Investigation On Material Defect Identification Algorithm And Development Of Mechanical Property Prediction Software

Defects are commonly present in material structures and can significantly impact their physical and chemical properties,including strength,hardness,toughness,conductivity,thermal conductivity,and corrosion resistance.The optimization of material physical properties can be achieved through the design of defects in materials.Molecular dynamics-based atomic scale calculation simulations are a critical tool for material design and physical properties research.This study focuses on the analysis and processing of structural data generated by molecular dynamics,proposing a set of defect identification algorithms that are universally applicable to all crystal systems.Furthermore,an interactive and visual Shell-DIS software has been developed to enable defect identification of all crystal system materials.Additionally,based on the physical modeling scheme of grain boundaries,an interactive software called VGB3D has been compiled for batch modeling.To achieve the prediction of mechanical properties,the study introduces the statistical analysis method into the machine learning method and establishes a database with multiple objectives.Overall,this study presents a comprehensive approach to identifying defects and predicting the mechanical properties of materials through the use of molecular dynamics simulations,software development,and statistical analysis techniques.In the field of materials science,identifying defects in material structures is a critical task,as it has a significant impact on the physical and chemical properties of materials.In this context,a defect identification algorithm that is applicable to all crystal systems is highly desirable.To achieve this,we have extended the existing defect identification algorithm and introduced a new concept of "shell nearest neighbor" by redefining the shell.Using this concept,we obtained the nearest neighbor table of the seven crystal systems,followed by the provision of precise selection criteria for existing algorithms such as centrosymmetric parameter(CSP),nearest neighbor distance analysis(NDA),and common nearest neighbor analysis(CNA).We have proposed two new algorithms,namely Shell-CSP and Shell-bond algorithm,which are efficient and applicable for the rapid identification of defect atoms in all crystal systems.To make these algorithms more user-friendly,we have developed an interactive and visual Shell-DIS software using C++ programming that compiles VTK and QT databases.This software not only identifies defective materials of all crystal systems but also allows researchers to analyze and visualize the identified defects using the window interactive mode.The software is a valuable tool for defect identification and analysis in materials science.The multi-objective prediction of material mechanical properties algorithms and software has been developed to classify the stress and strain of the grain boundary with symmetrical tilting and predict the optimal value.To establish a grain boundary model database with a large volume of data,we have compiled a batch modeling software VGB3D based on the theory of coincidence position lattice.The software can efficiently and accurately build a database of symmetrical tilted grain boundary atomic configurations of any material.Furthermore,we built the physical model database and determined the input items and input items of the prediction algorithm.The self-developed VGB3D software was used to build the tungsten symmetrical tilting grain boundary database A with ∑≤99.The uniaxial tensile behavior of database A was simulated by molecular dynamics method,its plastic deformation mechanism was analyzed,and the stress-strain curve was obtained.The grain boundary tensile behavior was classified based on the number of inflection points of the stress-strain curve,and the maximum value of the stress-strain curve was extracted as the yield strength.The yield strength was taken as the output item,and the six direction vectors needed for modeling were reduced to five scalars as input items based on the coincidence site lattice theory(CSL).Finally,by comparing the prediction accuracy and computational efficiency of various intelligent optimization algorithms applied to neural networks,the optimal whale neural network algorithm(WOA-BP)was obtained,and a multi-objective material performance prediction platform was constructed.This platform can realize the prediction of the stress-strain curve type and yield strength of arbitrary symmetrical tilted grain boundary structures.The selection of an appropriate material mechanical properties database is crucial in obtaining accurate prediction results.To this end,we propose a step-by-step screening process.Firstly,we conduct an analysis of the statistical parameters of the physical model database A to identify the main parameters that affect the database.Secondly,we expand the database by creating database B with 3 times the volume of A and database C with 5 times the volume of A based on the identified main parameters.Through the use of the WOA-BP neural network algorithm for prediction,we determine that the accuracy of the predictions is approximately the same for both databases B and C.Therefore,we select database B due to its smaller volume.Thirdly,we modify the output item of database B to predict the relevant optimal value,such as the maximum yield strength database D and the maximum strain length database E.By following this screening process,we can achieve accurate prediction results for various target databases using the existing prediction platform.This study has developed a software suite for material defect identification and mechanical property prediction,encompassing "multiple algorithms,two software,multiple databases".The software suite represents a powerful tool for analyzing and processing large-scale atomic-scale molecular dynamics simulations,offering a comprehensive data analysis and processing framework.