Closed-loop design methods, apparatus, equipment, media, and procedures for rare earth-doped solid electrolytes.
By employing a closed-loop design approach, utilizing performance prediction models for screening and iterative adjustments, and combining automated preparation and online characterization, the problem of low efficiency in the development of rare-earth-doped solid electrolytes has been solved, achieving an efficient and standardized material development process from design to implementation.
Patent Information
- Application Number
- CN202611367187.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-04
- Publication Date
- 2026-11-17
AI Technical Summary
In the development of rare earth-doped solid electrolytes, the existing technology lacks a systematic intelligent linkage mechanism, resulting in low R&D efficiency, difficulty in achieving a complete process from design to implementation, and disconnect between material performance characterization results and the design stage, making it impossible to correct design schemes and process parameters in real time.
By adopting a closed-loop design approach, candidate design schemes are screened through performance prediction models, samples are prepared and performance indicators are measured, and iterative adjustments are made based on deviations. Combined with automated preparation and online characterization, a closed-loop process from design to implementation is achieved, thereby improving R&D efficiency.
It achieves a closed-loop process from design to implementation, improves the efficiency of material research and development and the consistency of sample preparation, reduces individual differences and random errors introduced by manual operation, and ensures the standardization of experimental data and efficient iterative optimization.
Smart Images

Figure CN123120100A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of materials design technology, and more specifically, to a closed-loop design method, apparatus, equipment, medium, and program product for rare earth-doped solid electrolytes. Background Technology
[0002] Garnet-type lithium lanthanum zirconium oxide (LLZO) solid electrolytes are considered important candidate materials for all-solid-state lithium batteries due to their high ionic conductivity, wide electrochemical window, and good thermal stability. Introducing rare earth or high-valence ion doping can stabilize the cubic phase structure, improve lithium-ion migration channels, and enhance the overall performance of the material.
[0003] Currently, existing research and development solutions mainly focus on the single step of "doping ratio design". On this basis, it is still necessary to rely on manual experience rules to screen the formula, adjust the process parameters such as sintering temperature and atmosphere through repeated experiments, and evaluate the material performance by combining offline characterization methods. The lack of a systematic intelligent linkage mechanism leads to low efficiency in the development of the final solution. Summary of the Invention
[0004] The purpose of this application is to provide a closed-loop design method, apparatus, device, medium, and program product for rare earth-doped solid electrolytes, so as to improve the research and development efficiency of solid electrolytes.
[0005] In a first aspect, embodiments of this application provide a closed-loop design method for rare-earth-doped solid electrolytes, including: Multiple solid electrolyte design schemes to be predicted are input into the performance prediction model to obtain the performance index prediction results corresponding to each solid electrolyte design scheme; wherein, each solid electrolyte design scheme to be predicted includes a multi-dimensional parameter field, which includes the type of rare earth doping element, doping concentration and preparation process parameters. The solid electrolyte design schemes whose performance index prediction results meet the preset target performance conditions are selected as candidate design schemes, and the current design scheme is determined based on the candidate design schemes. Prepare corresponding solid electrolyte samples based on the multidimensional parameter fields in the current design scheme; Obtain the measured performance indicators of the solid electrolyte sample; The measured performance indicators are compared with the predicted performance indicators of the current design scheme, and the multi-dimensional parameter fields of the current design scheme are adaptively and iteratively adjusted according to the comparison deviation until the comparison deviation between the measured performance indicators and the corresponding predicted performance indicators is less than a preset deviation threshold, thus obtaining the final solid electrolyte target design scheme.
[0006] In this embodiment of the application, by inputting the design scheme into the model to predict the performance, preparing samples and obtaining measured indicators after screening candidate schemes, and then iteratively adjusting based on the deviation between the measured performance and the predicted performance, a closed loop of the entire process from material design to implementation can be realized, thereby improving the overall efficiency of material research and development.
[0007] In some embodiments, the step of selecting solid electrolyte design schemes whose performance index prediction results meet preset target performance conditions as candidate design schemes, and determining the current design scheme based on the candidate design schemes, includes: Solid electrolyte design schemes whose performance index prediction results meet the preset target performance conditions are selected as candidate design schemes, and the credibility score corresponding to each candidate design scheme is obtained; wherein, the credibility score is calculated by combining the model prediction uncertainty, cross-validation error, distance between candidate samples and training set distribution, consistency of multi-model output and number of historical similar cases; The current design scheme is selected from the candidate design schemes by combining the credibility score.
[0008] In this embodiment of the application, by obtaining the credibility scores of each candidate design scheme and selecting the current design scheme accordingly, the reliability of the prediction results can be quantified, thereby improving the accuracy of the current design scheme selection and further improving the overall material scheme development efficiency.
[0009] In some embodiments, the closed-loop design method of the rare earth-doped solid electrolyte further includes: When the number of iterations of the adaptive iterative adjustment exceeds a preset threshold, the backup design scheme is used as the current design scheme in the next round; wherein, the backup design scheme is the design scheme with the highest performance index prediction result among the solid electrolyte design schemes to be predicted, excluding the candidate design scheme.
[0010] In this embodiment, by automatically using the backup design scheme with the second-best performance prediction as the current design scheme in the next round when the number of iterations exceeds the limit, it is possible to prevent a single scheme from falling into an iterative dead loop, thereby improving the fault tolerance and convergence efficiency of the closed-loop system.
