Digital Twin and Artificial Intelligence Framework for Real-Time Optimization of Honeycomb Panel Geometry and Mechanical Performance

Abstract

Light-weight sandwich panels filled with phase-change material (PCM) with honeycomb cores have been found to have a great potential for aerospace, automotive, construction, and thermal-management applications due to their combination of mechanical stiffness, energy absorption, and high thermal-energy-storage capacity. The thermo-mechanical coupled response of such panels is, however, strongly nonlinear and depends on the cellular geometry, the thickness of the walls, the height of the cores, the relative density, and the applied pressure and thus forcing conventional trial-and-error or single-physics finite-element based design cycles to be slow and costly. An integrated Digital Twin and Artificial Intelligence (AI) framework for the real-time optimization of the geometry and mechanical performance of honeycomb panels is presented in this paper. A parametric nonlinear analytical-numerical model was developed that correlates deformation, von Mises stress, safety factor, stiffness, PCM heat storage capacity, thermal resistance and a composite score of multi-objectives to cell side length, wall thickness, core height, relative density and pressure load. A large design of experiments (DoE) database containing 4320 nonlinear designs was created and mined to generate nonlinear response surfaces, Pareto type performance maps and a ranked set of optimum designs. On top of the physics-based model, a “digital twin” layer was then developed, and a lightweight artificial-intelligence (AI) surrogate proof-of-concept was generated that reproduces the time-dependent stress, deformation, and temperature histories with less than 2% instantaneous deviation from the same physics-based reference signals used to define it; this illustrates the feasibility of near real-time monitoring and prediction of the panel's behavior under service loading, pending training and validation on independent, real sensor data. The results reveal that the reduction of the cell side length coupled with the increase of wall thickness results in a dramatic increase of stiffness and safety factor, but at the same time the mass increases; the von Mises stress field shows a strong nonlinear interaction between cell geometry and pressure load, while the optimal 20 designs have a common range of cell sizes (3–5 mm) and wall thicknesses (more than 1.0 mm) and core height (around 10 mm), with optimized scores exceeding 20. The proposed digital twin/AI framework greatly reduces the frequency with which full-scale simulation is required for design screening and lays groundwork toward real-time structural health monitoring and thermal monitoring once experimentally validated, and offers a transferable methodology for concurrent mechanical-thermal optimization of cellular sandwich structures, presented here as an analytical-parametric proof-of-concept rather than a fully validated finite-element or experimentally confirmed digital-twin deployment. A dedicated verification study (Section 3.11) further shows the top-ranked design and overall ranking to be reasonably robust to the score weights and correction-factor constants (Spearman rank correlation ≥ 0.94), reports a reduced-order finite-difference cross-check of the stiffness formula that motivates further three-dimensional FEM validation, and demonstrates a genuinely trained AI surrogate achieving below 2% relative error on held-out, previously unseen design cases.