Increasing immersion in machine learning pipelines for mechanical engineering

Machine learning has become a vital tool in the toolbox of any discipline that regularly interacts with data. This holds especially true for mechanical engineering, where the demand for data driven approaches to engineering challenges has recently increased drastically. While promising results have already been achieved by applying machine learning models in this field, the opaqueness of modern machine learning models, such as artificial neural networks, brings its own challenges. A lack of trust in these models often prevents their application and a lack of intuition for their inner workings prevents them from facilitating scientific discoveries about manufacturing processes. This thesis aims to approach these challenges from the unique angle of increasing the amount of immersion used when interacting with models and data throughout the whole machine learning pipeline. Immersion in virtual environments has the ability to facilitate a phenomenon called presence, which can make users perceive the presented stimuli as more real and engage with them much more directly. However, research regarding the effect of immersion on abstract data visualization is lacking. It is argued that a similar phenomenon, in this thesis called data presence, can be achieved with proper immersive visualization tools. Hence, multiple demonstrators were developed that showcase the possibility to facilitate better understanding of data, models, and the underlying processes using increased degrees of immersion. First, general considerations are presented on developing immersive applications for use in everyday work. Then, multiple immersive applications are presented for data labeling, data understanding and model visualization. Each of these applications is evaluated, showing indications that they can indeed lead to benefits for their users. Finally, the ANNtoNIA framework is presented, which was designed to simplify the development of immersive visualizations for artificial neural network models. The design of this framework is presented in detail, and a brief outlook is given on potential applications it could be used for. Overall, this thesis lays a foundation for future work in the area of applying immersive visualization to machine learning and manufacturing data.