High frequency finite element modeling and condition assessment of power transformers

Zhang, Ziwei
(2015) High frequency finite element modeling and condition assessment of power transformers. PhD thesis, University of Liverpool.

[img] Text
Available under License Creative Commons Attribution.

Download (13MB)


Due to the reforming and deregulation of electric power industry, investments in transmission equipment have drastically decreased to meet the economic needs of the competitive market. The electrical utility sector was forced to cut the costs in maintenance and operation without endangering steady supply of electrical power. With this trend, the maintenance strategy desires advanced methods for condition monitoring and assessment of in-service power transformers. Among the common condition assessment techniques, Frequency Response Analysis (FRA) is considered as an efficient off-line diagnostic technique for fault detection in transformer windings. Precise interpretation of the FRA output has proven a great challenge and attracted much effort in recent years. There is also a strong need in this research field to develop an intelligent interpretation procedure for automatic assessment of power transformer conditions. This thesis focuses on two main aspects of power transformer condition assessment: developing an accurate transformer model for the interpretation of transformer FRA responses, and establishing FRA-based novel algorithms to automatically identify transformer failure modes. Reviewing existing transformer modeling methods, this thesis explicitly introduces a simplified distributed parameter model (hybrid winding model) for FRA. The hybrid winding model has the advantage of less computational complexity and high accuracy in simulation results, even in the frequency range above 1 MHz. Analytical expressions for calculating key electrical parameters of winding models are presented. The electrical parameters of transformer windings with a complex or deformed structure are difficult to calculate using analytical formulations. Therefore, computational models based on Finite Element Method (FEM) are developed in this thesis to calculate the frequency-dependent parameters of transformer windings especially with deformed structures. These parameters are then applied to the transformer winding model for frequency response analysis. By applying the electrical parameters obtained from the FEM models, the accuracy of the hybrid winding model has been improved for cases with incipient winding faults. This methodology is implemented in simulation studies of radial winding deformation and minor axial winding movement to reveal the characteristic features of these two types of winding fault. Results show that: (1) frequency-dependent inductances and structure-dependent shunt capacitance derived from FEM models can be used in FRA analysis, (2) by using the proposed methodology, characteristics of frequency response above 1 MHz can be analyzed, (3) regarding radial winding deformation and minor axial winding movement, the changes in the electrical parameters also affect the frequency response in the high frequency range (>1 MHz). A Hierarchical Dimension Reduction (HDR) classifier built on FRA results is proposed in this thesis for condition assessment of power transformers. The algorithms of this classifier make advantageous use of advanced image processing technologies including image binalization and binary erosion in the first step of the procedure. This preprocessing procedure optimizes the measured FRA data by filtering the frequency sections with minor deviations which can effect the calculating results of the indices. Also in this step, FRA diagrams are re-scaled in a linear coordinate system for the convenience of calculating the indices in later step. Subsequently, based on the correlation between electrical properties and features of FRA responses, a division approach is proposed to dynamically divide the frequency range into 5 sub-bands. This division method of frequency range is more reasonable than the conventional methods of fixed frequency sub-bands and more applicable than other existing methods. Then the proposed algorithms of hybrid quantitative indices which include four indices are employed. The dimension reduction for the FRA data is processed by these algorithms in the 5 dynamic frequency sub-bands. It is the first time to establish the algorithms of the hybrid quantitative indices, which include two classical indices and two novel indices, for reducing the dimensions of the FRA data. Two new algorithms of indices, Area Ratio Index (ARI) and Angle Difference (AD), are proposed based on knowledge of FRA interpretation with respect to typical transformer failure modes. They are able to improve the classification performance in terms of the ability to identify electrical failures and the condition of residual magnetization. Based on these advantageous processes, the HDR classifier can aggregate related expertise and approved statistical indices in furtherance of automatic decision analysis on identifying transformer failure modes or conditions. The performance of this classifier has been verified by 32 sets of experimental FRA data, in which 20 sets are primarily used for determination of threshold values of the related algorithms and the rest 12 are purely used for the verification. Results of this implementation of the HDR classifier are 100% accuracy with using the 20 sets of training data and 95.83% accuracy with using the rest 12 sets.

Item Type: Thesis (PhD)
Depositing User: Symplectic Admin
Date Deposited: 06 Sep 2016 14:26
Last Modified: 01 Aug 2018 01:30
URI: http://livrepository.liverpool.ac.uk/id/eprint/2052862
Repository Staff Access