A framework for structural shape optimization based on automatic differentiation, the adjoint method and accelerated linear algebra

Shape optimization is of great significance in structural engineering, as an efficient geometry leads to better performance structures. However, the application gradient-based shape for form finding limited, which partly due difficulty and complexity gradient evaluation. In this work, framework based on automatic differentiation (AD), adjoint method accelerated linear algebra (XLA) proposed promote implementation finding. The realized by high-performance computing (HPC) library JAX. We leverage AD sensitivity analysis stage. XLA feature exploited programming architecture that we proposed, can boost evaluation via just-in-time compilation computer programs. also supports hardware acceleration such GPUs. applied arches different free-form gridshells: gridshell inspired Mannheim Multihalle, four-point supported gridshell, canopy-like Two geometric descriptive methods are used: non-parametric parametric description Bézier surface. Unconstrained constrained problems considered, where former solved descent latter sequential quadratic (SQP). Through these examples, shown be able provide engineers with tool finding, enabling design built environment.