Abstract: Carbon fiber composites produce a large number of acoustic emission signals in tensile damage experiments. The data of acoustic emission signals are analyzed to find out the damage evolution law of carbon fiber composites. The data is clustered and analyzed, and is further divided into multiple clusters composed of similar objects to find the correspondence between the cluster and the damage. The model of tensile damage identification of carbon fiber composites was obtained by modeling the data after clustering. Since the feature of the acoustic emission signal is a multi-dimensional vector, there is a certain relationship between the features. In order to improve the modeling speed, the data needs to be dimension-reduced to select the characteristics of the main influencing factors. To this end, the genetic algorithm is used to reduce the dimensionality of the data, and the redundant features are removed, and the main features are retained. Finally, the data before and after processing are brought into the BP neural network to identify the damage. The experimental results show that more optimized results can be achieved by genetic algorithm for dimensionality reduction with shorter modeling time and better recognition efficiency.