Aiming at the difficulties of weld stress concentration and magnetic memory quantitative identification on weld latent defect levels, a dynamic fuzzy clustering model based on immune optimization algorithm is presented. The steel Q235 plate specimens with preformed incomplete penetration weld were used as the test materials, and the fatigue tensile experiments were carried out. By comparing with the Xray synchronous test results and quantitative standard, the metal magnetic memory(MMM) signal characteristic parameter vectors for different defect levels are extracted. Considering the fuzzy and uncertainty of MMM test data in critical state identification of different weld defect levels, a dynamic fuzzy clustering algorithm (DFCA) is introduced. By outputting the threshold λ, the initial fuzzy clustering classification is obtained. Furthermore, considering the problem that the DFCA is easy to fall into the local optimum, the immune algorithm with global search and parallel ability is used to optimize the DFCA to obtain an optimal threshold λ. Finally, the dynamic fuzzy clustering model based on immune optimization is established. The verification results show that the defect level prediction accuracy of the proposed model reaches 90%, which provides a new idea for the evaluation on the weld defect levels and the quantitative evaluation on the equipment safety in practical engineering.