Viscoelastic Analysis Of Solid Rocket Propellant Based On The BP Neural Network

Solid rocket motor technology is a technology-intensive, integrated and cutting-edgetechnology. While the viscoelastic analysis of solid rocket motor is an essential link. For along time, the viscoelasticity of propellant is always an important subject for researchers.Due to the complex properties and too many parameters of the propellant，it is verynecessary to study on its viscoelastic properties from theory and engineering practice. Thispaper focuses on the viscoelastic numerical response of the propellant, which simulatedand predicted by the BP neural network, and then compared with the results of finiteelement analysis, studies the application of neural network in the viscoelastic analysis ofpropellant.In this paper, the constitutive equations of the linear viscoelastic models andfractional derivative models are presented, based on the methods of differential operatorand Laplace transform. Further analysis has been done for the static and dynamicconstitutive equation of elastic materials, for the time-varying load and alternating load ofviscoelastic response, and for the transformation mechanism of energy dissipation.The viscoelastic parameters of propellant have been measured by an advanceddynamic thermal mechanical analyzer. Combined with the time-temperature equivalenceprinciple, the master curve data of the propellant is obtained, and then fitting the data to theform of Prony series by the ANSYS own function of curve fitting. At the same time,fractional derivative model is used to fit the experimental data,and then compared with theresults of Prony series. The results show that the fractional derivative model fitting is lessfitting parameters and higher precision than Prony series.The solid rocket engine is simplified into a combined sleeve, and then we deduce itsviscoelastic response Laplasse transform under the condition of plane strain.The numerical results(NUM) of sleeve calculated based on inverse Laplace transform function in Matlab software by a sample, which provide data source to the neural networktraining set. The constitutive equation and finite element equations of three-dimensionalviscoelastic incremental is derived in detail based on the three dimensional viscoelastic unified constitutive theory, spherical tensor and partial tensor theory of stress and strain,principle of minimum potential energy, etc. The finite element simulation analysis, such asmodal analysis and the propellant viscoelastic response under plane strain condition, hasbeen carried out on the solid rocket sleeve by ANSYS software. The finite element analysisresults (FEA) can be used to compare the simulation results (ANN) of the merits of neuralnetwork.The weights and threshold of BP neural network are corrected according to thenegative gradient direction of performance function. In this paper, this learning algorithmhas been improved, then fitting the frequency-modulus experimental data of differenttemperature by the improved Levenberg—Marquardt algorithm of BP neural network, andthe simulation results show that the fitting process is very good.At the same time, combined with the uniform design method and Monte-carlo method,the numerical results have been trained by BP neural network, and then compared thenetwork simulation results and the finite element analysis results. After comparing thesimulation results of neural network and numerical results we found that the error is verysmall, and part of the neural network simulation results are even better than the finiteelement analysis, which indicates that neural network can effectively solve the problem ofthe viscoelastic property of propellant.In this paper, the research results and conclusions of the rocket propellant viscoelasticanalysis has certain significance.