AI-based design methodologies for hot form quench (HFQ®)

Abstract

This thesis aims to develop advanced design methodologies that fully exploit the capabilities of the Hot Form Quench (HFQ®) stamping process in stamping complex geometric features in high-strength aluminium alloy structural components. While previous research has focused on material models for FE simulations, these simulations are not suitable for early-phase design due to their high computational cost and expertise requirements. This project has two main objectives: first, to develop design guidelines for the early-stage design phase; and second, to create a machine learning-based platform that can optimise 3D geometries under hot stamping constraints, for both early and late-stage design. With these methodologies, the aim is to facilitate the incorporation of HFQ capabilities into component geometry design, enabling the full realisation of its benefits. To achieve the objectives of this project, two main efforts were undertaken. Firstly, the analysis of aluminium alloys for stamping deep corners was simplified by identifying the effects of corner geometry and material characteristics on post-form thinning distribution. New equation sets were proposed to model trends and design maps were created to guide component design at early stages. Secondly, a platform was developed to optimise 3D geometries for stamping, using deep learning technologies to incorporate manufacturing capabilities. This platform combined two neural networks: a geometry generator based on Signed Distance Functions (SDFs), and an image-based manufacturability surrogate model. The platform used gradient-based techniques to update the inputs to the geometry generator based on the surrogate model's manufacturability information. The effectiveness of the platform was demonstrated on two geometry classes, Corners and Bulkheads, with five case studies conducted to optimise under post-stamped thinning constraints. Results showed that the platform allowed for free morphing of complex geometries, leading to significant improvements in component quality. The research outcomes represent a significant contribution to the field of technologically advanced manufacturing methods and offer promising avenues for future research. The developed methodologies provide practical solutions for designers to identify optimal component geometries, ensuring manufacturing feasibility and reducing design development time and costs. The potential applications of these methodologies extend to real-world industrial settings and can significantly contribute to the continued advancement of the manufacturing sector.