Smart material simulation system and method combining big data and expert experience

Abstract

The invention discloses a smart material simulation system and method combining big data and expert experience. The method comprises the steps that a user inputs data; judging whether the structure of the input data is reasonable or not, and if not, prompting a user to pay attention to related items; generating a calculation file based on the input data, and generating a corresponding task record library at the same time; reading a task state in the task record library, submitting calculation if an uncalculated task exists and an unoccupied computer node exists, and modifying the task state into a submitted state; calculating the calculation file; task states in the task record library are read regularly, calculation result analysis is carried out on the submitted tasks based on the experience knowledge base, and a chart is drawn. The method has the advantages that the technical problems that in traditional material simulation, efficiency is low, experience is difficult to inherit, and results are unstable are solved, a material calculation platform with autonomous learning and intelligent evolution capacity is further constructed, and the intelligent process of new material design and discovery is greatly promoted.

Description

Technical Field

[0001] The present invention relates to the field of material simulation technology, and in particular to an intelligent material simulation system and method that combines big data with expert experience. Background Art

[0002] As an important research method in the current field of materials science, material simulation is widely used in the design and performance prediction of new materials. The setting of calculation parameters in the material simulation process plays a key role in the accuracy of the results. If the parameters are set unreasonably, it will directly lead to deviations or even errors in the calculation results, thereby misleading the research direction, extending the R&D cycle and increasing costs. At present, there is no mature intelligent automation system widely used in the field of material simulation, and traditional manual operation methods are generally used to simulate and calculate material properties. The implementation scheme of traditional manual operation methods is as follows: First, researchers manually set simulation parameters based on experience or relevant literature, such as crystal structure optimization parameters, electronic calculation accuracy parameters, energy convergence criteria, and K-point sampling grid density. These parameters are then manually entered into computational software (such as VASP and Quantum ESPRESSO) and gradually adjusted to achieve stable results.

[0003] Secondly, during the calculation process, researchers need to closely monitor the progress of the calculation and judge whether the parameters are appropriate based on the rationality of the output results. If the calculation results are unreasonable or do not meet the expected accuracy, they need to readjust the parameters and repeat the simulation until a satisfactory result is obtained.

[0004] Finally, researchers need to manually analyze the calculation results in detail, summarize the properties of the materials, and guide subsequent experiments or calculation plans based on the analysis results.

[0005] While widely adopted, this traditional method has numerous drawbacks. These include: parameter setting is highly dependent on the individual experience of researchers, which is difficult to transfer, limiting the accuracy and repeatability of the results; the extensive and repetitive manual adjustment process is inefficient and time-consuming; and the lack of the ability to effectively utilize big data to aid parameter optimization, preventing systematic improvement in the accuracy and intelligence of calculated parameters. Furthermore, with the rapid accumulation of materials data, the effective use of big data to assist in parameter setting has become a pressing issue.

[0006] Therefore, there is an urgent need to develop an intelligent system that integrates big data analysis and expert experience to automatically and efficiently optimize calculation parameters and improve simulation accuracy and efficiency. The present invention proposes an intelligent material simulation system and method to address the above technical problems. Summary of the Invention

[0007] In response to the shortcomings of the existing technology, the purpose of the present invention is to provide an intelligent material simulation system and method that combines big data and expert experience. By automatically optimizing and setting calculation parameters, it can effectively improve the accuracy, stability and efficiency of material simulation, and significantly reduce the cost and R&D cycle in the material research and development process.

[0008] In order to achieve the above object, the technical solution adopted by the present invention is as follows: In a first aspect, the present invention proposes an intelligent material simulation system that combines big data with expert experience. The key aspects of the system include: Task record library, used to store and record task records of material simulation process; Initial database, used to store the initial data required for the material simulation process input by the user; An experience knowledge base, used to store experience data required for the material simulation process, wherein the experience data and the data in the task record have the same ID number; The execution module is used for the user to input tasks and goals, and to determine whether the data structure of the tasks and goals input by the user is reasonable. If reasonable, the initial data is called based on the input data to generate a calculation file, and the corresponding calculation task library is generated at the same time. Otherwise, the user is prompted to pay attention to relevant matters or modify the input data; it is used to read the task status in the calculation task library, submit the calculation if there are uncalculated tasks and there are free computer nodes, and modify the task status to "submitted"; it is also used to calculate the calculation file, if the calculation is successful, modify the task status to "completed" and read the task status in the calculation task library at regular intervals, analyze the calculation results of the "submitted" tasks based on the experience data, draw a chart if the calculation results are reasonable, otherwise modify the task status to "calculation error", and recalculate the calculation file after adjusting the corresponding calculation parameters according to the error type, otherwise recalculate the calculation file after adjusting the parameter settings based on the active learning algorithm and the experience data.

[0009] Furthermore, the task record includes the following fields: task number ID, folder creation time creation_times, task type task_type, folder address folder_path, file name of the original POSCAR structure provided by the user poscar_original_name, task status status, task priority task_priority, task parameter setting method parameter_set_basis, ID number of the related calculation in the parameter setting basis related_calculation_ID, and experimental goal task_goals.

[0010] Furthermore, the status field includes five states: pending, submitted, completed, error pending, and error_pending.

