1,721,164 research outputs found
Neural network based ILC with application to FES electrode arrays
No full textFunctional electrical stimulation (FES) is a technology that can help paralysed or impaired subjects regain their lost movement after stroke. A large number of FES elements can be combined to form FES arrays which are capable of activating the multiple muscles needed to perform functional arm movement. However, the control of FES arrays is challenging since high precision is required but there is little time available in a clinical or home setting to identify a model. To addressthis problem, an approach is developed that can deliver high accuracy with minimal identification overhead. It is based on iterative learning control (ILC), a technique that exploits the repeated nature of rehabilitation training. The method uses a parameterised model form that adapts to new data and replaces identification tests while maintaining accuracy. Conditions are derived to ensure convergence is preserved while minimising identification time. Numerical results show that four references can be tracked using only 10.8% of the experimental tests required by conventional ILC algorithms. The approach is then applied experimentally to four unimpaired subjects using a realistic rehabilitation scenario, with results showing mean tracking accuracy within 5, while requiring only between 25% and 64:9% of the experimental tests of conventional ILC
Artificial Neural Network based Iterative Learning Control for Stroke Rehabilitation
No full textAn artificial neural network (ANN) is combined with gradient descent to form a model-free iterative learning control (ILC) approach than can be applied to a wide range of nonlinear discrete-time systems. The ANN is recursively trained on the entire set of past data collected from the system and uses a passivity condition to determine when the ANN can be used to compute the next ILC update, or if an identification test is needed. Convergence properties are established alongside design selections that ensure the passivity condition is fulfilled. By minimizing the reliance on identification tests, this methodology is substantially faster than existing model-free ILC algorithms. It is tested on a key stroke rehabilitation problem using functional electrical stimulation (FES) for hand/wrist tracking.Experimental results using the new ILC approach with eight participants show that three hand/wrist references can be tracked using an average of 56% fewer experimental inputs compared with the most accurate previous approach. As the first approach to combine ILC and machine learning in upper limb rehabilitation, the results demonstrate how their combination addresses their individual deficiencies
Decentralised collaborative iterative learning control for multi-agent systems point-to-point channel tracking
No full textApplication of learning to collaborative tracking control of multi-agent systems has addressed a wealth of problems across transportation, manufacturing, rescue, aerospace and medical care area. Iterative Learning Control (ILC) algorithmshave been proposed to address synergistic objectives in general optimisation problems, achieving a transparent balance between convergence speed, tracking error and robustness
Experimental evaluation of iterative learning control performance for non-minimum phase plants
No full textThis thesis describes the design and construction of a Single Input Single Output (SISO) non-minimum phase experimental test facilityan d the subsequent testing of a number of Iterative Learning Control (ILC) strategies. The system can be configured in three different ways in order to test the effect of increased plant complexity and non-linearity. The implementation of a number of both existing and new ILC strategies is detailed and results and analysis of their performance are presented. A principal objective has been to find the ILC controller that is most effective in forcing the output of the test-bed to follow a repetitive trajectory. The design and construction of the test-bed is explained in full and both linear and non-linear models of the system are produced. P-type, D-type and Delay-type ILC algorithms have been tested on the simplest form of the system. The phase-lead algorithm has been implemented and a method of establishing the optimum lead found, as well as a procedure to estimate unstable frequencies. Both causal and noncausal filters have been assessed for use with the algorithm. Phase-lead ILC has been implemented on the more complex plant and comparisons made with previous results. The use of a forgetting factor has been found to overcome the problem of instability, but at the expense of increased final error. The phase-lead algorithm has been vastly improved using additional phase-leads and this technique has been generalised to produce an novel optimisation routine which uses a large number of phase-leads. Its success has been confirmed with experimental results. A learning law utilising the plant adjoint fits naturally into this framework and practical results are presented. This method has been both reformulated into one which needs little plant knowledge, and also combined with deadbeat control to avoid truncation in the course of its implementation. Results are presented using these techniques and practical guidelines produced and tested. A simple method of increasing the learning at higher frequencies has been proposed and verified experimentally. An optimality based Repetitive Control algorithm has also been rigorously tested and the use of a relaxation parameter found to increase its robustness. Finally, a graphical method that represents both the robustness and the stability of an ILC algorithm applied to a known plant has been developed. This tool may find wide application when designing and developing future ILC strategies
