- 271 views
- 473 downloads
Data-driven Process Monitoring and Fault Detection with Convex Geometry
-
- Author / Creator
- Zeng, Qi
-
Nowadays, industrial processes are becoming highly complex and integrated due to the applications of advanced distributed control systems. As multiple production units with thousands of actuators are operating at the same time, the reliability issue of process plants naturally arises. To ensure the operational safety and maintain the product quality, process monitoring is one of the most critical and challenging topics in today's industrial control designs. In recent decades, multivariate data-driven monitoring techniques have gained a lot of attentions, due to their relatively inexpensive implementation and good performance. However, these technique based on the statistical modeling, are mostly built upon certain assumptions on the process data. Once the measurements violate these assumptions, the monitoring performance is hard to be guaranteed. To seek for a feasible solution of such potential issue, this thesis is to develop an assumption-free data-driven fault detection method which could be more applicable in the industrial practices. To avoid data distribution fitting in process monitoring designs, we propose a new approach to model the process normal operating behavior with a geometrical enclosure, namely, a convex hull, from the normal process data. In order to achieve on-line monitoring, an appropriate detection metric with the corresponding threshold has been developed based on the property of convex hulls. We also introduced a parallel coordinates based high-dimensional visualization tool to facilitate the visual presentations of the detection results. The performance of the proposed method is demonstrated with simulation experiments of a continuous stirred tank heater and a benchmark plant from the Tennessee Eastman challenge problem. The results have been compared with three conventional data-driven techniques, including the principle components analysis, partial least square and one-class support vector machine, in terms of the detection rates and detection delays. Both case studies reveal the advantages of the proposed method in detecting randomized disturbances over the other methods.
-
- Graduation date
- Fall 2016
-
- Type of Item
- Thesis
-
- Degree
- Master of Science
-
- License
- This thesis is made available by the University of Alberta Libraries with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.