Abstract
Modern connected manufacturing environments generate vast amounts of data from sensors, machines, and quality control systems. This creates an opportunity for manufacturers to leverage Industry 4.0 pillars for improvements to quality control and operational efficiency. While AI techniques such as machine learning have been popular choices in the data analysis field, their lack of explainability make them less ideal for analyzing both why a defect happened and how the technique identified the defect. This dissertation introduces a practical approach to manufacturing data analysis by combining QARM with GA to analyze a live production PCB dataset. QARM provides interpretable and explainable results. As part of my approach, I provide a framework and techniques for reducing the computational cost of running a QARM algorithm, reducing the impact of noise on the input data, and a novel application of QARM in the PCB manufacturing domain on a big data problem. My methodology includes both "smart start" and "smart mutation" to provide a heuristic-based approach to population initialization and mutation operations. Additionally, my approach provides up to 19x computational speed improvements across several aspects of the QARM-GA algorithm including generation time, population initialization, and inner loop statistic calculations. These improvements demonstrate that with thoughtful architectural design and optimization to address the challenges associated with mining quantitative association rules, these techniques can become a viable method for big data problems.