Development of an Artificial Neural Network-Based Tool for Predicting Failures in Composite Laminate Structures

Area of Science:

• Materials Science
• Mechanical Engineering
• Computational Science

Background:

• Composite materials offer high strength-to-weight ratios and design flexibility, crucial for aerospace and automotive industries.
• The anisotropic and layered nature of composites complicates structural analysis and failure prediction, often requiring computationally intensive methods like Finite Element Analysis (FEA).

Purpose of the Study:

• To develop a bio-inspired machine learning approach to accelerate structural analysis and failure prediction in composite materials.
• To reduce the reliance on computationally expensive FEA by utilizing an artificial neural network (ANN) model.

Main Methods:

• An artificial neural network (ANN) model was developed to predict key structural parameters (thickness, weight, stress, displacement, deformation, failure criteria).
• The ANN model utilizes stacking sequence and geometry as inputs for a given load case.
• The methodology was validated using a specific composite beam, comparing ANN predictions against FEA results.

Main Results:

• The ANN model achieved prediction deviations under 15% compared to FEA results.
• The ANN model drastically reduced prediction time from hours/days (FEA) to seconds.
• The developed approach demonstrates significant computational time savings while maintaining prediction precision.

Conclusions:

• The bio-inspired machine learning approach offers a viable alternative to traditional FEA for rapid structural assessment of composite materials.
• The methodology shows potential for broader applications, including aircraft maintenance, to enhance decision-making and structural reliability.
• Further refinement of the logic-driven ANN model could lead to more advanced applications in structural integrity assessment.