![]() | Only 14 pages are availabe for public view |
Abstract Most of balancing systems like inverted pendulum, ball and beam, and ball on plate system (BOPS) suffer from control problems like point stabilization and Trajectory tracking problems also these systems suffer from system problems like high nonlinearity, unstable in open loop, multi-inputs and multi outputs so many control algorithms is used to solve these problems like PID (ProportionalIntegral-Derivative), Fuzzy, sliding mode, and many other algorithms but there is a problem is noticed in the BOPS (Ball on Plate System) that most of scientific papers used a small range of plate angle and this small plate angle make the system slower and has an overshoot so this thesis uses a large plate angle to make the system faster, decrease or eliminate the overshoot, enhancement the point stabilization and trajectory tracking. When using a large plate angle or large control signal range, the nonlinearity of the system become larger and normal linear controller can’t run with efficient way, so a T-S (Takagi-Sugeno) fuzzy model is implemented to treat with the high nonlinearity of the system and improve the tracking and stabilization of the BOPS. The T-S fuzzy model is compared with the state space mode to validate it. The servo motor equation is deduced using identification. The relations between servo motor and plate angle in x and y direction contain Trigonometric Functions and when solving these relations in real time with control algorithms like PID of fuzzy or any other algorithms it takes a time larger than the sample time of the system so this problem is solved by using the Trigonometric Functions in the form of lookup table but using the lookup table has some problems like time delay and missing angles so the Trigonometric Functions is used in the form of ML (machine learning) model with the designed algorithm. The system depends on machine vision to determine the position of the ball in x and y direction. The hybrid T1FLPD/ML (type 1 fuzzy logic proportional derivative / Machine learning) design depends on a machine learning (ML) algorithm that detect the angle of the servo motor required to correct the position of the ball on the plate also the parameters of the PD controller is changed online using the fuzzy logic to enhance the performance of the trajectory and point tracking of the system. This paper introduces three different machine learning techniques for predicting servo motor angle that obtain higher accuracies of 99.95%, 99.908%, and 99.998% for support vector regression, decision tree regression, and random forest regression, respectively. Simulation and practical results on a BOPS dynamic model are both demonstrated to justify the validation of the proposed design scheme. The experimental practical results justify the feasibility and effectiveness of the proposed controller design strategy. |