In order to reduce testing costs and shorten the design cycle, a data-physics-driven performance prediction method for fiber-reinforced composites was proposed. Three directions of stress were regarded as the output variables, while three directions of fiber-reinforced composite strain were considered as the input variables. To enhance the accuracy and efficiency of strength performance prediction, an artificial neural network was used to develop a constitutive model, which was then incorporated into the finite element simulation analysis process. The computation time is 143 times faster than with pure finite element analysis, and the outcomes conclusively demonstrate the viability of a data-driven approach for predicting the performance of composite materials.