Issn 1392 - 1207

Model identification is one of the most important requirements for advanced control systems. Fast response and quick disturbance recovery of an advanced controller [1, 2] cannot be achieved without an accurate model of the plant. Although after years of development linear methods for model identification [3 5] have become mature, nonlinear identification methods are still needed in high performance applications to maintain sufficient identification accuracy over the entire operation range. When analytical nonlinear methods, e.g. leastsquared algorithms based on quadratic error functions [6], are not able to find a global solution for nonlinear systems, a promising alternative to these traditional methods is the swarm intelligence based optimization algorithms. In references [7-11], Genetic Algorithm (GA) is applied to some optimization problems and good results are achieved, however, the number of manipulations and required memory increases with and increase in the dimension of searching space. Therefore it is difficult for the GA algorithms to maintain searching velocity and convergence for the realtime identification applications. Referring to [12-15] Particle Swarm Optimization (PSO) algorithm acts well when dealing with multimodal and multidimensional problems. PSO algorithm is theoretically simple, and computationally efficient. It provides many advantages for complex engineering problems. In this paper, a parameter identification approach using Adaptive Particle Swarm Optimization (APSO) is utilized, which is a development of PSO based identification approach [13]. Furthermore, the proposed method is applied to a real ball on plate setup, which is a typical example of highly coupled, nonlinear electromechanical system. The ball on plate system is an extension of the traditional ball and beam [16] that moves a ball on a rigid or flexible surface. The mechatronic design of a ball on plate, which uses inexpensive materials with a simple setup, can be valuable. In addition, control of such a 2D coupled and nonlinear dynamical model is a very interesting and challenging example for studying and testing various control methods. In principle, rolling resistance of a ball on plate system is required for modeling the system, but in practice, the varying nature of such a resistance with the ball’s material, mass and dimensions make it difficult to be measured. To the best of our knowledge, such a real-time identification solution has not been reported before. It is the first report of identifying this critical value for modeling a ball on plate system. The rest of this paper is organized as follows: Section 2 provides a brief introduction on ball on plate mechatronic design. System modeling and problem formulation will be discussed in Section 3. APSO will be briefly introduced in Section 4. Section 5 shows an APSO approach for ball on plate parameter identification. Simulation results and experimental verification are presented in Section 6 and 7 respectively and concluding remarks are given in Section 8.