[0011] In some embodiments, before inputting the multiple solid electrolyte design schemes to be predicted into the performance prediction model, the method further includes: The parameter field candidate set corresponding to the multidimensional parameter field is determined based on the preset parameter field range; wherein, the parameter field candidate set includes a rare earth doping element type candidate set, a doping concentration candidate set, and a preparation process parameter candidate set; The elements and parameter values in each of the candidate sets of the parameter fields are traversed and combined to generate the multiple solid electrolyte design schemes to be predicted.
[0012] In this embodiment of the application, by constructing a candidate set based on a preset parameter field range and traversing and combining it to generate the design scheme to be predicted, the efficiency and comprehensiveness of the scheme design can be further improved.
[0013] In some embodiments, the preparation of the corresponding solid electrolyte sample based on the multidimensional parameter fields in the current design scheme includes: Using an automated preparation system, the ingredients are sequentially prepared, mixed, shaped, and sintered based on the multidimensional parameter fields in the current design scheme to obtain a solid electrolyte sample corresponding to the current design scheme.
[0014] In the embodiments of this application, by using an automated preparation system to sequentially perform batching, mixing, molding and sintering, it is possible to achieve standardized and reproducible preparation from design scheme to solid electrolyte sample, thereby improving the consistency and efficiency of sample preparation.
[0015] In some embodiments, the performance prediction model is constructed in the following ways: A sample dataset of solid electrolyte materials is obtained from a multi-source database; wherein, each sample data in the sample dataset contains the multidimensional parameter field and the corresponding performance index parameter; The pre-constructed initial machine learning model is trained and cross-validated using a sample dataset to obtain a trained performance prediction model; wherein the initial machine learning model includes at least one of the following: random forest model, gradient boosting regression model, support vector regression model, and Bayesian neural network model.
[0016] In this embodiment, the reliability and generalization ability of the model prediction are improved by training an initial machine learning model with sample data containing multidimensional parameter fields and corresponding performance indicators obtained from a multi-source database.
[0017] In some embodiments, the categories of the performance index parameters include ionic conductivity, activation energy, phase type, and phase purity; The process of training and cross-validating a pre-built initial machine learning model using a sample dataset to obtain a trained performance prediction model includes: For different categories of performance index parameters, corresponding candidate model sets are established; among them, a corresponding regression model set is established for continuous performance indexes, and a corresponding classification model set is established for categorical performance indexes; each candidate model set includes multiple models such as random forest model, gradient boosting regression model, support vector regression model, and Bayesian neural network model. For each set of candidate models, the candidate model with the best performance is selected as the final prediction model corresponding to that performance index parameter based on at least one of the following evaluation metrics: validation set error, classification accuracy, stability, and generalization ability.
[0018] In this embodiment of the application, by constructing candidate model sets for different categories of indicators and selecting the optimal model from each set based on multiple evaluation indicators, the accuracy and reliability of performance prediction are further improved.
[0019] In some embodiments, the adaptive iterative adjustment of the multidimensional parameter fields of the current design scheme based on the comparison deviation includes: If the comparison deviation is greater than a preset deviation threshold, the adjustment strategy will be executed in the following priority order: First priority: Adjust the doping concentration; Second priority: Change the type of rare earth doping element; Third priority: Adjust the sintering temperature and / or holding time in the preparation process parameters.
[0020] In this embodiment, the overall efficiency of the closed-loop iteration is further improved by executing each adjustment strategy in a specific priority order according to its characteristics.
[0021] Secondly, embodiments of this application provide a closed-loop design device for a rare-earth-doped solid electrolyte, comprising: The performance prediction module is used to input multiple solid electrolyte design schemes to be predicted into the performance prediction model to obtain the performance index prediction results corresponding to each solid electrolyte design scheme; wherein, each solid electrolyte design scheme to be predicted includes a multi-dimensional parameter field, which includes the type of rare earth doping element, the doping concentration, and the preparation process parameters. The scheme screening module is used to screen out the solid electrolyte design schemes whose performance index prediction results meet the preset target performance conditions as candidate design schemes, and determine the current design scheme based on the candidate design schemes. The sample preparation module is used to prepare corresponding solid electrolyte samples based on the multidimensional parameter fields in the current design scheme. The performance testing module is used to obtain the measured performance indicators of the solid electrolyte sample. The scheme optimization module is used to compare the measured performance indicators with the predicted performance indicators of the current design scheme, and to adaptively iteratively adjust the multi-dimensional parameter fields of the current design scheme according to the comparison deviation, until the comparison deviation between the measured performance indicators and the corresponding predicted performance indicators is less than a preset deviation threshold, thereby obtaining the final solid electrolyte target design scheme.
[0022] Thirdly, embodiments of this application provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, can implement the method described in any embodiment of the first aspect.
[0023] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program, which, when executed by a processor, can implement the method described in any embodiment of the first aspect.
[0024] Fifthly, embodiments of this application provide a computer program product, the computer program product including a computer program, wherein when the computer program is executed by a processor, it can implement the method described in any embodiment of the first aspect. Attached Figure Description
[0025] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.
[0026] Figure 1 A flowchart illustrating a closed-loop design method for a rare-earth-doped solid electrolyte provided in this application embodiment; Figure 2 This is a schematic diagram of the closed-loop design system provided in the embodiments of this application; Figure 3 A schematic diagram of a closed-loop design device for a rare earth-doped solid electrolyte provided in an embodiment of this application; Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation
[0027] The technical solutions in the embodiments of this application will now be described with reference to the accompanying drawings.
[0028] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.
