Description

This thesis advances the field of production processes, specifically injection molding, toward the goal of Zero Defect Manufacturing (ZDM), with an emphasis on improving quality prediction methods through Artificial Intelligence (AI). It explores innovative strategies for improving the efficiency and quality of manufacturing processes, highlighting the importance of minimizing defects in industrial production. The research begins with an analysis of the current state of predictive quality systems in injection molding processes, identifying the critical need for advances to achieve ZDM. By collecting data from legacy and modern equipment, this work establishes a basis for a comprehensive analysis, using standard protocols and new methodologies for data collection. Central to this thesis is the application of feature selection algorithms, which employ a combination of filter, wrapper, embedded, and hybrid approaches. This framework is designed to accurately identify the key parameters that influence the quality of the injection molding process, thereby facilitating more effective and predictive modeling of manufacturing results. In addressing quality prediction, the thesis introduces both supervised and unsupervised models to predict manufacturing quality. It particularly focuses on enhancing these models with human knowledge, integrating expert insights into the predictive algorithms to better adapt to the complex dynamics of the manufacturing environment. This approach not only improves the accuracy of predictions but also enriches the models with practical, real-world applicability. Through a series of detailed use cases, the effectiveness of the proposed methodologies is demonstrated across various equipment and manufacturing scenarios. The results highlight significant improvements in process predictability, contributing to the reduction of defects and moving closer to the ideal of ZDM. The conclusion of this thesis reiterates the importance of its contributions to the field of AI used in production processes, with a focus on injection molding, providing a solid foundation for future research aimed at improving the integration of data-driven and human-centered approaches in manufacturing. The thesis outlines potential pathways for continued innovation in digitalization, feature selection, and quality prediction, emphasizing the ongoing pursuit of excellence in manufacturing processes.