Ionic liquid intelligent design method for extreme high-temperature lubrication requirement

Abstract

The invention discloses an intelligent ionic liquid design method for extreme high-temperature lubrication requirements. The method comprises the following steps: constructing an ionic liquid lubrication performance data set oriented to extreme high temperature, and setting a data confidence interval to clean abnormal data; carrying out ion structure feature coding by adopting an SMILES format and extracting a structure descriptor, and dividing a data set based on stratified sampling of a temperature interval; determining a machine learning model and defining a hyper-parameter space of the machine learning model, applying the training set to machine learning, and reducing a hyper-parameter vector value range at the same time; inputting the test set into the trained model to perform precision evaluation, and adopting an improved bird flock algorithm to improve the retrieval quality of the machine learning model on data in the data set, so as to further optimize the model precision; according to extreme high-temperature environment parameters and lubricating surface characteristics provided by a user, the model screens out a plurality of ionic liquids which can stably play a role in lubrication under corresponding working conditions. According to the invention, the technical problem of high efficiency and high adaptability design of the ionic liquid lubricant in an extreme high-temperature environment is solved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an intelligent design method for ionic liquids oriented to extreme high-temperature lubrication requirements. Background Art

[0002] With the rapid development of technologies in high-end equipment manufacturing, aerospace, and nuclear clean energy, critical equipment often faces extremely high temperatures, placing higher demands on the lubrication and thermal stability of lubricants. Traditional solid lubricants, when exposed to long-term high temperatures, can cause irreparable wear on mechanical workpieces. Molecular liquid lubricants, when exposed to high temperatures for extended periods, are susceptible to thermal decomposition, leading to lubrication failure and severely impacting the stable operation and service life of equipment.

[0003] Ionic liquids, room-temperature molten salts composed of anions and cations, are ideal lubricants for extreme operating conditions due to their excellent lubrication properties, load-bearing capacity, wide temperature range, and unique designability. Especially in extreme high-temperature lubrication scenarios, ionic liquids exhibit excellent properties such as strong thermal stability, long-term stable lubrication performance, and low friction coefficient, offering broad application potential for extremely high-temperature mechanical operating conditions.

[0004] However, despite the significant performance potential of ionic liquids, rapid screening for ionic liquids that combine strong thermal stability with excellent lubrication properties remains a major challenge in current research and development. Specifically, current research and development faces the following bottlenecks: 1) Reliance on iterative trial-and-error methods, resulting in low R&D efficiency and high costs: Currently, the development of ionic liquid lubricants resistant to extreme temperatures relies primarily on the traditional "Edison-style" trial-and-error method. Researchers must invest considerable time and resources to individually formulate ionic liquids of varying compositions and conduct tedious lubrication performance tests under various extreme high-temperature conditions. This process is not only time-consuming and resource-intensive, but also faces extremely limited experimental coverage with tens of thousands of potential combinations, making comprehensive and efficient screening difficult. 2) The complex structure-property relationship of ionic liquids limits the effectiveness of traditional prediction methods: Due to the highly tunable molecular structure of ionic liquids, their performance is significantly influenced by their microstructure and exhibits a high degree of nonlinearity. Current performance prediction methods based on rule-based derivation and molecular simulations struggle to fully and accurately describe their lubrication behavior under extreme temperatures, lacking an efficient and systematic performance prediction and screening mechanism. 3) Lack of a systematic intelligent design and screening platform: Existing research mostly focuses on local performance prediction or experimental verification, and lacks an intelligent generation path that integrates molecular structure encoding, performance prediction and combinatorial optimization. It is difficult to support the targeted matching and precise recommendation of lubrication requirements under complex working conditions such as extremely high temperatures.