[0029] It should be noted that the current traditional R&D technology for rare earth-doped garnet-type lithium lanthanum zirconium oxide solid electrolytes mainly relies on manual empirical rules to screen formulas, adjust process parameters such as sintering temperature and atmosphere through repeated experiments, and evaluate material performance using offline characterization methods. This approach lacks a systematic intelligent linkage mechanism and mainly suffers from the following problems: 1. Limited technical coverage: The comparative documents only focus on the single step of "doping ratio design" and do not involve subsequent processes such as material preparation, performance characterization and parameter optimization. They cannot provide a full-chain solution and cannot support the complete R&D needs from design to implementation. Traditional R&D technologies also lack the synergy between various links and the process is fragmented.
[0030] 2. Low R&D efficiency: Traditional R&D techniques rely on manual experience to screen formulations, and process parameter adjustments depend on repeated experiments. Furthermore, the nonlinear coupling relationship between formulations and process parameters is difficult to decipher manually, resulting in long formulation screening and parameter optimization cycles with low efficiency. Although the comparative documents optimize the proportions through first-principles calculations, they have not solved the efficiency bottleneck of subsequent preparation and characterization.
[0031] 3. Lack of data linkage and feedback mechanism: In existing technologies, the results of material performance characterization are disconnected from the formulation design and process adjustment stages. It is impossible to correct the design scheme and process parameters in real time based on the characterization data, resulting in poor R&D repeatability and difficulty in ensuring performance consistency.
[0032] To address at least one of the problems existing in the prior art, this application provides a closed-loop design method for rare earth-doped solid electrolytes, aiming to improve the efficiency of solid electrolyte research and development.
[0033] like Figure 1 As shown in the figure, this application provides a closed-loop design method for rare earth-doped solid electrolytes, which may include the following steps: S1. Input multiple solid electrolyte design schemes to be predicted into the performance prediction model to obtain the performance index prediction results corresponding to each solid electrolyte design scheme; wherein, each solid electrolyte design scheme to be predicted includes multi-dimensional parameter fields, including rare earth doping element type, doping concentration and preparation process parameters.
[0034] In this step, multiple solid-state electrolyte design schemes to be predicted are first constructed. Each design scheme is a combination of multi-dimensional parameter fields including the type of rare earth dopant, doping concentration, and preparation process parameters. The type of rare earth dopant can include at least one of neodymium, gadolinium, samarium, etc.; the doping concentration can be selected from multiple candidate values between 0.1 mol% and 20 mol%; the preparation process parameters can include sintering temperature, heating rate, holding time, and sintering atmosphere.
[0035] Then, the multiple design schemes to be predicted are input one by one into the pre-trained performance prediction model. For example, the performance prediction model takes multi-dimensional parameter fields as input and outputs performance indicators such as ionic conductivity, activation energy, phase type, and phase purity. It calculates the predicted performance indicators for each design scheme through function mapping relationships. The model outputs a set of specific values or categories, such as predicting the ionic conductivity, activation energy, whether it is a pure cubic phase, and the percentage of cubic phase for each design scheme.
[0036] S2. Select the solid electrolyte design schemes whose performance index prediction results meet the preset target performance conditions as candidate design schemes, and determine the current design scheme based on the candidate design schemes.
[0037] After obtaining the predicted performance index results for each design scheme to be predicted, the system can filter them according to the user-preset target performance conditions. For example, the target performance condition can be set as "ionic conductivity ≥ 6.0 × 10⁻⁶". -4 If the predicted ionic conductivity and predicted cubic phase purity are both S / cm and ≥90%, then all design schemes that meet the above conditions will be selected as candidate design schemes.
[0038] It should be noted that if there are multiple candidate design schemes, the group with the highest comprehensive performance index (the comprehensive performance index is obtained by weighting the degree to which each performance index exceeds the preset threshold) can be selected as the current design scheme. Alternatively, the group with the highest single performance index can be selected as the current design scheme, for example, the group with the highest predicted ionic conductivity can be selected as the current design scheme.
[0039] S3. Prepare the corresponding solid electrolyte sample based on the multi-dimensional parameter fields in the current design scheme.
[0040] In this step, the corresponding solid electrolyte sample can be prepared manually based on the current design scheme; alternatively, the system can perform automated preparation based on the multi-dimensional parameter fields in the current design scheme. Specifically, firstly, according to the rare earth doping element types and doping concentrations specified in the current design scheme, the automated preparation system performs the material preparation operation; then, the raw materials are transferred to a high-energy vibration ball mill for mixing, so that the raw materials are uniformly mixed and refined to micron-level powder; subsequently, they are pressed into discs with a diameter of 10mm to 20mm by an automatic mold and tablet press; finally, the pressed sample is placed in a programmable tube furnace and sintered according to the heating rate, holding time and sintering atmosphere in the current design scheme to obtain the solid electrolyte sample.
[0041] S4. Obtain the measured performance indicators of the solid electrolyte sample.
[0042] It should be noted that the measured performance indicators and the predicted performance indicators are of the same type, such as ionic conductivity, activation energy, phase type, and phase purity.
[0043] For example, a flexible adaptive robotic arm can be used to remove solid electrolyte samples from the crucible and transfer them to an X-ray diffraction platform. A high-speed θ-2θ scanning module acquires the X-ray diffraction pattern of the sample, and a pre-trained convolutional neural network model automatically analyzes the diffraction pattern to identify the proportions of cubic, tetragonal, and impurity phases (referring to the relative content of crystalline phases other than the target garnet-type LLZO phase in the sample, including unreacted raw material phases and reaction intermediate phases, such as Li₂O, La₂Zr₂O₇, Li₂ZrO₃, La₂O₃, ZrO₂, etc.). Simultaneously, the ionic conductivity and activation energy of the sample can be measured using AC impedance spectroscopy as components of the measured performance indicators.
