ISSN 2384-5058
Abstract
One of the significant impacts of recent artificial intelligence (AI) advancements is their capability to expand structural design engineers' methodologies, especially in the field of topology optimisation for lightweight mechanical systems. Typically, structural topology optimisation done by methods such as Solid Isotropic Material with Penalisation (SIMP) or the level-set approach reaches highly efficient designs, but they require a lot of computation time and thus cannot be repeated many times for further exploration. This work presents an AI-supported topological optimisation strategy that includes the integration of deep learning surrogates with finite element analysis (FEA) to rapidly locate close to optimal material layouts under given boundary and load conditions. The convolutional neural network (CNN) model is developed with the help of a classical topology optimisation solutions dataset, which can yield starting density distributions, substantially speeding up the following optimisation steps or, in some instances, facilitating a direct solution without iterations. In silico trials reveal that the surrogate can help limit the computational budget by almost 80% while still keeping compliance within 5% of the fully optimised line. The proposed framework is especially relevant for lightweight structural parts in the aerospace and automotive sectors, where the rapid design iteration and manufacturability requirements are stringent. AI-powered topology optimisation, which integrates data-driven learning with physics-based constraints, is a step towards on-demand, generative design for advanced structural applications.
Keywords: topology optimisation; deep learning; surrogate modelling; finite element analysis; lightweight design; structural optimization