Research On Multifunctional Tactile Sensing Method Based On Piezoresistive Tactile Sensor

With the rapid development of sensor technology,computer science,artificial intelligence and other multi-disciplinary technologies,the relevant needs of intelligent robots have gradually spread across medical,service,industry and other fields.The research of intelligent robot is of great significance to modern society.The intelligent robot can use a variety of sensor devices it carries to complete the detection of external physical information.The tactile sensor is an important part to realize the interaction between robot and the external environment.Through the passive and active perception information feedback by the tactile sensor,the robot can not only achieve social interaction with people,but also evaluate the properties of objects.The combination of tactile sensor and deep learning technology can enable robots to complete specific tasks in a more intelligent way.Therefore,the analysis and research on the collection,processing and recognition of tactile information combined with the deep learning algorithm plays an important role in improving the tactile sensing function of robots,so as to better realize human-computer interaction and daily fine operation.The main contents of this dissertation are as follows:(1)This dissertation focuses on the multifunctional tactile perception of tactile sensor.In order to improve the performance of the tactile sensor and enable it to effectively perceive and collect the tactile information in the interaction process,the research is carried out from the active materials and flexible substrate structure of the flexible piezoresistive tactile sensor.The tactile sensor with graphene oxide as conductive active material and three-dimensional porous structure as flexible substrate structure is studied.The porous flexible substrate is prepared by using the economical and environmentally friendly elastomer template method.Graphene oxide is deposited on the porous flexible substrate by the solution impregnation method to prepare a porous graphene flexible tactile sensor sensing element.On this basis,the performance of the sensor is tested and analyzed from the perspectives of response time,sensitivity and hysteresis.The experimental results show that the porous graphene flexible tactile sensor proposed in this dissertation has excellent tactile signal sensing ability and is suitable for the detection and acquisition of tactile information.(2)In order to solve the problem that it is difficult to accurately identify the contact state in the process of passive tactile perception.This thesis proposes a contact state recognition method based on residual network model.Using the prepared porous graphene flexible tactile sensor to collect the tactile data of four contact states(slap,thump,stroke and push)applied by human hand,and a residual network model with excellent adaptability and generalization ability is constructed to classify and identify the four contact states applied to the surface of the tactile sensor,and the recognition accuracy is97.50%.On this basis,based on the same dataset,the multi-layer perceptron and convolutional neural network models are constructed respectively,and their recognition accuracy rates are 82.71% and 86.46% respectively.The experimental results show that the porous graphene tactile sensor prepared in this dissertation has good flexibility,high sensitivity and fast response,and can effectively sense the contact force in different states.The constructed residual network model can be effectively used for the classification and recognition of contact state of tactile sensors.(3)In order to solve the problem of distinguishing the hardness of objects,identifying the types of objects,and easily causing damage to objects in the process of grasping,this dissertation proposes a method for object hardness and type recognition based on residual network model.And this thesis prepared two 2 × 3 arrays of porous graphene tactile sensor with high flexibility,carries it on the two-finger mechanical claw,and controls the twofinger mechanical claw to slightly squeeze different objects by 2mm respectively,so as to collect corresponding tactile sequence characteristic signals;On this basis,the residual network model is constructed to realize the classification and recognition of four soft and hard(soft,soft,hard and hard)and twelve object types,with the average accuracy of 100%and 99.7% respectively.In order to further verify the classification ability of the residual network model to the tactile feature information detected by the sensor array,the multilayer perceptron,convolutional neural network,Le Net,multi-channel deep convolutional neural network and ENCODER model are constructed based on the same dataset.The average recognition accuracy of the five depth learning models for the four hardness categories is 93.6%,89.7%,98.3%,93.3% and 98.1% respectively.At the same time,the average recognition accuracy of 12 object types based on the five models is 94.7%,87.2%,98.9%,85.0% and 96.4% respectively.The experimental results show that the residual network model constructed in this dissertation can more accurately realize the hardness and type of object and object classification based on flexible tactile sensors.