Industry applications of multivariable control

Sammanfattning: In the face of environmental regulations, optimization of industrial processes becomes necessary. This doctoral thesis summarizes the results of three application-driven projects in automatic control that were aimed at process optimization in the steel industry. The objective of the projects was to apply advanced control strategies to two important processes in steel making, namely pulverized coal injection (PCI) in blast furnaces and LD converters. Firstly, an LQ multivariable controller with gas leakage detection system for PCI vessels is designed and analyzed. Secondly, a foam level control system for the LD converter process using an audio signal for measurement is designed. Thirdly, it is attempted to create a single line flow control system for PCI using a video camera. In the latter two cases the conservative approach of inferring unmeasurable physical quantities from the audio and video sources is used. Moreover, all the designs are tested through implementation or experiments at the industrial plant. The control and gas leakage detection system ended up as a full-scale industrial implementation, whereas the projects comprising audio and video information is still at an experimental stage. Work with implementation and experiments pays off in experiences and further insights in the application of control theory, and reveals weaknesses and gaps in the existing theory. Thus, application-driven projects lead to practical solutions and at the same time pose new theoretical challenges. Consequently, this chain of events is favorable to both practitioners and theoreticians, and in turn stimulates the collaboration of industry and academia. Unfortunately, in many research projects this sequence is reversed which complicates technology transfer into industry. As a spin-off effect from the multivariable control project of the PCI process two topics are addressed anew. In the problem of measurement/actuator pairs assignment for decentralized control, the geometrical background of Gramian-based interaction measures is clarified. It is shown that weighted Gramian-based interaction measures can be effectively used for control structure design. In control structure improvement of multivariable control systems, it is shown that improvement potentials can be deduced from coarse models of the closed-loop system. Finally, in the projects comprising audio and video signals in control applications, it is concluded that the theory is rather undeveloped and that these sources should be treated as a multivariable system.