Early Fault Diagnosis And Reliability Analysis Of Power Transformers Based On Kernel Density Esimation And FTA

Power transformer is the core of the energy conversion and transmission grid,smooth running of it without interruption is vital to efficiency,stability and reliability of power systems.Chinese power system facilities were mainly constructed in the 1980 s and 1990 s,which made many transformers have been in service for nearly 40 years.Those running transformers are facing increasingly serious problems of equipment failure and insulation aging,and have increasing probabilities of an accident.Transformer failures may cause significant economic losses and even serious social impacts.Therefore,effective fault diagnosis and reliability analysis for transformers have important theoretical and practical value for guiding transformer maintenance strategy and asset management,and reducing the probability of power accidents.This thesis focuses on early fault diagnosis of power transformers bassed on dissolved gas analysis and reliability analysis of power transformers based on fault trees and binary decision diagrams.The main work and contributions of this thesis are as follows,(1)Early fault diagnosis based on dissolved gas analysis can identify the exact fault type of transformers and provide information for the condition-based preventive maintenance decisions.This thesis proposes a new fault diagnosis model based on kernel density estimation and dissolved gas anlysis,in which the Bayesian formula is in combination of the diffusion-based kernel density estimator to describe the non-linear and multimodal chracteristics of fault data distribution and to enhance the correctness of fault diagnostics.The causes of different early faults are analyzed,the corresponding mataintenance measures are given out.(2)Fault tree analysis is a traditional method for reliability or risk assessment.Structure and function characteristics of transformers are analysed first,and the corresponding failure mode and effect analysis is conducted combined with the early fault analysis of power transformers in this thesis.A new fault tree model of transformers is constructed then,in which environment and human factors are both considered,while the structural dependency and the functional dependency between subsystems and components within transformers are also considered.(3)Modularization is an important preprocessing for analyzing large and complex fault trees,it can improve the efficiency while reduce the memory consumption of computers during the analysis process.Based on the concept of lowest common ancestor in graph theory,a new modularization algorithm,i.e.,the B&T algorithm(Branching and Transforming Algorithm)is proposed.The B&T algorithm is composed of two main steps,i.e.,the branching process and the transforming process.And the not-module nodes are traversed and labeled out during the transforming process,thus modules are obtained by deleting not-module nodes form gate events.The proposed B&T algorithm is easy to code and more efficient when compared with other linear-time modularization algorithms.(4)A new ordering heuristic is developed in this thesis based on the proposed B&T algorithm for fault tree analysis.Basic event nodes within the same module are ordered together by a post-order depth first traversal on the auxiliary tree graph generated by the B&T algorithm.Beides,structure of the original fault tree and the repeated nodes within it are both considered in this heuristic.And the new ordering heuristic is applied to transforming the fault tree model of transformers into the binary decision diagram model for verification.Based on the binary decision diagram model obtained,reliability analysis of transformers is conducted.The maximum time period between maintenance is given out for the preventive maintenance plans.