This conference paper was submitted for presentation at the NAFEMS World Congress 2025, held in Salzburg, Austria from May 19–22, 2025.

Abstract

Recycled short fiber reinforced polymers (SFRPs) offer promising solutions for sustainable design in industries like automotive, where environmental policies are increasingly enforcing recycled material quotas. However, the inherent variability in material and process performance for recycled SFRPs presents significant design challenges. This uncertainty can lead to either overdesign'”using excessive material to offset unknowns'”or underdesign, which compromises the product'™s reliability. Both approaches come with trade-offs: overdesign increases costs and environmental impact, while underdesign risks the integrity of the final product. To address these issues, a machine learning-based robust design approach for recycled SFRPs has been developed. This approach leverages machine learning to replace computationally costly structural finite element models with reduced-order models. These efficient models enable thousands of structural analyses to be performed across different uncertainty scenarios at a fraction of the cost of high-fidelity simulations. Using a Monte Carlo approach, this method incorporates variability from factors like local fiber orientation (due to process-related uncertainties), material performance fluctuations, and conditional factors. This machine learning integration ensures a comprehensive reliability assessment while minimizing computational costs, making robust reliability evaluation both feasible and practical. The machine learning-based robust design approach enables engineers to quantify uncertainties and accurately assess structural reliability, allowing for optimized material selection without the risk of overdesign or underdesign. Through this methodology, engineers can tailor the amount and type of recycled SFRPs used in specific applications, achieving both sustainability and efficiency. The paper will showcase the application of this method to a structural part, quantifying its reliability by accounting for different sources of uncertainty in both material and process parameters. This case study demonstrates how the machine learning-based robust design approach provides a clear understanding of how these uncertainties affect component reliability, offering insights into achieving optimal material and design choices without overdesign or underdesign. This advanced methodology enables engineers to balance sustainability and reliability for recycled SFRP applications, setting a new standard for robust and efficient design in industries where environmental goals are increasingly critical.