Research On Dynamic Contact Force Sensing Based On Piezoelectric Flexible Tactile Sensor

With the development of intelligent robot technology,flexible tactile sensor has been widely concerned for its good flexibility and real-time sensing advantages.In real life,the application scenarios of dynamic contact force perception such as contact pattern recognition and three-dimensional force recognition have put forward new requirements for flexible tactile sensors.At present,there is still room for improvement in the acquisition and processing of high-dimensional tactile information based on the dynamic contact pressure(stroking,patting,etc.)of flexible tactile sensors,such as the design of a new sensor structure,and the use of intelligent algorithms for feature extraction of highdimensional data collected by sensors.In this dissertation,on the basis of Polyvinylidene Fluoride material,we have prepared a flexible tactile sensor and its array which can acquire four contact patterns(stroking,patting,kneading,and scratching)and dynamic three-dimensional force tactile signals.It has good flexibility,dynamic response and repeatability.Four kinds of tactile signals and three-dimensional force signals are collected by this sensor,and a variety of intelligent algorithms are constructed to extract high-dimensional tactile signal features,so as to realize high-precision recognition of four kinds of tactile signals and accurate prediction of three-dimensional force.The main research contents of this dissertation are as follows:(1)Based on the constitutive equation of piezoelectric materials,the principle of piezoelectric effect is analyzed in detail.On this basis,the piezoelectric effect of PVDF films is simulated by finite element analysis technology,and the influence of different piezoelectric constants on the response voltage(positive piezoelectric effect)and sensor strain(inverse piezoelectric effect)are studied.The sensing mechanism of the flexible tactile sensor is analyzed combined with the simulation experiment of PVDF piezoelectric effect.(2)Based on simulation experiments,PVDF,PDMS and Cu materials are used to prepare a piezoelectric flexible tactile sensor for detecting dynamic tactile signals,and an experimental test platform is built to test its performance.The experimental results show that the sensor has fast response(3.2ms),low hysteresis(3.42%)and high repeatability(5.88%),which can be well applied to signal sensing and detection of different contact patterns.Based on the tactile sensor,the time series data of four contact patterns(stroking,patting,kneading,and scratching)are collected respectively.On this basis,a convolutional neural network-long short-term memory network(CNN-LSTM)fusion model with good spatial feature extraction and temporal feature extraction is constructed to classify and recognize the four contact patterns applied to the sensor surface.The average recognition accuracy is 99.43%.Based on the same data set,the convolutional neural network and random forest model are constructed to recognize the four contact patterns,and the average recognition accuracy rates are 96.67% and 91.39%,respectively.The experimental results show that the CNN-LSTM model constructed in this dissertation can be well applied to the classification and recognition of different contact patterns of flexible tactile sensor.(3)Based on a single tactile sensor,a 3D force sensor with 2×2 sensor array is designed.The variation curve of output voltage of four sensitive units is studied by simulation experiments when the sensor is subjected to horizontal pressure,and the influence of pressure application angle on the deformation displacement and electric field distribution of the sensor is analyzed in depth.The results show that the 2×2 sensor array constructed in this dissertation can effectively perceive the three-dimensional force-tactile information applied to the sensor surface,and can reverse the three-dimensional force information by extracting the voltage signal output from the sensor array.On this basis,2×2 sensitive unit array entity is prepared.PVDF voltage amplifier module and STM32single-chip microcomputer are used to collect the four-channel voltage signal output by the sensor,and the data is transmitted to the top computer for display and storage.Based on the collected temporal voltage signals with spatial characteristics,multi-channel fusion and single-channel independent CNN models are constructed respectively,so as to accurately predict the magnitude and direction of the three-dimensional force exerted on the sensor surface.The experimental results show that the multi-channel fusion CNN model(data layer fusion)can extract more spatial features from the four-channel voltage signal than the single-channel independent CNN model(feature layer fusion),that is,the multi-channel fusion CNN can predict the magnitude and direction of the threedimensional force(F,α,β)more accurately.The predicted errors are 7.31%,1.49% and4.45%,respectively.