Çevrimiçi model öngörülü kontrolün pratik endüstriyel uygulamalarıyla etkili bir uygulaması

The demand to utilize modern control algorithms for industrial applications is much more intensive. Model-predictive-controller (MPC), which is one of the modern optimal control policies, has gained more attention in servo drive and other industrial applications in recent years due to increased computational capabilities of embedded platforms and evident control performance benefits compared to more classical control methods. A digital MPC algorithm at each sampling instant produces the optimal control input sequence for a given prediction horizon while also guaranteeing that input and state-trajectories do not violate some set of constraints. Its optimization based capability brings more flexibility to include the additional requirements such as energy efficiency, quality of the systems' control input. Solving constraint optimization problems in each step requires excessive computational complexity and burden, which is the main drawback of online MPC over classical methods. In this thesis, we demonstrate the feasibility of online MPC in high sample frequency applications and provide some suggestions for practical implementation. We implemented the existing dual active set solver by replacing two common methods in the matrix update step to increase the performance in terms of execution speed. We also provide the linear approximation for the nonlinear constraints by taking the tradeoff between accuracy and speed into account. The proposed structure is successfully verified via both PIL simulation and experimental testing. In addition, two different processors, which are commonly used in motion control applications, perform the PIL simulation and experimental testing separately to certify the feasibility of our implementation in terms of execution speed and using minimal memory space.