Accelerating the Layup Sequences Design of Composite Laminates via Theory-Guided Machine Learning Models

Area of Science:

• Materials Science
• Mechanical Engineering
• Computational Science

Background:

• Composite laminate design is complex, requiring optimization of ply angles for desired strength and stiffness.
• Traditional methods can be time-consuming and data-intensive.

Purpose of the Study:

• To develop and validate a theory-guided machine learning (ML) model for optimizing composite laminate design.
• To accelerate the design process by integrating established composite theories with ML.

Main Methods:

• A finite element simulation incorporating Hashin failure theory (HFT) and classical lamination theory (CLT) generated training data.
• A multi-layer neural network (NN) system was designed following the theoretical sequence of composite mechanics.
• A genetic algorithm (GA) was used for inverse optimization, guided by the NN's forward predictions.

Main Results:

• The ML model successfully determined optimal layup sequences and fiber modulus for a composite tube under combined bending and torsion.
• The theory-guided ML approach demonstrated a faster optimization process compared to direct neural networks.
• Less training data was required when domain knowledge (composite theories) guided the ML model.

Conclusions:

• Integrating domain knowledge, specifically composite theories like HFT and CLT, significantly enhances ML model performance in engineering design.
• Theory-guided ML offers a more efficient and data-economical approach to optimizing composite structures.
• This methodology highlights the critical role of theoretical foundations in advancing ML applications for complex engineering problems.