The Optimization Of Material Property And Molecular Screen Based On Data Mining Methods

Materials data mining can be defined as the research work to utilize the computer,statistical theory and artificial intelligence for reorganizing,analyzing,evaluating,screening,modelling,predicting,optimizing and applying of complex materials data,in order to find the internal law and to discover the unknown system.Since the Material Genome Initiative?MGI?was proposed in 2011,the materials data mining researches had been developed very fast.Many research peoples would like to utilize data mining methods for exploring the quantitative relationships between the properties and features of compositions,structures and technical conditions,so the relationships can be used to predict,design and optimized the properties,thus to accelerate innovation process of new materials.In this work,the main contents can be summarized as follows:?1?Review on the recent developments of materials data miningThe close relationships were declared between the materials data mining and materials design,materials informatics,materials genome engineering and materials industrial optimizations.The advantages and limitations of widely-used data mining methods are summarized.The basic flow path of materials data mining is analyzed,while the recent developments are summarized and the developing trends are discussed.?2?Data mining on the specific surface area?SSA?of ABO₃-type perovskiteThe specific surface area?SSA?of ABO₃-type perovskite is related to its chemical compositions and technical parameters.It is a big challenge that machine learning could be used to predict SSA of ABO₃-type perovskite based on the experimental data available.In this work,the data mining methods were used to explore the relationship between the specific surface area?SSA?of ABO₃-type perovskite with its features including chemical compositions and technical parameters.The genetic algorithm?GA?-relevance vector machine?RVM?method was used to screen the main features for modeling.It was found that the machine learning model for SSA can be constructed based on the main features including Ra?atomic radius of the A position?,Rb?atomic radius of the B position?,IF?tolerance factor?,a O3?unit cell lattice edge?,B-Tb?normal boiling point of the B position?,DA?density of the A position?,DB?density of the B position?,CT?calcination temperature?and AH?calcination time?.The correlation coefficient?R?between predicted SSA and experimental SSA reached 0.84 for training set and 0.68 for independent test set,respectively.?3?Industrial optimization of fluororubber products based on data miningThe data mining methods were also successfully applied in the optimization of processing fluororubber?FKM?products.The optimal projection recognition technique was proposed to construct the pattern recognition model for classifying different kinds of fluororubber production concerning on Mooney viscosity.The optimization scheme of FKM production process available has been implemented in the production device.The results show the application of pattern recognition model improved the economic benefits of enterprises since the qualified rate of the products increased from53%to 85%.?4?Molecular screen of hypoglycemic efficacy of Dihydrochalcone based on data mining methodA series of dihydrochalcone derivatives having different positions and amounts of methoxy and hydroxyl groups were screened based on data mining method.The relationship between structure and effect was studied by using support vector machine regression algorithm,and the correlation between dihydrochalcone structure and hypoglycemic effect was established.It is expected that the compounds with stronger hypoglycemic effect could be screened based on the model available.