Scalable Stress-Constrained Topology Optimization: A Super-resolution and Simultaneous Approach Based on Machine Learning

Researchers proposed the simultaneous neural topology optimization (SNTO) framework, which uses custom neural networks to jointly optimize state, adjoint, and design variables in a single loop for stress-constrained topology optimization. The approach minimizes a multi-component loss function combining the design objective, aggregated local stress constraints, and potential energy functionals. A hardware-aware training scheme partitions the design domain across multiple GPUs to scale the method to millions of degrees of freedom.