An Integrated FEM-AI Methodology for Predictive Modelling and Optimization of Filled Bio-Composites

Laboratory of Modelling and Simulation of Intelligent Industrial Systems (M2S2I), ENSET, Hassan II University of Casablanca, Morocco

^* Corresponding author: maryame.lakrade-etu@etu.univh2c.ma

Abstract

This paper proposes a unified approach merging finite element modelling (FEM) and artificial intelligence (AI) to forecast the mechanical behavior of filled bio-composites, which are growing in importance for sustainable material applications. The proposed method applies Digimat software for multi-scale simulations to account for anisotropic behavior and interfacial effects, with AI algorithms, such as neural networks and nonlinear regression, creating predictive correlations between microstructural parameters and mechanical performance. The FEM simulations produce high-fidelity data on stress-strain distributions, which are subsequently employed to train the AI models, establishing a synergistic feedback loop that improves predictive accuracy. In contrast, differences between AI forecasts and FEM findings direct adjustments to the simulation framework, which leads to reliable modelling results. This cyclical merging resolves the constraints of isolated methods, where in finite element analysis by itself may be inadequate for investigating extensive parameter domains, and artificial intelligence frameworks may be compromised by limited or noisy numerical data. The hybrid FEM-AI model achieves high accuracy across mechanical properties, outperforming standalone FEM and AI approaches. Overall, the integrated approach offers a powerful and generalizable framework for efficient exploration of complex material behaviors.