Research On Structure Design And Energy Storage Performance Of Polyetherimide-Based Composite Dielectrics Based On Machine Learning

Polymer dielectric capacitors are widely used in the fields of new energy vehicles,pulse power devices,and portable electronic products due to their high-power density,fast charge-discharge rate,and excellent stability.The energy storage density of pure polymer dielectric is generally low and cannot meet the actual requirements for energy storage performance in the abovementioned fields.Polymer-based composite dielectrics provide an effective way to improve the energy storage density of polymers.However,the critical parameter that affects energy storage density,breakdown strength,is influenced by many factors and may have couplings between them,making it difficult to obtain the structure-property relationship solely from experiments.The solid dielectric breakdown time is short,and the physical process is difficult to capture by experimental observations.In addition,the composite forms of polymer-filler are diverse,the workload of exploring structure-property relationship through experiment is huge.Conventional characterization and testing methods cannot summarize the relevant laws of composite forms.To address these issues,the use of big data methods to assist in experimental design,optimize the microstructure and performance of polymer-based composite dielectrics,combine characterization testing and computational simulation to clarify the mechanisms of performance optimization is of great significance for understanding dielectric breakdown behavior,improving energy storage performance,and shortening the experimental cycle.Through high-throughput stochastic breakdown simulations,a breakdown performance database of 504 polymer-based composite dielectrics was established by taking the size,content,and dielectric constant of inorganic fillers as parameters.By using the control variable method,corresponding stochastic models were constructed to study the effects of blocking effect,distribution characteristic,and dielectric matching on the breakdown behavior of polymer-based composite dielectrics.Using the least squares regression(LSR)machine learning approach,an interpretable machine learning model(coefficient of determination R2=0.778)was constructed,and an energy storage density prediction function adaptable to most polymer-0D inorganic filler composite systems was obtained by combining the Maxwell-Garnett(MG)effective medium model.Polyetherimide(PEI)with high insulation properties and 0D alumina(Al2O3)were selected as the matrix and filler,respectively,for validation experiments outside the database.The experimental results of dielectric,breakdown,and energy storage were consistent with the breakdown prediction model,the MG model,and the energy storage density prediction function,respectively.Through literature review and data extraction from the Web of Science database,the structure-property relationship between the structural parameters(dielectric constant and layer numbers of the matrix;dielectric constant,size,orientation ratio,each layer concentration,gradient concentration of the inorganic filler;dielectric constant and thickness of the coating layer on the filler)and the breakdown performance was established.Other parameters affecting breakdown performance were further decoupled,and a database of 4646 polymer-based composite breakdown strengths was established through combining the high-throughput stochastic breakdown simulations.Machine learning approaches such as linear regression(LR),k-nearest neighbor(k-NN),random forest(RF),and gradient boosting(XGBoost)were used for database training and hyperparameter optimization,with the XGBoost model exhibiting the highest predictive ability(R2=0.966).PEI/2D boron nitride nanosheets(BNNS)single-layer(filled)and three-layer(layered)composite dielectrics,were prepared as verification experiments.Comparison of the experimental results for breakdown performance indicated that the prediction errors of k-NN,RF,and XGBoost models were all less than 2.1%,with the XGBoost model having a prediction curve that more closely matched the experimental test curve.The machine learning model was analyzed to obtain the feature importance ranking,and the gradient structure design of the PEI/BNNS composite dielectric was optimized.Through alternate high-speed directional electrospinning,composite dielectrics with symmetric gradient structures of different total concentration,gradient concentration,and gradient directions of BNNS were prepared.The test results showed that parallel-oriented BNNS and mesoscopic interfaces in the symmetric gradient structure can significantly increase the interface polarization effect and the breakdown path bending.The gradient structure design can homogenize the total electric field distribution,further increasing the withstand voltage capacity of composite dielectrics.In addition,the surface barrier layer in the symmetric inverse gradient structure can more effectively inhibit the charge injection from the electrode/dielectric interface,and create effective interception and secondary inhibition after the carriers cross the barrier.The symmetric gradient structure design synergistically optimized the polarization and breakdown performances of the composite dielectric,and the final energy storage performance was significantly improved.BNNS with gradient distribution forms a continuous thermal conduction channel in the matrix,and the composite dielectrics still maintain high energy storage performance at high temperatures.Adjusting the BNNS distribution in the PEI,PEI-based composite dielectrics with asymmetric gradient structure were prepared using alternating high-speed directed electrospinning,where the gradient direction of BNNS from the center to the bottom surface was consistent with that from the top surface to the center.The structure design was optimized by stochastic breakdown simulation to shorten the experimental cycle.The test results showed that the surface barrier layer constructed in the asymmetric inverse gradient structure could effectively inhibit carrier injection and early transport,the central hinder layer could further inhibit carrier transport,and the short distance between the barrier layer and the central hinder layer reduced the free acceleration distance of carriers,slowing down the acceleration energy of carriers.At room temperature,the energy storage density of PEI-based composite dielectric with 2 vol% BNNS total concentration,1 vol% BNNS gradient concentration,and BNNS asymmetric inverse gradient distribution is 8.26 J/cm3;at 100 ℃ and 150 ℃,the energy storage density is 7.80 J/cm3 and 4.78 J/cm3,respectively.This paper is based on big data and simulation calculations to assist in the optimization of the microstructural design of polymer-based composite dielectrics.Through the combination of characterization,testing,and simulation,the carrier injection/transport mechanism is elucidated.This has important theoretical significance and application prospects for the development of high-energy density dielectric materials.