[0011] Furthermore, the initial data includes the initial structure POSCAR, task type, task target, experimental data experiment_lattice.csv, and calculation rules rules.csv.

[0012] Furthermore, the format of the calculation rules rules.csv is as follows: Material name Material, rule type rule_type, corresponding parameter param_key in INCAR file, parameter setting value param_value, and note note.

[0013] Furthermore, the empirical data includes the following fields: task number ID, material name / chemical formula Materials, elements contained in the material Elements, whether there is a surface configuration Surface, whether it is a semiconductor Semiconductor, GGA parameter setting method GGA, whether there is a van der Waals structure VDW, van der Waals structure correction method IVDW, van der Waals structure correction atomic radius VDW_RADIUS, K point density denK in KPOINTS file, cutoff energy ENCUT, U value setting GGA+U in GGA+U, experimental value La-exp of a-axis lattice parameter, experimental value Lb-exp of b-axis lattice parameter, c-axis lattice parameter Experimental value of lattice parameter Lc-exp, calculated value of a-axis lattice parameter La-cal, calculated value of b-axis lattice parameter Lb-cal, calculated value of c-axis lattice parameter Lc-cal, error La of the calculated value of a-axis lattice parameter relative to the experimental value, error Lb of the calculated value of b-axis lattice parameter relative to the experimental value, error Lc of the calculated value of c-axis lattice parameter relative to the experimental value, calculated value WK-cal of work function, experimental value WK-exp of work function, error WK of the calculated value of work function relative to the experimental value, calculated value Eg-cal of band gap, calculated value Eg-exp of band gap, error Eg of the calculated value of band gap relative to the experimental value.

[0014] In a second aspect, the present invention provides a smart material simulation method based on the system of the first aspect, comprising the following steps: Step 1: User inputs data; Step 2: Determine whether the structure of the input data is reasonable. If so, proceed to step 3. Otherwise, prompt the user to pay attention to relevant matters or return to step 1. Step 3: Generate calculation files based on input data and generate corresponding task record library; Step 4: Read the task status in the task record library. If there are uncalculated tasks and there are free computer nodes, submit the calculation and change the task status to "Submitted". Step 5: Calculate the calculation file. If the calculation is successful, change the task status to "Completed" and proceed to Step 6. Otherwise, proceed to Step 7. Step 6: Read the task status in the task record library regularly, analyze the calculation results of the "submitted" tasks based on the experience knowledge base, draw a chart if the calculation results are reasonable, otherwise change the task status to "calculation error", adjust the corresponding calculation parameters according to the error type, and return to step 5; Step 7: Adjust the parameter settings based on the active learning algorithm and the experience knowledge base and return to step 5.

[0015] Furthermore, the user input data in step 1 includes data structure, tasks, experimental data, and goals.

[0016] Furthermore, the task records in the task record library include the following fields: task number ID, folder creation time creation_times, task type task_type, folder address folder_path, file name of the original POSCAR structure provided by the user poscar_original_name, task status status, task priority task_priority, task parameter setting method parameter_set_basis, ID number of the related calculation in the parameter setting basis related_calculation_ID, and experimental goal task_goals.

[0017] Furthermore, the empirical data in the empirical knowledge base includes the following fields: task number ID, material name / chemical formula Materials, elements contained in the material Elements, whether there is a surface configuration Surface, whether it is a semiconductor Semiconductor, GGA parameter setting method GGA, whether there is a van der Waals structure VDW, van der Waals structure correction method IVDW, van der Waals structure corrected atomic radius VDW_RADIUS, K point density denK in the KPOINTS file, truncation energy ENCUT, U value setting GGA+U in GGA+U, experimental value La-exp of a-axis lattice parameter, experimental value Lb-exp of b-axis lattice parameter , experimental value of c-axis lattice parameter Lc-exp, calculated value of a-axis lattice parameter La-cal, calculated value of b-axis lattice parameter Lb-cal, calculated value of c-axis lattice parameter Lc-cal, errorLa of the calculated value of a-axis lattice parameter relative to the experimental value, errorLb of the calculated value of b-axis lattice parameter relative to the experimental value, errorLc of the calculated value of c-axis lattice parameter relative to the experimental value, calculated value of work function WK-cal, experimental value of work function WK-exp, errorWK of the calculated value of work function relative to the experimental value, calculated value of band gap Eg-cal, calculated value of band gap Eg-exp, errorEg of the calculated value of band gap relative to the experimental value.

[0018] The remarkable effects of the present invention are: By integrating automated parameter optimization, historical experience retrieval, expert feedback mechanism, and machine learning model construction, the present invention has the following significant beneficial technical effects: 1. Improved accuracy and consistency of parameter settings: This invention uses a machine learning model to automatically fit experimental data or call on historical experience to achieve automated recommendation of simulation parameters, significantly reducing reliance on the personal experience of researchers, avoiding simulation result deviations caused by lack of experience or improper settings, and improving the consistency and reliability of the simulation.

[0019] 2. Significantly improved simulation efficiency and reduced manpower and time costs: It can automatically complete structural rationality judgment, parameter configuration, task execution and result judgment, greatly reducing manual operation and trial-and-error processes, improving the automation level of material simulation work, saving scientific researchers a lot of repetitive work time, and improving R&D efficiency.

