-based Fault Diagnosis Approach To Power Transformer Winding Deformation Detection

电力变压器的设计是为了承受各种运行事件产生的机械力，如过电压和雷击，这些事件可能会导致绕组变形或位移。在电力变压器故障诊断的各种技术中，频率响应分析(FRA)可以给出绕组变形故障的指示，而不需要昂贵的中断操作来打开变压器油箱，可以最大限度地减少对系统运行的影响和对客户的供电损失，从而节省数百万英镑的及时维护。然而，在工业实践中，FRA通常被用作一种比较方法，通过将测试频率响应与参考集进行比较，这不能提供对变压器内部故障的深入了解。为了在开发绕组模型中使用FRA，已经开展了一系列研究活动，但存在局限性，如模型过于复杂、计算时间过长以及在1 MHz至10 MHz的高频范围内响应不准确。建议的研究是建立在利物浦已经获得的经验的基础上，并开发一个准确的绕组模型和可靠的故障诊断方法。通过修正Rudenberg对每个圆盘的分析方法和结果，并将每个圆盘的行波方程连接成多导体传输线(MTL)模型，建立了一种新的混合绕组模型。这可以显著降低模型的阶数，但在高频范围内具有良好的建模精度，从而允许访问任何所需绕组匝数的电流和电压。混合模型的电参数将用有限元方法估计，并用基于实际FRA测量的进化算法进一步辨识。提取特定绕组故障与绕组参数之间的特征特征，可用于检测和区分绕组变形故障。然后，将混合模型的仿真用于提取高频故障指纹，以提高对绕组微小变化的检测，并将通过实验室研究进一步检验和验证。在典型绕组故障诊断中，一般采用定性和定量相结合的诊断方法，这些判断往往是不完整和不精确的，可以作为证据。证据推理(ER)算法非常适合于将这类证据与坚实的数学基础相结合。在本项目中，将构建一个基于证据的故障诊断系统，以收集诊断信息并处理不确定性，以实现可靠的绕组故障诊断。这项工作将作为利物浦大学、奥米克龙公司和NG公司之间的合作项目进行，汇集了变压器测试、建模和故障诊断领域的学术和工业专业知识。提出的研究成果将是新的混合绕组模型和基于证据的绕组故障诊断系统。这种新方法旨在提高对绕组多频信号传播的基本认识，从而能够在低频和高频范围内提取故障指纹，并为早期故障检测和定位提供新的诊断规则。提取的高频故障指纹将为早期故障检测提供可行的解决方案，这可以帮助法兰克福机场检测试剂盒制造商(如奥米克隆公司)充分了解法兰克福机场管理局，提高测试试剂盒的精度。开发的基于证据的绕组故障诊断系统可以为公用事业公司(如NG)在处理大量FRA记录时进行可靠的故障诊断提供有用的决策支持工具。

Power transformers are designed to withstand the mechanical forces arising from various in-service events, such as over-voltage and lightning, which may cause deformation or displacement of winding. Among various techniques applied to power transformer fault diagnosis, frequency response analysis (FRA) can give an indication of winding deformation faults without expensive and interruptive operations of opening a transformer tank, which can minimise the impact on system operation and loss of supply to customers and consequently save millions of pounds in timely maintenance. However, in industrial practice, FRA is always used as a comparative method, by comparing a test frequency response with a reference set, which cannot provide an insight understanding of transformer internal faults. A range of research activities have been undertaken to utilise FRA in the development winding models but with limitations, such as too complicated models, large computation time and inaccurate responses in the high frequency range between 1MHz and 10MHz. The proposed research is to build on the experience already gained at Liverpool and to develop an accurate winding model and a reliable fault diagnosis approach. A new hybrid winding model will be developed by modifying the analytical approach and results of transformer winding analysis obtained by Rudenberg for each disc, and subsequently connecting the travelling wave equation of each disc in a form of Multi-conductor Transmission Line (MTL) model. This can significantly reduce the order of the model yet with good modelling accuracy in the high frequency range, which allows access to the current and voltage at any desired turns of a winding. The electrical parameters of the hybrid model will be estimated with the finite element method (FEM), and further identified with evolutionary algorithms based on actual FRA measurements. The characteristic signatures between particular winding faults and winding parameters will be derived, which can be employed to detect and distinguish winding deformation faults. Then, the simulation of the hybrid model will be used to extract high frequency fault fingerprints of FRA for improving the detection of small winding changes, which will be further examined and verified through laboratory studies. For typical winding fault diagnosis, both the quantitative and qualitative judgements are generally considered, which can be treated as evidence and are often incomplete and imprecise. The Evidential Reasoning (ER) algorithm is very suitable for combining such evidence with a firm mathematical foundation. In this project, an evidence-based fault diagnosis system will be constructed to aggregate diagnosis information and deal with uncertainties for reliable winding fault diagnosis. The work is to be carried out as a collaborative project between the University of Liverpool, OMICRON and NG, bringing together academic and industrial expertise in the field of transformer test, modelling and fault diagnosis. The outcome of the proposed research will be the new hybrid winding model and the evidence-based winding fault diagnosis system. The new approach aims to improve the fundamental understanding of multi-frequency signal propagation across a winding, which will allow extracting fault fingerprints in both the low and high frequency ranges and provide new diagnostic rules for early fault detection and location. The extracted high frequency fault fingerprints will provide a feasible solution for early fault detection, which can assist a FRA test kit manufacturer, e.g. OMICRON, in fully understanding FRA and improving test kit precision. The developed evidence-based system for winding fault diagnosis can be a useful decision support tool for utility companies, e.g. NG, for reliable fault diagnosis yet with high efficiency, when processing numerous FRA records.