[0005] Therefore, an intelligent design method integrating molecular structure expression, deep learning modeling, performance prediction, and formulation generation and optimization is urgently needed to rapidly screen and predict ionic liquid lubricant formulations with excellent performance under extreme high-temperature conditions. This method can realize the intelligent design and optimization of ionic liquid lubricants from the perspective of "data-structure-performance". This method can significantly improve the R&D efficiency and reduce R&D costs of ionic liquid lubricants, and promote their engineering applications in key fields such as aerospace, military industry, and extreme manufacturing. Summary of the Invention

[0006] The purpose of the present invention is to provide a method for intelligently generating high-temperature resistant ionic liquids based on deep learning and intelligent optimization algorithms, in order to address the problems in the existing technology of relying on traditional experimental trial and error methods for the development of extreme high-temperature ionic liquid lubricants, such as low R&D efficiency and high cost, and inaccurate structure-performance relationship modeling and insufficient multi-objective optimization capabilities in performance prediction methods based on rule deduction or simple simulation. The method realizes the intelligence of the entire process from molecular structure and performance prediction to formula optimization, and efficiently screens ionic liquid formulas with both strong thermal stability and good lubrication performance, significantly improving the R&D efficiency of lubricants in extreme high-temperature environments, and promoting the application of ionic liquid lubricants under extreme working conditions in key fields such as aerospace, military industry and extreme manufacturing.

[0007] To achieve the above object, the technical solution adopted by the present invention is as follows:

[0008] An intelligent design method for ionic liquids for extreme high-temperature lubrication requirements includes the following steps:

[0009] 1) Initial Dataset Construction: Using the structured query language SQL, we searched the data platform and screened for data on the lubrication performance of ionic liquids under extreme high temperatures. We then systematically integrated the data into a four-category definition model: ion combination, environmental parameters, surface characteristics, and lubrication performance, to establish the initial dataset.

[0010] 2) Supplementation of missing entries in the dataset and cleaning of abnormal data: Further search for literature related to the lubrication performance of ionic liquids. The literature was uploaded to the natural language processing model BERT in .pdf format to identify and extract the ionic liquid lubrication performance data in the literature and supplement the blank data. Based on the calculation results of molecular dynamics simulation, a theoretical threshold for data confidence was set, and data points that deviated from the confidence interval were automatically cleaned to obtain an optimized dataset.

[0011] 3) Model training: The data processed in step 2) is pre-processed with ionic structure feature engineering to stack the lubrication performance data of the ionic liquid into 20 dimensions for ease of machine learning. The data set is stratified and sampled according to temperature ranges, and proportionally divided into training and test sets. The machine learning model is determined and its hyperparameter space is defined. The training set is input into the machine learning model for training, and the optimal value of the hyperparameter vector is further converged based on grid search to improve model accuracy.

[0012] 4) Model accuracy evaluation and optimization: Input the test set into the model obtained in step 3) and use R 2 The accuracy of the ionic liquid generated by the model was evaluated with RMAE. The improved bird flock algorithm (IBSA) was combined to perform a global search of the training set data points to ensure that the learning process converged to data points with higher accuracy. This optimized the model's description of the structure-activity relationship of the lubrication performance of ionic liquids, further improving the model's accuracy and resulting in a generative model capable of generating ionic liquid lubricants for extremely high-temperature lubrication requirements.

[0013] 5) Formulation Generation: The specified extreme high-temperature environmental parameters and lubrication surface characteristics are input into the generative model obtained after optimization in step 4). The model will determine the corresponding hyperparameter vector based on the specified temperature range and generate 10-20 candidate ionic liquid formulations for molecular dynamics simulation. The model calculates their mass loss at a heating rate of 10K / min and their friction coefficient under a normal stress of 1.0-2.5GPa, and further screens out ionic liquids that can meet the lubrication requirements of the specified extreme high-temperature working conditions.

[0014] Furthermore, in step 1), the decomposition temperature T when the mass loss is 5% in the thermogravimetric analysis TGA is screened. d The ionic liquid lubrication data showed a kinematic viscosity range of 20-40 cP measured by a rotational rheometer at 1450±50 rpm in a nitrogen atmosphere at test temperatures of ≥210°C and 240°C, and a friction coefficient μ<0.1 measured at a contact pressure of 1.0-2.5 GPa in a four-ball test.

[0015] Furthermore, in steps 1) and 2), the data sources of the data set include the NIST ILThermo and Web of Science data platforms, SCI-indexed literature, and ASTM standard experimental data; each data is systematically organized according to the class definition model of the four-tuple lubrication performance described by the anion and cation type, temperature and load environment parameters, roughness and charge density surface characteristics, viscosity and friction coefficient.

[0016] Furthermore, in step 2), the natural language processing model BERT is used to identify and extract the semantics of ionic liquids in the literature, so that the lubrication performance can be automatically included in the data set from the literature; for the ionic structural formulas and 3D model image data in the literature, the ionic structural features are mapped and extracted based on the graph neural network GNN; the obtained ionic structure data are uniformly encoded and stored in the SMILES format.

