基于振动信号多特征值的电力变压器故障检测研究

doi:10.19595/j.cnki.1000-6753.tces.221422

Abstract
Figure/Table
References (0)
Related Citation (15)

Download: PDF (9198 KB) HTML (1 KB)
Export: BibTeX | EndNote (RIS)

Abstract

Vibration signal analysis is an essential part of power transformer fault detection. The current researches mainly compare the transformers of a specific type or a single voltage level. The resulting detection algorithm is unsuitable for the transformers of different types, and the generality is poor. This paper proposed a power transformer fault detection algorithm based on vibration multi-eigenvalues to solve the above problem. The vibration signals of transformers under normal, winding deformation, winding loosening, core loosening, and DC magnetic bias statuses were collected, involving a voltage level range of 2~500kV. The above data were classified by using traditional eigenvalues of the 100Hz amplitude proportion, the total harmonic distortion (THD) considering high-frequency proportion and THD without considering high-frequency proportion, to verify whether the traditional eigenvalues have the ability of classification. The interval coincidence degree was used to represent the classification effect. The traditional eigenvalues were interfered by the harmonics of the vibration, and the interval coincidence degree of the classification results were 66%, 70% and 66%, respectively. The results showed that the traditional eigenvalues cannot be used to diagnose the transformers of different types.
According to the analysis of the transformer vibration spectrum, the vibration of core and winding is nonlinear and has an influence on the resonance of mechanical parts. This results in a particular proportion of harmonics in the transformer vibration spectrum under normal operation, which affects the classification effect of eigenvalues seriously.
In order to optimize the eigenvalues, the vibration harmonics of lower frequency were brought into normal range. Firstly, the main vibration frequency distribution of transformers under normal and fault conditions was obtained in this paper, which were distributed within 200~500Hz in normal situations, and were evenly distributed within 150~600Hz in cases of fault. By analyzing various frequency combinations, the 200~400Hz harmonic components were considered as normal situations to maximize the classification effect of the two eigenvalues. The optimized classification interval coincidence degrees of the low-frequency proportion and the upper THD were reduced to 14% and 16%, respectively. The classification accuracy was improved by 79% and 76%, compared with the original eigenvalues.
Because different eigenvalues have different sensitivities to various faults, a two-step fault detection process including primary and secondary diagnosis was proposed to avoid the influence of multiple eigenvalues. The primary diagnosis was based on low frequency proportion, upper THD and vibration entropy, and the SMOTE algorithm was used to expand the fault data to realize the balance of the data set. After that, the truncated normal distribution was used for data fitting and constructing the diagnosis function. By calculating the probability that the sample came from a normal transformer, the result was divided into three categories: normal, uncertain and fault. Based on the odd and even harmonics ratio and the low-frequency odd and even harmonics ratio, the data diagnosed as "uncertain" in the primary diagnosis were further classified by demarcating threshold in the secondary diagnosis. The testing set was used to test the overall algorithm classification effect, and the results showed that the fault diagnosis accuracy was up to 92.6%.
In summary, the method proposed in this paper is suitable for transformer detection with different types, different measuring point distributions and different working conditions.

Key words： Power transformer vibration eigenvalue transverse diagnosis two-step diagnosis

Received: 24 July 2022

PACS:

TM41

	Service

	E-mail this article
	Add to my bookshelf
	Add to citation manager
	E-mail Alert
	RSS
	Articles by authors
	Du Houxian
	Liu Hao
	Lei Longwu
	Tong Jie
	Huang Jianye
	Ma Guoming

Cite this article:

Du Houxian,Liu Hao,Lei Longwu等. Power Transformer Fault Detection Based on Multi-Eigenvalues of Vibration Signal[J]. Transactions of China Electrotechnical Society, 2023, 38(1): 83-94.

URL:

https://dgjsxb.ces-transaction.com/EN/10.19595/j.cnki.1000-6753.tces.221422 OR https://dgjsxb.ces-transaction.com/EN/Y2023/V38/I1/83