[0020] 3. Build a sustainably evolving intelligent system: By feeding back experts' analytical experience on abnormal results to the system and automatically updating the model and database, a cycle of "experts teaching AI, and AI assisting experts" is implemented, allowing the system to continuously optimize and improve its intelligence level with use.

[0021] 4. Enhanced traceability and controllability of results: The simulation process and parameters are recorded and traceable. The simulation results output by the system not only include numerical predictions, but also provide control paths and analysis of influencing factors, which helps researchers understand the mechanism of physical property changes and provide a decision-making basis for material design.

[0022] 5. Realize the rapid replication and promotion of scientific research knowledge: Once a cutting-edge material or key technology is solved by scientific researchers and input into the system, it can realize rapid high-throughput simulation and law mining of a large number of similar structures, greatly accelerating the transformation and promotion of advanced technologies.

[0023] In summary, this invention not only solves the technical difficulties in traditional material simulation, such as "low efficiency, difficulty in inheriting experience, and unstable results", but also builds a material computing platform with autonomous learning and intelligent evolution capabilities, greatly promoting the intelligent process of new material design and discovery. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a principle block diagram of the system of the present invention; Figure 2 It is a method flow chart of the method of the present invention. DETAILED DESCRIPTION

[0025] The specific implementation manner and working principle of the present invention will be further described in detail below with reference to the accompanying drawings.

[0026] Since the traditional parameter setting method of the material simulation process mainly relies on the personal experience and trial and error method of researchers, there are problems such as difficulty in inheriting experience and low work efficiency. In addition, with the rapid accumulation of material data, how to effectively use big data to assist parameter setting has become an urgent problem to be solved. That is, the technical problems to be solved by this application are as follows: Solve the technical problem that the traditional manual method of setting calculation parameters is highly dependent on the personal experience of researchers, resulting in poor repeatability, accuracy and stability of parameter settings; Solve the technical problem that traditional manual simulation methods require repeated manual adjustment of parameters during the calculation process, resulting in low calculation efficiency and a large amount of manpower and time costs; Solve the technical problem that traditional methods lack systematic data analysis and utilization methods, and it is difficult to effectively use existing big data resources to intelligently optimize calculation parameters and thus improve simulation accuracy. Example

[0027] like Figure 1 As shown, a smart material simulation system that combines big data and expert experience includes: Task record library, used to store and record task records of material simulation process; An initial database, used to store initial data required for the material simulation process input by the user, the initial data including the initial structure, experimental data and calculation rules input by the user; An experience knowledge base, used to store experience data required for the material simulation process, wherein the experience data and the data in the task record have the same ID number; The execution module is used for the user to input tasks and goals, and to determine whether the data structure of the tasks and goals input by the user is reasonable. If reasonable, the initial data is called based on the input data to generate a calculation file, and the corresponding calculation task library is generated at the same time. Otherwise, the user is prompted to pay attention to relevant matters or modify the input data; it is used to read the task status in the calculation task library, submit the calculation if there are uncalculated tasks and there are free computer nodes, and modify the task status to "submitted"; it is also used to calculate the calculation file, if the calculation is successful, modify the task status to "completed" and read the task status in the calculation task library at regular intervals, analyze the calculation results of the "submitted" tasks based on the experience data, draw a chart if the calculation results are reasonable, otherwise modify the task status to "calculation error", and recalculate the calculation file after adjusting the corresponding calculation parameters according to the error type, otherwise recalculate the calculation file after adjusting the parameter settings based on the active learning algorithm and the experience data.

[0028] In some specific implementations, the task record (task_list table) in the task record library includes the following parts: ID: Task number, automatically generated by the program, similar to the ID card of the task, each task is unique.

[0029] creation_times: folder creation time, automatically generated by the system.

[0030] task_type: Task type, which can be structure optimize (default), static_calculation (static calculation), lattice optimize (lattice optimization), molecular_dynamics (molecular dynamics), etc. Different task types require different parameter settings. Recording the task type here will facilitate later research on the corresponding program, parameter setting, and processing of calculation results.

[0031] folder_path: The folder path. This will be modified later as the folder is moved. The folder path is an important basis for finding information such as the progress of the corresponding task and calculation results.

[0032] poscar_original_name: The file name of the original POSCAR structure provided by the user, for tracing back to the source.

[0033] status: Task status, which can be pending (pending), submitted (submitted), completed (computation completed), error_pending (computation error pending recomputation), error_pending_adjust (computation error pending manual adjustment), etc. It is used to record the status of the task to facilitate subsequent automated operations.

[0034] task_priority: Task priority, which can be A (priority), B (intermediate), or C (not urgent). The default is B. For more urgent tasks, set them to A (priority) to avoid delaying work. For less important tasks, set them to C (not urgent). Prioritizing tasks in this way can improve work efficiency.

[0035] parameter_set_basis: Task parameter setting method, which can be default, created_ML, or inherit. For unknown systems, the parameter settings can use common default parameter settings. For systems with a certain research foundation, the system can search for relevant research records in the empirical knowledge base and use machine learning algorithms to obtain the corresponding parameter settings. For some calculations with clear correlations, inherited parameter settings are required. For example, when calculating the adsorption energy of Cu atoms on the graphene surface, when calculating the energy of Cu atoms separately, the parameter settings of the previous calculation should be used. This ensures that the calculation results are comparable.

