Application of CRISP-DM and DMME to a Case Study of Condition Monitoring of Lens Coating Machines

Condition monitoring industrial machinery by using data mining in industrial projects has been extensively extended in the last few years. Applying data science to industrial processes should be straightforward in theory, but very few instances in the literature deal with the actual practical issues encountered while carrying out industrial data science projects. The case study discussed in this paper was pursued in accordance with the steps outlined in the standard CRISP-DM (CRoss-Industry Standard Process for Data Mining) including its latest holistic approach for engineering applications called DMME (Data Mining Methodology for Engineering Applications). The industrial data was acquired as part of industrial cooperation from multiple anti-reflective lens coating machines. Various deep learning (DL) models like long short-term memory (LSTM), and machine learning (ML) models like Decision Trees and Support Vector Machines (SVM) were used as proof of concept for confirming the domain understanding of the process experts. Our main contribution is the description of deficiencies and gaps of the standard process framework CRISP-DM based on the issues faced in implementing each phase of the process in a real world case study. In addition, we propose future research ideas to close these gaps. This complements findings in the literature on gaps in CRISP-DM and DMME.