A data driven design strategy to improve quality in additive manufacturing

• Main Author: Zhang, Yongjie
• Other Authors: Moon Seung Ki
• Format: Thesis-Doctor of Philosophy
• Language: English
• Published: Nanyang Technological University 2022
• Subjects: Engineering::Manufacturing::Flexible manufacturing systems; Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence
• Online Access: https://hdl.handle.net/10356/157350
• Institution: Nanyang Technological University

Additively manufactured end part has been gaining lots of attention and interest in the recent years in aviation industry, especially parts manufactured using Fused Filament Fabrication (FFF). However, there still exist significant barriers to adopt AM parts for loaded components. End use part requires assurance of part mechanical strength, especially for high criticality parts and parts that requires certification. Extensive testing is required to take into account confounding and interactions between the process parameters on the responses of interest, due to property structure process linkage. As such, if AM designers were to understand and characterise the design space through conventional trial and error methods, in order to produce optimised AM components, the cost is extremely high. Thus, to alleviate the situation, a data-driven design strategy is proposed, that aid designers to characterise the design space and optimise AM component designs efficiently and effectively. In the proposed data-driven design framework, the following is proposed: 1) An overall data driven framework that is efficient and effective in characterisation of design space, prediction, and optimisation of FFF parts responses of interest. 2) Establishing a methodology for FFF part mechanical properties and aesthetic prediction and optimisation. The surrogate model utilises Gaussian Process Regression to model the responses of interest and using multi-objective optimisation to trade of conflicting response of interest and obtain the optimal FFF part design 3) Understanding the PSP linkage, deposition strategy, and their impact on mechanical properties of FFF parts. This enables greater understanding and prediction of localised features found in FFF parts. 4) Data- driven, physics-based prediction of mechanical properties of FFF parts. Bayesian hierarchical modelling is proposed to model the influence of process variability and uncertainty propagation on the mechanical performance. To demonstrate the effectiveness of the framework, the methodology is validated against a case study, which predicted and optimised the FFF part properties in both conflicting response of mechanical strength and surface roughness requirements. Further, framework enables the AM designers to visualise the variation in the response of interest due to changes in the process parameters. Limitations of current proposed approach includes technology specificity (i.e. it is only currently applicable to FFF) and the framework can be integrated with the toolpath slicing software for improved prediction. Although there are limitations to the proposed framework, the methodology forms the fundamentals which AM designers can utilise to efficiently and effectively characterise the design space, predict and optimise FFF part performance.