|M.Sc Student||Burshtein Uri|
|Subject||Design and Autotuning of Decentralized Multivariable Dead|
|Department||Department of Mechanical Engineering||Supervisors||Professor Emeritus Zalman Palmor|
|Professor Leonid Mirkin|
The Dead Time Compensator (DTC) controller has the potential to improve considerably the performance of control systems with significant delays. This was shown and demonstrated in numerous investigations and led to its widespread usage in industrial SISO (single input single output) applications. The DTC controller consists of a primary controller and an internal feedback that contains the plant’s model. The improvement in performance gained by the DTC largely depends on its tuning. This is quite a difficult task in the SISO case but many times more complicated in the MIMO (multi-input multi-output) case as the extension of the SISO DTC to its MIMO version is not straight forward.
This research presents the first step towards the development of a complete auto-tuning algorithm for MIMO DTCs with decentralized primary controllers. The goal of
this work was twofold, to identify the most suitable structure for those controllers and to develop simple, robust and practical methods for their tuning that are suitable for industrial applications.
Three decentralized DTC structures for MIMO systems were investigated. One was found to be incapable of extending the DTC properties from the SISO case to the MIMO one. While the second structure extends most of these properties, the third one achieves a full generalization by adding an artificial time delay to the plant. However,
through a comparative study it is demonstrated that despite its complicated structure the latter does not lead to improvements in performance, relative to the second structure, in the presence of uncertainties. Hence, it is concluded that the second structure is best suited for the development of the tuning algorithm.
Two design and tuning methods for DTCs with decentralized primary controllers were developed in this research. The first is the ELSC (Equivalent Loop Sequential Closing) method, where the primary controller can be designed using the conventional Bode or Nichols diagrams incorporating parametric uncertainties.
The second is the EVS (EigenValue Sensitivity) method. This method is suitable for automatic tuning since it requires a relatively small amount of calculations. The method consists of two stages: In the first stage, all the primary controllers are tuned as SISO controllers for SISO plants. In the second stage the algorithm automatically reduces the primary controllers’ gains to provide adequate robustness in the presence of interactions.
The EVS algorithm was tested via simulations on various plants and was found to perform satisfactorily despite of its simplicity.