红外热成像技术在电气设备故障识别中的应用研究

发布时间：2018-04-21 02:08

本文选题：红外热像 + 电气设备　；参考：《沈阳工业大学》2015年硕士论文

【摘要】：监测电气设备的热状况,在对于保持电气系统的可靠性上是非常必要的。电气设备中的老化等因素将导致过热状况的产生,而这些热问题最终会导致设备故障的产生。此外,设备故障产生后需要花费大量的维修成本、人力,还有可能变成灾难,造成人生伤害甚至是死亡。因此,识别设备运行状态是否处在正常状态下的这一过程,对于维持系统的可靠性和稳定性是非常重要的。如今,红热成像技术由于其快速、可靠、非接触性以及良好的经济性能,被广泛的应用于设备的故障检测及诊断上。本文提出了一种利用红外热像仪获取电气设备红外温度热像图,并根据数据进行分析来评价设备热状态的方法。电气设备的红外温度热像图,是在设备不解体、不影响设备运行的情况下通过红外热成像仪获取。选用的红外热像仪是由日气NEC公司出产的。获取红外温度热像图后,首先,对红外温度热像图进行手动分割获取有效区域,继而,进行预处理和图像增强。图像增强中,本文采用了三种方法进行对比,直方图均衡化、二维离散小波变换和二维经验模态分解。随之,提取故障部位与相关组件的不同的一阶直方图特征和灰度共生矩阵特征作为特征量,共计22个特征量。使用主成分分析与判别分析相结合的方法进行特征的优化选取,先利用主成分分析法对特征量进行初步的选择,从22个特征量中选择出15个特征量,再利用判别分析法从15个特征量中剔除5个特征量。最终从22个特征量中选择出10个特征量作为状态分类系统的输入。在最终的状态分类上采用两种方法：判别分析法和人工神经网络方法。将利用判别分析分类的结果与利用神经网络方法进行分类的结果进行对比。结果表明,利用判别分析进行故障状态分类相较于人工神经网络有更好的性能体现,最优结果为精确度82.6%。
[Abstract]:Monitoring the thermal condition of electrical equipment is necessary to maintain the reliability of electrical system. Factors such as aging in electrical equipment will lead to overheating, and these thermal problems will eventually lead to equipment failure. In addition, equipment failure will cost a lot of maintenance costs, manpower, and may become a disaster, causing life damage or even death. Therefore, it is very important to identify whether the equipment is in the normal state or not, to maintain the reliability and stability of the system. Nowadays, red thermal imaging technology is widely used in fault detection and diagnosis of equipment due to its rapid, reliable, non-contact and good economic performance. In this paper, a method of obtaining infrared thermogram of electrical equipment by infrared thermal imager and analyzing the data to evaluate the thermal state of the equipment is presented. The infrared thermogram of electrical equipment is obtained by infrared thermal imager without disintegrating the equipment and not affecting the operation of the equipment. The selected infrared thermal imager is produced by NEC. After obtaining the infrared thermogram, the infrared thermal image is segmented manually to obtain the effective region, and then the preprocessing and image enhancement are carried out. In image enhancement, three methods, histogram equalization, two-dimensional discrete wavelet transform and two-dimensional empirical mode decomposition, are used. Subsequently, different first-order histogram features and gray level co-occurrence matrix features of fault location and related components are extracted as feature quantities, and a total of 22 feature quantities are extracted. The method of combining principal component analysis (PCA) with discriminant analysis is used to optimize the selection of features. First, the method of principal component analysis is used to select the characteristic quantity, and 15 characteristic quantities are selected from the 22 characteristic quantities. Then the discriminant analysis was used to eliminate 5 characteristic variables from 15 characteristic variables. Finally, 10 of the 22 feature variables are selected as the input of the state classification system. Two methods are used in the final state classification: discriminant analysis and artificial neural network. The results of discriminant analysis and neural network are compared. The results show that the performance of fault state classification by discriminant analysis is better than that of artificial neural network, and the best result is 82.6 accuracy.
【学位授予单位】：沈阳工业大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP391.41;TN219

【参考文献】