FEM GeniusóAI-powered Intelligent Agent for Automated Finite-Element Modeling of Aerospace Structures

Traditional finite-element modeling (FEM) of airframe structures is labor-intensive, tribal-knowledgeñdriven, and prone to mesh-induced errors that require weeks of manual rework. EPIR will leverage its previously developed in-house advanced modeling capabilities to developÝFEM-GENIUSô, a self-learning AI agent that overcomes these bottlenecks by transforming raw CAD geometry into solver-ready, mixed-fidelity meshes in minutes and iteratively refines them until user-defined accuracy targets are met. The Phase I effort integrates four innovations: (i) an automated CAD-to-graph encoder that heals defects, defeatures non-critical details, and encodes topology as an attributed graph; (ii) a reinforcement-learning (RL) policy that assigns element types, local mesh sizes, and shell-to-solid transitions to minimize a cost-weighted error metric; (iii) a physics-informed surrogate network that predicts nodal stresses and calibrated uncertainty bounds four orders of magnitude faster than full FEM solves; and (iv) an active-learning loop that continuously retrains both surrogate and RL modules on newly validated data. The six-month program will deliver a TRL-3 prototype capable of meshing rib-stiffened panels and NASA wing-box benchmarks with < 5 % stress error while cutting manual setup time by > 80 %. Successful execution will compress Navy design-certification cycles, enhance fleet readiness, and democratize high-fidelity simulation across aerospace and defense sectors.