Intelligent generation method of radiation-resistant ionic liquid
By combining graph neural networks with multi-objective optimization, an ionic liquid database was constructed and multi-performance predictions were performed. This solved the problem of lubricant decomposition and cross-linking under high-energy irradiation, enabling efficient and accurate lubricant formulation generation that meets the stability and lubricity requirements under high-radiation environments.
Patent Information
- Application Number
- CN202510911010.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-02
- Publication Date
- 2025-10-31
AI Technical Summary
Existing lubricants are prone to decomposition and cross-linking under high-energy irradiation, making it difficult to work stably for a long time. Furthermore, the lack of systematic design and high-throughput prediction tools leads to low development efficiency and insufficient prediction accuracy.
A database of ionic liquids was constructed using a combination of graph neural networks and multi-objective optimization. Multi-performance prediction was performed using graph attention networks, and a non-dominated sorting genetic algorithm was used to generate formulations that meet the requirements of stability and lubricity. Molecular dynamics simulations were then used to verify the stability of the ionic liquids.
It enables efficient screening and design of lubricants under complex irradiation conditions, improves development efficiency and prediction accuracy, and ensures the structural stability and lubrication performance of ionic liquids in high-radiation environments.
Smart Images
Figure QLYQS_1
Abstract
Description
Technical Field
[0001] This invention belongs to the interdisciplinary field of lubricating materials and artificial intelligence, and relates to an intelligent method for generating radiation-resistant ionic liquids. Background Technology
[0002] With the development of technologies such as nuclear energy engineering, deep space exploration, and advanced particle devices, equipment faces high-intensity radiation environments, placing higher demands on the radiation resistance of lubricant materials. Traditional lubricants are prone to decomposition, cross-linking, or deterioration of lubrication performance under high-energy radiation such as gamma rays and neutron beams, making it difficult for them to operate stably for a long time.
[0003] Ionic liquids are considered potential radiation-resistant lubricants due to their designability, high thermal stability, and excellent lubrication properties. However, different ionic liquids exhibit significant differences in their response to irradiation, and their stability and lubrication behavior are influenced by multiple factors, including irradiation type, dose, and energy density. Current research relies heavily on empirical screening and single-factor experiments, which are inefficient and lack systematic design and high-throughput prediction tools. Therefore, there is an urgent need for a rapid screening technology for radiation-resistant lubricants that integrates multi-source data, molecular modeling, and intelligent optimization to support the development of lubricating materials under complex irradiation conditions. Summary of the Invention
[0004] This invention aims to address the problems of low efficiency, insufficient prediction accuracy, and lack of structural stability evaluation in the development of existing radiation-resistant lubricants, and provides an intelligent generation method for radiation-resistant ionic liquids. This method integrates graph neural networks and multi-objective optimization. By constructing an ionic liquid database containing irradiation response data, it extracts key molecular and irradiation environment features, trains a graph attention network for multi-performance prediction, and combines a non-dominated sorting genetic algorithm to generate formulations that meet stability and lubricity requirements under target irradiation doses.
[0005] The technical solution of the present invention is as follows:
[0006] A smart method for generating radiation-resistant ionic liquids includes the following steps:
[0007] (1) Data construction: Ionic liquid experimental data with irradiation stability indicators were collected from databases and literature, and missing irradiation stability indicator data were supplemented by experiments. Ionic liquid experimental data with no change in stability indicator data after irradiation were selected to construct training samples.
[0008] (2) Feature engineering: SMILES encoding is performed on the molecular structure of ionic liquids in the training samples, radiation-related molecular descriptors are extracted, and the optimized dataset is obtained.
