Research On Neural Network Modeling Of Hysteresis Characteristics And Estimation Of Gripping Force For Pneumatic Gripper Systems

Along with the continuous improvement of China’s industrial automation level,pneumatic grippers are highly favored in industrial manufacturing for their advantages of low cost,safety,reliability,and being clean and pollution-free.However,the gas compressibility,the inherent nonlinear friction of the system,and the uncertainty-induced interference such as expansion and deformation of the rubber tube of the air circuit during the filling and discharging processes make the pneumatic gripper have complex,strong nonlinear hysteresis characteristics,which limit the further application of pneumatic grippers.The pneumatic hysteresis characteristics lead to strong nonlinearity and timevarying output of the gripping force,so it is difficult to achieve effective control of the pneumatic gripping force based on the air pressure signal directly.Therefore,by modeling the strong nonlinear hysteresis characteristics of pneumatic grippers and constructing a pneumatic gripping force estimation model based on pneumatic hysteresis modeling to achieve indirect measurement of the gripping force of pneumatic grippers,it becomes an effective means to improve the accuracy of pneumatic fixture gripping force control.For the problem of high accuracy estimation of the gripping force of a pneumatic gripper based on a hysteresis characteristic model,two methods of gripping force estimation based on the hysteresis model are proposed in this paper:1.A strategy by CNN-LSTM modeling and estimating of pneumatic gripping force.According to the characteristics of pneumatic gripper gripping force related to the input and output of driving air pressure history,the Long Short-Term Memory(LSTM)network with memory characteristics is used to establish the air pressure based gripping force estimation model.To address the problem of large errors in modeling directly with the LSTM network,the Convolutional Neural Network(CNN)is used to extract the nonlinear relationship between air pressure and gripping force in the input information,and the CNNLSTM model is constructed by connecting the CNN and LSTM in series,the LSTM network structure is further optimized to improve the model estimation accuracy,effectively describe the multi-value correspondence characteristics and nonlinear hysteresis characteristics between air pressure and gripping force,and realize the effective estimation of the gripping force of the pneumatic gripper.The experimental data results show that the constructed CNN-LSTM gripping force estimation model has higher estimation accuracy compared with the LSTM model directly used for gripping force estimation.2.Combined continuous DBN with ELM for hysteresis modeling and gripping force estimation method.The above method has not fully considered the uncertainty of the multi-value correspondence of the pneumatic hysteresis characteristics,resulting in a significant estimation error.In order to further improve the estimation accuracy of the modeling,considering both the output uncertainty and the strong nonlinear characteristics of the positive and negative ranges unique to the pneumatic gripping system,the uncertainty corresponding to multiple values of the output exhibited by the hysteresis characteristics of the pneumatic gripping system is described with the help of the stochastic uncertainty property possessed by the Deep Belief Network(DBN)model structure.In order to further enhance the description of forward and backward strong nonlinear characteristics by DBN and improve the accuracy of pneumatic gripping force modeling estimation,Extreme Learning Machine(ELM)with strong nonlinear characteristics description capability is introduced to replace the regression layer in DBN,and DBN and ELM are docked and fused to construct a hysteresis model of pneumatic gripper fused by DBN and ELM for effective estimation of gripping forces.The experimental data validation results show that compared to the gripping force estimation methods based on the Prandtl-Ishlinskii(P-I)model and the CNN-LSTM model,the gripping force estimation method based on the fusion model of DBN and ELM has a high estimation accuracy.The two methods of estimating the gripping force based on the pneumatic hysteresis model proposed in this paper have been experimentally shown to be effective in describing the hysteresis nonlinear characteristics of pneumatic grippers,among which the gripping force estimation method based on the fusion model of DBN and ELM has higher modeling accuracy while the CNN-LSTM model outperforms the fusion model of DBN and ELM in terms of computational power.The two methods of estimating the gripping force proposed in this paper provide different options for environments with different requirements on accuracy and computational power.