In-depth analysis tagged Design Optimization — covering PLM history, vendor strategy, and the technical decisions reshaping engineering software.
3 articles
5 AI trends remaking CAE and engineering simulation. How neural surrogates, adaptive meshing, and AI-powered physics are turning simulation from a validation step into a design accelerant.
OpenBOM gives small and midsize hardware companies a cloud-native BOM and product data management platform that works in days rather than months. Leo AI applies artificial intelligence to optimize engineering designs against multiple objectives simultaneously — performance, cost, weight, manufacturability. Together they represent the state of the art in making product data not just accessible but actionable: a system that tells you what to do with the data, not just where to find it.
nTop and Neural Concept are both solving the same engineering design bottleneck — the gap between what engineers can imagine and what simulation can evaluate in reasonable time. nTop eliminates the CAD-to-simulation-to-manufacturing loop latency with computational geometry. Neural Concept, backed by $100M from Goldman Sachs, applies deep learning to reduce simulation cycle times by orders of magnitude. Together they represent where AI meets the fundamental physics of design.
Mesh refinement strategy where element size is adjusted based on predicted or actual stress concentration. Fine mesh where stresses are high, coarse mesh where they are low.
Automated search through design space (geometry, material, topology) to find solutions that meet performance goals (minimize mass, keep stress below threshold, maximize stiffness) or objectives (maximize efficiency, minimize cost).
Coupled simulation where two or more physics domains (structural mechanics, thermal, fluid dynamics, electromagnetics) interact. E.g., temperature rise causes material property changes, which affects structural strength.
A machine-learning model (neural network) trained to approximate the output of a computationally expensive simulation. Given design parameters, the surrogate predicts results (stress, temperature, frequency) in milliseconds. Also called a metamodel or emulator.
Neural networks trained not just on labeled data, but also constrained to satisfy physical laws (governing PDEs, boundary conditions). The training process incorporates domain knowledge about conservation laws.