Research On Fall Prediction And Balance Recovery Control Technology Based On Multi-sensor Data

Population ageing is becoming increasingly visible worldwide,and the health of older people is turning into a matter of concern.Falls can be very damaging to the health of older people.It seriously interferes with the normal life of the elderly,so fall detection for the elderly has become a research hotspot in the field of rehabilitation medication.Regular evaluation of the balance ability of the elderly can screen out the poor balance ability of the population,that is,the high risk of falls,which makes it easier to monitor their movement information to predict their falls.The fall prediction can detect the tendency of the fall in time,and help to restore the balance with the help of the fall-avoiding device during the effective time,so as to minimize the injury caused by falling.However,there are still some problems in the current research,such as complicated process of fall risk assessment,poor accuracy and timeliness of fall prediction,and lack of fall avoidance devices for protection.Therefore,the balance ability evaluation method based on simple walking task is deeply studied in this paper.The fall prediction model based on multi-sensor is constructed and the balance restoration control strategy is developed and applied to the knee joint exoskeleton robot.The main research work includes:(1)Comprehensive evaluation model of balance ability based on multi-modal motion data.The principle of human posture control is analyzed,and an evaluation method based on dynamic balance and reactive balance is proposed.The human motion parameters of gait test and obstacle-crossing test are selected,and the balance ability is tested in normal gait and two kinds of simulated abnormal gait.A motion capture system is used to collect multi-modal motion data,which is then combined with quantitative analysis to find the best feature set to input nonlinear regression.The obstacle level is taken as the reference value to get a value that represents the balance of the subject,based on which the high fall risk population is screened and the assist torque of fall avoidance is calculated.(2)Multi-sensor-based fall prediction method for human body.The regularity of physical signs in the process of human body falling is studied.The inertial measurement unit and the plantar pressure insole are used to collect multi-sensing information including falls and various other normal motion states.A fall prediction framework based on the decision tree model is proposed.The attitude angle is analyzed and calculated using the Kalman filter based on the acceleration and angular velocity.The data samples are extracted using the sliding time window.An off-line training model is established to improve the detection lead time and verify the accuracy of the prediction results.(3)Knee exoskeleton balance restoration control strategy for fall avoidance.The dynamic equation of human lower limb is established to calculate the moment of knee joint in normal motion.And the balance ability of the subjects and the degree of unbalance in the course of motion are combined,based on the fuzzy logic reasoning.The mapping relation between the balance and torque of the knee exoskeleton is established.The knee exoskeleton is control based on impedance control algorithm to exert force on the wearer,and the effectiveness and timeliness of the balance restoration control strategy in the application of fall avoidance are verified by experiments.