Development of data augmentation technique for predicting strain rate effects in the mechanical behaviour of fibre-reinforced polymer (FRP) composites (Masterarbeit)

Kurzfassung

The mechanical response of Fiber Reinforced Polymers (FRPs) under dynamic loading is crucial for structural applications, particularly in aerospace, where lightweight, high-strength materials are essential. However, accurately predicting strain-rate-dependent behavior remains challenging due to the complex material structure and the limitations of experimental and numerical methods, which are often costly, time-intensive, and sensitive to material inconsistencies. This research investigates whether a Constitutive Artificial Neural Network (CANN) can effectively model and predict the strain-ratedependent behavior of FRPs while ensuring accuracy, generalizability, and computational efficiency. A CANN model is developed and trained using dynamic compressive testing data from cross-ply IM7/8552 composites. Unlike conventional data-driven approaches, CANNs integrate constitutive laws governing anisotropic materials, enhancing model reliability and extrapolation capabilities. The methodology involves data acquisition through high-speed uni-axial compression tests, preprocessing, empirical curve fitting, and model training. The performance of the trained CANN is evaluated against experimental results and compared with existing analytical and numerical methods. Results show that the CANN model accurately captures the stress-strain response of FRPs across varying strain rates. By incorporating physics-based constraints, the model improves extrapolation beyond the training data. These findings demonstrate that CANNs offer a viable and computationally efficient alternative for predicting strain-rate-dependent behavior in FRPs, with potential applications to full the gap of material behavior for simulations, crashworthiness analysis, and material design.