[0009] (3) Model Training: The optimized dataset was divided into training, testing, and validation sets in a 7:2:1 ratio. A multi-task learning model was constructed based on a graph isomorphic network (GIN) to jointly predict the three properties of ionic liquids: irradiation stability, friction coefficient, and viscosity. The model adopted a three-branch output structure, corresponding to the three performance indicators respectively. During the training process, a weighting strategy based on task uncertainty was introduced, and the system automatically allocated the loss weights for each task. The loss weight of each task was determined by its prediction uncertainty. The overall loss function α is expressed as follows:
[0010] a = (1 / 2σ1) 2 )·ɑ1+logσ1+(1 / 2σ2 2 )·ɑ2+logσ2+(1 / 2σ3 2 )·ɑ3+logσ3,
[0011] Where: α represents the total loss function, α1, α2, and α3 are the predicted losses for irradiation stability, friction coefficient, and viscosity, respectively, and σ1, σ2, and σ3 are the uncertainty parameters of the corresponding tasks, which are introduced as learnable parameters;
[0012] The optimizer uses AdamW, which combines early stopping mechanism and regularization to avoid overfitting, and loads weight parameters pre-trained in radiation-related materials tasks to improve initial performance.
[0013] (4) Formulation generation: Input the target irradiation dose and lubrication performance requirements, and search for candidate ionic liquids that meet the requirements through evolutionary optimization algorithm;
[0014] (5) Structure verification and screening: Molecular dynamics and irradiation reaction simulation were used to evaluate the lubrication performance of candidate ionic liquids under the target irradiation dose. The candidate ionic liquids were ranked according to the lubrication performance evaluated by the simulation, and the preferred ionic liquids were screened.
[0015] (6) Output display: Output the structure of the preferred ionic liquid, the predicted lubrication performance and the confidence score.
[0016] Furthermore, in step (1), the database is IONILBASE, ILThermo database, etc.; the irradiation stability index data includes the ionic liquid performance data after being exposed to different types of radiation sources such as γ-rays, α particles, and fast neutrons, and the ionic liquid performance data includes the friction coefficient and viscosity.
[0017] Furthermore, in step (2), the molecular descriptor includes π-electron density, aromaticity distribution, charge mobility, etc., and the molecular descriptor calculation is performed using RDKit and Gaussian.
[0018] Furthermore, in step (3), a graph isomorphic network architecture is used, and transfer learning is employed to initialize the parameters.
[0019] Furthermore, in step (4), the non-dominated sorting genetic algorithm (NSGA-II) is used for multi-objective optimization.
[0020] Furthermore, in step (5), the molecular dynamics and irradiation reaction simulation experiment is a stability test under a target irradiation dose impact of 10 ns.
[0021] Furthermore, in step (5), the candidate ionic liquids ranked in the top 10% are selected as preferred ionic liquids.
[0022] Compared with the prior art, the present invention has the following advantages:
[0023] (1) Irradiation stability is introduced as a prediction target for the first time and is modeled together with friction performance; (2) The graph neural network design based on structural tunability can be generalized to multiple types of ionic liquid structures; (3) A simulation verification mechanism is introduced to ensure the stability of the screened formulation under irradiation conditions. Detailed Implementation
[0024] The present invention discloses a smart method for generating radiation-resistant ionic liquids, comprising the following steps:
[0025] Step 1: Collect post-irradiation performance data from IONILBASE, ILThermo and literature, and screen ionic liquids whose stability index data do not change under irradiation doses of γ, α and neutron beams of 10-100 kGy to form a training dataset;
[0026] Step 2: Calculate molecular descriptors using RDKit and Gaussian, including molecular polarizability, maximum π orbital density, charge distribution entropy, etc., as input features;
[0027] Step 3: Construct a graph neural network model. A GIN network is built based on PyTorch Geometric. The input nodes are the atom types and the descriptors mentioned above. The output branches predict: irradiation stability indices (such as relative mass loss rate), friction coefficient, and viscosity. During training, a multi-task loss function is used, and task weights are dynamically adjusted using a weighting strategy based on task uncertainty.
