Three-dimensional force simulation prediction of flexible sensor based on BP neural network

Abstract

Abstract When the robot grabs the object, the force information detection of the object is the basis for the smooth grabbing process. The force information of the object can be fully reflected by detecting the force in the three-dimensional direction. In this paper, with polydimethylsiloxane (PDMS) as the substrate and embedded into the sensitive unit prepared by conductive rubber, a new flexible sensor which can detect three-dimensional force is designed. Firstly, based on the piezoresistive effect of conductive rubber, COMSOL Multiphysics software was used to carry out multi-physical field simulation experiment for the sensor. Furthermore, based on the nonlinear approximation ability of BP neural network and the resistance of simulation output, the training sample set and the test sample set were constructed by using the 5-fold cross validation method, and the BP neural network model was constructed to achieve the accurate prediction of three-dimensional force. Finally, the number of hidden layer neurons was adjusted to optimize the BP network model. The results of cross-validation experiments show that the sensor designed in this paper can effectively detect the three-dimensional force information, and the optimized BP neural network can significantly improve the accuracy of the three-dimensional force prediction.