Parameterised function ILC with application to stroke rehabilitation
No full textFunctional electrical stimulation (FES) is a popular assistive technology that uses electrical impulses to artificially stimulate muscles to help paralysed or impaired subjects regain their lost movement after stroke. A large number of FES elements can be combined to form FES arrays which are capable of activating the multiple muscles needed to perform functional arm movements. However, the control of FES arrays is challenging since high precision is required but there is little time available in a clinical or home setting to identify a model. To date, by far the highest accuracy has been achieved using iterative learning control (ILC), a technique that mirrors the repeated nature of rehabilitation task practice. In particular, high accuracy has been achieved using a well-known ILC law for a general class of nonlinear systems which computes the updated control input using a linearised plant model. Since a global system model is unavailable, this is identified on every ILC trial by running an identification test. This adds many time-consuming identification tests, making it infeasible for clinical deployment. To solve this problem, an approach is developed that can deliver high accuracy with minimal identification overhead. It introduces a parameterised plant model that is updated in parallel with the ILC using all available data, and then applied to replace identification tests. Rigorous conditions are derived to ensure convergence is preserved while minimising identification time. Numerical results show that four references can be tracked using only 10.8% of the experimental tests required by standard ILC algorithms. The approach is then applied experimentally to six unimpaired subjects using a realistic rehabilitation scenario. In particular, a novel stereo camera system is used to measure hand joint angles in a manner that can transfer to home use. Results show mean joint angle tracking accuracy within 5°, while requiring only between 25% and 64.9% of the experimental tests of standard ILC.</p
Robust iterative learning control for unstable MIMO systems
No full textIterative learning control (ILC) is a well-established technique to successively improve tracking accuracy for systems that repeatedly perform the same task. Most current literature imposes constraints on the nature of the system, such as requiring it to be full-rank, or inherently stable. This paper presents a generalised ILC framework that can handle non-linear, unstable, MIMO systems with rank deficiency. This involves the minimisation of a cost function that balances tracking performance and input effort, extending previous approaches to include a 'robustness filter' within the optimisation. Gap metric analysis is then applied to examine the robustness of the resulting system, with performance bounds derived for both serial and parallel ILC architectures. A design procedure is presented that allows the designer to transparently trade-off robustness and convergence properties. The design framework is illustrated via application to the inverted pendulum problem, a classic example of a highly nonlinear, unstable, and under-actuated system
Evaluating the feasibility of chaos-based guidance systems in swarm projectiles
No full textThis study investigates the feasibility and performance implicationsof chaotic trajectory tracking for guided projectiles using PIDcontrollers under realistic environmental disturbances. Specifically, Lorenz,Sprott-A, and Halvorsen chaotic systems were used as reference trajectoriesin a three-dimensional projectile framework incorporating windperturbations and actuator dynamics. The methodology involved simulatingprojectile dynamics with gimbal angle feedback to compute pitchand yaw variations, angular velocities, and torque-based control effort.Simulation results revealed that all three chaotic trajectories were trackableusing PID control. Among them, the Sprott-A system demonstratedthe most favorable profile for energy-constrained applications, requiringthe least control effort (198.43 Nm·s) and angular change (4805.74°),along with the lowest energy consumption (18.24 kJ). In contrast, theHalvorsen system exhibited the highest torque demand (122696.79 Nm)and angular variability (31246.31°), making it less suitable for systemswith limited actuation capability. Lorenz presented intermediate performancein all metrics. The results confirm that chaotic references canenhance evasiveness while remaining within mechanical and energeticconstraints. These findings support the integration of chaos-informed trajectoryplanning into autonomous swarm projectile systems for improve
Multiple model iterative learning control of FES electrode arrays
No full textStroke is a common cause of hand and upper limb disability, but current rehabilitation approaches do not adequately support successful recovery. Functional electrical stimulation (FES) is the most widely used assistive technology, and is able to support accurate hand and wrist motion when applied using multi-element electrode arrays. However, accurate movements have only been possible using an iterative learning control (ILC) approach involving many repeated model identification tests. This lengthy process limits wide-spread use. This paper presents a solution for FES electrode array control using estimation-based multiple-model ILC (EM-MILC), in which a set of parameterised models is used to automatically update the stimulation applied to each array element every time a task is carried out. This removes the need for model identification, significantly improving system usability whilst maintaining high performance. Experimental results demonstrate that EM-MILC reduces the average number of tests from 16 to 3, compared to the most accurate existing approach
- …