[0044] S5. Compare the measured performance indicators with the predicted performance indicators of the current design scheme, and adaptively iterate and adjust the multi-dimensional parameter fields of the current design scheme according to the comparison deviation until the comparison deviation between the measured performance indicators and the corresponding predicted performance indicators is less than the preset deviation threshold, and obtain the final solid electrolyte target design scheme.
[0045] In this step, the measured performance indicators are compared with the predicted performance indicators of the corresponding current design scheme, and the comparison deviation of each indicator is calculated. If the deviation is greater than a preset deviation threshold (e.g., a relative deviation exceeding 10%), it is determined that there is a deviation in the current manufacturing process, and the multi-dimensional parameter fields of the current design scheme need to be adjusted. It should be understood that when the deviation of any performance indicator exceeds the threshold range, it is necessary to trigger the adjustment of the parameters of the next round of design scheme. When the deviations of all performance indicators are within the threshold range, no adjustment is triggered and the current scheme is output.
[0046] For example, the adjustment strategy may include at least one of the following: adjusting the doping concentration (e.g., keeping the doping element type unchanged and adjusting in ±0.05 mol% increments), changing the doping element type (e.g., fixing the doping mole fraction and replacing it with a metal element of the same valence state), adjusting the sintering temperature (e.g., adjusting in 50°C increments within the range of 900°C to 1300°C), and adjusting the holding time (e.g., setting it to 2h, 4h, 6h, or 8h).
[0047] After each updated design scheme is generated, steps S3 to S5 are executed again until the deviation between the measured performance index and the model's prediction result for the current design scheme in a certain iteration is less than the preset deviation threshold. At this point, it is determined that the design scheme has stably reproduced the theoretical performance predicted by the model, and the current design scheme is output as the final solid electrolyte target design scheme.
[0048] This application embodiment achieves a closed-loop process from material design to implementation by inputting the design scheme into the model to predict performance, preparing samples and obtaining measured indicators after screening candidate schemes, and then iteratively adjusting based on the deviation between the measured performance and the predicted performance, thereby improving the overall efficiency of material research and development.
[0049] It should be noted that during the implementation of this application, automated preparation operations and online characterization processes are all automatically executed by robotic arms, precision equipment, and sensors according to preset programs. This avoids individual differences and random errors introduced by manual operation, ensuring a high degree of consistency in preparation and testing conditions in each iteration. The deviation between the standardized measured data and the model prediction results can accurately reflect the optimization direction of the formulation and process parameters. Simultaneously, the entire closed-loop process is uniformly scheduled by the control platform, allowing for 24-hour uninterrupted operation without manual intervention. This effectively improves the iterative optimization efficiency of the solution development and continuously produces standardized, low-error experimental data.
[0050] In some embodiments, step S2, selecting the solid electrolyte design schemes whose performance index prediction results meet the preset target performance conditions as candidate design schemes, and determining the current design scheme based on the candidate design schemes, may include: The solid electrolyte design schemes whose performance index prediction results meet the preset target performance conditions are selected as candidate design schemes, and the credibility score corresponding to each candidate design scheme is obtained. The credibility score is calculated by combining the model prediction uncertainty, cross-validation error, distance between candidate samples and training set distribution, consistency of multi-model output, and number of historical similar cases. The current design scheme is selected from the candidate design schemes by combining the credibility score.
[0051] It should be noted that the credibility score is a quantitative evaluation index of the reliability of the prediction results of candidate design schemes. This score comprehensively considers multiple dimensions such as model prediction uncertainty, cross-validation error, distance between candidate samples and training set distribution, consistency of multi-model outputs, and the number of historical similar cases. Among them, model prediction uncertainty reflects the variance of the model's prediction output for the scheme; cross-validation error measures the model's consistency performance across different data subsets; distance between candidate samples and training set distribution characterizes the closeness of the scheme's parameter combination to the sample space already learned by the model; consistency of multi-model outputs reflects the dispersion of different models' prediction results for the scheme; and the number of historical similar cases refers to the number of existing experimental samples in the training set with similar parameter characteristics to the scheme. Based on the above multiple factors, a weighted fusion and integration is performed and normalized to a value between 0 and 1 to obtain the credibility score corresponding to each candidate design scheme. The higher the score value, the more reliable the prediction result of the candidate scheme.
[0052] For example, after obtaining the confidence scores for each preferred design scheme, the system combines the predicted performance index values and confidence scores of each candidate design scheme to jointly determine the current design scheme. For instance, if a candidate scheme has the highest predicted ionic conductivity among all candidates, but its confidence score is low (e.g., 0.5), it indicates that the model's prediction reliability for this scheme is insufficient. This may be due to a lack of similar samples in the training set for the dopant element combination in this scheme, or significant discrepancies in the prediction results of different models. In this case, the system can prioritize selecting the scheme with the second highest prediction performance but a higher confidence score (e.g., a confidence score of 0.9) as the current design scheme to reduce the risk of iteration failure caused by model prediction bias.
[0053] Based on this, by obtaining the credibility scores of each candidate design scheme and selecting the current design scheme accordingly, the reliability of the prediction results can be quantified, thereby improving the accuracy of the current design scheme selection and further improving the overall material scheme development efficiency.