[0017] Furthermore, in step 3), the SMILES-encoded ionic structural features were converted into 208 feature descriptors using the RDKit 2022.09 toolkit. Based on the one-hot encoding and Pearson correlation coefficient heat map shared or not shared by the anion and cation feature descriptors, the 208 feature descriptors were further stacked into 20 dimensions that could accurately reflect the differences in the ionic structural features of the data set.

[0018] Furthermore, in step 3), the data in the dataset are stratified and sampled according to a temperature interval of 10K to ensure that the data can be evenly divided into the training set and the test set according to the temperature, thereby reducing the underfitting or overfitting of the model caused by the division difference; the training set and the test set are divided in a ratio of 7:3.

[0019] Furthermore, in step 3), the dimensions of the constructed hyperparameter space include depth, number of iterations, learning rate, number of early stopping rounds, and L2 regularization coefficient; a conditional generative machine learning model that integrates the CatBoost framework and the VAE encoder is used for training, and grid search and 10CV cross-validation methods are used to determine the value range of a better hyperparameter vector, narrowing the value range of the hyperparameter vector to improve the accuracy of the model.

[0020] Further, in step 4), by R 2 The accuracy of the ionic liquid generated by the quantitative model was ≥0.95 and RMSE ≤0.02; the bird flock algorithm was used to mark each data point one by one and share the data to exclude a few abnormal data points.

[0021] Furthermore, in step 5), the generated ionic liquid will be filtered through RDKit syntax verification, ΔG<0 thermodynamic criterion, and CatBoost performance prediction in sequence, and 10-20 ionic liquid candidate solutions will be output, and the output results will be encoded in SMILES.

[0022] Compared with the prior art, the present invention has the following significant advantages:

[0023] (1) Achieve efficient screening of extreme high-temperature ionic liquid lubricants: Using AI's high computing power to learn the structure-activity relationship of ionic liquid lubrication properties instead of traditional trial-and-error experiments can shorten the R&D cycle by 75%;

[0024] (2) Multimodal data fusion improves design reliability: The domain-specific BERT model is used to accurately extract literature performance parameters, and GNN is used to analyze ionic structures and 3D models, which are uniformly converted into SMILES encoding and storage to solve the problem of heterogeneous data integration;

[0025] (3) Generative model-driven active design: The CatBoost machine learning model is integrated with the VAE encoder to build a conditional generative model, which automatically rolls back and generates candidate recipes based on the input working conditions. The improved bird flock algorithm (IBSA) is introduced to globally optimize data point selection, allowing the model to converge to a higher accuracy area, breaking through the accuracy bottleneck of traditional models.

[0026] (4) Triple filtering mechanism ensures the feasibility of the formula: through RDKit syntax verification, ΔG<0 thermodynamic criterion, and CatBoost performance prediction, invalid solutions are triple filtered. The output SMILES code can be directly converted into a structural formula and 3D model to accelerate experimental verification;

[0027] (5) Good scalability: The method framework is universal and can be adapted to more complex material screening tasks by expanding descriptors, improving model structure, or integrating other lubrication performance indicators. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 is a flow chart of this method.

[0029] Figure 2 It is an automatic extraction platform for ionic liquid lubrication performance data from literature based on the BERT model.

[0030] Figure 3 It generates the ionic liquid SMILES output by the model and displays it in the form of images in chemical software. DETAILED DESCRIPTION

[0031] A specific embodiment of the present invention is: an intelligent design method for ionic liquids for extreme high temperature lubrication needs, comprising:

[0032] 1) Initial dataset construction: SQL was used to search the NIST ILThermo and Web of Science data platforms to screen for data that meet the decomposition temperature T when the mass loss is 5% in thermogravimetric analysis (TGA). d Lubrication data for ionic liquids with kinematic viscosities ranging from 20 to 40 cP, measured by rotational rheometry at test temperatures of ≥210°C and 240°C, and friction coefficients of μ <0.1 at contact pressures of 1.0 to 2.5 GPa in a four-ball test were collected. This initial dataset was constructed by integrating data based on four categories: anionic and cationic structure, environmental parameters, surface characteristics, and lubrication performance.

