A Machine Learning Boosted Data Reduction Methodology for Translaminar Fracture of Structural Composites

Abstract This work explored a machine learning (ML) algorithm as a fast data reduction method for translaminar fracture energy in composite laminates. The method was validated with translaminar fracture tests on compact tension (CT) specimens on AS4/8552 and IM7/8552 cross-ply lay-ups. Experimental fracture energy and R-curves for both materials were determined using the most common data reduction methods, such as the compliance calibration (CC), the area (AM) and the Irwin relationship (IM). Our new data reduction method uses a surrogate model based on an artificial neural network (ANN) trained with synthetic data generated with the cohesive crack finite element model. Such a surrogate model maps the cohesive properties with the corresponding load–displacement, crack-displacement and energy-displacement curves with interrogation times in the order of 20 ms and relative errors in the load–displacement and crack growth less than 2%. Such performance enabled its encapsulation to approximate the inverse problem to infer the cohesive parameters with the maximum likelihood estimator (MLE) directly from the experimental load–displacement and crack-displacement curves. The results demonstrated the ability of the model to deliver cohesive parameter inference directly from the macroscopic tests carried out at the laboratory level.