ADVANCED DEEP LEARNING IN OPTIMAL DESIGN FOR ADDITIVE MANUFACTURING

Abstract

In this thesis, we propose several deep learning-based frameworks to address challenges of both manufacturability assessment and optimal design in additive manufacturing (AM), intending to facilitate the end-to-end product cycle in AM processes from design to manufacturing.

The first half of the thesis discusses manufacturability assessment algorithms at both pre-manufacture and in-process stages. The objective is to derive efficient algorithms to predict the manufacturability of a given part before the manufacturing process starts and detect anomalies that can potentially affect the manufacturability of the part during the manufacturing process. Specifically, we explain how existing design for additive manufacturing (DfAM) schemes become ineffective when the structure of the parts and the AM process become complex and how existing in-situ anomaly detection methods can be improved with uncertainty measures. Therefore, we propose efficient algorithms to address both of these two challenges. The second half of the thesis discusses optimal design problems in AM. Focusing on a hierarchical design approach, we adopt multiscale topology optimisation (MSTO) as the central methodology to solve for the optimal structure given the application-specific requirements. We explain the limitations of existing MSTO methods and propose a practical framework to improve MSTO methods with general voxel-based microstructures.