Study on impact condition identification of composite laminate based on deep learning and peridynamics

Composite is often subjected to various loads (such as impact load) from outside, resulting in fracture, delamination and other damages. Traditional continuum mechanics theory based on partial differential equation is difficult to deal with discontinuous problems such as fracture and damage because it involves space derivation. Peridynamics (PD) is a nonlocal theory based on integral equation. It uses space integration to describe the material function, which has great advantages in dealing with the above problems. But there is "surface effect" in traditional PD methods. That is, when discretizing the material points, the horizon of the material points in the boundary area is incomplete, which will cause calculation errors. Based on this, the paper analyzes the problem of the problem, a more concise surface correction factor is proposed. In the design process of composite materials, accurate load information is needed, such as the direction and velocity of external impactor, for example, for aircraft, it is convenient for engineers to design enough strength in appropriate positions, or estimate the residual strength of structures subjected to load and evaluate the probability of its continued use. Therefore, it is of great significance to identify the impact condition based on the damage data, and to improve the design of composite materials and ensure its safe use. Based on this problem, this paper develops a set of impact condition identification model based on deep learning, which can use the impact damage evolution data of composite materials under different impact conditions for training, and realize the identification of unknown impact conditions, so as to provide more detailed reference and basis for improving the design method of laminated plates. The main contents of this paper include: (1) Aiming at the problem of impact damage discontinuity of composite laminates, a numerical analysis model of impact damage evolution of composite laminates based on peridynamics theory is established, and the corresponding calculation program is developed. Moreover, in order to solve the problem of incomplete horizon of material points in the boundary region, an improved "surface correction factor" is proposed, which can improve the calculation accuracy. On the basis of the above model, the damage evolution of composite materials under different impact conditions of cylindrical and spherical rigid bodies is analyzed. (2) In order to identify the impact condition of composite laminates, a model based on machine learning convolution neural network (CNN) is developed under the framework of TensorFlow and Jupyter Notebook. The recognition model uses the impact damage evolution data of peridynamic composite laminates under different impact conditions for training, realizes the recognition of unknown impact conditions, and can control the relative error within 5% and reach a high accuracy.