Pattern Recognition and Modelling for Transformer FRA Data Management

Abstract

Mechanical faults of transformer winding is one of the main transformer failure modes. Frequency Response Analysis (FRA) has been found as the most sensitive and effective technique for detecting winding deformation. However, the key of FRA applications lies in the correct interpretation of FRA measurement results. Sensitivity study, through either experimental or simulation approaches, is undoubtedly a productive method to develop interpretation guidelines. Simulation approach based on a valid transformer model is preferred as it provides a flexible test bed for the investigation of FRA physics. The study of FRA characteristics in the low frequency region is a grey area. Transformer core has a dominant effect on FRA in this region, however, only a small number of studies have researched into the impact of changing magnetisation states on FRA fingerprints, most of which are through the experimental methods. Therefore, a transformer model specifically for studying FRA low frequency characteristic is developed in this PhD work. Based on Principle of Duality, it provides a topology correct transformation from core magnetic path to its equivalent electrical circuit. Transformer series capacitance, ground capacitance and interwinding capacitance are considered and included into the model. In particular, the core magnetising inductance is represented by nonlinear inductor with embedded λ-i curve so that the nonlinear characteristics of the core can be reflected. FFT is needed to covert the obtained time domain signals to the frequency domain to obtain the FRA fingerprint. The influencing mechanism of transformer core is investigated through simulation scenarios with different core magnetisation states, which are achieved by quantitatively assigning different initial values of residual magnetisation to core magnetising inductance. The effect of the magnitude of the residual magnetisation, the different distributions of the residual magnetisation in the core and the magnitude of the injected voltage are investigated by means of the sensitivity study. The medium and high frequency regions of FRA characteristics tend to be controlled by transformer winding itself and the interaction with other windings. In practice, transformer modelling is often highly dependent on the design parameters, which are often not available, and this constrains the FRA interpretation based on white box modelling. Therefore, a corresponding grey box model is developed in this PhD work. The measured FRA at the winding terminals, especially the key information of resonance and antiresonance points, are used as the model inputs. The relationships between the transformer internal geometric dimensions and the electrical parameters of the circuit model are described by analytical equations, furthermore, the Genetic Algorithm (GA) is utilised for searching the unknown geometric dimensional data, the square of relative error of FRA magnitude is used in the fitness function to quantify the similarity between the reference and estimated FRA. The developed model is validated by applying itself on two single windings with different structure, one is a continuous disc winding (the low series capacitance winding type) and the other is an interleaved disc winding (the high series capacitance winding type), and a single phase two winding structure, respectively. Such a grey box modelling approach provides an alternative for developing FRA interpretation when white box modelling technique is constrained. It is envisaged that such a model can be then used to produce an FRA fingerprint library of different types of winding deformation for fault identification, localisation and severity.

Date of Award	13 Nov 2023
Original language	English
Awarding Institution	The University of Manchester
Supervisor	Haiyu Li (Co Supervisor) & Zhongdong Wang (Main Supervisor)