[0036] related_calculation_ID: The ID number of the related calculation in the parameter setting basis (parameter_set_basis), for easy traceability.

[0037] task_goals: Records experimental values ​​(such as band gap, magnetic moment, etc.) and the expected calculation error. If the calculated results do not match the experimental values, the system can use machine learning algorithms to adjust the parameter settings.

[0038] The task record library needs to be used in combination with corresponding programs and machine learning models to form the central command brain of the entire system and form an intelligent body, so as to achieve the effect of automatic adjustment of computing parameters and intelligent computing.

[0039] In some embodiments, the initial database user inputs initial data such as initial structure (POSCAR), task type, task target (what range of calculated value should be reached), experimental data (experiment_lattice.csv), and calculation rules (rules.csv).

[0040] In order to make the system more user-friendly when in use, the system is connected to large language models such as chatGPT and deepseek. Users can directly enter tasks and goals in the dialog box. The system automatically recognizes keywords and selects the corresponding program to generate a calculation folder, and at the same time feedback is given to the user to confirm the task details again before submitting the calculation. For example, the user enters in the large language window (voice call is also possible): Help me calculate the energy band of graphene. The system automatically calls the graphene structure in the database based on the keywords "graphene, energy band" and feeds back the relevant parameter settings to the user. After confirmation, it automatically generates KPOINTS, INCAR, POTCAR calculation files and folders, and records the tasks (three: including structural optimization, static calculation, and energy band calculation) in the task_list table, waiting for submission of the calculation.

[0041] In addition to the initial data mentioned above, other initial data input is divided into two categories: parameter fitting and property testing, as follows: Parameter fitting class (file name distinction): poscar_original_name, type 1 (property 1), 5% (error requirement), 4.5 (experimental measurement value). www.xxxxx.com (source of experimental data, which can be a website or a literature volume, issue, or page number); Type 2 (property 2), 5% (error requirement), 4.5 (experimental measurement value), www.xxxxx.com; Type 3 (property 3), 5% (error requirement), 4.5 (experimental measurement value), www.xxxxx.com; Type 4 (property 4), 5% (error requirement), 4.5 (experimental measurement value), www.xxxxx.com.

[0042] Property test class (file name distinction): poscar_original_name, task_type [possibly multiple], associated calculation [possibly multiple]: The initial setup should be as simple as possible, preferably automated.

[0043] Later, you can directly tell deepseek to calculate the electronic state density for XXX.

[0044] In this embodiment, the format of the calculation rules (rules.csv) is as follows: Material (material name), rule_type (rule type), param_key (parameter in the corresponding INCAR file), param_value (parameter setting value), note (notes), for example: NiO, GGA+U, U_Ni, 5.0~7.0, NiO is a strongly correlated electron system, and U is recommended to be 5-7; Fe3O4,GGA+U,U_Fe,4.0~5.5,Fe3O4 contains Fe2+ / Fe3+, recommended U range is 4.0~5.5; MoS2, VDW, IVDW, 12, MoS2 are layered materials and it is recommended to add van der Waals interaction.

[0045] In some specific implementations, the content of the experience data in the experience knowledge base (vasp_opt_db.sqlite) is as follows: ID: task number; Materials: material name / chemical formula; Elements: the elements contained in the material; Surface: whether there is a surface configuration; Semiconductor: whether it is a semiconductor; GGA: How to set GGA parameters; VDW: presence or absence of van der Waals structure; IVDW: van der Waals correction method; VDW_RADIUS: van der Waals corrected atomic radius; denK: K point density in KPOINTS file; ENCUT: cutoff energy; GGA+U: U value setting in GGA+U; La-exp: experimental value of a-axis lattice parameter; Lb-exp: experimental value of the b-axis lattice parameter; Lc-exp: experimental value of c-axis lattice parameter; La-cal: calculated value of a-axis lattice parameter; Lb-cal: calculated value of b-axis lattice parameter; Lc-cal: calculated value of c-axis lattice parameter; errorLa: the error of the calculated value of the a-axis lattice parameter relative to the experimental value; errorLb: the error of the calculated value of the b-axis lattice parameter relative to the experimental value; errorLc: the error of the calculated value of the c-axis lattice parameter relative to the experimental value; WK-cal: calculated value of work function; WK-exp: experimental value of work function; errorWK: the error of the calculated value of the work function relative to the experimental value; Eg-cal: calculated value of band gap; Eg-exp: calculated value of band gap; errorEg: The error of the calculated band gap relative to the experimental value.

[0046] Regarding the above content, the setting rules of the experience knowledge base in this embodiment are as follows: 1) It must have an ID number, which cannot be changed, can only be expanded, and must be consistent with the ID in task_list; 2) The data content, especially experimental values, can be supplemented and expanded according to the needs of later research; 3) Calculations that require comparison with experimental data can be incorporated into the vasp_opt_db.sqlite database.

