7.11.2024
Doctoral thesis on control performance analysis
M.Sc. (Tech.) Mehmet Yağcı’s doctoral thesis in Process and Energy Technology (Process and Systems Engineering will be put forth for public defence at the Faculty of Science and Engineering at Åbo Akademi University.
The thesis is entitled Control Performance Analysis with Industrial Scale Applications.
The public defence of the doctoral thesis takes place on 15 November 2024 at 1PM in auditorium Helikon, Arken, Tehtaankatu 2, Turku. Professor Paweł Domański, Politechnika Warszawska, Poland, will serve as opponent and Docent Jari Böling, Åbo Akademi University, as custos.
Summary
Control Performance Monitoring (CPM) has been an active research topic as automation systems have become more complicated, along with the increasing number of control loops in industries. A continuous monitoring solution is needed as the controllers deteriorate for various reasons. The industries build their periodic health-check activities, but unfortunately, loss of potential is unavoidable as the control engineers are fully occupied with multiple tasks. In addition, a few hundred controllers per control engineer multiplies the problem of resource scarcity. These emphasize the need for online or automatic monitoring applications.
This thesis roots back to an in-house CPM development project, where the author was the primary developer. This thesis work addresses the weaknesses identified during that project. The overall aim is to ease the need for user input by improving established methods. Minimum variance control (MVC), a widely used benchmark in performance assessment, fails in the presence of oscillations. The integrated oscillation detection method proposed in this thesis solves that problem by identifying the characteristics of disturbances. Apart from that, a fractal-based and delay-free assessment approach, the Hurst exponent, suffers from the crossover effect in the presence of oscillations. The proposed method in this thesis solves this issue and automates window length selection to avoid the crossover effect. The second paper deals with plant-wide disturbances and helps to build a complete map of disturbance propagation paths by integrating a frequency domain-based method called spectral envelope and a model-based method Granger causality. In addition to that, in the last paper, a brand-new CPM philosophy is presented by utilizing classifiers from machine learning. It helps to create a more user-friendly and less coding-required CPM tool. All the proposed methods have been tested on industrial controllers and proved to work well in different cases where the conventional methods fail to detect issues. The studies have been presented at various conferences and published in the associated proceedings.
Mehmet Yağcı can be reached by email yagci.mehmet@hotmail.com.
The doctoral thesis can be read online through the Doria publication archive.