[0033] 2) Supplementation of missing entries in the dataset and cleaning of abnormal data: For the missing entries in the initial dataset, we searched from the SCI-indexed literature, and extracted lubrication performance parameters from the literature using the BERT model to supplement the blank data. For the image data in the literature, we used the graph neural network (GNN) to map the ionic structure features and converted them into the SMILES format for encoding and storage. Based on the calculation results of molecular dynamics simulation, we set the theoretical threshold of data confidence, and automatically cleaned the data points that deviated from the confidence interval to obtain the optimized dataset.

[0034] 3) Model training: The data was preprocessed with ion feature engineering, and the SMILES-encoded ion structure features were converted into 208 feature descriptors using the RDKit 2022.09 toolkit. Based on the one-hot encoding and Pearson correlation coefficient heat map of the common or non-common anion and cation feature descriptors, the 208 feature descriptors were further stacked into 20 dimensions that can accurately reflect the differences in the ion structure characteristics of the dataset; the dataset was divided into training and test sets in a 7:3 ratio based on stratified sampling of each temperature interval of 10K to reduce model underfitting or overfitting caused by division differences; the dimensions of the hyperparameter space include depth, number of iterations, and learning rate, and the value of the hyperparameter vector is affected by the temperature interval; the conditional generative machine learning model that integrates the CatBoost framework and the VAE encoder was used for training, and the grid search and 10CV cross-validation method were used to determine the value range of the better hyperparameter vector, narrowing the value range of the hyperparameter vector to improve the accuracy of the model.

[0035] 4) Model accuracy evaluation and optimization: using R 2 The accuracy of ionic liquids generated by the RMAE evaluation model is evaluated; the improved bird flock algorithm IBSA is combined to perform a global search of the data points in the training set, mark each data point one by one and share the data, exclude a few abnormal data points, ensure that the learning process converges to data points with higher accuracy, optimize the model's description of the structure-activity relationship of the lubrication performance of ionic liquids, and further improve the model accuracy.

[0036] Formula generation: The specified extreme high-temperature environmental parameters and lubricating surface characteristics are input into the ionic liquid generation model for extremely high-temperature lubrication needs. The model will determine the corresponding hyperparameter vector based on the specified temperature range. The generated ion combinations will be triple-filtered in sequence through RDKit syntax verification, ΔG<0 thermodynamic criterion, and CatBoost performance prediction, and finally output 10-20 ionic liquid candidate formulas for feedback to the user end; the output results are encoded in SMILES and can be directly converted into ionic structural formulas and 3D models in ChemDraw and Chem3D to facilitate structural feature analysis.

[0037] In order to better understand the technical content of the present invention, the following is a detailed description of the present invention in conjunction with the specific embodiments and the appended drawings. Figure 1-3 Further explanation is given.

[0038] Various aspects of the present invention are described herein with reference to the accompanying drawings, in which a number of illustrative embodiments are shown. The embodiments of the present invention are not necessarily intended to encompass all aspects of the present invention. It should be understood that the various concepts and embodiments described above, as well as those described in more detail below, may be implemented in any of a number of ways, as the concepts and embodiments disclosed herein are not limited to any particular implementation. In addition, some aspects disclosed herein may be used alone or in any appropriate combination with other aspects disclosed herein.

[0039] The present invention provides an intelligent design method for ionic liquids for extreme high-temperature lubrication requirements, comprising the following steps:

[0040] S1: Initial dataset construction

[0041] a) Through the coordinated call of Structured Query Language (SQL) and Application Programming Interface (API), joint search is carried out on the following databases:

[0042] -NIST Ionic Liquid Database (ILThermo v2.0)

[0043] -Web of Science Core Collection (indexed between 2010 and 2025)

[0044] b) Establish a three-level screening system, using a logical AND relationship to connect the following conditions:

[0045] - Thermal stability condition: decomposition temperature T when the mass loss is 5% in thermogravimetric analysis (TGA) d ≥210℃

[0046] - Rheological conditions: The kinematic viscosity range of the rheometer is 20-40 cP at a test temperature of 240 °C

[0047] -Tribological conditions: Friction coefficient μ<0.1 measured at 1.0-2.5 GPa contact pressure in four-ball test

[0048] c) Data standardization - building a four-dimensional feature matrix:

[0049] -Chemical structure: cation / anion type (IUPAC nomenclature)

[0050] -Environmental parameters: temperature, load, humidity, etc.

