Vojislav D. Kalanovic

Feedback error learning neural network for trans-femoral prosthesis

Feedback-error learning (FEL) neural network was developed for control of a powered trans-femoral prosthesis. Nonlinearities and time-variations of the dynamics of the plant, in addition to redundancy and dynamic uncertainty during the double support phase of walking, makes conventional control methods very difficult to use. Rule-based control, which uses a knowledge base determined by machine learning and finite automata method is limited since it does not respond well to perturbations and environmental changes. FEL can be regarded as a hybrid control, because it combines nonparametric identification with parametric modeling and control. This paper presents simulation of a powered trans-femoral prosthesis controlled by a FEL neural network. Results suggest that FEL can be used to identify inverse dynamics of an arbitrary trans-femoral prosthesis during simple single joint movements (e.g., sinusoidal oscillations). The identified inverse dynamics then allows the tracking of an arbitrary trajectory such as a desired walking pattern within a multijoint structure. Simulation shows that the identified controller responds correctly when the leg motion is exposed to a perturbation such as a frequent change of the ground reaction force or the hip joint torque generated by the user. FEL eliminates the need for precise, tedious, and complex identification of model parameters.

Output space tracking control for above-knee prosthesis

The control of a knee joint in an active above-knee prosthesis has been designed using the Lyapunov tracking method. Locomotion was simulated to prove that the tracking control in output space is a valuable real-time control method for artificial legs. The data used for simulation were collected from able-bodied subjects while they walked on a powered treadmill. Human volunteers were braced with an ankle splint (limiting dorsi- and plantar flexion) and with a knee cage (limiting knee movements to the lateral plane). The achieved tracking of the prescribed knee motion, deviations of the thigh movement from the prescribed trajectory, maximal angular deviations from the desired trajectory, and the power consumption were studied as functions of a limited maximal knee torque and a damping constant in the knee actuator. It was found that the use of output tracking is suitable for the design of appropriate hardware and for real-time control of an above-knee prosthesis.<<ETX>>

Practical aspects of precision membrane antenna shape control

A theoretical elasticity model of an inflatable membrane reflector is being developed, which predicts how the structure responds to perturbations at its boundary. The objective in the process of model development is to use interconnected cells to spatially discretize the membrane structure. Each cell would consist of three basic components-a spring, mass, and damper-while forming a symmetric cell configuration. Such cells will allow a multi-input, multi-output (MIMO) linear model formation where either displacements and/or forces would be considered as inputs and/or disturbances. Feedback error learning method is used for local state estimation and local parameter identification of the membrane. In this method a learning algorithm is proposed that uses the output of a feedback controller as the error for training an adaptive feedforward neural network model. In other words, feedback error learning is control strategy, which incorporates a neural network in a feedforward path, and allows this artificial system to learn the inverse dynamics of the controlled plant in real-time. A scanning laser velocimeter is used for measuring the local displacements and local transfer functions (LTF) of the membrane. The feedback on the membrane shape is coupled with a mathematical model of boundary perturbation effects for control efforts.

Fuzzy Control of Membrane Wrinkling

ABSTRACTThe impact of this work is reflected in combining two different areas of research, one of which deals with intelligent control strategies, while the other focuses on the mechanics of highly flexible structures, such as membranes. True membranes are “no-compression” structures which exhibit the unique response of wrinkling. Accurate measurement of membrane wrinkling has heretofore not been presented in the literature; among other requirements, noncontact methods must be used. First, some background information on membrane wrinkling prediction and measurement is given. Then, an experimental apparatus is discussed within which a membrane was subject to planar deformations. Finally a control strategy, based on a fuzzy PD controller, is discussed in detail together with experimental results.

Fuzzy tuned nonlinear rate controller for manipulators

This paper addresses the problem of controlling a class of nonlinear dynamic systems, including robots, represented by autonomous nonlinear models without functional input derivatives. Industrial and space robots are controlled with PID or PD controllers which are difficult to tune for aperiodic response over a wide range of spatial motion, since the controller gains are functions of changing plant parameters such as effective inertia. A controller which can achieve fast robot performance despite these shortcomings, and with high pay-loads, is a disclosed four stage variable-structure nonlinear rate controller (NRC). This work expands the NRC application with a fuzzy logic tuner resulting in the systems robust performance with respect to variable loads and command input. The fuzzy tuner consists of nine rules which are logical functions of the system error and its derivative.<<ETX>>

Thermal Control of Laser Powder Deposition: Heat Transfer Considerations

Laser based solid free-form fabrication is an emerging metallurgical forming process aimed at rapid production of high quality, near net shape products directly from starting powders. Laser powder deposition shares, with other free-form technologies, the common characteristic that part fabrication occurs directly from a 3-D computer aided design (CAD) model. The microstructure evolution and resulting material properties of the component part (strength, ductility, etc.) fabricated using laser deposition are dependent upon process operating parameters such as melt pool size, laser power, head (manipulator) speed, and powder flow rate. Presently, set points for these parameters are often determined through manual manipulation of the system control and trial and error. This paper discusses the development of a path-planning, feed-forward, process-driven control system algorithm that generates a component part thermal history within given constraints, thereby assuring optimal part quality and minimizing final residual stresses. A thermal model of the deposition process drives the control algorithm. The development of the thermal model is the subject of this paper. The model accounts for temperature-dependent properties and phase change processes. Model validation studies are presented including comparisons with known analytic solutions as well as comparisons with data from experiments conducted in the laser laboratory at SDSM&T.Copyright

Adaptive closed loop material testing using fuzzy logic control

One of the major difficulties in the design of control systems for the hydraulic electromechanical systems, is presented by the fact that the specimen stiffness may change, especially, during the low cycle fatigue tests. Thus, there is a need for frequent tuning of the existing controller. Typically, an operator will change the controllers parameters if the performance deteriorates during testing. This problem may be circumvented with adaptive techniques but also, as shown in this paper, an approach using fuzzy logic can be as successful. Further, a fuzzy compensator can overcome this structural disturbance, and still retain a simple controller configuration which is readily applicable in real time.

Feedback Error Learning Neural Network for Above-Knee Prosthesis

Abstract The purpose of this paper is to show how Feedback Error Learning can be applied to control hip and knee joints in an active Above-Knee Prosthesis. Feedback Error Learning is a control scheme based on neural networks and has its origin in physiological studies. A total energy method is used, thus eliminating the need for a known mathematical model of the prosthesis. The systems total energy is extracted from kinematic analysis, and used as an equivalent mathematical model. The findings conclusively show Feedback Error Learning is a suitable control strategy for an active above-knee prosthesis.