Thesis for the Degree of Master of Science


Year 2006
Esa Järvinen

Fault Tolerant Control of a Petroleum Refinery Unit

Fault tolerant control (FTC) aims to mitigate and/or prevent the harmful effects of faults in sensors, actuators and plant equipment, especially in control systems, and to thereby increase the availability of plants.

The aim of the master's thesis was to apply the concept of FTC to a petroleum refinery unit operated under Model Predictive Control (MPC). Furthermore, performance of different FTC strategies was evaluated in the face of incipient faults in online process analysers. The studied process unit is the solvent de-aromatisation unit located at the Naantali oil refinery of Neste Oil Oyj.

In the literature part of the thesis, the concepts of fault tolerant control and Model Predictive Control are discussed. Fault-tolerant MPC applications reported in scientific literature are also presented. The experimental part is focused on developing and comparing alternative FTC strategies for the studied process unit. The comparisons between the strategies are finally made by performing simulation studies.

Fault tolerant control strategies were implemented by exploiting the features of the multivariable MPC controller currently used for controlling the process unit. The following strategies were developed and tested: feedback deactivation, measurement reconstruction, target manipulation, constraint manipulation and online re-tuning of the controller.

Slowly increasing, incipient faults were introduced to the results of online process analysers. It is shown that without FDI/FTC, the quality control objectives were finally lost because of the faults. Also, it is stressed that the final consequences of the faults depended on the direction of the faults.

It was discovered that feedback deactivation offered satisfactory performance in the face of downward faults, whereas target manipulation was also recommended for upward faults to remove a harmful steady state offset. Fault-tolerant capabilities of the controller were further improved by introducing early automatic online re-tuning actions. Preventive re-tuning of three different MPC parameters was based on a fault detection reliability index provided by the related FDI system. The usage of a fault detection reliability index and the strategy of preventive re-tuning proved effective and both are recommended for systems with slow dynamics and/or infrequent sampling rates of analysers.

This info last modified 21 Aug 2018 by Jerri Kämpe-Hellenius