When analyzing the fracture behavior of unidirectional carbon fiber-reinforced polymer (CFRP), it is important to consider the interfacial strength between the reinforcing fiber and the base resin, and the strength of the base resin. Therefore, the adhesiveness of the base material and the compatibility with the sizing material and fibers are important design parameters in the development of CFRPs. However, a quantitative method for estimating the interfacial strength and the strength of the base resin has not been established. In this study, we propose a method to evaluate the interface strength of unidirectional CFRPs by creating learning data through a series of numerical material tests and by constructing a neural network that outputs the interface strength based on a homogenization method from the results of off-axis tensile tests. We adopt a general feed forward neural network whereby parameters are learned by employing a backpropagation method. The interfacial strength and the matrix resin strength is predicted and evaluated from the results of the off-axis tensile test to demonstrate the effectiveness of this system.
