Determination of parameter regions for diagonal dominance and stability of MIMO systems

Most of the industrial plants include more than one input and output variable. Compared to Single Input Single Output (SISO) systems, such systems include different structural properties. For instance, an output variable is effected by all input variables in general. On the other hand, in terms of controller structures, researchers have focused on two main approaches for such systems, which are "centralized" and "decentralized" controllers. However, it can be proposed that decentralized controllers are preferred more in practice due to various reasons like less number of tuning parameter, possibility to apply single loop controller design methods, ease of use for operators etc. Whereas, in general, performance and efficiency of such controllers reduce when there are significant interactions between different input-output pairs in a Multi Input Multi Output (MIMO) system. Reducing the interactions between different input-output pairs in MIMO systems is crucial in terms of decentralized controller design due to the previously mentioned reasons. Diagonal dominance which is a weaker condition compared to decoupling, is one of the approaches that can be used to reduce interactions in MIMO systems. One input variable is strongly related with one specific output variable in diagonal dominant systems. One of the main aims of this thesis is to determine controller parameter regions that achieve diagonal dominance conditions. Additionally, it is also aimed to determine stabilizing parameter spaces, since diagonal dominance does not indicate stability in general. As a result, controller parameter regions that achieve both diagonal dominance and stability conditions in closed loop are determined in this thesis as the first step of decentralized controller design. In literature, the diagonal dominance concept has gained attraction since the pioneering studies of Rosenbrock in early 1970s. However, in the meantime most of the researchers focused on determining a specific controller parameter pair that optimizes a predetermined condition. Such a case may restrict the designer in the next steps of the design process. Additionally, the number of studies are limited that investigates the diagonal dominance characteristics of the determined controller parameters in case uncertainties or checks how the system is close to the diagonal dominance boundaries. Two Input Two Output (TITO) systems are special subset of MIMO systems since in practice many MIMO systems can be treated as several TITO subsystems as proposed in literature. In terms of diagonal dominance, particularly, TITO systems and diagonal type controllers are discussed in detail, since it is aimed to determine necessary and sufficient conditions on diagonal dominance in terms of controller parameters. For such systems, exact conditions on the controller parameters in terms of both column and row diagonal dominance are derived at a given fixed frequency. Derived results are also valid for finite number of frequencies and practically applicable for a given frequency range. Moreover, weighting factors are added to the original definition of diagonal dominance in order to derive controller parameter regions that achieve better diagonal dominance ratios. Necessary and sufficient conditions on diagonal type controllers are also derived for the weighted diagonal dominance problem. Lastly, critical frequencies that may possibly change the interval characteristics of static diagonal controllers for the column diagonal dominance are derived. Effectiveness of the derived results in terms of diagonal dominance are demonstrated over several case studies using Gershgorin Disc plots and diagonal dominance ratio plots. On the other hand, a Lyapunov equation based stability mapping approach is proposed within the scope of this thesis to derive stabilizing controller parameter spaces of a given MIMO system. In the present approach, it is not necessary to calculate singular frequencies or apply frequency sweeping that most of the frequency based approaches require. From the Lyapunov point of view, positive definiteness of the Lyapunov matrix P(k) is necessary and sufficient for LTI systems. However, considering the numerators and denominators of the leading principal minors it is required to solve 2n parametric equation in order to determine positive definiteness of P(k). This number is reduced to n+1 at the first step. After that, Lyapunov matrix equation is reduced to the standard set of equation representation using the Kronecker products and vectorization operator. At this point, a new matrix M(k) is defined over the Kronecker products and it is shown that determinant of M(k) is the product of binary combinations of A(k). Using the relations between the system matrix A(k), Lyapunov matrix P(k) and M(k), it is shown that it is sufficient to solve at most 2 parametric equations which are |M(k)|=0 and |M(k)|->infinity. Determinant of M(k) includes redundant multiplications of binary combinations of eigenvalue pairs of A(k) due to the matrices P(k) and Q that are used in Lyapunov formulation are symmetric. In order to eliminate the redundant multiplications and reduce the computational