Withit Chatlatanagulchai

Motion control of two-link flexible-joint robot, using backstepping, neural networks, and indirect method

We present a state-feedback control of a two-link flexible-joint robot. The control algorithm does not require the mathematical model representing the robot. Three-layer neural networks approximate the unknown plant functions. The neural network weights are adapted on-line. We use backstepping control structure together with variable structure control to provide robustness to all uncertainties. We have included experimental results to show the effectiveness of the control algorithm

Command shaping applied to a flexible robot with configuration-dependent resonance

Joint flexibility is an intrinsic property in most industrial robot manipulators. Because of the great complexity of its model, the resonant frequencies of the two-link flexible-joint robot manipulator vary with the configuration of the manipulator. Our objective is to move the manipulator from point to point with the least amount of residual vibration at the end point in the shortest time possible. Two shaped command profiles, based on ramped sinusoidal basis function and segmented versine basis function, are compared against one another as well as against the unshaped bang-bang command profile. The segmented versine basis function was proved to be the most effective

Intelligent control of a two-link flexible-joint robot, using backstepping, neural networks, and direct method

We present a state-feedback control of a two-link flexible-joint robot. First, we obtain desired control laws from Lyapunovs second method. Then, we use three-layer neural networks to learn unknown parts of the desired control laws. In this way, the control algorithm does not require the mathematical model representing the robot. We use smooth variable structure controller to handle the uncertainties from neural network approximation and external disturbances. To show the effectiveness and practicality of this control algorithm, we performed an experiment on one of the robots in our laboratory.

Backstepping high-order differential neural network control of flexible-joint manipulator

We present an output-feedback control design of a two-link, flexible-joint manipulator. The control system is a combination of the Luenberger-type observer, backstepping control, variable structure control, and high-order differential neural network. Using the neural network as model identifier, we can control this complicated system without using its closed-form mathematical model. The observer provides us with the ability to design the control system from the output signals. The variable structure controller handles uncertainties arising from model identification and state estimation. Backstepping structure provides a way of applying the robust control to each subsystem. Simulation of the two-link flexible-joint manipulator is included.

Motion control of two-link flexible-joint robot with actuator nonlinearities, using backstepping and neural networks

We present a state-feedback control of a two-link flexible-joint robot. The control algorithm does not require the mathematical model representing the robot. Three-layer neural networks approximate the unknown plant functions. The neural network weights are adapted on-line. We use backstepping control structure. We use variable structure control to provide robustness to all uncertainties. For simulation, we obtain parameter values of the Euler-Lagrange model from real experiment. We, then, add backlash, deadzone, and additive disturbances to the Euler-Lagrange model to closely replicate the actual robot. We show through simulation that our controller can handle these actuator nonlinearities effectively.

Model-free observer backstepping control design for nonlinear systems in strict feedback form

A model-free control system design for nonlinear system in strict feedback form is presented. A type of neurofuzzy system is used as a system identifier. A Luenberger-type observer is designed from the identifier structure. The controller is based on backstepping and Lyapunov design schemes. Variable structure control is added to deal with uncertainties arising from estimation errors. Together, this control system is capable of controlling a system where the system model is not known and only output is measurable. However, some assumptions of the actual system are required. A tracking problem example is provided together with stability proofs of the closed loop signals.

Backstepping high-order differential neural network control of a type of nonlinear systems

This paper presents an output feedback control. The control algorithm does not require plant mathematical model. However, the actual plant is assumed to be affine with respect to local inputs. High order differential neural networks are used to identify the unknown plant. Luenberger-type observer provides estimated states. Controller is based on backstepping and Lyapunov direct method. Variable structure control handles uncertainties arising from the estimation processes. Closed loop errors are proved to be bounded. A trajectory tracking example demonstrates the effectiveness of the design.

Output-feedback model predictive control with constraints - improve robustness using past information

We present nonlinear model predictive control design for a type of discrete-time nonlinear system. Range constraints are imposed on input, output and input rate. When states are not available, an observer based on Newtons algorithm is used to estimate all states. We also present an idea to improve robustness of the control system when plant is subjected to modeling uncertainties. By using past information, the correction to the plant model can be computed from the difference between past actual states and past model states using a simple linear least-square algorithm. The correction is applied to the plant model in the future time step. Two illustrative examples are included

Observer Backstepping Neuro-Fuzzy Control Design for a Type of Nonlinear System

This paper presents a control system design for a type of time-varying nonlinear system. The control system comprises neuro-fuzzy system identifier, Luenberger observer, backstepping controller and variable structure controller. We use adaptive neuro-fuzzy inference system to identify the plant in real time without the need of underlying mathematical model. However, some knowledge about the plant structure and upper bounds is required. With the use of observer, the control system can be designed from plant output and input alone while plant states are assumed unmeasurable. Controller is designed based on backstepping scheme and uncertainties from the plant identification and state estimation processes are handled by variable structure controller. Under some important assumptions, the control system is proved to be able to track a smooth desired trajectory with uniformly ultimately bounded tracking error. A simulation based on one-link flexible-joint robot manipulator is provided.Copyright