Explainability in advanced manufacturing: leveraging the interpretability of multi-criteria decision making and neutrosophic logic

Interpretability has been a vital aspect of modelling since the emergence of machine learning. Despite this, the dominance of non-transparent models due to their arguable superior performance meant that interpretable transparent models were seldom used, especially in data-driven applications. Interpretability is key to reaching explainability. Thus, interpretable models are the most promising way to achieve this. In this thesis, a new class of interpretable models (multi-criteria decision making (MCDM)) is investigated, for the first time, in a series of industrial and academic applications toward achieving explanation. The MCDM model is shown to achieve enhanced interpretability following its extension with fuzzy logic. The Fuzzy-MCDM model's interpretability enables the generation of output explanations. Consequently, the model's interpretability is improved further by introducing neutrosophic logic. The proposed models are applied to benchmark and industrial pipe inspection datasets. The experimental results demonstrate the framework's capability for generating meaningful explanations; while maintaining good performance. The purpose of explaining a model's result is to pave the way for broad adoption in fields where a decision's accountability and transparency are paramount due to the high stakes of the decisions. These areas include biomedical, aviation, nuclear and advanced manufacturing. Machine learning adoption is lacking in high stake areas due to the lack of explanation, an obstacle preventing the acquisition of trust from the experts. The thesis describes how an interpretable modelling framework is adapted to reduce the performance trade-off often attributed to transparent models while exploiting the advantages to generate useful explanatory information - a clear advantage over opaque models.