complexity, elimination and duplication matrices are introduced as transformation matrices. In addition to MIMO systems, the proposed stability mapping approach is applicable to a broad range of systems, further system classes and sub problems where Lyapunov formulation is possible. In order to demonstrate these properties of the proposed approach, firstly, controller integrity problem of MIMO systems is discussed in detail. An approach is proposed to determine stabilizing controller parameter regions even in case of possible failures related with controller parameters. A benchmark case study is included and effectiveness of the proposed approach is shown over a comparative study with a currently existing approach. Additionally, discrete time systems is also discussed in detail to demonstrate the further application areas of the proposed Lyapunov equation based stability mapping approach. In this case, the structure of the Lyapunov equation varies slightly compared to the continuous time case. Another benefit of the proposed Lyapunov equation based approach is the opportunity to determine analytical expressions of stability boundaries. So that, it becomes possible to use Lyapunov equation based stability mapping approach in optimization based approaches by inserting the stability boundaries as constraints on such approaches. This case is also addressed through the robust Model Predictive Control (MPC) problem. Analytical stability boundaries which is derived in the off-line phase using the proposed stability mapping approach is inserted to the robust MPC problem formulation to achieve stability. In this way, robust MPC problem is transformed into the nominal MPC problem. The effectiveness of the proposed method is also demonstrated through a benchmark system that is frequently used in the literature. Diagonal dominance proposes weaker conditions compared to decoupling. As a result, it becomes possible to determine controller parameter regions that achieve diagonal dominance in case of parametric uncertainties. Within the scope of this thesis, two conservative approaches which are based on triangular inequality and griding are proposed for the systems that include interval type uncertainties in Transfer Function Matrix (TFM) elements. Using these approaches diagonal dominance problem of a parametric uncertain system is transferred to the weighted diagonal dominance problem of the nominal plant. After that, previously derived results are used to determine static diagonal controller parameter regions. Lastly, stability of parameter uncertain multivariable systems is discussed in order to determine robustly stabilizing parameter spaces. There are two main assumptions on uncertain parameters in literature. In the first assumption, there is no restriction on uncertain parameters and it is aimed to determine all uncertain parameter spaces that preserve stability of the closed loop system. In this case, proposed Lyapunov equation based stability mapping approach is directly applicable. Contrary to this approach, many methods that is currently available in the literature include the results obtained by making some assumptions on the number and the type of uncertain parameters. The validity of the Lyapunov equation based method has been demonstrated through different benchmark case studies. On the other hand, in some cases, it is assumed that upper and lower bounds of uncertain parameters are known. It is aimed to determine whether the whole polynomial family is stable in all cases where the uncertain parameters take any value between these known intervals. In some special cases, it was shown in literature that stability of finite number fixed polynomials guarantee the stability of whole uncertain polynomial family in case of SISO systems. However, the characteristic polynomial of MIMO systems includes the multiplication of free controller parameters and individual transfer functions even in the simplest cases. As a result, it can be proposed that compared to SISO systems, it is more difficult to determine the controller parameter areas that provide robust stability in such systems. In the discussed problem characteristic equation includes both uncertain parameters that have known upper and lower bounds and free controller parameters. In this thesis, an approach is presented to determine robustly stabilizing parameter spaces using the Kharitonov Theorem in accordance with the Lyapunov method by applying overbounding method on characteristic polynomial coefficients. The proposed method reduces the computational complexity significantly, since Kharitonov Theorem is used. However, it must also be noted that calculation of invariant controller parameter sub regions in terms of overbounding also introduces additional analysis steps. As a conclusion, in this thesis, it is mainly focused on determining controller parameter regions of the diagonal type controllers that make both nominal and parametric MIMO systems diagonal dominant and stable. The results are derived through TITO systems from the standpoint of diagonal dominance, since it is aimed to determine the necessary and sufficient conditions. On the other hand, there is no restriction on the system and controller type for the proposed stability mapping approach.