Interface-Aware Deep Energy Method for the Stress Analysis of Tapered Composite Laminates

Authors:
Preprint
Publisher website
Google Scholar
Abstract
This study introduces the Interface-Aware Deep Energy Method (IDEM), a physics-informed neural network framework for predicting the mechanical linear response of tapered composite laminates. The proposed methodology incorporates an interface property-averaging scheme within a deep energy minimization framework to improve the representation of stress discontinuities across different material interfaces. The proposed formulation embeds governing physics directly into the learning process, enabling accurate prediction of displacement and stress fields without requiring any training data.The proposed method is evaluated on tapered composite laminate configurations with varying stacking sequences and taper angles. IDEM's predicted full-field displacement and stress distributions within the tapered laminate cross sections are validated against high-fidelity finite element simulations. Results demonstrate that the proposed IDEM framework accurately captures both displacement and stress fields with high accuracy. IDEM achieves high correlation with finite element results ($R^2 \geq 0.99$) for displacement predictions and significantly reduces stress prediction errors compared to a baseline Deep Energy Method, with mean absolute error reductions of over 50\%. Therefore, the proposed interface-aware formulation extends the applicability of Deep Energy Methods to composite laminates with material discontinuities and complex geometries. By accurately resolving localized stress concentrations, the framework provides a robust physics-informed approach for the analysis and design of tapered composite structures.

This publication has 43 references indexed in Scilit: