Research On Adaptive Control System Of Lower Limb Exoskeleton Based On Identification Of SEMG

Human limb exoskeleton is a mechanical leg which is placed in the outside of the lower limb bone. Combined with human lower limb, it can help people walk, extend lower limb’s movement, relieve muscle force and increase muscle endurance at the same time. Therefore, human limb exoskeleton has good promotion value in civil and military markets, which is the reason that the limb exoskeleton gets fast development in recent years.As an optional controlling signal in controlling limb exoskeleton, sEMG has special advantages. Because it is the reaction of brain consciousness, which can response the states of human movement directly, using sEMG as limb exoskeleton control signal will be more flexible and effected. This thesis introduces sEMG to limb exoskeleton control system to realize adaptive control of limb exoskeleton. It includes two aspects:firstly we need to solve information coupling problem between the limb exoskeleton and human body, that is to say, the movement pattern of eoskeleton must keep up with the body movement mode; secondly we need to solve the problem that how much power the limb exoskeleton shoud export, that is how to control the auxiliary force dynamically according to the needs of the human body. Only do these two aspects, can we truly achieve the body limb exoskeleton adaptive control.According to the adaptive control of the two main contents, the following several aspects of research are proposed.Firstly, traditional pattern classification methods are used to realize the classification of the three models of the human walking on the ground and up and down the stairs, as well as the identification of the human body in normal status, moderate and severe fatigue conditions. What’s more, the thesis introduces process neural network which has space weighted aggregation effect and the time cumulative effect. The thesis uses particle swarm algorithm to solve local minimum problem in the training process and it improves the lower limb motion pattern recognition rate.Secondly, the research is focused on the lower limb in this thesis and we set up two rigid bodies and two degrees of freedom model of human lower limb to collect the different speed of leg angular velocity information when walking on the ground. It has applied the theorem of angular momentum to calculate the lower limb joint torque.Thirdly, this thesis has established RBF neural network of feed-forward model and Elman neural network of feedback model to predict joint torque. We combine EMG and various motion parameters as input to get best forecasting combination and get different forecasting effect by changing the training and test sample data. Summing up two prediction models in various cases of lower limb joint torque prediction, this thesis finally achieve the result that the Elman neural network has better prediction effect in the prediction of time series.