Abstract:To enhance the intelligentization of power system equipment and augment the fault diagnostic competencies of technical personnel，this paper addresses the research gap in acoustic signal analysis for fault diagnosis of transformer fans. This paper presents a method that the voiceprint recognition of motor based on two-stage cerebellar model neural network（TCMNN） and feature extraction. Firstly，the experimental data of voiceprint is collected from a small fan in main transformer. The methodology proposed in this paper employs discrete wavelet transform and principal component analysis for feature extraction，where the selection of wavelet basis functions and decomposition levels is optimized based on wavelet energy analysis to enhance computational efficiency. Secondly，TCMNN is designed for fault detection and fault type identification， effectively enhancing classification performance. Experimental results demonstrate that the proposed framework achieves superior classification accuracy and enhanced robustness compared to conventional methods under white noise conditions.