Pajunen, Grazyna

This research is concerned with the application of system identification and adaptive control to Functional Electrical Stimulation. The work consists of developing a model which describes EMG (Electromyogram) activity to forearm motion. Although several EMG models presently exist, the goal was to produce a model more suitable for on-line applications while also taking into account the system nonlinearities. The parameters of this model were estimated using a least squares algorithm. The model was tested by simulation and experimentally collected data. The developed model explains well the forearm movement. From the developed model, an adaptive controller was designed using a model reference control scheme. This adaptive controller was used for generating the suitable stimulus pattern. The simulation results showed good tracking and indicated the controllers ability to adapt to changes in the arm's nonlinear gain.

Many positioning systems with varying loads or geometries, such as robotic systems, could take advantage of the class of non-linear controllers known as Adaptive Controls. Model Reference and Pole Placement Adaptive Controllers are usually the preferred techniques for position control systems. Pole Placement is the more universally applicable technique. Adaptive controllers must be able to change control parameters as the system's parameters change (i.e., as is the case with a load or geometry change). The most common and perhaps the fastest converging technique uses the Least Squares Identification Algorithm. Many positioning systems cannot tolerate overshoot. These systems should use an adaptive velocity controller in conjunction with a conventional position controller. This will minimize system overshoot during the learning period. Adaptive controllers tend to be very complex and require a great number of computations. With today's advances in computer technology, adaptive controllers can now be economically considered for many industrial, consumer and military positioning applications.

This dissertation is concerned with the relevant research in developing finite dimensional indirect adaptive schemes to control vibrations in flexible smart structures based on the finite element approximation of the infinite dimensional system. The advantage of this type of modeling is that the dominant modes of vibrations wherein the total energy is concentrated are accommodated thereby avoiding the so-called "spillover" phenomenon. Further, the mass, stiffness and damping coefficients associated with each element appear explicitly in the model facilitating the derivation of the ARMA parametric representation which is suitable for on-line estimation of the structural parameters. The state-space representation of the finite dimensional model is used to design an indirect linear quadratic self tuning regulator algorithm using the parameter estimation, indicated above. Further, a method to choose the control and state weighting matrices (required to design the controller) to yield a stable closed-loop system, is presented. Simulation results demonstrating the performance of the adaptive control system are presented. Another algorithm based on the model reference technique is also developed by considering the discrete time approximation of the finite dimensional model. This control algorithm in conjunction with the parameter estimation constitute an indirect model reference adaptive control system. Simulation results are presented to demonstrate the effect of the reference model parameters, which may impose certain constraints on the force requirements causing actuator saturation and thereby affecting the stability of the closed-loop system. In order to overcome the problem of using bulky and expensive sensors to measure transverse displacement and velocity, a new spatial recursive technique to estimate these variables alternatively by using a distributed set of (measured) strain data, is developed. Relevant algorithm enables the use of smart materials to sense the strain developed at various locations along the length of the structure leading to the development of flexible smart structures. Experimental results on the personal computer based control of vibrations in an aluminum beam using patches of polyvinyldene fluoride (PVDF), and lead zirconate titanate (PZT) as sensors and control actuators respectively, are furnished to demonstrate the feasibility of real-time implementation of the above mentioned control algorithms.

Implementation of the Optimized Policies for Adaptive Control (OPAC) strategy in conjunction with a vehicle velocity controller offers the potential for significantly improving the control strategies used at isolated intersections with respect to measured vehicle delays. The exhaustive sequential search procedure by OPAC provides the optimal switching policies for the intersection while the vehicle velocity controller varies vehicle velocities to reduce vehicle stopping delays. The OPAC algorithm implemented with the vehicle velocity controller was found to have substantially lower delays than OPAC alone.

The I.B.M. Electric Drive Robot (E.D.R.) is a six-link manipulator originally controlled by a classical analog P.I.D. controller. Its performance is not satisfactory because of its poor tracking capabilities and a considerable vibration during arm movement. This is the central motivation for designing an adaptive computed torque controller for this system. In order to accomplish this the physical model of the robot is first reparameterized such that it is linear with respect to a set of uncertain parameters. Once this is accomplished the adaptive controller is then formulated. Next methods of computer simulation are developed and employed. These simulation results show the superior performance of the proposed scheme over both a classical computed torque controller and the current P.I.D. controller.

Several adaptive controllers have been designed to control the infusion of the drug sodium nitroprusside for the purpose of reducing high blood pressure in post surgical patients. Most of these controllers have not considered one or more factors of the controlled system including the stochastic background noise present in the patients, the time varying nature of the patient, and the constraints imposed on the control input and system output. This thesis presents a model reference adaptive controller which takes into account all of these factors. Through simulations on a personal computer, the robustness of the controller is demonstrated in the presence of noise, time varying parameters, and deterministic disturbances. Furthermore, this performance is achieved without requiring any prior knowledge of the system delay.