Low velocity impact performances prediction and classification of damages for fiber reinforced composites by machine learning models

Due to the unique strength-to-weight ratio and durability fibre-reinforced composites (FRCs) are utilised in the aerospace, automotive and defence industries. Disregarding their high-tech advance the microstructure produces broad behaviours which remain challenge to predict. Of these behaviours, low-velocity impacts (LVI) which can produce internal damages that are not visually detectable and can greatly vary the overall performance of the structure. Characterisation of these impacts by empirical measurements has a high cost and relies on a prone-to-failure set of assumptions which has necessitated the search for better approach. To that end, the study aims to establish a machine learning (ML) approach to predict impact responses and to determine the damage modes in pure and hybrid laminates. Data collected from experiments with varying stacking sequences, ply counts, thicknesses and impact energies based on mono, bi and tri composition of carbon, glass and Kevlar composites. Several models including baseline and ensemble and neural network were implemented and validated. For regression tasks, the Gradient Boosting (GB) performed the best (R 2 = 0.98, MSE = 0.87, MAE = 0.45) while same model gained high accuracy and F1 score of 0.98 for failure classification. Analysis of model workings indicated impact energy and thickness were the most influential in determining the predictions. Despite of limited experimental augmented dataset this approach has opened faster horizons and new efficiencies along with greater reliability and a non-intrusive method of enhancing the design and performance analysis of FRCs through ML.