Research On Vibration Model Of Power Transformer Based On Blind Separation

Power transformers are vital components of power network infrastructure. Failures of transformers may cause considerable economic losses and disruption of power supply. The evolution of transformer faults can often be related to changes in the transformer tank vibration. This is why several vibration-based techniques have been developed for transformer condition monitoring and fault diagnosis. Transformer tank vibration has been wildly used for the condition monitoring and fault diagnosis of power transformers. In both applications, seeking a correlation between transformer tank vibration and the vibration of the transformer windings and core, and finding proper models to reason about transformer vibration from the electric inputs, become challenges for fault identification. Since the main sources of vibration in a transformer are the magnetostrictive forces in the core and the electrodynamic forces in the windings, the vibration measured on the transformer tank is a combination of the contributions from these two forcing components, which makes the challenges even harder.In the cases of short and open circuit, the transformer vibrates caused by single forcing component (winding or core), which eliminates the effects of combination of contributions from these two components, and allows us to directly access the individual information of the core and windings from the tank vibration. However, for safety and economic concerns, these two conditions are not desirable for transformers in service. In order te solve this problem, the Blind Source Separation (BSS) techniques were adopted in this paper to infer and retrieve the contribution of each component from combined vibration under normal conditions. Then the construction of input-output models, based on the separated components vibrations, were performed to reason about components vibrations from the electric inputs and make predictions about in what manner the vibrations will behave. Simulation and experiments were also undertaken to verify the proposed methodology. The details of the content are listed as follow:In Chapter one:The research works on seeking a correlation between transformer tank vibration and health state of the internal components, and finding proper models to reason about transformer vibration from the electric inputs were both reviewed as two main issues in live monitoring and fault identification of transformer. In order to find a solution with high effectiveness and wide acceptance to fault diagnosis, a BSS based tank vibration modeling method for condition monitoring and fault diagnosis of power transformer was proposed. Moreover, the arrangements and the innovations of the thesis were both listed.In Chapter two:The purpose of this chapter is to experimentally investigate the winding radial vibration of an electrically live power transformer and characterize the changes in the spatial and frequency features of the radial vibration as various mechanical faults are introduced to the transformer winding. To avoid the effects of transducer loading and electromagnetic fields on the measurement results, a laser Doppler vibrometer was used to make non-contact measurements of the winding vibration. The results indicated that changes in spatial and frequency features of the radial vibration often corresponded to failures in the transformer.In Chapter three:This chapter discussed the generation and propagation of vibration in a power transformer, and presented a BSS model based on the vibration signals collected by individual sensors on a transformer tank.In Chapter four:This chapter presented a BSS method for separating the vibration components respectively contributed by the core and winding from transformer tank vibration collected by an individual sensor. This method is based on the method of time-frequency ratio of mixtures (TIFROM), BSS, clustering, and delay detection. The experimental results demonstrated the effectiveness of the proposed method for separation of transformer vibration.In Chapter five:This chapter aims to construct an appropriate model to represent the nonlinear transformer vibration system, with electric applies as the system inputs and the vibration of the transformer tank contributed by the core vibration or winding vibration as the outputs. A single-input and single-output (SISO) Hammerstein-type model was developed for identifying the nonlinear transformer vibration system, when the observed vibration on the transformer tank was derived into two components contributed by the individual vibration sources. The nonlinear system was identified by the Fourier neural network, which consists of a nonlinear element and a linear dynamic block. The order determination method based on the Lipschitz criterion as well as the back-propagation algorithm for weights update were both presented. The Hammerstein-type Fourier neural network-based model was tested on several transformers, giving promising results for prediction of the transformer vibration.In Chapter six:The contents of the thesis were summarized, and a future outlook of research work was also addressed and discussed.