[0028] Step 4: Based on the user-input target radiation type, dose, ambient temperature, and friction performance requirements, call the NSGA-II optimization algorithm to search for the optimal ionic liquid in the pre-screening library;
[0029] Step 5: Perform molecular dynamics simulations using GROMACS, set the irradiation excitation state and evaluate the structural integrity and thermal stability within 10 ns, and eliminate reactive dissociated ionic liquids.
[0030] Step 6: Output the candidate formulations that meet the conditions, including their SMILES structure, predicted values of each performance, and stability scores under simulation conditions. The candidate formulations with scores in the top 10% are the preferred formulations.
[0031] Example 1: Screening and Validation of Radiation-Resistant Lubricant Formulations for Aerospace Applications
[0032] This embodiment addresses the lubrication requirements of geosynchronous orbit satellites under space radiation doses greater than 200 krad (Si), and uses the method of this invention to screen and evaluate radiation-resistant ionic liquid lubricants.
[0033] Step 1: Screen ionic liquid records containing radiation experimental data from the ILThermo and IONILBASE databases, and construct an initial sample set by combining literature expansion, totaling 489 records;
[0034] Step 2: Encode the sample structure using SMILES expressions, extract molecular descriptors (TPSA, LogP, aromaticity atomic ratio, number of hydrogen bond donors / acceptors, etc.) using RDKit, and divide the training set, validation set, and test set in a ratio of 7:2:1;
[0035] Step 3: Construct a GIN graph isomorphic network model. The GIN graph network structure can accurately capture the topological information of ionic liquid molecules and improve the characterization ability of performance characteristics under extreme radiation environment.
[0036] Add a three-branch output structure to predict: (1) radiation stability factor; (2) tribological properties (friction coefficient, wear rate); (3) viscosity;
[0037] The AdamW optimizer is used for training, and a weighted strategy based on task uncertainty is introduced, where the loss weight for each task is determined by its prediction uncertainty.
[0038] Step 4: Input the space operating conditions requirements (radiation dose > 200 krad, target friction coefficient < 0.1), and call the NSGA-II optimization algorithm combined with simulated annealing search optimization combination;
[0039] Step 5: The optimal formulation was obtained through screening: [(C6mim)(FAP)](1-hexyl-3-methylimidazolium fluorophosphite); this formulation maintained more than 90% lubrication performance after irradiation with a total γ-ray dose of 300 krad (Co-60 source), with wear volume increase not exceeding 5%, and the error between prediction and actual measurement was <4.5%.
[0040] Example 2: Lubricant Design for Main Pump Sealing System of Nuclear Power Plant
[0041] To address the long-term lubrication stability requirements of the shaft seal of the primary circuit main pump in a high-temperature and high-radiation (80℃, 100krad) operating environment in a nuclear power plant, the intelligent generation method of this invention is used for design.
[0042] Step 1: Construct a multi-source dataset containing thermal stability, radiation stability, and tribological properties, collecting a total of 362 composite property records;
[0043] Step 2: Extract molecular features including thermal decomposition temperature (Td), charged surface adsorption energy (Es), number of hydrogen bonds, LogP, etc., and divide the training set, validation set and test set in a ratio of 7:2:1;
[0044] Step 3: The model structure incorporates a GIN graph isomorphic network model to enhance its ability to perceive the stability changes of molecular structures under the combined effects of heat and radiation.
[0045] The model adopts a multi-task regression architecture with three output branches, which correspond to the predicted decomposition temperature (Td), the irradiation stability factor, and the friction coefficient, respectively.
[0046] A weighted strategy based on task uncertainty is introduced, and the weights of each task are dynamically updated based on the performance of the validation set to achieve balanced performance optimization. To ensure that the model's prediction results meet the requirements of extreme applications, the following constraints are introduced in the subsequent formulation optimization stage:
[0047] Decomposition temperature Td > 300℃ (meets high thermal stability requirements);
[0048] Irradiation stability factor > 0.85 (ensuring structural integrity under radiation conditions);
[0049] The coefficient of friction is <0.12 (meets the standard for friction reduction and wear resistance).
