Study On Acoustic Diagnosis Method For Mechanical Faults Of Power Transformer

As an important hub equipment in power grid,power transformer plays a key role in the safe operation of power system.During the operation of transformer,the vibration of transformer will produce acoustic signal,which contains rich transformer status information.Transformer fault diagnosis can be realized by accurately detecting and analyzing the transformer acoustic signal,thereby reducing the probability of major accidents in the power grid.Aiming at the existing problems in acoustic diagnosis of transformers such as acoustic denoising and fault diagnosis methods,this paper reveals the time-frequency characteristics of transformer acoustic signal based on transformer vibration mechanisms and mechanical fault simulation experiment,proposes a blind source separation method for transformer acoustic signal based on sparse component analysis,and establishes a transformer mechanical fault identification model based on stacked auto-coder and incremental extreme learning machine.The specific research content is as follows:By analyzing the acoustic mechanism of power transformer core and winding,the characteristics of transformer acoustic signals under different mechanical conditions are revealed.A transformer acoustic acquisition system and a mechanical fault simulation test platform were designed to simulate the normal operation of the transformer,iron core looseness,and winding looseness.The work provided theoretical basis and data support for subsequent transformer acoustic signal denoising and mechanical fault identification.In order to solve the problem of many interference signals in transformer acoustic signal,a blind source separation method for transformer acoustic signal based on sparse component analysis is proposed.Firstly,the substation mixed acoustic signal collected by the acoustic sensor is converted to the time frequency domain through short time Fourier transform,and the single source point of the signal is obtained by detecting the phase difference of each component of the mixed acoustic signal;Then,a spatial density clustering algorithm is used to cluster the time-frequency single source points of the mixed acoustic signal to obtain the number of acoustic sources and the estimation of the mixing matrix;Finally,based on compressed sensing theory,the separation of transformer sound signals and their interfering signals is realized.Simulation and field test results show that this method can effectively separate transformer sound signals from mixed sound signals in substations under undetermined conditions.An acoustic diagnosis model for transformer mechanical faults based on stacked spares auto-encoder and incremental extreme learning machine is established for transformer voiceprint feature extraction and mechanical faults recognition.The original acoustic signal of the transformer is dimensionally reduced by using complementary ensemble empirical mode decomposition,and the obtained time-frequency energy matrix is used as the input of the stacked spares auto-encoder network to adaptively mine the deep voiceprint features of the acoustic signal,solving the problems of strong subjectivity and easy loss of feature information in traditional feature extraction methods.Based on the supervised learning mechanism of transformer voiceprint features and incremental extreme learning machine,a mapping relationship between transformer voiceprint features and mechanical status is established,and intelligent recognition of transformer mechanical faults is realized.The acoustic diagnosis model for transformer mechanical faults is optimized and trained using the sound data set of the mechanical fault simulation experimental platform.The results show that the model has good performance in transformer mechanical fault identification.