A data-driven approach to improve process robustness for bolted critical joints

Sammanfattning

In the current world, the companies are inclining towards advanced, rapidly evolv ing industry 4.0 technologies for meeting dynamic customer demands and producing products at a much higher volume, less cost, and more efficiently. These technologies facilitate communication links between different systems that talk with each other to use intelligent manufacturing techniques, industrial digitalization, cloud systems, Internet of Things (IoT), and autonomous machines to acquire, store and efficiently handle big data. Some companies hold an edge over their competitors due to their extensive knowledge of processing raw data into valuable insights. The exploration and exploitation of the analytics and visualization techniques using Data Science, Data Mining (DM), and Machine Learning (ML) have become the critical enablers for data-related technologies, forming the base of this thesis. This thesis study aims to develop a data-driven approach using ML and visualization techniques to build a predictive model and a dashboard for improving the robustness of the bolted critical joint process in a truck production assembly line. Additionally, the study will shed some light on the harmful effect of poor data quality. Moreover, the non-robust process was directly and indirectly impacting various departments in manufacturing such as Quality, Production, Audit, Maintenance, Procurement, and internal information technology (IT). Additionally, it became evident that the existing solution by the third-party provider was not used for developing data in sights and used only as a reactive solution. A supervised ML classification model was developed to identify critical parameters influencing the tightening process outcome and predict this outcome by extracting data from different data sources. A visualization solution with a dashboard is de signed to direct the attention of all levels of management to reduce the use of manual actions/backup routines. The results of this study, along with discussions, are pre sented. The ML model results were promising, with high accuracy for the training dataset, test dataset, and cross-validated with live data. The ML model and dash board can be used as a standard framework for developing sustainable models in the future, provided the data for the model is automated. The limitation of the study is that some of the results can be generalized, whereas others are not. Furthermore, this study’s practical and academic contributions are highlighted along with future recommendations.