Abstract
By now, the theory of Model Predictive Control is well known for several decades. In the beginning, it was utilized to control processes which posses large time constants. It therefore has been used most notably in the chemical process industry. Over the years, the development of computer systems enabled the application to plants with much smaller time constants due to the available computing power required. Hence, it also has drawn interest within the field of electric drives and machines. Since then, Model Predictive Control has evolved to a practical control method, for which reason it is commonly used beside conventional control methods.
Disregarding some drawbacks, this theory also has its outstanding strengths which in turn can be utilized advantageous.
In this master thesis, Model Predictive Control is applied to a Squirrel Cage Induction Machine and a DCMachine. Thereby, especially the current control is investigated. To be more precise, a current controller for the Induction Machine as well as an armature current controller for the DCMachine is inspected. The excitation current controller is discounted since it is not really challenging. For the applied control strategy, Continuous Control Set as well as Finite Control Set is probed. For the Induction Machine, an additional distinction is made for Finite Control Set since Exhaustive Search as well as Modified Sphere Decoding is investigated.
After giving a short Introduction to Model Predictive Control and providing some fundamental definitions which are of particular importance in this thesis, an appropriate mathematical model for the Induction Machine is derived. In order to compensate any inaccuracy due to a model fault, the model is augmented by an embedded integrator. Afterward, a convenient cost function is designed in order to fulfill the desired control purposes. Subsequently, this allows it to formulate an optimization problem accordingly. In a further step, the optimization problem
is adapted onto the inspected control strategies. For each strategy, an algorithm to solve the problem in order to find the optimal solution is gathered. Since Model Predictive Control allows an expedient consideration of constraints, the latter are also integrated into the optimization process. Following the same course of action for the DCMachine finally also allows it to the design an armature current controller. Merely the mathematical model as well as the constraint consideration has to be adapted. The cost function design as well as the optimization problem formulation in general stays the same.
Since the mathematical model allows it to predict the future states, one is able to compensate the unit delay which is inevitably present due to the discrete nature of real systems. This is one of the big advantages that comes along with Model Predictive Control, especially in comparison to linear PIcontrollers. However, also for linear control some countermeasures can be clasped by taking advantage of a so called SmithPredictor. As long as there is a sufficiently accurate estimate available for the plant, one can obtain similar results as with the Model Predictive Controller. If the estimate is inaccurate though, thinking of a model fault for example, this strategy provides rather suboptimal results. Due to the proper acting of the embedded integrator, the strength of the Model Predictive Controller appears here again. This comparison is also addressed in this thesis by means of simulations.
In the end, beside providing a small overview on the measurement setup, the measurements performed on the machine test bench in the laboratory of the Electric Drives and Machines Institute at Graz University of Technology are presented. The intention is to depict some selected measurements in order to validate the function of the investigated controllers, as well as to highlight appearing differences between the individual control strategies.
Disregarding some drawbacks, this theory also has its outstanding strengths which in turn can be utilized advantageous.
In this master thesis, Model Predictive Control is applied to a Squirrel Cage Induction Machine and a DCMachine. Thereby, especially the current control is investigated. To be more precise, a current controller for the Induction Machine as well as an armature current controller for the DCMachine is inspected. The excitation current controller is discounted since it is not really challenging. For the applied control strategy, Continuous Control Set as well as Finite Control Set is probed. For the Induction Machine, an additional distinction is made for Finite Control Set since Exhaustive Search as well as Modified Sphere Decoding is investigated.
After giving a short Introduction to Model Predictive Control and providing some fundamental definitions which are of particular importance in this thesis, an appropriate mathematical model for the Induction Machine is derived. In order to compensate any inaccuracy due to a model fault, the model is augmented by an embedded integrator. Afterward, a convenient cost function is designed in order to fulfill the desired control purposes. Subsequently, this allows it to formulate an optimization problem accordingly. In a further step, the optimization problem
is adapted onto the inspected control strategies. For each strategy, an algorithm to solve the problem in order to find the optimal solution is gathered. Since Model Predictive Control allows an expedient consideration of constraints, the latter are also integrated into the optimization process. Following the same course of action for the DCMachine finally also allows it to the design an armature current controller. Merely the mathematical model as well as the constraint consideration has to be adapted. The cost function design as well as the optimization problem formulation in general stays the same.
Since the mathematical model allows it to predict the future states, one is able to compensate the unit delay which is inevitably present due to the discrete nature of real systems. This is one of the big advantages that comes along with Model Predictive Control, especially in comparison to linear PIcontrollers. However, also for linear control some countermeasures can be clasped by taking advantage of a so called SmithPredictor. As long as there is a sufficiently accurate estimate available for the plant, one can obtain similar results as with the Model Predictive Controller. If the estimate is inaccurate though, thinking of a model fault for example, this strategy provides rather suboptimal results. Due to the proper acting of the embedded integrator, the strength of the Model Predictive Controller appears here again. This comparison is also addressed in this thesis by means of simulations.
In the end, beside providing a small overview on the measurement setup, the measurements performed on the machine test bench in the laboratory of the Electric Drives and Machines Institute at Graz University of Technology are presented. The intention is to depict some selected measurements in order to validate the function of the investigated controllers, as well as to highlight appearing differences between the individual control strategies.
Translated title of the contribution  Model Predictive Control für elektrische Antriebe 

Original language  English 
Awarding Institution 

Publication status  Published  2022 