UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Automatic model structure determination for adaptive control Kotzev, Anat


This work is a study of adaptively controlled systems with plant model structures that may vary due to changing operating conditions. Most closed loop adaptive control algorithms use identification methods for determination of the parameters in fixed structure models. Those parameters, once estimated, are assumed to be correct and uncertainties in the values are ignored. If the structure of the plant dynamics changes on-line, the incorrect model can lead to poor performance and instabilities. The adaptive algorithm used i n this work is the Generalized Predictive Control (GPC ) algorithm. It is reported to be capable of handling a number of simultaneous problems and therefore was chosen. Along with handling on-line changes of parameters, it claims to overcome nonminimum-phase plants, open loop unstable plants, plants with badly damped poles, plants with variable or unknown time delay, and plants with unknown order. The goal of this research is to investigate and study GPC with the on-line changes in the model structure of the plant, and corresponding changes in the order of the estimated model for G P C and the structure of the controller and as well as to propose a method that detects on-line, the need for model order changes and determines the correct one. There are at least two major sources for structure variations in the estimated model. The first is the model actually being time variant and the second resulting from the use of inherently nonlinear systems and mis-modeling. Two applications exemplifying these variants were selected to examine the techniques developed in the thesis. The first is a single flexible link manipulator, whose changes in model structure are due to new excited vibration modes. The second is a two link rigid manipulator with hydraulic actuators causing the system to be highly nonlinear, whose model could change due to changes in operating points. The effect of mis-modeling on the total system performance and stability was assessed. A cost function was used as a measure of the closed loop controlled system reaction to under, correct and over-modeling. Its effectiveness in terms of stability and performance was measured i n context of the two applications. In addition, experimental data from open loop identification of the dynamic model of a 215B Caterpillar, an excavator type machine, confirms the study of the behavior of the cost function for those conditions. Based on the behavior of the cost function a new algorithm was developed. The MOD (Model Order Determination) algorithm detects, determines and executes, on-line, changes to the model order. It was implemented for both application which were controlled with the GPC algorithm. The results show that good performance and stability can be achieved. The main contributions of this work are: • The MOD algorithm which based on the behavior of a cost function, corrects on-line mis-modeling of adaptively controlled systems while maintaining good performance. • GPC was successfuly implemented for hydraulically actuated manipulators. On-line automatic change of the GPC output horizon was introduced to achieve sufficiently fast transient response and avoid overshoots. • Experimental data from a 215B Caterpillar manipulator proved the need for a closed loop approach.

Item Media

Item Citations and Data


For non-commercial purposes only, such as research, private study and education. Additional conditions apply, see Terms of Use https://open.library.ubc.ca/terms_of_use.