Research On Mechanical And Thermal Properties Of Thermal Conductive Gasket Based On Artificial Neural Network

Thermal pad is a kind of heat-conducting composite material which is easy to process and excellent in performance,and has good thermal and mechanical properties.Numerical simulation has achieved great success in the research of composite materials.As a representative method,FEM(Finite Element Method)has the advantages of high calculation efficiency and good fitting accuracy,and is widely used in various model calculations.However,there are many parameters involved in the design process of thermal pad,and many factors can’t be quantitatively considered.It takes a lot of work and a long time for engineering calculation only by FEM,and it is difficult to give consideration to the prediction of mechanical properties and thermal properties at the same time.Therefore,a new research method needs to be introduced as a supplement to this problem.In this paper,based on FEM,the thermal pad composite material with epoxy resin as matrix,carbon fiber and aluminum oxide as filler is numerically simulated to establish a model.The data sets of mechanical and thermal properties are obtained through simulation,and an artificial neural network is established to predict the mechanical and thermal properties of the thermal pad.The research method of this paper is the preliminary research work on the optimal design of new materials,which is used to guide the selection and gradation of fillers in composites,and has important value for preparing composites with good mechanical and thermal properties.The main work contents are as follows:(1)An artificial neural network was established to accurately predict the mechanical properties of thermal pad composites.Firstly,the influence of the volume fraction of total filler and carbon fiber volume fraction on the mechanical properties of thermal pad was explored.Then,a large number of heat-conducting gasket models were established and mechanical simulation was carried out.From the simulation results,transverse Young’s modulus,longitudinal Young’s modulus,transverse plane shear modulus,and longitudinal plane shear modulus were selected as the indexes to evaluate the mechanical properties of thermal pad.Descriptors that may affect the mechanical properties of the thermal pad are extracted from the model,and 63 features of the data are composed with the input parameters.Eight of the better features are selected to form feature subsets for the training of artificial neural network.The model with the smallest comprehensive error is selected from the structures with different hidden layers and neurons as the result,and the test set is used to evaluate it.The results show the fitting degree of artificial neural network on this data set is very good,R2 is 0.992,MAPE is 0.02,ACC is 91.51%,and it has a high prediction accuracy,which is obviously superior to other machine learning models.(2)An artificial neural network was established to accurately predict the thermal properties of the thermal pad.Firstly,the influence of interfacial thermal resistance between different materials on thermal conductivity was explored by using the main effect analysis.Then,a large number of thermal pad models were established.At the same time,the thermal performance of the thermal pad was simulated by taking the interfacial thermal resistance between different materials as an input parameter,and the thermal conductivity was used as an index to evaluate the thermal performance of the thermal pad.A data set consisting of 7 features and 3 labels was established.Based on the data set,an artificial neural network model was established,and the model was evaluated by using a test set.The results show artificial neural network has high prediction accuracy on this data set,R2 is0.9962,and MAPE is 0.0457,and the error is much lower than other machine learning models.