Research On On-line Monitoring And Fault Diagnosis Of The Insulation Condition For Capacitive Equipments

To monitor and diagnose the insulation condition of capacitive equipments on-line is the precondition and basis to realize the condition-based maintenance (CBM) for capacitive equipments. So it is very useful and meaningful to study and apply advanced monitoring and diagnosis techniques. In this dissertation, the high accuracy measurement of tanδ, preprocessing of the on-line data, analysis of the data regularity and application of the novel diagnosis approaches for capacitive equipments are investigated, and two practical on-line monitoring systems have been developed. The main researches and achievements are as follows:Based on the Blackman-harris window, an improved harmonic analytical algorithm is presented to measure dissipation factor (tanδ). Through the simulation and practical tests, compared with the normal harmonic analytical algorithm, the improved algorithm can restrain the leakage problem of asynchronous sampling caused by the fluctuation of system frequency, and has higher accuracy.Due to large amount and redundancy of the on-line monitoring data, a processing approach based on the principle component analysis is presented to compress the on-line monitoring data. By the orthogonal transformation, few principle components of the data which reflect the basic variation regularity of the data, can be extracted. Then the efficiency of the information management and further diagnosis can be improved.The on-line monitoring data series of capacitive equipments are analyzed in frequency domain. Based on the frequency spectrum characteristic of the series, the model of the tanδ data series is constructed. Then a preprocessing approach based on the adaptive generalized morphological filter is proposed. The results of the simulation study and field data processing show that the generalized morphological filter can suppress different kinds of the white noise and pulse noise among the data. The principle of the trend-extracting approach through relative comparison is introduced and its problems are pointed out too. So the wavelet transform and morphological filter used to extract the data trend are proposed. Both approaches are based on the filtering idea. One is constructing the low-pass filter through the wavelet transform. The other is constructing the filtering window through the basic morphological transform. Finally, analysis results of the practical data verify the effectiveness of the above approaches. The influential regularity of each factor on tanδ series is analyzed, and the diagnosis idea based on the regularity is presented. Then the grey relational analysis is introduced to quantify the similarity of tanδ data series and each factor data series, through sorting grey relational degree to realize the fault diagnosis. Meanwhile, the analytical results show that the approach is valid.To mining the latent valuable information from a mount of the on-line monitoring data, based on the rough set theory (RST), a regularity-mining approach for the on-line monitoring tanδ data is proposed. First, the continuous data are discretized to form the information table. Then through setting the support and confidence to preprocess the discretized data, the general insulation regularity of the analyzed equipment can be obtained through the RST reduction. Finally, two advanced hierarchical and distributed on-line insulation monitoring systems are developed. In hardware design, the sampling unit (SU) based the voltage-divided capacitor sampling is developed to obtain leakage current signals, and an RS-485 fieldbus is applied to realize the distributed data sampling. The modularization design idea is employed in the software development. According to the demand of the functions, three individual software systems are developed. Each keeps relative independence, which is easy for the maintenance and update. It can be seen from the detection and preliminary operation that the system is safe and reliable, and the sampling result is accurate.