Högskolan i Skövde

Within the industrial sector, there has not been much research from the scientific community on how to manage data to create a well-structured dataset. The majority of case studies, and articles in magazines and conferences, work with data that is already classified and structured. The present project aims to expose how to manage multi-source and multi-format datasets through Python coding and statistical analysis to categorise the stop data of machines from a production line, working as a continuation of the approach developed in the previous thesis from Soman (2021). The machines oroperations may be stopped for a variety of reasons, including scheduled and unscheduled stops. Scheduled stops are lunch breaks, weekends, maintenance, tool changes etc. Unscheduled stops are real breakdowns, bottlenecks etc. It is critical to maintain track of these stops to diagnose inefficiencies such as low throughput and significant cycle time fluctuation in subsequent production simulation analyses. Having an automated procedure to filter these stops will save time and improve simulation accuracy increasing the productivity of the company. Several approaches are proposed to combine the different sources of data in order to obtain automation. As a result of creating a new feature called strength and combining the shift, maintenance and stop data, further progress in detecting faulty stops was achieved. Maintenance data combined with stop data is found to capture true stops from the machines. Shift data combined with stop data capture false stops. The strength feature serves as an approach to capture if a group of machines in the line stopped together, indicating that there is a low chance of randomness and therefore a high probability of a faulty stop.