Analyzing Critical Process Parameters Influencing Product Quality Defects using Machine Learning:A Real-World Case Study from a Foundry Line
Examensarbete för masterexamen
Production engineering (MPPEN), MSc
Manufacturing has grown in importance with the goal of reducing production disrup tions, minimizing unplanned downtime, minimizing quality defects, and increasing the efficiency of production systems. This project emphasizes quality control in manufacturing, where traditional quality control processes are reactive but modern processes utilizing Artificial Intelligence (AI) and Machine Learning (ML) techniques are proactive. These modern processes have the potential to reduce waste, improve efficiency, and decrease costs, thereby ensuring the success of a manufacturing com pany. AI/ML utilizes large datasets from multiple sources, such as machines, pro cesses, and products, to predict deviations in the future and make better data-driven decisions, allowing organizations to stay competitive in this digitalized era. This thesis employs CRISP-DM, a comprehensive tool used for data mining and ML projects, to provide a structured approach for solving business problems and facilitating data-driven decision-making. The case company specializes in produc ing castings for power-train components in heavy vehicles a process governed by a range of process parameters crucial to product quality. Poor control of these pa rameters can lead to casting defects. Leveraging ML models, this study aims to establish correlations between various process parameters in two core-making ma chines (Machine A & Machine B) along with their maintenance logs, with the goal of identifying root causes for different casting defects in the core-making process. The ML models utilized for identifying faulty cores attained accuracy rates of 66.2% for Machine A and 56% for Machine B. These levels of accuracy were deemed satisfac tory by the domain experts within the company. The results from the thesis would aid human operators in determining whether to manually eliminate defective cores from the production line before casting. Therefore, this would effectively prevent the production of flawed castings and contribution to the production of high-quality products.
Prediction of Quality Defects, Process Parameters, Maintenance Logs, Machine Learning, CRISP-DM, Manufacturing, Foundry