Research On Flexible Tactile Sensor Array And Object Recognition Based On Micro-airbag Structure

In unstructured environments such as intelligent robots,rehabilitation medicine,and human-computer interaction,the scientific problem of tactile perception in the work environment has received extensive attention from the academic community.Tactile perception is a key technology to enhance the autonomy of robot operations and its "empowerment".As a medium for intelligent robots to perceive the external environment,tactile sensors is an indispensable part of robot to complete intelligent and delicate tasks,and also the key to realize human-machine security interaction.The inherent scientific problems and the challenging technology of tactile sensor,have attracted domestic and foreign experts to continuously carry out theoretical research and technical research.This thesis focuses on a series of scientific and technical issues related to the tactile sensor's perception of the external environment,analyzes the tactile features such as target surface morphology,target flexibility,constructs a micro-airbag flexible array tactile sensor to sense contact distribution force,and discuss the sensing mechanism of the tactile sensor.Based on the tactile information,the research on the recognition of target contact surface shape,static contact state and motion state is studied by using deep learning method such as convolutional neural network and long-short-term memory neural network.The specific research contents and related work are as follows.(1)Research on the mechanism of tactile feature perception.Starting from the requirements of application scenarios and task types,the tactile expressions form is explored.The normal contact force Fz is processed by filtering and Fourier transform,and then the frequency domain and time domain features of Fz signal are extracted respectively to determine the softness of the operation target.Based on the different surface morphology of the target,the contact area and contact force distribution law of the manipulation target are detected,and the contact force distribution features are extracted to realize the recognition of the target surface morphology.The deep learning algorithm is used to fuse the contact force distribution characteristics to realize the target slip state judgment.(2)Research on the characteristics of the micro-airbag flexible tactile sensor.Based on the theory of materials mechanics and elasticity mechanics,the static characteristics of the sensor unit and its array are analyzed by using analytical methods.The physical properties of the linear relationship between the external load and the internal pressure of the airbag are revealed.The influence of the structure size of the micro-airbag on the performance of the sensor is discussed.The relationship between the internal diameter of the airbag d2 and the sensitivity of the sensor are obtained.The relationship between the wall thickness,the internal height of the airbag and the sensitivity limit of the sensor are obtained.The theoretical research results provide a solid theoretical basis for the structural optimization and perfomance improvement of the tactile sensor,and have obvious engineering application value.(3)Development of micro-airbag flexible array sensor.On the basis of tactile feature perception and the theoretical research of micro-airbag flexible tactile sensor array,the micro-airbag flexible tactile sensor array is developed.In this thesis,the preparation technology and the assembly process of micro-airbag flexible tactile sensor array are described in detail,the sensor hardware and host computer configuration are expounded,the tactile sensor signal acquisition experimental system is established,and the performance indicators of micro-airbag flexible sensor are tested.The test results show that the resolution of the micro airbag flexible sensor is not less than 0.26%,the measuring range is not less than 26 N,the sensitivity is better than 60.8 hpa/N,the nonlinearity is less than 3.07%F.S.,and the hysteresis rate is 9.779%F.S..The sensor has excellent performance such as long life and good repeatability.(4)Research on the identification and classification of the target static contact state based on force tactile information.The distribution characteristics of the pressure value array signals of the micro-airbag flexible sensor under different contact states are studied in detail.The tactile image data set of the micro-airbag flexible sensor array is constructed.The spatial characteristics of the tactile image are extracted by using convolutional neural network(CNN),which solves the classification problem of the static contact state of the target.The test set results show that the accuracy of target static contact state classification can reach more than 96%by using CNN technology.The research results show that the generalized feature extraction function of CNN can effectively improve the accuracy of operation target classification.(5)Research on the targets motion state recognition and classification based on force tactile information.With the help of the tactile image sequence information of the micro-airbag flexible sensor array,the variation characteristics of the tactile sensor pressure value under different dynamic conditions are discussed.The Convolution-Long-Short-Term Memory(CNN-LSTM)composite network structure is adopted to extract the temporal and spatial characteristics of tactile image sequence to solve the problem of the target motion state recognition.The experimental results show that the accuracy of dynamic behavior classification can reach 92%by CNN-LSTM composite network model,which provides a feasible technical path for improving the handling stability of dexterous hand.The characteristics and highlights of the research work in this thesis are as follows.According to the analysis of sensing mechanism of micro-airbag structure sensitive unit,a new type of micro-airbag flexible sensor array is developed,which solves the key technical problems of intelligent robot perception.The research on the tactile features of sensor information has laid a solid theoretical foundation for the intelligent robot to recognize and stabilize the manipulated object.Based on the deep learning method,the effective recognition of the static contact state and the motion state between the target and the flexible tactile sensor array can be realized,which can further improve the accuracy of the intelligent robot to grasp and manipulate the target.