Data science for additively manufactured stainless steel : developing tools to aid the efficient characterisation, simulation and analysis of 3D printed structures

Abstract

Additive manufacturing has the potential to revolutionise the construction industry, promising benefits such as reduced material usage and relative freedom of geometry. However, novel manufacturing methods have drawbacks, stemming from limited understanding and relative technological immaturity. A case study concerning a bridge constructed from wire and arc additively manufactured stainless steel is presented to the reader. Attention is given to two principal sources of variation in the material, the surface geometry and the material properties. The variation is only realised at print time and varies spatially across a printed component. This presents a significant challenge to the use of 3D-printed stainless steel in commercial applications; performance of a given design is too much of an unknown. To try to overcome this problem, the thesis explores a statistical approach to the analysis and quality control of 3D-printed stainless steel structures. Part I develops statistical methods to enable the creation of a generic, generative model to allow the simulation of ‘realised’ components from a notional design. These results are carried forward to part II where the computational and practical aspects to conduct structural analysis on the ‘realised’ components is explored, with a workflow developed to enable the analysis of a notional stub column which was used to conduct a buckling test - a difficult and unstable procedure with a characteristic response observed in 3D-printed steel. It is seen that while there are dissimilarities to reality in both the generative model and the analysis, a reasonable ball park estimate can be obtained through a statistical treatment of the problem. Part III takes a different focus, and instead focuses on a quality control problem which aims to be solved as a predictive machine learning task. This task is shown to be possible, though significant improvement of the models is required in future works.