Research On Fall Detection And Control Strategy For Walking Training Robot

With the aging of the global population and the increasing population with lower limb disabilities,contemporary medical care conditions can no longer meet the nursing needs of the elderly and the disabled.Intelligent rehabilitation equipment can provide better services for the elderly and the disabled,and has become a research hotspot in the field of robotics.However,because of the poor physical fitness of this group and the lack of anti-fall ability of intelligent rehabilitation equipment.Therefore,it is of great significance to improve the functions of fall detection and fall prevention control strategies of intelligent rehabilitation equipment for restoring the walking ability of the elderly and the disabled.In this thesis,the structure of omni-directional walking training robot(WTR),and the principles of two dimensions(2D)laser sensor and postur sensor were introduced firstly.By observing the posture characteristics of people in different falling states during walking training,an experimental model was built to ensure that the remarkable characteristics of people in different falling postures could be identified during walking.By analyzing the physical characteristics of people falling forward with WTR,this thesis proposed a fall detection method based on fuzzy reasoning.The posture sensor and 2D laser sensor were mainly used to detect the user's upper body posture and the movement of two legs.A fuzzy knowledge base was established to distinguish between normal walking and forward falling based on the information fusion of two sensors,and a new fuzzy reasoning mechanism was used to analyze the user's falling state.Finally,the effectiveness of this method was verified by experiments.Because falls are sudden and uncertain,the fall detection method based on fuzzy reasoning has limitations,and its recognition effect on sudden falls is poor.Therefore,a fall detection method based on convolutional neural network(CNN)was proposed.The recognition effect of this method was positively correlated with the number of samples.Therefore,the samples in this thesis included different walking states of people in different environments,which greatly solved the problems of sudden fall and non-repetition of fall posture.In this method,two posture sensors and a 2D laser sensor were used to detect the leg motion information of people in different walking conditions in different environments.CNN model was established according to the data information.Model parameters were constantly adjusted to improve the the model accuracy.In addition,a fall detection strategy based on Softmax activation function was adopted to distinguish different states.Finally,the effectiveness of this method was verified by experiments.In order to avoid the secondary injuries caused by the user falling forward during walking training,this thesis proposed an adaptive impedance control anti-fall strategy with safety and flexibility.The dynamic model and adaptive impedance control model of man-machine when the user fell forward were established.When the user fell forward,WTR could enter the anti-fall system immediately and quickly adjusted to the safe posture according to the user's needs.The feasibility of adaptive impedance control anti-fall strategy was verified by modeling and simulation.