[0054] In some embodiments, the closed-loop design method for rare earth-doped solid electrolytes may further include the following steps: When the number of iterations for adaptive iterative adjustment exceeds a preset threshold, the backup design scheme will be used as the current design scheme for the next round; among them, the backup design scheme is the design scheme with the highest predicted performance index other than the candidate design scheme among the solid electrolyte design schemes to be predicted.
[0055] It should be noted that the backup design scheme refers to an alternative scheme used to continue the closed-loop iteration when the current design scheme fails to achieve deviation convergence within a preset number of iterations due to inherent model deviations or process limitations. The preset number of iterations threshold is a preset maximum allowed number of iterations to prevent a single scheme from iterating indefinitely.
[0056] During the closed-loop iteration process, the system can count the number of adaptive iteration adjustments for the current design scheme in real time. When the number of iterations exceeds a preset threshold, it indicates that the current design scheme is unlikely to achieve convergence of the measured and predicted deviations under the preset adjustment strategy. Therefore, to avoid infinite iteration, the system should use the backup design scheme as the current design scheme for the next round. The backup design scheme is the design scheme with the highest predicted performance index among all solid electrolyte design schemes to be predicted, excluding those already identified as candidate design schemes (i.e., schemes whose predicted performance meets the target conditions). After switching to the backup design scheme, the system re-executes the sample preparation, online characterization, and iterative adjustment steps (i.e., steps S3-S5).
[0057] Based on this, by automatically using the backup design scheme with the second-best performance prediction as the current design scheme in the next round when the number of iterations exceeds the limit, it is possible to prevent a single scheme from falling into an iterative dead loop, thereby improving the fault tolerance and convergence efficiency of the closed-loop system.
[0058] In some embodiments, prior to step S1, the following may also be included: The parameter field candidate set corresponding to the multidimensional parameter field is determined based on the preset parameter field range; among which, the parameter field candidate set includes the rare earth doping element type candidate set, the doping concentration candidate set, and the preparation process parameter candidate set; The elements and parameter values in the candidate sets of each parameter field are traversed and combined to generate multiple solid electrolyte design schemes to be predicted.
[0059] It should be noted that the system can determine the candidate set corresponding to the multi-dimensional parameter field based on the preset parameter field range. The parameter field candidate set refers to the preset value range or candidate set of parameters in each dimension of the multi-dimensional parameter field. Among them, the rare earth doping element type candidate set includes a variety of lanthanide elements that can be selected; the doping concentration candidate set includes multiple candidate concentration values divided into a certain step size from the lowest value to the highest value; the preparation process parameter candidate set may include sintering temperature candidate set, holding time candidate set, and sintering atmosphere candidate set, etc.
[0060] Then, the elements and parameter values in the candidate set of parameter fields are paired one by one to generate all possible permutations and combinations, resulting in multiple solid electrolyte design schemes to be predicted.
[0061] Based on this, by constructing a candidate set based on the preset parameter field range and traversing and combining them to generate the design scheme to be predicted, the efficiency and comprehensiveness of the scheme design can be further improved.
[0062] In some embodiments, a corresponding solid electrolyte sample is prepared based on the multidimensional parameter fields in the current design scheme, including: Using an automated preparation system, the ingredients are sequentially prepared, mixed, shaped, and sintered based on the multidimensional parameter fields in the current design scheme to obtain a solid electrolyte sample corresponding to the current design scheme.
[0063] It should be noted that the system can parse the types of rare earth doping elements, doping concentrations, and values of various preparation process parameters from the current design scheme, and generate an automated preparation instruction sequence accordingly.
[0064] like Figure 2 As shown, exemplarily, the automated preparation system (i.e., the automated preparation module of the closed-loop design system) consists of four parts: batching, mixing, molding, and sintering, as detailed below: 1. Batching system: Based on the recommended formula, the system automatically weighs Li source (LiOH·H2O, Li2CO3), La2O3, ZrO2 and rare earth oxides and other precursor materials in proportion, using a precision balance and mechanical gripper for control; Specifically, after receiving parameters from the control console, the batching machine automatically weighs the lithium source (such as lithium hydroxide monohydrate, lithium carbonate), lanthanum oxide, zirconium oxide and corresponding rare earth oxides and other precursor materials according to the formula ratio through a precision balance and a robotic gripper.
[0065] 2. Mixing system: A high-energy vibrating ball mill is used for uniform mixing. Different frequencies and times can be set to ensure uniform mixing of raw materials and reduce powder radius. Specifically, the weighed raw materials are automatically transferred to a ball mill / crusher; the ball mill / crusher is started according to a preset time and rate to perform high-energy mixing treatment on the raw materials, so that the raw materials are mixed to form uniform micron-sized powder, which meets the requirements of subsequent molding.
[0066] 3. Molding system: including automatic assembly / demolding machine and tablet press, which compresses powder raw materials into round tablets with a diameter of 10-20mm.
[0067] Specifically, a robotic arm (equipped with flexible grippers) picks up the mixed powder in the ball mill / crusher and transfers it to a mold loading / demolding machine; the mold loading / demolding machine automatically completes mold assembly, receives the powder and accurately fills it into the mold, and then transfers the mold to a tablet press; the tablet press applies pressure to the powder in the mold according to a preset pressure to form a tablet sample; after pressing is completed, the mold loading / demolding machine starts the demolding program, and the robotic arm simultaneously picks up the demolded sample.
[0068] 4. Sintering System: Equipped with a programmable automated tube furnace, which allows setting the heating rate, holding time, and atmosphere (air, O2, Ar, etc.), and uses sensors to provide feedback on the temperature gradient for real-time adjustment. Specifically, the robotic arm grasps the sheet-like sample output from the molding system and temporarily stores it in a crucible on a crucible rack. The robotic arm then transfers the crucible containing the sample from the crucible rack to the sintering furnace. The sintering furnace starts the sintering process according to the parameters set on the control panel (e.g., 1250-1280℃, heating rate 5-10℃ / min, holding time 2-6h, air / oxygen atmosphere). Built-in sensors provide real-time feedback on the temperature gradient, and the control panel dynamically adjusts the temperature control parameters. After sintering, the robotic arm removes the crucible and returns it to the crucible rack for cooling. Once the sample has cooled to room temperature under a protective atmosphere, a solid electrolyte sample corresponding to the current design is obtained.
[0069] Based on this, by using an automated preparation system to sequentially carry out batching, mixing, molding and sintering, it is possible to achieve standardized and reproducible preparation from design scheme to solid electrolyte sample, thereby improving the consistency and efficiency of sample preparation.
[0070] In some embodiments, the performance prediction model is constructed in the following ways: Sample datasets of solid electrolyte materials were obtained from a multi-source database; each sample in the dataset contains multidimensional parameter fields and corresponding performance index parameters. The pre-built initial machine learning model is trained and cross-validated using a sample dataset to obtain a trained performance prediction model; wherein the initial machine learning model includes at least one of the following: random forest model, gradient boosting regression model, support vector regression model, and Bayesian neural network model.
[0071] It should be noted that multi-source databases mainly come from the following channels: 1. Open-source experimental databases, such as Materials Project (MP), Open Quantum Materials Database (OQMD), and AFLOW, can be used to obtain intrinsic information about compounds; 2. Extract information such as experimental formulations and preparation processes of LLZO and its rare-earth doped variants from published literature in this field.
[0072] The process of building a performance prediction model is as follows: First, a sample dataset of solid electrolyte materials is obtained from a multi-source database. Each sample in the dataset contains a complete set of multidimensional parameter fields (as model input) and corresponding measured values of performance indicators (as model output labels). For example, the multidimensional parameter fields in the database may include crystallographic parameters (lattice constant, symmetry, space group, etc.), elemental physicochemical parameters (atomic radius, relative mass, electronegativity, first ionization energy, affinity, etc.), and preparation process parameters (sintering atmosphere, sintering temperature, holding time, test temperature, etc.); performance indicator parameters may include ionic conductivity, activation energy, phase type, and phase purity.
[0073] Then, the sample dataset is preprocessed, which includes at least one of the following operations: missing data handling; outlier handling; numerical encoding of qualitative parameters; normalization or interval mapping of parameters with different dimensions. The preprocessed sample data is used for subsequent model training.
[0074] The pre-processed sample dataset is used to train a pre-built initial machine learning model. Multidimensional parameter fields are used as model input, and performance metric parameters are used as model output. Optimization algorithms such as backpropagation are used to adjust the model parameters, minimizing the error between the model output and the sample labels. During training, cross-validation can be used to evaluate the model's generalization ability. For example, five-fold cross-validation can be used, dividing the sample dataset into five subsets. One subset is used as the validation set, and the other four subsets are used as the training set in turn. This process is repeated five times, and the average validation error is used as the model's evaluation metric. When the model's prediction error meets a preset accuracy requirement (e.g., mean absolute percentage error on the validation set ≤ 10%), training stops, and the trained performance prediction model is obtained.
[0075] It should be noted that the initial machine learning model can be one or more of the following: Random Forest, Gradient Boosting Regression, Support Vector Regression, and Bayesian Neural Network. Random Forest is an ensemble learning model based on multiple decision trees, suitable for small to medium-sized sample data, and still exhibits high predictive stability and robustness even with noise or incomplete parameters. Gradient Boosting Regression constructs a prediction model by iteratively optimizing an ensemble of multiple weak learners, possessing strong fitting ability for complex nonlinear relationships, suitable for high-dimensional parameter input scenarios, and can output high-precision performance prediction results. Support Vector Regression is based on kernel functions to achieve nonlinear regression modeling, suitable for prediction tasks with a small number of samples and high parameter dimensionality. Bayesian Neural Network introduces model parameters in the form of probability distributions to evaluate the uncertainty of prediction results, thus characterizing the reliability of performance prediction results.
[0076] Based on this, the reliability and generalization ability of the model prediction are improved by training the initial machine learning model with sample data containing multi-dimensional parameter fields and corresponding performance indicators obtained from multi-source databases.
[0077] In some embodiments, the categories of performance index parameters include ionic conductivity, activation energy, phase type, and phase purity; The pre-built initial machine learning model is trained and cross-validated using a sample dataset to obtain a trained performance prediction model, including: For different categories of performance index parameters, corresponding candidate model sets are established; for continuous performance indexes, corresponding regression model sets are established, and for categorical performance indexes, corresponding classification model sets are established; each candidate model set includes multiple models such as random forest model, gradient boosting regression model, support vector regression model, and Bayesian neural network model. For each set of candidate models, the candidate model with the best performance is selected as the final prediction model corresponding to that performance index parameter based on at least one of the following evaluation metrics: validation set error, classification accuracy, stability, and generalization ability.
