Prediction of Mechanical Properties for tape-layered Specimen using Finite Element Method and Machine Learning

Kurzfassung

The production of fibre composite components using the Automated Fibre Placement (AFP) process is often confronted by manufacturing-related defects, such as gaps or overlaps between tapes. These defects can influence the material properties in different ways, depending on their characteristics. While the Finite Element (FE) Method can be used to investigate these effects, it requires expert knowledge and significant computational resources. The aim of this work is therefore to predict the mechanical properties using machine learning (ML) methods trained on compression tests simulated with LS-DYNA. First, an FE modelling approach and material model are developed with the goal of minimising both computation time and deviation from experimental test data. Subsequently, 1000 FE models are generated with random defect configurations by varying parameters such as position, size, and number of defects with the help of a developed Python routine. Two different ML methods, namely Random Forest (RF) and Support Vector Regression (SVR), are then applied and further improved in terms of their performance. Both models achieved promising results, with stiffness and strength predictions reaching a Mean Absolute Percentage Error (MAPE) of 3 to 7%. However, the prediction of post-failure behaviour showed limited accuracy, with a MAPE of around 50% due to numerical issues.