The mechanical properties of ceramic matrix composite structures are predominantly influenced by their micro-scale damage evolution features, given that material failure fundamentally stems from the initiation and progression of microstructural damage. To precisely characterize the micro-damage behavior of composites, multi-scale analysis methods grounded in parameter cross-scale transfer have been extensively employed in the study of composite mechanical properties. Nevertheless, traditional multiscale methods are plagued by low computational efficiency and a geometric increase in computational effort with nested scales, rendering them ill-suited for real-time engineering applications. To tackle this issue, this work puts forward a concurrent multiscale modeling method based on self-consistent clustering analysis. This method utilizes clustering to reduce the dimensionality of the microscale stress/strain field. It substitutes full-scale microscale finite element computations with homogenized response results, thereby substantially enhancing the efficiency of multiscale computations. Numerical results reveal that for both unidirectional and woven composite material cases, the proposed method achieves an overall improvement in computational efficiency of approximately one order of magnitude (10-15 times faster) while keeping computational errors below 3%. By effectively cutting down on computational costs while maintaining high accuracy, this approach provides new insights and technical means for efficient damage analysis and life prediction of composite structures.