[0051] -Surface characteristics: roughness, charge density, etc.

[0052] -Lubrication properties: viscosity, friction coefficient

[0053] S2: Supplementation of missing entries in the dataset and cleaning of abnormal data

[0054] a) Supplement missing data through literature data mining:

[0055] -Using the natural language processing model BERT to automatically analyze the literature on ionic liquid lubrication performance data

[0056] - For image data such as structural formulas and 3D models appearing in the literature, their structural features are mapped based on graph neural networks (GNN) and converted into unified IUPAC for ion naming

[0057] -Convert the extracted molecular structure data into SMILES format for encoding and storage

[0058] b) Abnormal data cleaning:

[0059] - Based on the calculation results of molecular dynamics simulation, set the theoretical threshold of data confidence and remove abnormal data for data points that deviate from the confidence interval

[0060] c) The final dataset outputs a structured CSV file, including the following fields:

[0061] -Cation / anion SMILES strings

[0062] -Environmental parameters such as temperature, load, humidity, etc.

[0063] -Surface characteristics such as roughness and charge density

[0064] -Viscosity, friction coefficient

[0065] S3: Generative Model Training of Ionic Liquids for Extreme High-Temperature Lubrication Requirements a) Ionic Feature Engineering and Dataset Partitioning:

[0066] - Calculate 208 molecular descriptors using the RDKit 2022.09 toolkit (Wiener index for topology, highest occupied molecular orbital for electronic states, moment of inertia for geometrical features, etc.)

[0067] -Min-Max normalization of environmental parameters (temperature, load, humidity, etc.) will be performed, and the structural feature descriptors and surface characteristics of ionic liquids will be further screened and simplified based on One-Hot encoding and Pearson correlation coefficient heat map, and all ionic liquid data dimensions will be stacked to 20 dimensions.

[0068] - The dataset was divided into training, validation, and test sets in a 7:3 ratio, based on stratified sampling of 10K temperature intervals. b) Model training for intelligent design of ionic liquids for extreme high-temperature lubrication requirements:

[0069] -Define the model hyperparameter space, use a CatBoost regression model with a depth of 8, set the number of early stopping rounds to 50, the learning rate to 0.03, and the L2 regularization coefficient to 5

[0070] - Use grid search and 10CV cross-validation to determine the value range of the better hyperparameter vector, narrow the value range of the hyperparameter vector to improve the accuracy of the model

[0071] S4: Model Accuracy Evaluation and Optimization

[0072] a) Accuracy evaluation indicators:

[0073] -Using R 2 The accuracy of the ionic liquid generated by the RMAE evaluation model is plotted, the actual value of the friction coefficient is plotted against the calculated value, and the learning accuracy of the training set and test set models is compared.

[0074] b) Model optimization:

[0075] - Combined with the improved bird flock algorithm (IBSA), a global search is performed on the data points of the training set to ensure that the learning process converges to data points with higher accuracy, optimize the model's description of the structure-activity relationship of the lubrication performance of ionic liquids, and further improve the model accuracy.

[0076] S5: Recipe Generation:

[0077] After receiving the environmental parameter input, the VAE model first generates 100 candidate molecular structures. These structures are then filtered through three stages: the first stage uses RDKit to verify the validity of the SMILES syntax; the second stage uses ORCA to perform DFT calculations to screen for thermodynamically stable structures with Gibbs free energy ΔG < 0; the third stage uses the trained CatBoost model to predict lubrication performance, retaining candidate molecules that meet both viscosity and friction coefficient requirements.

[0078] -Each run generates 10-20 optimal candidate solutions and sorts them by CatBoost prediction performance. Evaluation metrics require R 2 ≥0.95 and RMSE≤0.02. For batches that do not meet the quality requirements, the VAE latent space sampling radius is automatically adjusted and regenerated until the quality requirements are met or the maximum number of iterations (default 5) is reached.

[0079] Example: Development of extremely high temperature lubricants for aircraft engines

[0080] This embodiment applies the method of the present invention to develop an aviation engine lubricant suitable for an environment of 260°C.

