Research On Hand Grip Strength Prediction Based On Flexible Deformation Sensor

As the global population of people with upper extremity disabilities increases,the demand for prostheses is also increasing with the development of society.The emergence of smart prostheses facilitates the life and work of disabled patients.The interaction of prosthetic limbs has become a research hotspot in the field of rehabilitation robots.However,most of the current prosthetic interaction methods focus on the use of EMG to classify gestures,and the research on grip is less,which greatly limits the use of prosthetic limbs.The EMG signal itself is susceptible to interference and weakness,which makes it difficult to estimate the continuous grip.In contrast,muscle deformation signal has unique advantages for continuous estimation.In this project,a self-designed flexible deformation sensor is used to collect the muscle distortion and wrist rotation information of human arms,and continuous grip strength is estimated at different wrist angles.The specific research contents are as follows:Firstly,a flexible deformation sensor is developed and designed independently,which uses the internal photoelectric effect and Hall effect to convert the amount of distortion and rotation into a measurable voltage signal.A signal acquisition system is built using integrated circuits for real-time acquisition of sensor data.A series of deformation tests are performed on the sensor to ensure that the information on the human arm can be accurately collected by the sensor during subsequent use.Secondly,considering the viscoelastic properties of human muscle,the grip strength was estimated using Voigt and Kelvin Voigt models,which were improved based on Hill muscle model.The flexible strain sensor and the grip force sensor are used to collect the muscle strain and the grip force on the user’s arm synchronously.The information collected by the above sensors is pre-processed,and the pre-processed deformation data and grip strength data are fitted to the model using the least squares method.Finally,the validity of grip strength estimation is verified experimentally and the differences between models are compared.Finally,a multi-wrist angle grip estimation method is proposed to solve the limitation of grip estimation for fixed wrist angle.This method mainly uses a time series convolution network（CNN）to detect wrist angle,which improves the detection accuracy and anti-jamming ability.Then,different muscle model parameters were designed at different wrist angles.The model parameters are selected by the detected wrist angle,and the wrist angle and grip force are output simultaneously.The effectiveness of this method is verified by a large number of experiments.