Research On Predictive Control Of Active Above-knee Prostheses

Active above-knee-prostheses(AKP) can provide amputees with walking assistance and can follow the gait of healthy legs. Compared to conventional passive prostheses, the research on AKP has important theoretical significance and application value.In this paper, based on the motion data which collected and analyzed from the human lower limb, a piecewise affine model can be established to describe the correspondence between the knee torque and the knee angle. Then a model predictive controller based on multi-parametric programming theory is built. This control algorithm can provide a new idea for the active AKP motion control.The main content of the dissertation could be presented as follows:1. Firstly, the data is collected and analyzed when the subject walks in different road conditions. Then, the knee torque can be calculated by the tested data and the dynamic model of human lower limb.2. Extreme learning machine classifier is built for human gait recognition after analyzing the character human limbs motion. The motion models under different gait are established based on piecewise affine systems theory and linear ization theory of nonlinear system.3. By multi-parameters programming theory, control laws for different gaits can be calculated off- line. Then the online optimal control problem can be changed to a looking- up table method. In this process lots of works, such as modeling, solving the optimal quadratic programming problem, can be completed offline, the online computation burden should be reduced heavily. Some time consuming work, are arranged offline. The online work includes: determining when a gait begins from the foot pressure signal, collecting enough data for gait recognition, selecting the correct controller, fixing the walking state by signal data collected, calculating the control signal by looking- up table method, exerting control signal to the AKP. This is the whole control process of the active AKP. This scheme computation burden is not heavy and has higher control accuracy. It is a promising way for active AKPs.