基于时频特征融合的电力变压器绕组故障声纹诊断OA

To address the challenges of limited data on faults in power transformer windings and the insufficient accuracy of traditional feature extraction diagnostic methods,this study proposes a diagnostic approach based on the fusion of acoustic fingerprint and time-frequency domain features,aiming to enhance the early identification capability of transformer fault risks under complex operating conditions.A physical model of power transformer winding is constructed in the laboratory to collect acoustic fingerprint signals under fault conditions.Fast Fouri-er transform(FFT)is employed to obtain frequency domain features,and a convolutional neural network(CNN)model is developed for fault identification.Comparative experiments are designed to validate the diag-nostic effectiveness,using accuracy as evaluation metrics,and comparing with traditional methods that only in-put time-domain features or frequency-domain features.Experimental data indicate that,in the experiment to determine the presence of faults,the accuracy rate has increased by 4.00 percentage point compared with the case when only time-domain features are used as input,and it has increased by 3.34 percentage point compared with the case when only frequency-domain features are used as input.The diagnostic method based on the fu-sion of acoustic fingerprint and time-frequency domain features overcomes the accuracy bottleneck of single fault feature extraction under small sample conditions.The collaborative extraction mechanism of time-frequen-cy domain features significantly enhances the expressive power of fault features.Experimental results demon-strate that this method can effectively diagnose faults in power transformer windings.