[0047] 4) For the calculation of physical and chemical properties in the task_list table (without experimental data comparison), it is necessary to apply the parameter design in the vasp_db.sqlite database and indicate the ID number of the referenced database calculation.

[0048] 5) For experimental data, the source of experimental data, literature, etc. must be provided and recorded in this database.

[0049] 6) For calculations that do not match experimental values, their corresponding results can also be entered into this database for easy learning.

[0050] The empirical knowledge base, compiled from past calculation results, consolidates the computational experience across all systems, facilitating direct learning and incorporation of relevant parameter setting experience for subsequent calculations of similar systems. The introduction of this database greatly facilitates automated parameter setting. Compared to traditional research approaches, which typically require beginners to learn for over six months to develop reasonable computational parameter settings for a system, this approach significantly improves work efficiency and lowers the barrier to entry for beginners.

[0051] When using the system, please pay attention to the following details: 1. The task_list file needs to be modified when the system generates a calculation folder, submits a calculation, completes a calculation, encounters a calculation error, abandons a calculation, deletes or modifies a folder (programmed processing, not manual).

[0052] 2. Associated calculations (arranged in parallel, just mark the ID of the previous associated calculation), such as: structure optimization electronic state density calculation, adsorption, transition state, etc.

[0053] 3. The experimental data should indicate the source, literature, or website link, etc.

[0054] 4. For bandgap calculations, you need to link several calculations. When there is a discrepancy between the results and the experimental results, you need to process these results simultaneously, and the parameter settings for these results must be consistent. Parameter inheritance mechanism: Design an intelligent script so that when a new calculation task is created, it automatically inherits the parameter settings from the associated previous task.

[0055] 5. Regularly reduce the size and relocate folders.

[0056] 6. The p602_folder_creator.py program generates a calculation folder based on POSCAR and modifies the task_list file.

[0057] 7. Same structure, different parameter settings, same POSCAR name, folder names need to be added with subscripts v1, v2, v3, and different ID numbers.

[0058] The case shows: Example 1: Calculate the adsorption energy of a water molecule using this structure and draw a density of electronic states diagram for each adsorbed state.

[0059] Example 2: Fit the calculated parameters for this structure, the work function is 5.1 eV, and the error is less than 5%.

[0060] Example 3: Fit the calculated parameters for this structure. The work function is 5.1 eV. The error is uncertain, the lower the better.

[0061] The scripts and functions involved in the execution module are as follows (partial): p100_Mpdown.py: Download POSCAR of the corresponding crystal from the materials project website.

[0062] p101_atomremove.py: Delete some atoms in POSCAR.

[0063] p102_CONTCAR2POSCAR.py: Convert CONTCAR file to POSCAR p105_xsd2POSCAR.py: Convert XSD file to POSCAR p106_atomadsite.py: Constructs various adsorption configurations of certain atoms / molecules on the substrate surface.

[0064] p107_primitivecell.py: Converts POSCAR into a primitive cell with the fewest atoms, which helps reduce the amount of computation.

[0065] p108_suppercell.py: Converts POSCAR into a supercell and computes various special cases.

[0066] p109_nearAtom.py: Finds the nearest neighbor atoms, coordinated atoms, and bond lengths to facilitate analysis of atomic characteristics, such as surface adsorption activity and magnetic moment.

[0067] p110_AFM-2d.py: Finds possible antiferromagnetic configurations of the corresponding two-dimensional POSCAR structure.

[0068] p110_AFM-3d.py: Searches for possible antiferromagnetic configurations of the corresponding three-dimensional POSCAR structure.

[0069] p200_VASPstropt.py: Calculates the lattice structure of the structure, including lattice constants, bond angles, etc.

[0070] p204_Tcread.py: Read the magnetic ordering temperature during Monte Carlo simulation.

[0071] p205_Jmatrix.py: Calculates the spin exchange interaction energy of magnetic materials.

[0072] p207_charge.py: Automatically generates folders for calculating charge density and INCAR parameters, using inheritance mode.

[0073] p208_adsorption.py: Automatically generates folders for calculating surface atomic or molecular adsorption energies and INCAR parameters, using inheritance mode.

[0074] p209_dos.py: Automatically generates folders for calculating electronic state density and INCAR parameters, using inheritance mode.

[0075] p301_POTCARcreate.py: Automatically generate POTCAR files based on elements.

[0076] p302_GGAU-modify: Automatically adjusts the U-value settings of the GGA+U method in the INCAR file according to the element.

[0077] p303_KPOINTScreate.py: Automatically generates and calculates KPOINTS files based on K-point density requirements.

[0078] p304_INCARmodify: Modify INCAR file as required.

[0079] p305_charge: Read the charge distribution of the calculation result file OUTCAR system.

[0080] p305_MAGMOM: Read the magnetic moment of the OUTCAR system from the calculation result file.

[0081] p306_bondlength: Read the bond length around an atom in the calculation result file.

[0082] p401_vaspout: Determines whether the calculation is completed successfully.

[0083] p408_property-dos: Read the band gap of the calculated result system.

[0084] p409_atompdos: Extract the projected electronic density of states of an atom.

[0085] p412_dosplot: Plot the projected electronic density of states of an atom.

[0086] p413_doscenter: Calculate the D-band center.