[0078] It should be noted that performance indicators can be divided into continuous performance indicators and categorical performance indicators. Continuous performance indicators refer to performance parameters with continuous values, such as ionic conductivity (unit: S / cm, value: 6.8 × 10⁻⁶). -4 7.1×10 -4 Continuous values are used to predict continuous variables, and activation energy is used to predict categorical variables (unit: eV, values can be continuous values such as 0.35, 0.38, etc.). Categorical performance indicators refer to performance parameters that take discrete categories or levels, such as phase category (cubic phase, tetragonal phase, or a mixture of both), phase purity level (high purity, medium purity, low purity, etc.), and impurity risk (high risk, medium risk, low risk). Correspondingly, regression models are used to predict continuous variables, and their output is a specific numerical value; classification models are used to predict categorical variables, and their output is a category label or category probability.
[0079] In this embodiment, the system first classifies the performance index parameters: ionic conductivity and activation energy are classified as continuous performance indicators, and a regression prediction task is established accordingly; phase type, phase purity level and impurity risk are classified as categorized performance indicators, and a classification prediction task is established accordingly.
[0080] For regression prediction tasks, a candidate regression model set is established, which can select from multiple models such as random forest regression, gradient boosting regression, support vector regression, and Bayesian neural network regression. Each regression model is trained using training sample data, and then the model with the best overall performance is selected as the final prediction model for the corresponding prediction task (including ionic conductivity and activation energy) by comparing at least one evaluation metric of each model on the validation set.
[0081] The evaluation metrics may include at least one of validation set error, classification accuracy, stability, and generalization ability. Specifically, validation set error is a metric for evaluating the prediction accuracy of the regression model, such as root mean square error or mean absolute percentage error; classification accuracy is a metric for evaluating the prediction accuracy of the classification model; stability and generalization ability reflect the model's consistency in performance on different subsets of data and its ability to predict unseen data, respectively.
[0082] For classification prediction tasks, a candidate classification model set is established, which can include various models such as random forest, gradient boosting, support vector, and Bayesian neural network. Each classification model is trained using training sample data. Then, by comparing at least one evaluation metric of each model on the validation set, the model with the best overall performance is selected as the final prediction model for the corresponding prediction task (including phase category, phase purity level, and risk of impurities).
[0083] Based on this, by constructing candidate model sets for different categories of indicators and selecting the optimal model from each set based on multiple evaluation indicators, the accuracy and reliability of performance prediction can be further improved.
[0084] In some embodiments, adaptive iterative adjustment of the multidimensional parameter fields of the current design scheme based on the comparison deviation includes: If the comparison deviation is greater than the preset deviation threshold, the adjustment strategy will be executed in the following priority order: First priority: Adjust the doping concentration; Second priority: Change the type of rare earth doping element; Third priority: Adjust the sintering temperature and / or holding time in the preparation process parameters.
[0085] It should be noted that the comparison deviation refers to the degree of difference between the measured performance index and the performance index predicted by the model for the current design scheme. When the difference between the measured value and the predicted value of a certain performance index exceeds a preset deviation threshold, the parameter adjustment strategy is triggered.
[0086] For example, this embodiment executes the adjustment strategy according to a specific priority order: the first priority is to adjust the doping concentration, the second priority is to change the type of rare earth dopant, and the third priority is to adjust the sintering temperature or holding time in the preparation process parameters. It should be understood that the doping concentration is the most direct factor determining the crystal structure and lithium-ion migration channels of the material, and should be adjusted first; if the adjustment of the doping concentration cannot converge, it indicates that the doping element system itself may not be suitable for the target system, and the doping element should be replaced; only lastly are process parameters such as sintering temperature and holding time considered.
[0087] Specifically, adjusting the doping concentration includes: Keeping the types of rare earth doping elements and preparation process parameters unchanged in the current design scheme, only adjusting the doping mole fraction in increments of ±0.05 mol% to generate multiple candidate doping amounts. The new design scheme corresponding to each candidate doping amount is then used as the "current design scheme" for the next round, and S3-S5 are repeated.
[0088] Changing the types of rare earth doping elements includes: If the comparison deviation fails to converge after multiple iterations following adjustments to the doping concentration, then fix the doping mole fraction (e.g., at 10 mol%) and replace the original doping element with another metal element of the same valence state (e.g., M). n+ This led to the development of several new design schemes.
[0089] Adjusting the sintering temperature and / or holding time in the preparation process parameters, including: If the comparison deviation still fails to converge after multiple iterations of changing the dopant element type, then keep the dopant element type and dopant concentration in the current design unchanged, and adjust the sintering temperature in 50℃ increments within the range of 900℃ to 1300℃ (for example, adjust from 1250℃ to 1200℃, 1250℃, 1300℃), while setting the holding time to discrete combinations of 2h, 4h, 6h, and 8h to form multiple new design schemes.
[0090] It should be noted that the adjustment strategies of the three priorities mentioned above are executed sequentially, with each subsequent priority being activated only if the previous priority has been adjusted multiple times without success. If the comparison deviation is less than a preset threshold after any strategy is executed, the adjustment stops and the final solution is output; if convergence is still not achieved after the third priority adjustment, the system can trigger an alarm, prompting manual intervention or automatically switching to a backup design solution.
[0091] Based on this, the overall efficiency of the closed-loop iteration can be further improved by executing each adjustment strategy in a specific priority order according to its characteristics.
[0092] like Figure 2As shown, the following specific examples illustrate the solution of this application: The closed-loop design system for rare-earth-doped solid electrolytes in this application mainly includes a material design module, an automated preparation module, an online characterization module, an intelligent feedback optimization module, and a control platform. Each module achieves data interoperability and task linkage through an industrial Internet of Things (IoT) architecture, forming a closed-loop R&D system for rare-earth-doped solid electrolytes. Specifically: (1) Materials Design Module: This is used to integrate open-source experimental databases, literature, and historical experimental data to build a multi-source database, and to train machine learning models based on this database. It recommends rare earth doping element types, doping concentrations, and corresponding preparation process parameters according to preset target performance.
