Reconstruction and prediction of Mode-I cohesive law using artificial neural network

Abstract

Cohesive zone model (CZM) is a widely used tool for simulating both static and fatigue damage propagation for bonded area in composite materials. Under static loading, the damage propagation behavior is governed by a cohesive law (CL), the shape of which is an important property especially when non-linear damage mechanisms such as fiber bridging is involved. In this work, a CZM driven by artificially neural network (ANN) for simulating mode-I damage propagation is proposed, where the traction-separation relationship is calculated by a multi-layer perceptron (MLP). More importantly, a reconstruction method is proposed to extract the CL through the training of the neural network using a simple load-displacement (P–U) relation of a double cantilever beam (DCB) as the only inputs, without the need to measure the crack opening displacement (COD). The proposed method is first validated using finite element generated virtual experimental data with various CL shapes before being applied to experimental data, where good correlations are obtained, proving the effectiveness of the proposed method.