[0087] p416_workfuncation: Calculates the work function of the system.

[0088] p501_removeinvalidFolder: Delete invalid calculation paths and folders.

[0089] p502_secondsearch: Search for related folders.

[0090] p601_poscar_checker.py: Checks whether the POSCAR structure is reasonable.

[0091] p602_folder_creator.py: Generates a calculation folder and writes the path in task_list.

[0092] p604_db_manager.py: Integrates all calculations and writes experience to the vasp_opt_db.sqlite database.

[0093] p605_import_experiment.py: Imports experimental data into the vasp_opt_db.sqlite database.

[0094] p606_rules_import.py: Import widely recognized and mature calculation rules into the vasp_opt_db.sqlite database.

[0095] p607_parmselect.py: Selects appropriate parameter settings in the vasp_opt_db.sqlite database based on task characteristics.

[0096] p608_ml_model.py: Uses a computational learning algorithm to select appropriate parameter settings from the vasp_opt_db.sqlite database based on the task characteristics. Parameter search is performed using Bayesian optimization or genetic algorithms.

[0097] p609_feature_engineer.py: Feature extraction for the task.

[0098] p610_generate_incar.py: Automatically generate INCAR files.

[0099] p611_task_submit.py: Intelligently reads the task_list file, finds available nodes for pending tasks, and submits the calculations.

[0100] These scripts, combined with the effective use of task libraries and experience databases, realize the automation and intelligence of material simulation. They have the following innovations: 1) Automatically generate calculation files and folders according to different systems and task types.

[0101] 2) Automatically submit calculations based on task status.

[0102] 3) Automatically determine the next step based on the calculation results.

[0103] 4) User-friendly interaction. Traditional methods require learning Linux system operations and mastering various parameter setting techniques. This approach directly connects to a large language model, providing a user-friendly interface through chat, and submitting computing tasks.

[0104] 5) Record experience and constantly look for suitable parameter designs. Example

[0105] like Figure 2 As shown in the figure, a smart material simulation method combining big data and expert experience is shown in the figure. The specific steps are as follows: Step 1: The user enters data, including structure, tasks, experimental data, etc. This can be input directly in the dialog box in a chat-like manner, making it user-friendly. This is grafted with the current mainstream international language models, such as ChatGPt, Deepseek, Wenxin Yiyan, iFlytek Spark, Doubao, etc. Step 2: Determine whether the structure of the input data is reasonable based on the system's built-in initial database. If so, proceed to Step 3. Otherwise, prompt the user to pay attention to relevant matters or return to Step 1. Step 3: Based on the system's built-in initial database and the large language model, a calculation file is generated based on the input data. At the same time, a corresponding task record library (task_list) is generated to schedule the calculation status of all tasks. Here, details such as the structure, task type, address, status, and parameter settings of each task are recorded in detail; During the implementation process, we primarily used mainstream international material simulation software (VASP, Quantum ESPRESSO, etc.) and high-performance computer servers running on Linux systems to perform scientific calculations on these materials. Therefore, the generated files were based on formats that conform to these calculations.

[0106] Step 4: Read the task status in the task record library. If there are uncalculated tasks and there are free computer nodes, submit the calculation and change the task status to "Submitted". Step 5: Calculate the calculation file. If the calculation is successful, change the task status to "Completed" and proceed to Step 6. Otherwise, proceed to Step 7. Step 6: Read the task status in the task record library regularly, analyze the calculation results of the "submitted" tasks based on the experience knowledge base, draw a chart if the calculation results are reasonable, otherwise change the task status to "calculation error", adjust the corresponding calculation parameters according to the error type, and return to step 5; Step 7: Adjust the parameter settings based on the active learning algorithm and the experience knowledge base and return to step 5.

[0107] The following is a specific example of the method described in this embodiment: S1. The user enters "Help me calculate the single-atom adsorption energy of Cu on the surface of monolayer graphene" in the chat interface; S2: Based on the prompts "graphene," "Cu single atom," and "adsorption energy," the system matches the materials involved to graphene and Cu atoms, and the task is to calculate the adsorption energy. S3. The system analyzes the initial database and provides feedback: "Task: Calculate adsorption energy; Substrate: Monolayer graphene; Adsorbate: Cu single atom. Are there any other calculation rules or task supplements (such as U value setting)?" S4, user replies “none”; S5. The system calls the monolayer graphene structure and constructs a 2×2 supercell (default). It calls the script p106_atomadsite.py to search for high-symmetry adsorption sites on its surface and place Cu atoms at various high-symmetry positions. The system then informs the user, "We constructed a 2×2 monolayer graphene supercell and found three high-symmetry adsorption sites for Cu atoms: bridge, hollow, and top. Please confirm whether this is acceptable." S6. The user replies “yes”; S7. The system calls the p208_adsorption.py program to automatically generate various possible high-symmetry adsorption sites. Construct three POSCAR models: POSCAR_GCutop, POSCAR_Gcubridge, POSCAR_Gcuhollow, POSCAR_Gsingle, and POSCAR_Cusingle. S8. The system calls the p602_folder_creator.py program to generate a calculation folder based on the above structure. It then calls the vasp_opt_db.sqlite database to find the appropriate C-Cu bond length and set parameters using the machine learning algorithm. It also uses the parameter inheritance mechanism to set INCAR and creates the corresponding tasks in task_list. S9. The p611_task_submit.py program regularly checks the tasks waiting for calculation in the task record library task_list. If it finds free nodes and tasks waiting for calculation, it automatically submits the calculation. S10. The p612_task_result.py program periodically checks the tasks in task_list that have been submitted for calculation. If it finds a completed task, it modifies the task status to prompt the user that the calculation has been completed.

