Production Analytics for a novel additive manufacturing system

Additive manufacturing, also known as 3D printing, has emerged as a promising solution in the context of technological advancements and innovative production processes. This thesis focuses on two additive manufacturing processes, namely Selective Laser Sintering (SLS) and Selective Thermoplastic Electrophotographic Process (STEP), which have the potential to revolutionize industrial production. However, their current status requires extensive research and development efforts to upscale them for widespread application.
Several key challenges need to be addressed to achieve the industrial-scale implementation of 3D printing. Firstly, the understanding of the 3D printing process is limited, necessitating the expansion of the engineering knowledge base, its functioning and underlying mechanisms. Secondly, optimizing the entire 3D printing process is crucial for ensuring reproducibility and consistently high-quality output. Lastly, the inherent instability of the 3D printing process poses a significant challenge to its integration into large-scale production. Three main sources of data can be identified: input data (machine parameters), process data (data acquired during the production process) and output data (the quality of the final products). During the PhD journey, some methodologies have been developed to address the challenges faced by the new 3D printers.
The Ph.D. thesis begins with an introductory chapter that discusses the context of the Ph.D. project as well as its contributions. A brief introduction to additive manufacturing, its limitations, and the different types of data related to the machines is described in the next chapter.
The following chapters summarise the contributions produced during the project. The first contribution of the thesis focuses on building a link between the various types of data. In particular, it takes into consideration the process and output data, and it focuses on the identification of relevant process variables that mostly affect the quality. It proposes a variable selection algorithm based on the Random Forest model to address the case of input variables highly correlated. The second contribution introduces a hybrid approach that combines correlation analysis using observational data and machine learning techniques with designed experiments to establish causality. This approach is particularly useful when investigating unknown phenomena with a large input space, enabling insights and correlations to be gathered from observational data before conducting experimental designs. Next, a novel methodology for analyzing the printer’s process data that are organized in a three-way array of data with a multi-group structure is presented. This methodology is applied to the problem of process modelling, where batches are structured according to multiple groups. By extending the PARAFAC model, the proposed approach accounts for the grouping structure of three-way data sets, enabling the estimation of a model representative of all the groups simultaneously. Following that, a model for the supervised analysis of multi-group three-way data and multi-group output data is described. Finally, some of the collaborative projects conducted with the research group during the Ph.D. project are presented.
Overall, this thesis addresses the challenges and limitations of additive manufacturing processes, contributes to the understanding and optimization of 3D printing, and provides innovative methodologies for data analysis in manufacturing applications.