基于多指标优化 TQWT 稀疏表示的轴承故障诊断

[Objective] Aiming at the problems of low precision in bearing vibration fault feature extraction, fault information easily submerged by strong noise, lack of reasonable evaluation indices for adaptive signal construction, and data redundancy, a study on bearing fault feature enhancement and fault diagnosis was conducted. [Methods] A sparse decomposition method with non-convex regularization coefficients wass proposed, and adaptive signal reconstruction is realized based on a multi-index feature matrix. Firstly, the signal is sparsely decomposed via tunable Q-factor wavelet transform (TQWT) to obtain its most concise representation. A non-convex penalty function is constructed to improve the sparsity of the reconstructed signal while retaining key fault feature information. Secondly, a fused index parameter matrix was established by integrating harmonic-to-noise energy ratio, kurtosis-skewness ratio, and sparsity, which was adopted to optimize the parameters of tunable wavelets and the threshold of sparse penalty terms. Finally, signal denoising and reconstruction were completed using the optimized parameters, and demodulation analysis was performed on the reconstructed signal to extract bearing fault feature information. [Results] Simulation and test results show that the proposed method can effectively extract the impact features of early bearing faults and achieve accurate diagnosis of bearing vibration faults.