[0108] S11. The user issues the command: "Calculate adsorption energy." The system calls the p613_adsorptionresult.py program, obtains the corresponding calculation result according to the set calculation formula, and reports it to the user.

[0109] The present invention also considers the following alternative technical solutions, which, although different in technical principles and implementation paths, can achieve the same or similar invention objectives: Alternative parameter optimization methods: In addition to the machine learning models used in this invention (such as random forest, support vector machine, neural network), the parameter optimization module can also use intelligent optimization algorithms such as Bayesian optimization, genetic algorithm, particle swarm optimization, etc. to realize the automatic parameter recommendation function, so as to improve parameter accuracy and simulation efficiency.

[0110] Alternative structural assessment mechanisms: In terms of structural rationality assessment, in addition to using standard databases and structural analysis software for initial judgment, deep learning models can be combined to predict structural stability, or molecular dynamics simulation can be used as a method for initial structural screening, thereby enhancing the system's adaptability to complex structures.

[0111] Alternative data sources and access methods: In addition to accessing internally built historical experience databases and experimental data, the system can also be designed to support access to external databases (such as Materials Project, OQMD, AFLOW, etc.), and implement dynamic data reading through APIs to assist in model fitting and parameter generation.

[0112] Alternative expert intervention mechanisms: In addition to setting up manual review and marking interfaces, it is also possible to automate some expert judgment processes by establishing a rule base or expert system, presetting common problems and expert processing strategies as knowledge graphs or logical reasoning rules.

[0113] Alternative system deployment methods: In addition to being deployed on local servers, the system can also achieve remote access and distributed task scheduling through cloud computing platforms (such as based on HPC clusters or commercial cloud services such as AWS and Aliyun), thereby adapting to different user scales and computing power requirements.

[0114] Alternative user interaction methods: In addition to using a natural language interface, you can also use a graphical user interface (GUI), command line interface (CLI), or mobile app for human-computer interaction to achieve functions such as structure input, task management, and result visualization.

[0115] Alternative Software Integration Methods: This invention can be used in conjunction with existing first-principles software such as VASP and Quantum ESPRESSO. Alternatively, the simulation module can be replaced with a proprietary semi-empirical model (e.g., TB, DFTB) to reduce computing resource requirements and improve high-throughput computing efficiency. These alternatives can complement and be used in combination with the primary solution to further enhance the system's versatility, stability, and intelligence, and constitute the integral technical protection scope of this invention.

[0116] In summary, the present invention not only solves the technical difficulties of "low efficiency, difficulty in inheriting experience, and unstable results" in traditional material simulation by integrating automated parameter optimization, historical experience retrieval, expert feedback mechanism and machine learning model construction, but also constructs a material calculation platform with autonomous learning and intelligent evolution capabilities, which greatly promotes the intelligent process of new material design and discovery, improves the accuracy and consistency of parameter settings, significantly improves simulation efficiency, reduces manpower and time costs, enhances the traceability and controllability of simulation results, and realizes the rapid replication and promotion of scientific research knowledge.

[0117] The technical solution provided by the present invention is introduced in detail above. Specific examples are used herein to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. It should be pointed out that for ordinary technicians in this technical field, without departing from the principles of the present invention, the present invention can also be improved and modified in a number of ways, and these improvements and modifications also fall within the scope of protection of the claims of the present invention.

Claims

1. An intelligent material simulation system that combines big data and expert experience, characterized by: include: Task record library, used to store and record task records of material simulation process; Initial database, used to store the initial data required for the material simulation process input by the user; An experience knowledge base, used to store experience data required for the material simulation process, wherein the experience data and the data in the task record have the same ID number; The execution module is used for the user to input tasks and goals, and to determine whether the data structure of the tasks and goals input by the user is reasonable. If reasonable, the initial data is called based on the input data to generate a calculation file, and the corresponding calculation task library is generated at the same time. Otherwise, the user is prompted to pay attention to relevant matters or modify the input data; it is used to read the task status in the calculation task library, submit the calculation if there are uncalculated tasks and there are free computer nodes, and modify the task status to "submitted"; it is also used to calculate the calculation file, if the calculation is successful, modify the task status to "completed" and read the task status in the calculation task library at regular intervals, analyze the calculation results of the "submitted" tasks based on the experience data, draw a chart if the calculation results are reasonable, otherwise modify the task status to "calculation error", and adjust the corresponding calculation parameters according to the error type and recalculate the calculation file, otherwise adjust the parameter settings based on the active learning algorithm and the experience data and recalculate the calculation file.