[0081] Step S1:

[0082] Database source: v3.2 (12,340 records)

[0083] SQL filter code:

[0084] SELECT * FROM ionic_liquids

[0085] WHERE decomposition_point>=260

[0086] AND viscosity_240c BETWEEN 20AND 40

[0087] AND friction_coef <= 0.1;

[0088] ·632 valid records were obtained

[0089] Construct a four-dimensional feature matrix

[0090] Step S2:

[0091] Supplement 15 sets of ASTM D4172-22 standard experimental data

[0092] The BERT model recognizes the semantics of text in .jsonl format. The following is an extraction example:

[0093] {"text":"micro and nanoactuators,which operate with sliding speeds of5-50μm / s","meta":{"source":"DOI:10.1021 / la3010807"}}

[0094] {"text":"for PAF,PAN,EtAN,EAF,and DMEAF as a function of slidingvelocity for a silica colloid probe sliding against a mica surface","meta":{"source":"DOI:10.1039 / c4cp02320j"}}

[0095] GNN structure feature extraction:

[0096] gnn_model = IonicGNN()

[0097] smiles=gnn_model.predict(image)

[0098] Example of the final CSV output:

[0099] cation_smiles,anion_smiles,temp(℃),load(GPa),roughness(μm),charge_density(e / nm 2 ),viscosity(cP),friction_coefficient,data_source

[0100] "[P+](C)(C)(C)CCCC","[N+](=O)([O-])[S](=O)(=O)C(F)(F)F",300,2.5,0.1,1.2,28,0.07,DOI:10.1039 / c4cc00979g

[0101] "[C8mim+]","[PF6-]",280,1.8,0.05,0.8,35,0.09,USPTO_Patent:US2023018376A1 ...

[0103] (A total of 1,248 data items were supplemented and 43 abnormal data items were cleaned)

[0104] Step S3:

[0105] Calculate ion feature descriptors:

[0106]

[0107] Hyperparameter space definition:

[0108] hpspace={

[0109] 'depth':[6,7,8,9,10],

[0110] 'iterations':[500,800,1000],

[0111] 'learning_rate':[0.01,0.05,0.1]

[0112] }

[0113] Step S4:

[0114] Calculate the index and draw the fitting line:

[0115] r2=r2_score(y_true,y_pred)

[0116] rmae=mean_absolute_error(y_true,y_pred) / np.mean(y_true)

[0117] print(f"

{model_name}Model Evaluation

[0118] print(f"r2={r2:.3f}|RMAE={rmae:.3f}")

[0119] plt.figure(figsize=(8,6))

[0120] sns.regplot(x=y_true,y=y_pred,

[0121] scatter_kws={'alpha':0.4},

[0122] line_kws={'color':'red','linestyle':'--'})

[0123] plt.xlabel("Actual friction coefficient μ")

[0124] plt.ylabel("Predicted friction coefficient μ")

[0125] plt.title(f"{model_name} Fitting curve (r2={r2:.3f})")plt.grid(True)

[0126] plt.show()

[0127] IBSA optimizes the main process:

[0128]

[0129] Step S5:

[0130] Optimal formula: [P 6,6,6,14 ][BMB]

[0131] Comparison between prediction and actual measurement:

[0132] Kinematic viscosity: predicted value 20-40 cP, measured value 26.8 cP

[0133] Thermal decomposition temperature: predicted value 260℃, measured value 271.1℃

[0134] Friction coefficient: predicted value 0.1, measured value 0.089.

