Abstract: Broad learning by variational Bayesian is proposed to address the issue of overfitting in the model during fault diagnosis.Firstly, the combination of wavelet packet transformation and fast Fourier transformation is used to decompose the original vibration signal, reconstruct its basic waveform, and extract fault sensitive features.Subsequently, a fault diagnosis model is constructed using a broad learning system, and the weight matrix is optimized using variational Bayesian methods to update the variational distribution parameters and estimate the posterior distribution of its weight matrix.Fault diagnosis is then performed, effectively avoiding overfitting issues.The experimental results show that this method has improved accuracy, precision, and recall by 9.34%,5.64%,and 7.65%,respectively, and has better stability.