Çok değişkenli PID kontrol döngülerinin performans değerlendirmesi ve kök neden analizi

Industrial processes are controlled with different types of controllers. Among those, PID controllers are very popular. These controllers manipulate the process inputs to keep the process stable, to reject disturbances and to regulate the process outputs at the desired set-points. PID loops include different digital and mechanical hardware. In general, control engineers at the plants are responsible for the maintenance and performance of these loops. Nevertheless, with the increasing number of the control loops at the plants, diagnosis of each loop becomes tedious. Control Loop Performance Monitoring (CPM) philosophy aims to solve this problem. Incorporating basic to advanced algorithms, CPM tries to deliver the most accurate assessment of control loops discovering the underlying causes for performance degradation. While doing that, CPM also does not manipulate the processes, instead it analyzes the already generated data. These analyses can be listed as performance assessment, oscillation detection, non-linearity detection, stiction detection and quantification, controller tuning assessment and root cause analysis. In this study, each component building the CPM philosophy is examined. The methods for each analysis component are selected based on industrial popularity. The methods are described and their strengths and weaknesses are discussed extensively. The analysis starts with the assessment of control loop performances in terms of both stochastic and deterministic techniques. Then, a common problem, oscillation, is explained and the possible methods to detect oscillations in control loops are discussed. After that, non-linearities which are the reasons behind the oscillations are discussed. Then, stiction problem which is the most frequent cause of non-linearities in control valves is described. Later, multivariable PID controllers are assessed to detect the root causes. Finally, the examined methods are connected in a single algorithm aiming to detect the root cause of performance degradation. The examined methods and the overall algorithm were tested in a plant of T¨ UPRAS¸ ˙ Izmit Refinery. The advantages and disadvantages of the methods examined are explained. The analysis showed that 47% of the loops have acceptable performance. The reasons of performance degradation were found out as oscillations, stiction and tuning problem. Finally, the root cause analysis shows that oscillations propagate and affect the other loops. 4 possible candidates for plant-wide disturbance are detected in the plant.