- 21 views
- 21 downloads
Oscillation Detection and Root Cause Analysis with Machine Learning
-
- Author / Creator
- Memarian, Amirreza
-
Abnormal events in chemical processes present significant challenges, often disrupting the performance of industrial operations. Accurate detection and diagnosis of abnormal behavior among control system elements are crucial to prevent potential process degradation and associated economic losses. This study aims to address the complex task of identifying and characterizing such anomalies, with a particular emphasis on oscillations and their two common root causes: control valve stiction and poor controller tuning. Oscillations, a recurrent issue in process industries, can have
far-reaching consequences by compromising stability and undermining product quality. Control valve stiction, another major concern, introduces inefficiencies that not only hinder smooth operations but also degrade the overall quality of the end product. Furthermore, the importance of proper controller tuning cannot be overstated, as it directly impacts control loop performance and, consequently, the efficiency and stability of the entire process.
In response to these multifaceted challenges, the study aims to provide innovative solutions that go beyond traditional approaches. By introducing novel methodologies rooted in cutting-edge techniques and leveraging data-driven analyses and machine-learning approaches, the study seeks to empower industries to proactively address these issues. The research showcases the potential of shape-based pattern recognition, deep learning, and advanced data processing techniques to revolutionize the detection and diagnosis of abnormalities in chemical processes. With an unwavering
commitment to enhancing industrial operations, the study strives to establish a path toward achieving heightened precision, efficiency, and promptness in the identification
of abnormal events. Through a comprehensive exploration of the complexities associated with oscillations, control valve stiction, and poor controller tuning, the objective is to provide industries with the necessary resources and understanding to ensure process stability, elevate product quality, and maximize operational efficiency. The study delves into shape-based pattern recognition for detecting oscillations in process control loops. Two distinct methods are proposed, one utilizing nonlinear algebraic functions and the other harnessing deep convolutional neural networks (CNN). Both methods aim to identify oscillations by recognizing distinct trianglelike
patterns in data plots. Evaluation across diverse industries demonstrates their effectiveness in detecting oscillatory control loops. In the realm of control valve stiction
detection, a unique method is introduced, combining Markov Transition Field (MTF) and CNN. By transforming process variable data into images using MTF and training CNN with the MTF images, the proposed methodology effectively distinguishes stiction-induced oscillations from other types of oscillations. The integration of transfer learning enhances the stiction detection capability. Application to benchmark
control loops confirms the robustness of this approach. The study also explores poor controller tuning detection using Gramian Angular Field (GAF) and Stack Autoencoder (SAE) techniques. These advanced tools offer real-time monitoring and alert operators about poorly tuned controllers. Case studies conducted on diverse datasets validate the accuracy and practicality of the proposed method. -
- Subjects / Keywords
-
- Graduation date
- Spring 2024
-
- 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.