Study On Molecular Structure Design Of High Thermal Conductivity Polymer Insulating Materials Assisted By Machine Learning

Polymers are widely used in high voltage electrical equipment because of their excellent insulating and mechanical properties.With the development of power equipment to high voltage,large capacity and miniaturization,the thermal aging of polymer materials leads to insulation failure,which threatens the safety of power system more and more seriously.It is difficult to improve the thermal conductivity of polymer by conventional means because of its complex morphology and changeable molecular chain structure.The emergence of material informatics provides new ideas for the development of new high thermal conductivity polymer insulating materials.This thesis combines molecular simulation and Machine learning(ML)to discover new high thermal conductivity polymer insulating materials,and carries out research on the relationship between its molecular structure and thermal conductivity.A series of simulation models of polymer chains with different molecular structures were established by selecting typical polymer groups.The thermal conductivity of polymer chains with different molecular structures was calculated by molecular dynamics(MD).The thermal conductivity distribution of polymer chains with different molecular structures was summarized by statistical calculation results.The results show that the thermal conductivity of polymer chains is significantly dependent on its morphology,and the polymer chains formed by some monomers show high thermal conductivity potential.The molecular fingerprints of the samples were constructed by integrating the monomer structure information and the spatial structure information of the polymer as the input of the ML model,including two two-dimensional fingerprints representing the monomer structure information and one three-dimensional fingerprint representing the spatial structure information.Then,different ML models were established,and the parameters and structures of the models were adjusted to obtain the optimal prediction model,so as to realize the rapid prediction of the thermal conductivity of polymer chains.The results show that these prediction models can effectively predict the thermal conductivity with only the input of molecular fingerprints representing the structural information of the unknown new material.Compared with the MD simulation,the calculation cost of these prediction models is reduced by several orders of magnitude with a relatively high accuracy.The relationship between polymer monomer structure and thermal conductivity was studied by Pearson correlation coefficient method,and the reasons for the different thermal conductivity of polymer chains with different structures were analyzed according to the conformational energy parameters.An interpretable ML model based on neural network was established to identify the relationship between the monomer structure and spatial structure information of polymer chains and the key data of thermal conductivity,and to find the key factors governing heat transfer in polymer chains.The results show that the polymer chains with large conformational energy parameters show high thermal conductivity,which means the monomer configuration is stable and the spatial structure is compact..