基于红外热像技术的电接触故障检测方法研究

发布时间：2018-02-20 04:50

本文关键词： 电接触红外热像故障等级识别图像去噪图像分割概率神经网络　出处：《湖南大学》2015年硕士论文　论文类型：学位论文

【摘要】：电接触是指电气设备和电气线路中一切导电的连接部位,它具有数量庞大、类型繁多、故障高发等特点。因此,电接触的状态检测与故障诊断对整个电力系统的安全稳定运行至关重要。本文研究了基于红外热像特征与概率神经网络的电接触故障检测的新方法。主要针对高压隔离开关、液压式耐张线夹、高压套管将军帽三类电接触进行故障研究与分析,在综合考虑空气污染指数、环境温度对检测结果准确性影响的基础上,从理论、实验等方面系统地验证了该方法的准确性和有效性。本文采用高分辨率红外热像仪采集电接触图像。针对电接触红外热像对比度低、噪声大的特点,提出了一种基于总体最小二乘法(Total Least Squares,TLS)估计的遗传小波红外热像去噪方法。该方法采用总体最小二乘法获得图像的小波系数估计,小波逆变后得到的去噪图像作为父本,然后与通过维纳滤波后得到母本进行遗传运算,最终得到的子代作为去噪后图像。该方法充分利用两个亲本的优势,逐代进化产生优良个体,使去噪后图像同时具有以上两种方法的去噪优势。仿真结果表明,去噪后的红外图像相比于常规的去噪方法结果具有更高的信噪比和更小的最小均方误差。针对现场所拍摄的红外图像所包含的信息繁多,为了能从复杂的背景中提取目标物,本文提出了一种改进的自适应遗传算法对电接触红外热像进行分割。该方法综合考虑了适应值与相似度的适应度选择,引入进化因子,根据个体适应度和进化代数自动调整交叉概率和变异概率。把图像分割最佳阈值选取转换成最优问题,利用改进算法的寻优高效性求解最优阈值实现图像分割。仿真结果表明,新算法比常规的阈值分割算法更能清晰、完整地分割出目标区域,并缩短了最佳阈值寻优时间,提高图像分割的质量和速度。针对电接触热故障等级判别,本文设计了概率神经网络分类器。在综合考虑了空气污染指数、环境温度以及红外热像的温度信息基础上,利用概率神经网络分类器建立了三类不同电接触在不同空气污染等级和不同环境温度下与故障等级的映射关系。利用概率神经网络具有较强的容错能力和结构自适应性,实现了多个故障等级的准确快速识别。
[Abstract]:Electrical contact refers to the electrical equipment and electrical circuit in all conductive connection parts, it has a large number, various types, high frequency of faults and so on. The state detection and fault diagnosis of electric contact is very important for the safe and stable operation of the whole power system. In this paper, a new method of electrical contact fault detection based on infrared thermal image feature and probabilistic neural network is studied. Three kinds of electric contact of hydraulic tension clamp and high pressure bushing general cap are studied and analyzed. On the basis of considering the influence of air pollution index and ambient temperature on the accuracy of the test results, the theory is presented. The experimental results show that the method is accurate and effective. In this paper, the high resolution infrared thermal imager is used to collect electrical contact images. In this paper, a genetic wavelet infrared thermal image denoising method based on Total Least SquaresTLS) estimation is proposed, in which the wavelet coefficients of the image are estimated by the total least square method, and the denoised image obtained by the wavelet inversion is used as the parent. Then genetic operation is carried out with the female parent obtained by Wiener filter, and the final offspring is used as the denoised image. This method makes full use of the advantages of the two parents and evolves to produce good individuals from generation to generation. The de-noised image has the advantages of the above two methods simultaneously. The simulation results show that, Compared with the conventional denoising method, the denoised infrared image has higher SNR and smaller minimum mean square error. The infrared image taken in the field contains a great deal of information, in order to extract the object from the complex background. In this paper, an improved adaptive genetic algorithm is proposed to segment the infrared thermal image of electric contact. The crossover probability and mutation probability are automatically adjusted according to individual fitness and evolutionary algebra. The optimal threshold selection of image segmentation is transformed into an optimal problem, and the optimal threshold is solved by the improved algorithm. The simulation results show that, Compared with the conventional threshold segmentation algorithm, the new algorithm can segment the target area more clearly and completely, shorten the optimal threshold searching time, and improve the quality and speed of image segmentation. In this paper, a probabilistic neural network classifier is designed. On the basis of considering the air pollution index, ambient temperature and the temperature information of infrared thermal image, Using probabilistic neural network classifier, the mapping relationship between three kinds of electrical contact and fault level under different air pollution levels and different ambient temperatures is established. The probabilistic neural network has strong fault tolerance and structural adaptability. The accurate and fast identification of multiple fault levels is realized.
【学位授予单位】：湖南大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TM501.3

【参考文献】