Research On Method Of Database Platform Design And Machine Learning Performance Prediction Of Lubricating Materials

Thedemand for high-quality lubricating oil in China is increasing year by year,while high-end lubricating oil and its development technology are heavily dependent on foreign countries,independent research and development is facing great difficulties.In the development and improvement of lubricating products,performance detection and evaluation is an essential important link.The performance evaluation of lubricating oil mainly depends on the current situation of experience or experimental results,which seriously restricts the design and development cycle of lubricating oil.The combination of material database and machine learning is a method for research and development of lubricating materials,which provides a new way for the rapid evaluation of the performance of lubricating oil,it is of great significance for improving the reliability of equipment operation and guiding the development and application of lubricants.In this thesis,typical lubricating oils and its additives are as the research object.By constructing a diversified data storage and analysis material database for experimental testing and simulation calculations,an integrated data platform for automatic data entry,retrieval,simulation calculation and performance prediction has been established;Based on molecular simulation calculations,the molecular structure parameters of lubricants and their additives have been analyzed;The weights of the influence of the molecular parameters of the lubricant oil on the tribological properties and thermal oxidation properties are determined.Combined with machine learning algorithm,molecular structure performance machine learning models of lubricant wear established,the accuracy and applicability of different machine learning models for predicting the thermal oxidation properties and tribological properties of lubricant were studied,and an efficient performance prediction ensemble learning algorithm was proposed,by combining the lubricating material database with machine learning,it provides strong evidence for the development of high-performance lubricants.Based on the above research work,the main conclusions of the paper are as follows:(1)An integrated platform for data storage and analysis of lubricating materials is designed to realize the diversified data import of lubricating material test detection data and simulation calculation result files;The data retrieval function based on the conversion between different data formats(tables and documents)and the extraction of key information;The call of simulation software and the performance prediction of rapid evaluation of support materials.The database covers commercial lubricants,base oils,additives,lubricating greases,solid lubricating films,etc,and the data content includes the chemical structure of lubricants and their additives,physical parameters of materials,test parameters,and main properties of materials.The establishment of the database meets the requirements of high-throughput calculation and design results induction and data mining,It provides data base and analysis development platform for rapid evaluation of lubricating material performance.(2)Four typical ester oils(diester,pentaerythritol ester,trimethylolpropane ester and trimellitic ester)were taken as the research objects,molecular parameters of ester oils with different chain length structures were simulated and calculated and effects of bonding properties,chemical activity and molecular orbital on their working properties were analyzed.The results show that the total molecular energy and dipole moment of different ester oils are significantly different,which are important parameters affecting the antioxidant,hydrolytic stability and lubricity of ester oils.The results of HOMO-LUMO energy levels show that the ester group(or the conjugated structure formed by benzene ring and ester group)in the molecular structure of ester oil has the highest activity.In the process of lubricating metal,ester oil will form a solid lubricating film on the metal surface,which ensures good friction performance.Increasing the length of carbon chain at both ends of diester molecule can improve its anti-wear performance.The results of electrostatic potential show that the C = O bond in the ester group is negatively charged,and it is easy to combine with metal cations or be attacked by H+ in water,which will destroy its molecular structure and affect the stability of lubrication or hydrolysis.In addition,the results of electronic structure elucidate the contribution and distribution of molecular orbital of ester oil.The simulation results can provide a scientific explanation for the importance of feature parameters selected by machine learning.(3)Based on the molecular structure parameters of lubricating oil calculated by molecular simulation,the influence weight of the structural parameters of lubricating oil relative to the amount of wear was calculated,and the molecular characteristic parameters of lubricating oil are determined as follows: LUMO energy and Total dipole;Similarly,according to the influence weight of the structural parameters of lubricating oil relative to the initial oxidation temperature,the molecular characteristic parameters of lubricating oil are determined as follows: Total energy,LUMO energy,HOMO-LUMO energy,Total dipole,and Alo P.After that,combined with the multiple linear regression machine learning method,the preliminary research on the prediction methods of tribological properties and oxidation resistance of lubricating oil was carried out.The machine learning models between the calculated characteristic parameters and the wear amount,and between the calculated characteristic parameters and the initial oxidation temperature of lubricating oil were established,and the accuracy of the prediction set was verified,the results show that the predicted values are in good agreement with the experimental values.(4)Based on multiple linear regression,support vector machine and neural network machine learning algorithm,the prediction of tribological properties and antioxidant properties of lubricating oil materials was carried out,and the differences of different machine learning methods for predicting lubricating oil properties were clarified.On this basis,the integrated learning algorithm of lubricating oil machine learning performance prediction based on stacking theory is explored,and an efficient prediction method of lubricating oil tribological performance based on machine learning is proposed.Finally,an efficient prediction system of lubricating oil tribological performance was established,enrich and improve the lubricating material database,Promote the performance evaluation technology of lubricating oil,and accelerate the development and application of lubricating oil.