Claims

1. An intelligent design method for ionic liquids for extreme high temperature lubrication requirements, characterized in that: The following steps are involved: 1) Initial Dataset Construction: Using the structured query language SQL, we searched the data platform and screened for data on the lubrication performance of ionic liquids under extreme high temperatures. We then systematically integrated the data into a four-category definition model: ion combination, environmental parameters, surface characteristics, and lubrication performance, to establish the initial dataset. 2) Supplementation of missing entries in the dataset and cleaning of abnormal data: Further search for literature related to the lubrication performance of ionic liquids. The literature was uploaded to the natural language processing model BERT in .pdf format to identify and extract the ionic liquid lubrication performance data in the literature and supplement the blank data. Based on the calculation results of molecular dynamics simulation, a theoretical threshold for data confidence was set, and data points that deviated from the confidence interval were automatically cleaned to obtain an optimized dataset. 3) Model training: The data processed in step 2) is pre-processed by ion structure feature engineering to stack the lubrication performance data of the ionic liquid into 20 dimensions for easier machine learning. The data set is stratified and sampled according to the temperature range, and divided into training set and test set in proportion; Determine the machine learning model and define the model's hyperparameter space. Input the above training set into the machine learning model for training. Based on grid search, further converge on the optimal value of the hyperparameter vector to improve model accuracy. 4) Model accuracy evaluation and optimization: Input the test set into the model obtained in step 3) and use R 2 The accuracy of the ionic liquid generated by the model was evaluated with RMAE. The improved bird flock algorithm (IBSA) was combined to perform a global search of the training set data points to ensure that the learning process converged to data points with higher accuracy. This optimized the model's description of the structure-activity relationship of the lubrication performance of ionic liquids, further improving the model's accuracy and resulting in a generative model capable of generating ionic liquid lubricants for extremely high-temperature lubrication requirements. 5) Formulation Generation: The specified extreme high-temperature environmental parameters and lubrication surface characteristics are input into the generative model obtained after optimization in step 4). The model will determine the corresponding hyperparameter vector based on the specified temperature range and generate 10-20 candidate ionic liquid formulations for molecular dynamics simulation. The model calculates their mass loss at a heating rate of 10K / min and their friction coefficient under a normal stress of 1.0-2.5GPa, and further screens out ionic liquids that can meet the lubrication requirements of the specified extreme high-temperature working conditions.

2. The method according to claim 1, wherein: In step 1), the decomposition temperature T that meets the mass loss of 5% in thermogravimetric analysis TGA is screened. d The ionic liquid lubrication data showed a kinematic viscosity range of 20-40 cP measured by a rotational rheometer at 1450±50 rpm in a nitrogen atmosphere at test temperatures of ≥210°C and 240°C, and a friction coefficient μ<0.1 measured at a contact pressure of 1.0-2.5 GPa in a four-ball test.

3. The method according to claim 1, wherein: In steps 1) and 2), the data sources of the dataset include NIST ILThermo and Web of Science data platforms, SCI-indexed literature, and ASTM standard experimental data; each data set is systematically organized according to the class definition model of the four-tuple lubrication performance described by anion and cation type, temperature and load environment parameters, roughness and charge density surface characteristics, and viscosity and friction coefficient.

4. The method according to claim 1, wherein: In step 2), the natural language processing model BERT is used to identify and extract the semantics of ionic liquids in the literature, so that the lubrication performance can be automatically included in the data set from the literature. For the ionic structural formulas and 3D model image data in the literature, the ionic structural features are mapped and extracted based on the graph neural network (GNN). The obtained ion structure data are uniformly encoded and stored in SMILES format.

5. The method according to claim 1, wherein: In step 3), the SMILES-encoded ionic structural features were converted into 208 feature descriptors using the RDKit 2022.09 toolkit. Based on the one-hot encoding and Pearson correlation coefficient heat map of the common or non-common anion and cation feature descriptors, the 208 feature descriptors were further stacked into 20 dimensions that can accurately reflect the differences in the ionic structural features of the data set.

6. The method according to claim 1, wherein: In step 3), the data in the dataset are stratified and sampled according to a temperature interval of 10K to ensure that the data can be evenly divided into the training set and the test set according to the temperature, thereby reducing the underfitting or overfitting of the model caused by the difference in the division; the training set and the test set are divided in a ratio of 7:

3.

7. The method according to claim 1, wherein: In step 3), the dimensions of the constructed hyperparameter space include depth, number of iterations, learning rate, number of early stopping rounds, and L2 regularization coefficient; a conditional generative machine learning model that integrates the CatBoost framework and the VAE encoder is used for training, and grid search and 10CV cross-validation methods are used to determine the value range of the better hyperparameter vector, narrowing the value range of the hyperparameter vector to improve the accuracy of the model.

8. The method according to claim 1, wherein: In step 4), through R 2 The accuracy of the ionic liquid generated by the quantitative model was ≥0.95 and RMSE ≤0.02; the bird flock algorithm was used to mark each data point one by one and share the data to exclude a few abnormal data points.

9. The method according to claim 1, wherein: In step 5), the generated ionic liquids will be filtered through RDKit syntax verification, ΔG < 0 thermodynamic criterion, and CatBoost performance prediction in sequence, and 10-20 ionic liquid candidate solutions will be output. The output results will be encoded in SMILES.

Images