[0050] Step 4: Input the space operating conditions requirements (radiation dose > 100 krad, target friction coefficient < 0.05), and call the NSGA-II optimization algorithm combined with simulated annealing search optimization combination;
[0051] Step 5: The final screening formula is [P] 6,6,6,14 [BMB] (Trihexyltetradecylphosphonium bis(2,4,4-trimethylpentyl)phosphonium); Molecular dynamics simulations verified its good adsorption stability at 80℃ and 100 krad, with no obvious desorption of the lubricating film. The measured friction coefficient was 0.098, and the wear was 0.63 mm. 3 .
Claims
1. A smart method for generating radiation-resistant ionic liquids, characterized in that, Includes the following steps: (1) Data construction: Collect experimental data of ionic liquids with irradiation stability indicators from databases and literature, supplement the missing irradiation stability indicator data through experiments, and screen the experimental data of ionic liquids whose stability indicator data did not change after irradiation to construct training samples. (2) Feature engineering: SMILES encoding is performed on the molecular structure of ionic liquids in the training samples, radiation-related molecular descriptors are extracted, and the optimized dataset is obtained; (3) Model Training: The optimized dataset was divided into training, testing, and validation sets in a 7:2:1 ratio. A multi-task learning model was constructed based on a graph isomorphic network to jointly predict the three properties of ionic liquids: irradiation stability, friction coefficient, and viscosity. The model adopted a three-branch output structure, corresponding to the three performance indicators respectively. During the training process, a weighting strategy based on task uncertainty was introduced, and the system automatically allocated the loss weights for each task. The loss weight of each task was determined by its prediction uncertainty. The overall loss function α is expressed as follows: ， in: Represents the total loss function. , , The predicted losses are for irradiation stability, friction coefficient, and viscosity, respectively. , , The uncertainty parameters for the corresponding task are introduced as learnable parameters; The optimizer uses AdamW, which combines early stopping mechanism and regularization to avoid overfitting, and loads weight parameters pre-trained in radiation-related materials tasks to improve initial performance. (4) Formulation generation: Input the target irradiation dose and lubrication performance requirements, and search for candidate ionic liquids that meet the requirements through evolutionary optimization algorithm; (5) Structure verification and screening: Molecular dynamics and irradiation reaction simulation were used to evaluate the lubrication performance of candidate ionic liquids under the target irradiation dose. The candidate ionic liquids were ranked according to the lubrication performance evaluated by the simulation, and the preferred ionic liquids were screened. (6) Output display: Output the structure of the preferred ionic liquid, the predicted lubrication performance and the confidence score.
2. The intelligent generation method according to claim 1, characterized in that, In step (1), the databases are IONILBASE and ILThermo databases; the irradiation stability index data include the ionic liquid performance data after being exposed to different types of radiation sources such as γ-rays, α-particles and fast neutrons, and the ionic liquid performance data includes the friction coefficient and viscosity.
3. The intelligent generation method according to claim 1, characterized in that, In step (2), the molecular descriptor includes π-electron density, aromaticity distribution and charge mobility. The molecular descriptor calculation is performed using RDKit and Gaussian.
4. The intelligent generation method according to claim 1, characterized in that, In step (3), a graph isomorphic network architecture is used, and transfer learning is employed to initialize the parameters.
5. The intelligent generation method according to claim 1, characterized in that, In step (4), a non-dominated sorting genetic algorithm is used for multi-objective optimization.
6. The intelligent generation method according to claim 1, characterized in that, In step (5), the molecular dynamics and irradiation reaction simulation experiment is a stability test under a target irradiation dose impact of 10 ns.
7. The intelligent generation method according to claim 1, characterized in that, In step (5), the candidate ionic liquids that are ranked in the top 10% are selected as preferred ionic liquids.