Application Of Machine Learning To The Design Of TiB_2p/TiB_W Reinforced Copper Matrix Composites

The advanced copper matrix compositcs with high strength,high conductivity and multiple functions can meet the design requirements under special working conditions or harsh service conditions.It is an irreplaceable key base material in the field of high performance metal matrix composites.By adding high-performance second phase material TiB₂ particles and TiB whiskers into the copper matrix,the mechanical properties of the copper matrix materials can be improved,but the conductivity will be greatly reduced.Therefore,while greatly improving the mechanical propertics of copper matrix compositcs.how to control the decrease of the conductivity of the composites within the allowable range is the key link in the research of copper matrix composites.On the other hand,composite materials naturally have a large amount of data and a complex design process due to their characteristics of complex composition.complex structure and various preparation and processing means.With the development of machine learning and neural network,how to apply machine learning to accelerate the design process of composite materials has become a new development direction of composite materials design.In this paper,the microscopic morphology of TiB₂ particles and TiB whiskers was determined according to scanning electron microscopy（SEM）images,and the finite element reconstruction modeling was carried out.The mechanical and electrical properties were obtained through the construction of mechanical and electrical models.The data set was unified and combined through data screening and feature extraction,and the mechanical and electrical properties prediction model of copper matrix composites was established using the neural network based on machine learning.It was substituted into the intelligent optimization algorithm to construct the unified multi-objective regression prediction model.which realized the prediction of the characteristic parameters to the performance parameters of the composite materials,and mined the complex relationship between the high-dimensional material modeling and improve the cost performance of model calculation,the intelligent and accurate design of composite material structure is realized.The results show that:（1）For pure TiB₂ particle reinforced copper matrix composites,the conductivity of TiB₂p/Cu composites decreases with increasing the volume fraction of TiB₂ particles.TiB grain whisker orientation Angle θ increases from 0° to 90°.The conductivity of TiBw/Cu composites decreases gradually.The conductivity of the model with whisker orientation Angle θ 0° is the highest,and the conductivity of the model with whisker orientation Angle θ 90° is the lowest.This is because TiB₂ particles and TiB whisker intensifiers reduce the electron relaxation time by increasing the probability of continuous collision in the electron composites,thus affecting the change of electrical conductivity.（2）For the（TiB₂p+TiBw）/Cu composite with a volume fraction of 6vol.%,the conductivity of the model is the highest when the whisker orientation Angle θ is 0°.For the randomly distributed（TiB₂p+TiBw）/Cu copper matrix composites,different proportions of TiB whisker and TiB₂ particles have no significant effect on the conductivity,and the type of reinforcement has little effect on the conductivity.The porosity is an important factor leading to the decrease of（TiB₂p+TiBw）/Cu composite conductivity.The higher the porosity,the less rounded the pore morphology,the greater the current obstruction and scattering ability,and the lower the conductivity of the composite.（3）Based on the above model,appropriate computational results of the electro-mechanical model are selected as the sample set,and feature engineering and normalization are used to preprocess the sample set.BP neural network is used to predict the nonlinear relationship,and the mean square error is within a reasonable range compared with the actual results.The BP neural network regression prediction model can provide reference for the prediction of yield strength,elastic modulus,tensile strength,elongation and conductivity of copper matrix composites and the selection of input characteristic parameters.（4）Global optimal algorithms（genetic algorithm GA and particle swarm optimization algorithm PSO）are used to improve the problem that neural networks are prone to fall into local optimal.The M value was used to characterize the synthesis and properties of copper matrix composites.The prediction network was used to calculate the optimal properties.The results showed that the overall integral number of（TiB₂p+TiBw）/Cu composite was 13.5%and the whisker content was 12%.The verified results are close to the actual simulation results.