Kittipong Ekkachai

Force control of a magnetorheological damper using an elementary hysteresis model-based feedforward neural network

An inverse controller is proposed for a magnetorheological (MR) damper that consists of a hysteresis model and a voltage controller. The force characteristics of the MR damper caused by excitation signals are represented by a feedforward neural network (FNN) with an elementary hysteresis model (EHM). The voltage controller is constructed using another FNN to calculate a suitable input signal that will allow the MR damper to produce the desired damping force. The performance of the proposed EHM-based FNN controller is experimentally compared to existing control methodologies, such as clipped-optimal control, signum function control, conventional FNN, and recurrent neural network with displacement or velocity inputs. The results show that the proposed controller, which does not require force feedback to implement, provides excellent accuracy, fast response time, and lower energy consumption.

Swing Phase Control of Semi-Active Prosthetic Knee Using Neural Network Predictive Control With Particle Swarm Optimization

In recent years, intelligent prosthetic knees have been developed that enable amputees to walk as normally as possible when compared to healthy subjects. Although semi-active prosthetic knees utilizing magnetorheological (MR) dampers offer several advantages, they lack the ability to generate active force that is required during some states of a normal gait cycle. This prevents semi-active knees from achieving the same level of performance as active devices. In this work, a new control algorithm for a semi-active prosthetic knee during the swing phase is proposed to reduce this gap. The controller uses neural network predictive control and particle swarm optimization to calculate suitable command signals. Simulation results using a double pendulum model show that the generated knee trajectory of the proposed controller is more similar to the normal gait than previous open-loop controllers at various ambulation speeds. Moreover, the investigation shows that the algorithm can be calculated in real time by an embedded system, allowing for easy implementation on real prosthetic knees.

Knee torque estimation in non-pathological gait using dynamics modelling and feed-forward neural network

This paper presents a knee torque estimation in non-pathological gait cycle at stance phase. Comparative modelling by using dynamics model and neural network model is discussed. Dynamics modelling is constructed by using simple two degree of freedom dynamics with Newtonian calculation approach and more complex four degree of freedom dynamics with Lagrangian calculation approach. Neural network based model is constructed with feed-forward neural network (FNN) structure using six different kinematic and kinetic input from gait experiments to provide one output in the form of knee joint torque. Six cases of different combination of input is presented to test which model give lower complexity in input data while maintain the performance criterion. The available dataset used for simulation was also divided into five different speed categories. The performance of FNN model simulation is given by the normalized root mean squared error (NRMSE). We find that in estimating knee torque, hip angle data is noticeably insignificant, while ankle angle data is more important than ground reaction force in medio-lateral and antero-posterior direction combined. Using only two kinematics input and one kinetic input can achieve 2.67% of NRMSE.

On-line neural network predictive control for swing phase control of MR damper prosthetic knee

The prosthetic knees have been improved and developed to support the amputee to be able to walk as normal people and help them on a daily basis. This research is concerned with a swing phase of a semi-active prosthetic knees utilizing magnetorheological (MR) damper. Although the referred work which use a neural network predictive control (NNPC) has a satisfying results with low error, it has a possibility to improve the error at certain walking speeds. Therefore, an alternative algorithm for a semi-active prosthetic knee is proposed. The modified version of NNPC using online training scheme with scaled conjugate gradient training algorithm is applied to substitute the initial version of NNPC for the prosthetic knee model to estimate appropriate knee angle. The results from simulation using proposed controller show the improvement in error over the initial version of NNPC at absolutely slow and absolutely fast walking speeds while the results using other walking speeds are comparable.

Variable walking speed controller of MR damper prosthetic knee using neural network predictive control

Semi-active prosthetic knees utilizing magnetorheological (MR) damper have been introduced to help transfemoral amputees to improve their quality of life. However, they lack the ability to generate power required to achieve the normal gait of healthy people. The aim of this paper is to propose a novel control algorithm for MR damper prosthetic knee, focusing on swing phase of the gait cycle. The proposed controller is developed by using a neural network predictive control optimized using constrained nonlinear optimization. It provides the ability to predict future knee angle trajectory when a certain command voltage is applied in order to determine the optimal command voltage. Moreover, the proposed algorithm is designed to support amputees at walking speeds. Performance of the proposed controller has been evaluated in an offline simulation compared to the normal gait of healthy people. The results show that the proposed controller can generate the knee trajectory with minimal errors.

A control of MR damper using feed-forward neural network without force sensor

This paper proposes an inverse model feed-forward neural network (FNN) that does not require any force sensor to control magneto-rheological (MR) dampers. The system is designed by using time-histories of displacement and velocity in combination with desired force to predict voltage input to control MR damper. Unlike conventional MR damper controller, the proposed system does not require force inputs, providing more economical control system. Additional dead zone filter is also introduced here to reduce errors at near zero state of velocity. Using training and validation data sets generated by a modified Bouc-Wen model, the results of the proposed system with and without dead zone filter are also presented.