Research On Parameters Inversion Of Metal Powder Forming Model Based On BP Neural Network

As the most promising technology at present,powder metallurgy technology can quickly and efficiently produce high-performance structural materials and complex parts,with features such as less cutting and easy processing.The key point of powder metallurgy part forming is the powder pressing process.The numerical simulation of the pressing process can reduce the cost of product development.Numerical simulation is the key to the establishment of the material constitutive model and the yield stage in the process of powder compaction can be well described by the modified Drucker –Prager Cap(DPC)model.However,due to the complexity of the powder itself,its parameters need to be determined through multiple tests such as uniaxial compression test,Brazilian disc test and compression test.At the same time,there are measurement errors in the parameter determination experiment,which makes it impossible to accurately obtain the constitutive structure of the powder.The model parameters cannot accurately describe the powder compaction.How to determine the constitutive model parameters quickly and accurately is the basis of further research on metal powder forming,with a more important application of powder metallurgy process optimization.Firstly,the parameters of the modified Drucker-Prager Cap model by the material were measured by multiple experiments,and the mathematical model of FC0205 metal powder compaction was established.The numerical simulation of the FC0205 metal powder unidirectional compaction process was realized based on the finite element analysis platform of ABAQUS and its subroutine USDFLD.And the validity of the numerical simulation was verified by the compression experiment data.Secondly,the parameters inversion prediction of the modified Drucker-Prager Cap model of metal powder was realized based on the numerical simulation of the modified Drucker-Prager Cap model and the BP neural network optimized by genetic algorithm.The results show that the prediction errors of FC0205 metal powder parameters are all lower than 8%,which shows the feasibility of the inversion method.In addition,Al6061 metal powder was taken as an example for parameters prediction.The results show that the average absolute percentage error between the numerical simulation results and the die compaction tests data after parameters inversion is only 5.10%.The BP neural network model can predict the parameters of the modified Drucker-Prager Cap model quickly,effectively and accurately.Finally,forming process parameters of FC0205 powder metallurgy parts were analyzed based on the optimized modified Drucker-Prager Cap model.The influence of compaction mode and friction coefficient on cylinder forming was discussed.The relative density uniformity of the cylindrical shaped parts is improved by bidirectional compaction.The friction coefficient is negatively related to the maximum relative density of the green compacts.The difference in relative density is weakly correlated with the friction coefficient,and the overall relative density distribution of the green parts can be improve d by appropriately increasing the friction coefficient.Taking the relative density as the evaluation standard,an orthogonal experiment is carried out on the process parameters of the synchronizer cone ring forming.The results show that the extreme difference of the compaction method is 0.2430,the extreme difference of the friction coefficient of the outer wall of the mandrel is 0.0557,and the extreme difference of the friction coefficient of the upper and lower punches is 0.0049.The suppression mode has the greatest influence on the relative density distribution.The optimum process scheme is determined as the bidirectional compaction,the friction coefficient of the outer wall of the mandrels was 0.06,and the friction coefficient of the upper and lower punches was 0.06.