An artificially intelligent space-filling trajectory planning for wire arc additive manufacturing

Abstract

This thesis systematically explored the implementation of a Space-Filling strategy for an artificially intelligent trajectory planning to be used in Wire Arc Additive Manufacturing (WAAM) and investigated its benefits and challenges. The Pixel strategy, as the focus, was proposed and developed as an innovative and flexible computerized tool in trajectory planning for complex geometries. Pixel was intended to provide multiple applicable trajectories for part printings and the subsequent optimized trajectory selection for each case. To achieve this target, a basic version was offered using a space-filling approach, by formulating a grid of nodes, and, simultaneously, four heuristics for node connections. Computational evaluations demonstrated the effectiveness of the "Basic-Pixel" strategy for various part geometries. Experimental builds using Gas Metal Arc (GMA) and plain carbon steel confirmed the practical viability of this basic version, enabling the deposition and construction of intricate shapes, including polygonal nonconvex geometries with holes. To boost the algorithm's performance, the "Enhanced-Pixel" strategy was introduced, incorporating a new node sorting method and four trajectory planning heuristics. Comparative analyses in specific case studies validated the operational efficiency and effectiveness of the "enhanced" version compared to commercially applied conventional strategies. The study further explored the following "Advanced-Pixel" strategy, utilizing reinforcement learning techniques (artificial intelligence) to optimize the selection of trajectory planning heuristics and ordering methods. Experimental analyses revealed that the "Advanced-Pixel" strategy outperforms the "Enhanced-Pixel" strategy in terms of performance gains and response quality, demonstrating reduced printing time and trajectory distance, particularly for larger components. Additionally, the thesis work investigated the "Fast-Pixel" strategy, leveraging clustering techniques with "k-means" to reduce the dimensionality of the optimization problem. The "Fast-Pixel" strategy implementation demonstrated improved performance across all tested parts, significantly reducing computational time while improving response quality. At last, the thesis text outlines future research directions, including expanding to different materials, optimizing computational efficiency, mitigating non- conformities, exploring hybrid strategies, and developing real-time monitoring and quality control systems. In conclusion, the research and development work in this thesis, by introducing the Pixel strategy and its improvements, provided an option for trajectory planning in WAAM. The experimental validations, computational evaluations, and practical demonstrations highlighted the effectiveness and viability of the proposed strategies. These scientific-oriented developments have significant implications for the efficient and effective production of complex parts using additive manufacturing technologies, paving the way for further advancements in the field.