Clustering analysis is a commonly used unsupervised method in data processing. However， the difficulty in accurately de termining clustering parameters limits the application of this method in structural damage identification. To address this issue， a non-parametric Bayesian clustering method is proposed in this study， which combines structural modal parameters for structural damage identification and quantitative analysis， thereby expanding the application range of the non-parametric Bayesian model. First， the natural excitation method is used to extract the natural frequency from the measured vibration data of the structure. Then， the non-parametric Bayesian clustering method is employed to cluster the data. Finally， maximum likelihood heteroscedastic Gaussian process regression and Bayesian factors are combined to quantitatively analyze the clustering results for damage quantita tion analysis. The results of the damage identification method are verified by the actual engineering case of Yonghe Bridge in Tian jin. The results show that this method can accurately cluster the natural frequency data and identify the different damage states of the structure without the need to pre-set clustering parameters.