2. The intelligent material simulation system combining big data and expert experience according to claim 1, characterized in that: The task record includes the following fields: task number ID, folder creation time creation_times, task type task_type, folder address folder_path, file name of the original POSCAR structure provided by the user poscar_original_name, task status status, task priority task_priority, task parameter setting method parameter_set_basis, ID number of the related calculation in the parameter setting basis related_calculation_ID, and experimental goal task_goals.

3. The intelligent material simulation system combining big data and expert experience according to claim 2, characterized in that: The status field includes five states: pending, submitted, completed, error pending, error_pending, and error_pending_adjust.

4. The intelligent material simulation system combining big data and expert experience according to claim 1, characterized in that: The initial data includes the initial structure POSCAR, task type, task target, experimental data experiment_lattice.csv, and calculation rules rules.csv.

5. The intelligent material simulation system combining big data and expert experience according to claim 4 is characterized in that: The format of the calculation rules rules.csv is as follows: Material name Material, rule type rule_type, corresponding parameter param_key in INCAR file, parameter setting value param_value, and note note.

6. The intelligent material simulation system combining big data and expert experience according to claim 1, characterized in that: The empirical data includes the following fields: task number ID, material name / chemical formula Materials, elements contained in the material Elements, whether there is a surface configuration Surface, whether it is a semiconductor Semiconductor, GGA parameter setting method GGA, whether there is a van der Waals structure VDW, van der Waals structure correction method IVDW, van der Waals structure correction atomic radius VDW_RADIUS, K point density denK in KPOINTS file, truncation energy ENCUT, U value setting GGA+U in GGA+U, experimental value La-exp of a-axis lattice parameter, experimental value Lb-exp of b-axis lattice parameter, c-axis lattice parameter Experimental value Lc-exp of the number, calculated value La-cal of the a-axis lattice parameter, calculated value Lb-cal of the b-axis lattice parameter, calculated value Lc-cal of the c-axis lattice parameter, error La of the calculated value of the a-axis lattice parameter relative to the experimental value, error Lb of the calculated value of the b-axis lattice parameter relative to the experimental value, error Lc of the calculated value of the c-axis lattice parameter relative to the experimental value, calculated value WK-cal of the work function, experimental value WK-exp of the work function, error WK of the calculated value of the work function relative to the experimental value, calculated value Eg-cal of the band gap, calculated value Eg-exp of the band gap, error Eg of the calculated value of the band gap relative to the experimental value.

7. A smart material simulation method based on the system according to any one of claims 1 to 6, characterized in that: The steps include: Step 1: User inputs data; Step 2: Determine whether the structure of the input data is reasonable. If so, proceed to step 3. Otherwise, prompt the user to pay attention to relevant matters or return to step 1. Step 3: Generate calculation files based on input data and generate corresponding task record library; Step 4: Read the task status in the task record library. If there are uncalculated tasks and there are free computer nodes, submit the calculation and change the task status to "Submitted". Step 5: Calculate the calculation file. If the calculation is successful, change the task status to "Completed" and proceed to Step 6. Otherwise, proceed to Step 7. Step 6: Read the task status in the task record library regularly, analyze the calculation results of the "submitted" tasks based on the experience knowledge base, and draw a chart if the calculation results are reasonable. Otherwise, change the task status to "calculation error", adjust the corresponding calculation parameters according to the error type, and return to step 5; Step 7: Adjust the parameter settings based on the active learning algorithm and the experience knowledge base and return to step 5.

8. The intelligent material simulation system combining big data and expert experience according to claim 7, characterized in that: In step 1, the user input data includes data structure, tasks, experimental data, and goals.

9. The intelligent material simulation system combining big data and expert experience according to claim 7, characterized in that: The task records in the task record library include the following fields: task number ID, folder creation time creation_times, task type task_type, folder address folder_path, file name of the original POSCAR structure provided by the user poscar_original_name, task status status, task priority level task_priority, task parameter setting method parameter_set_basis, ID number of the related calculation in the parameter setting basis related_calculation_ID, and experimental goal task_goals.

10. The intelligent material simulation system combining big data and expert experience according to claim 7, characterized in that: The experience data in the experience knowledge base include the following fields: task number ID, material name / chemical formula Materials, elements contained in the material Elements, whether there is a surface configuration Surface, whether it is a semiconductor Semiconductor, GGA parameter setting method GGA, whether there is a van der Waals structure VDW, van der Waals structure correction method IVDW, van der Waals structure correction atomic radius VDW_RADIUS, K point density denK in KPOINTS file, truncation energy ENCUT, U value setting GGA+U in GGA+U, experimental value La-exp of a-axis lattice parameter, experimental value Lb-exp of b-axis lattice parameter, c-axis Experimental value of lattice parameter Lc-exp, calculated value of a-axis lattice parameter La-cal, calculated value of b-axis lattice parameter Lb-cal, calculated value of c-axis lattice parameter Lc-cal, error La of the calculated value of a-axis lattice parameter relative to the experimental value, error Lb of the calculated value of b-axis lattice parameter relative to the experimental value, error Lc of the calculated value of c-axis lattice parameter relative to the experimental value, calculated value WK-cal of work function, experimental value WK-exp of work function, error WK of the calculated value of work function relative to the experimental value, calculated value Eg-cal of band gap, calculated value Eg-exp of band gap, error Eg of the calculated value of band gap relative to the experimental value.

Images