Not logged in.
Quick Search - Contribution
Contribution Details
Type | Master's Thesis |
Scope | Discipline-based scholarship |
Title | Bottlenecks in a Multi-Variant Production Setting: Detection, Analysis, Visualization and Prediction. |
Organization Unit | |
Authors |
|
Supervisors |
|
Language |
|
Institution | University of Zurich |
Faculty | Faculty of Business, Economics and Informatics |
Date | 2021 |
Abstract Text | In this digital age, manufacturing companies generate large volumes of machine data that can be used to make data driven decisions using computerized algorithms. It is well-known that the productivity of any production line is constrained by throughput bottlenecks. These bottlenecks if detected at the right time can allow maintenance engineers to proactively plan resources to eectively mitigate such bottlenecks and achieve a higher throughput across the production line. In order to provide support to such an exercise, this masters thesis dealt with building a bottleneck detection model using real-world production data, implementing a working formulation of the detection model on a real production line, further analysing the nature of the observed bottlenecks, and using machine learning algorithms to predict future bottlenecks in the production line. A detection mechanism was build to compare the dierence between consecutive products arriving at all machines with their takt time (threshold time). Through a practical demonstration, the said framework was used to detect not just the bottlenecks but also its severity. For predicting future bottlenecks on production machines, a threshold mechanism was developed to predict if the next time step will be a bottleneck or not. In order to do this, four machine learning algorithms were used to learn their distribution, both on single-step and multi-step (auto-regressive) frameworks for all machines. All models were implemented on a validation set and the model with the lowest MSE score was used for prediction on the test data at each machine respectively. On the single-step framework, we observed an overall accuracy score of 76.7%, while an accuracy of 90.4% was observed on the multi-step framework. |
PDF File | Download |
Export | BibTeX |