Automation and Optimisation of Wavelet Transform Techniques for Partial Discharge Denoising, and Pulse Shape Classification, in Power Plant

在高压设备失效之前，通常存在绝缘系统发生劣化的时期，这可能是若干年。改进设备状况和预期寿命评估的关键在于确定和描述退化阶段。人们普遍认为，无论原因如何，降解阶段都会在降解部位产生小火花。这些电火花通常被称为局部放电（PD）。火花的特征受到故障点材料和应力的影响。改进其检测和特性鉴定将提供关于降解的位置、性质、形式和程度的信息。这些PD脉冲非常小，因此，无论应用何种特定策略来检测它们（电或声），检测设备必须非常灵敏。在现场，这使得它容易受到周围环境和电气/机械基础设施的影响或外部干扰或“噪音”。最好的情况下，这会导致数据损坏并损害状况评估的效率。在最坏的情况下，它停止了任何使用的技术作为“噪声”信号超过局部放电活动的水平。为了解决与噪声相关的问题，已经尝试了许多方法，如：筛选和过滤，模拟带通滤波，匹配滤波器，极性鉴别电路，时间窗口方法和数字滤波器的应用。本文作者在最近的一项研究中比较了匹配滤波器、传统滤波器和离散小波变换（DWT）在局部放电测量去噪中的应用，证明了DWT在强噪声存在的实际测量中是最好的解决方案。小波变换是唯一一种可以重构局部放电脉冲的方法，它是从傅里叶变换（FT）发展而来的，特别适合于分析瞬态、不规则和非周期信号。理想地，如果可以选择小波以匹配PD脉冲形状，则可以从任何强噪声信号中提取PD脉冲。虽然WT比FT产生更多的信息，但它本质上比FT更复杂，并且涉及依赖于从噪声数据中提取的信号的形状、记录长度和采样率的过程。GCU绝缘诊断小组的Zhou博士是第一个研究最合适小波的最佳选择、最佳层数和基于电平的阈值算法的人，用于使用DWT从电噪声环境中自动提取PD脉冲。这一创新性的工作已被证明是有效的，在一些测量平台。然而，在不同形状的脉冲存在的时刻，DWT的应用仍然需要大量的经验。本研究的目的是在GCU已有的经验和成功的基础上，开发一种方法，使离散小波变换能够应用于各种局部放电测量系统，而不管它们的机理和带宽如何，用于局部放电数据去噪和局部放电脉冲重建和分类，研究的结果将是能够识别分析数据中包含的所有类型的瞬态脉冲并将其呈现出来的算法分别在时间域。这将允许从PD测量和phi-q-n图的产生中识别和分类各种PD活动，phi-q-n图与脉冲形状结合，提供了用于植物诊断的显著改进的手段。

Before high voltage plant fails there is generally a period when degradation of the insulation system occurs, this may be a number of years. The key to improving the assessment of the equipment condition and life expectancy lies in identifying and characterising the stages of degradation. It is widely recognised that the degradation phase, irrespective of the cause, results in small sparks being generated at the site(s) of degradation. These electric sparks are generally referred to as partial discharges(PD). The characteristics of the sparks are influenced by the materials and stresses at the fault site. Improvement in their detection and characterisation will provide information on the location, nature, form and extent of degradation.The current detection process is severely compromised in practical on-site testing. These PD pulses are extremely small and hence, irrespective of the particular strategy being applied to detect them(electrical or acoustic), detection equipment must be very sensitive. In the field, this makes it prone to the influence or external interference or 'noise' from the surrounding environment and electrical/mechanical infrastructure. At best, this results in data corruption and compromises the efficiency of the condition assessment. At worst, it stops the technique from being of any use as the 'noise' signal exceeds the level of partial discharge activity.To solve the problems associated with noise a number of methods have been tried such as: screening and filtering, the application of analogue band-pass filtering, matched filters, polarity discrimination circuitry, time-windowed methods and digital filters. Each of these is, however, applicable to only certain types of noiseIn a recent study the author compared the matched filter, the traditional filter and the Discrete Wavelet Transform (DWT) in PD measurement denoising and has proven DWT provides the best solution in practical measurement when strong noise is in presence. Furthermore, DWT is the only method which allows reconstruction of the PD pulse.Having evolved from the Fourier Transform(FT), WT is particularly designed to analyse transient, irregular and non-periodic signals. Ideally, if a wavelet can be selected to match the PD pulse shape, the PD pulse could be extracted from any strong noise signals. Though the WT generates more information than the FT, it is inherently more complex than the FT and involves procedures dependent on the shape of the signals to be extracted from noisy data, the record length and the sampling rate. Dr. Zhou in the Insulation Diagnostics Group at the GCU was the first to study the optimal selection of the most appropriate wavelets, the optimal number of levels and level-dependent thresholding algorithm for automatic PD pulse extraction from electrically noisy environments using DWT. This innovative work has been proved to be effective in a number of measurement platforms. However, the application of DWT still requires significant experience at the moment when pulses of different shapes exist. The proposed research is to build on the experience and success already gained at GCU and to develop a methodology which allows the DWT to be applied to various PD measurement systems irrespective of their mechanism and bandwidth for PD data denoising and PD pulse reconstruction and classification.The outcome of the proposed research will be algorithms which can identify all types of transient pulses contained in data under analysis and present them separately in time domain. This would allow the identification and classification of various PD activities from PD measurements and production of phi-q-n diagrams which, in conjunction with pulse shapes, provides significantly improved means for plant diagnosis.