| In outdoor or field situations,some devices can be powered by energy generated by the human body during motion.As the knee joint energy harvesting device using dielectric elastomer(DE)material,its power generation principle is to periodically charge and discharge according to the stretching and compression form of the material,and convert mechanical energy into electrical energy.One of the keys to achieve high-efficiency energy harvesting is to estimate the human attitude,i.e.,the knee angle,and determine the charge and discharge time of the energy harvesting circuit according to the change of the angle,so as to accurately control the charge and discharge of the circuit.According to the requirements of knee joint energy harvesting,this subject uses inertial sensors to acquire the motion data of human legs.Based on these data,the attitude information is obtained by attitude calculation algorithm.At the same time,the control circuit based on STM32 processor(STM32F405)is designed,which can realize the intelligent control of charge and discharge of energy harvesting circuit based on the obtained attitude information.The main work implemented in this subject is listed as below:(1)The main control circuit with the STM32F405 processor as the core is designed,which can realize the charging and discharging control function,as well as the collection and transmission of sensor data.(2)The inertial sensor(MPU9250)is selected to collect the human motion data,its systematic errors are modeled respectively,and the calibration parameters of each sensor are calculated through the electric turntable experiment,which establishes the foundation for the calculation of human motion attitude.(3)The extended Kalman filter algorithm based on quaternion is used to calculate the human attitude,the extended Kalman filter model is realized by MATLAB software using the sensor data.And,the model is programmed on the STM32 processor to realize the real-time attitude calculation.(4)The method of human activity recognition based on inertial sensor data is studied,and a hybrid network structure combining convolutional neural network and deep feedforward sequential memory network is proposed,which realizes high accuracy of human activity classification.This research can provide support for the further improvement of attitude calculation and intelligent control performance.The research work of this paper has obtained satisfactory experimental results.The designed circuit and data processing method can be applied to the knee joint energy harvesting device,which has a certain application value for high-efficiency energy harvesting. |
