Research On Gait Recognition And Gait Planning Of Lower Limb Exoskeleton Based On The Control Of SEMG

Human lower limb exoskeleton is a kind of intelligent auxiliary equipment. Combined with human lower limb, it can help and strengthen the movement ability of the human body, therefore, human limb exoskeleton has good promotion value in civil and military markets. Lower limb exoskeleton can assist the disabled to walk, extend the movement ability of lower limbs, great convenience to the daily life and it can also help soldiers to alleviate their lower extremity muscle force, enhance the weight and improve their endurance.sEMG is the direct reaction of brain consciousness, with sEMG as the control signal source of lower limb exoskeleton, The lower limb exoskeleton will be more flexible and intelligent. In this thesis, through the acquisition of sEMG, road conditions and the level of fatigue will be recognized and joint angle will be outputted according to the different road information so that the exoskeleton can walk like human and realize human-machine coordination. It includes two aspects:firstly, this thesis need to solve information coupling problem between the limb exoskeleton and the human, whether it can effectively identify the external environment information is the basis of the adaptive function of human lower limb exoskeleton; secondly, this thesis need to solve the problem of adaptive limb exoskeleton joint angle output, that is exoskeleton output joint angle adaptively according to the external environment information timely and walking like a man. This is the key step of human-machine coordination and control system of lower limb exoskeleton. This paper was investigated in terms of sEMG signal pattern recognition and joint angle outputs of three road conditions, the specific contents can be summarized as follows:Firstly, this thesis uses EMD and wavelet denoising methods to filter the sEMG signal. sEMG is periodic, so this thesis uses the moving window single threshold multi threshold technique to obtain the starting and ending positions, then make feature extraction. In order to select the better characteristics, this thesis presents a dynamic analysis of relevance and redundancy of the forward feature selection method, which improves the classification performance and generalization ability in the unknown set.Secondly, traditional classifiers such as BP and RBF has long training time using gradient learning algorithm, this thesis introduces the ELM algorithm which has fast learning speed and good generalization performance. It is particularly suitable for online learning. In the actual process, all the sample could not be getted and may change over time and the environment, this paper uses OS_ELM extreme learning machine for online learning to adapt to different environment. At the same time in order to further improve the accuracy, this thesis adopts an integrated online classifier to improve the accuracy of the experiment.Thirdly, lower limb exoskeleton assist person walking that need output joint angle adaptivly based on the outside world. this thesis uses fuzzy neural network based on Particle Swarm Optimization to make modeling. It stablishes the knee joint angle information mapping stride length and pace and solves the gait planning problem under different road conditions. The scheme applied to intelligent lower limb exoskeleton can be used as an online gait planning method and also can be used as a criterion for model regulating the hip and the knee.