A Data-Driven Downtime Reduction Framework for Traditional Plastic Extrusion Machines
Pysyvä osoite
Kuvaus
Opinnäytetyö kokotekstinä PDF-muodossa.
Unplanned machine downtime is one of the major challenges in production environments. It reduces productivity, increase operational costs, and affects delivery deadline. Large manu-facturing companies usually deal with this problem by implementing smart machine mainte-nance strategy, but small and medium-sized enterprises (SMEs) often lack the resources to implement such advanced maintenance strategy. Therefore, SMEs struggle a lot to keep the machine downtime under a tolerance limit. The opportunity here is that the SMEs already collect some operational data, such as interruption log and maintenance log, which are rarely analyzed together. This thesis aims to develop a practical decision-support framework that integrates these existing sources of data to reduce the downtime without any additional investment. The research is conducted at a local plastic profile manufacturing company. The study is based on three main theoretical perspectives, such as reliability engineering theory, root cause analysis theory, and decision theory. The methodology of this study follows a very structured approach. First, the downtime notes are categorized using text mining. Then, Pareto analysis and KPI analysis are performed to identify the dominant downtime category and evaluate machine performance. Next, a predictive model is developed to anticipate the failure risk of each machine. Finally, all the results from descriptive analysis and predictive analysis are combined into a framework to support maintenance decision-making. The Pare-to analysis has found that only five categories, including equipment failure, tooling fault, component failure, calibration issue, and material issue were responsible for majority of the total downtime. It has been shown through KPI analyses that machine performance varies dramatically and the KPIs are highly interconnected. The accuracy rate of Random Forest model is high, but the recall rate is low, which stemmed from the class imbalance of the da-taset. Nevertheless, the risk probabilities are useful in prioritizing maintenance task in ma-chines. Feature importance analysis identifies that among the five features of the predictive model, rolling downtime is the most important feature. The main contribution of this study is to develop a decision support framework. The machines are ranked based on their failure probability achieved from the predictive model. Also, the underlying downtime reason of each machine is identified through machine-specific Pareto analysis. By combining these two information, actionable recommendations are generated for each machine. The thesis also provides a practical implementation roadmap for the company. The decision-support framework generated by this study is not just an analytical tool, it’s a practical solution for SMEs or resource constrained manufacturing companies.