Generative inverse design of steel gridshell joints with multi-objective optimisation

Abstract

The design of steel gridshell joints, simultaneously minimising weight, maximising stiffness and ensuring a uniform stress distribution, is a challenging multi-objective problem. This paper presents a generative inverse design framework integrating topology optimisation (TO), data-driven surrogate modelling and multi-objective optimisation to automatically generate high-performance steel joint designs. A parametric workflow links a BESO-based TO module with a Bayesian-optimised XGBoost surrogate model for predicting joint compliance and stress variation. An NSGA-II parametric evolutionary optimiser then explores trade-offs among competing objectives, while K-means clustering extracts representative Pareto-optimal solutions. The effectiveness of the framework is validated by a case study, with the generated joints achieving up to 40% weight reduction and improved stiffness and stress uniformity relative to a conventional hollow joint. One selected design was successfully fabricated via selective laser melting 3D printing, demonstrating practical manufacturability. The proposed framework is also adaptive to other steel gridshell joint forms.