[0093] The multi-source database includes at least the following data fields: crystallographic parameters, elemental physicochemical parameters, preparation process parameters, and performance index parameters; among which, the performance index parameters include ionic conductivity, activation energy, phase type, and phase purity.
[0094] The machine learning model used can be one or more of the following: random forest model, gradient boosting regression model, support vector regression model, and Bayesian neural network model. The model is trained on preprocessed sample data to establish the mapping relationship between doping schemes, process parameters (multi-dimensional parameter fields) and performance indicators, and outputs candidate design schemes and their credibility evaluation.
[0095] (2) Automated preparation module: The automated preparation module includes a batching system, a mixing system, a molding system, and a sintering system. The batching system automatically weighs precursors such as lithium source, lanthanum oxide, zirconium oxide, and rare earth oxides according to a recommended scheme; the mixing system uses ball milling to improve the uniformity of raw materials; the molding system completes mold loading, tableting, and demolding; and the sintering system performs programmable sintering of the samples according to the set heating rate, holding time, and atmosphere conditions.
[0096] In one embodiment, the batching system uses a precision balance and a mechanical gripper to automatically weigh the powder; the molding system presses the powder into discs with a diameter of 10 mm to 20 mm; and the sintering system supports high-precision temperature control under air, oxygen, or inert atmosphere conditions.
[0097] (3) Online characterization module: The online characterization module is used to automatically transfer sintered samples to an X-ray diffraction platform and perform rapid analysis of their crystal phase composition. This module includes a flexible adaptive robotic arm, a high-speed XRD platform, and an automatic spectrum interpretation unit.
[0098] It should be noted that the sample transfer by the robotic arm faces the following technical challenges: 1. LLZO ceramic samples are mostly small-sized sheet-like or columnar structures with fragile edges, brittle and hard texture, rough surface, and irregular shape, which makes conventional rigid clamps unsuitable and poses a risk of slippage and breakage; 2. Some rare earth-doped LLZO samples have hygroscopic and easily pulverized problems, requiring controlled humidity in the transfer environment (RH<20%), and the contact parts of the robotic arm must be made of corrosion-resistant, non-absorbent materials (such as PTFE pads).
[0099] Therefore, considering the characteristics of solid electrolyte samples—brittleness, irregular edges, and high hygroscopicity in some systems—the robotic arm preferably employs flexible grippers, real-time force feedback, and a humidity-controlled transport environment. Specifically: 1) Flexible adaptive gripper: It adopts a three-finger pneumatic flexible gripper with an embedded strain sensor and vision assistance system to adaptively position itself according to the edge morphology of the sample. 2) Real-time clamping force feedback: A closed-loop force control unit is introduced, combined with control algorithms (such as impedance control or fuzzy control), to automatically adjust the clamping force to the set range (such as 0.3–0.5 N) to prevent breakage; 3) Obstacle avoidance path planning: Use real-time path planning algorithms (such as RRT* or DWA) in conjunction with camera recognition of the precise location of the sample tray to improve the success rate of grasping and positioning accuracy.
[0100] The XRD platform preferentially employs a rapid θ-2θ scanning method. The scanned spectra can be automatically analyzed and feature peaks extracted using a pre-trained CNN model, and compared with a database to identify the proportions of cubic, tetragonal, and impurity phases. It should be noted that the high-speed θ-2θ scanning module can complete crystalline phase quantification for each sample within 3 minutes; the accompanying anti-vibration structure avoids mechanical vibration interference with signal acquisition; and the XRD optical platform features automatic focusing and horizontal adjustment to adapt to focus alignment for samples of varying thicknesses, ensuring high-quality spectra.
[0101] (4) Intelligent feedback optimization module: The intelligent feedback optimization module compares the online characterization results with the machine learning prediction results and automatically adjusts the experimental parameters for the next round based on the deviations in ionic conductivity, activation energy, and phase purity.
[0102] In one embodiment, if the deviation between the measured value and the predicted value exceeds a preset threshold, the step size of the doping amount is adjusted first; if the target index still cannot be met after adjustment, a candidate doping element with the same valence state or similar radius is replaced; then the sintering temperature and holding time are optimized in combination until the error converges to the set range.
[0103] (5) Control platform: The control platform, based on an industrial IoT architecture, is used to uniformly receive target performance data, schedule various sub-modules, record equipment status and experimental data, and display the formula-performance relationship, iterative convergence curves, and abnormal alarm information through a visual interface. The control platform also enables remote monitoring, access control, and experimental report output.
[0104] Specifically, the control platform can integrate data acquisition, analysis and management functions, display key performance indicators and optimization progress through a human-machine interface, and realize remote monitoring and operation of the system.
[0105] 1. Based on the Industrial Internet of Things (IIoT) architecture, the control platform can achieve the following functions: 1) Data acquisition: Embedded sensors acquire information such as temperature, pressure, counterweight, and spectral data; 2) Data storage and synchronization: It adopts an edge node + cloud database architecture and supports standardized JSON format interfaces; 3) Visual interface: Real-time display of formula-performance relationship graphs, iterative convergence curves, sample performance trends, etc. 4) Control command execution: One-click start / stop / interruption of each submodule, with remote access control capability; 5) Abnormal alarm mechanism: Automatically triggers early warning in cases of over-temperature / error deviation, etc.
[0106] 2. Given the unique properties of rare earth oxides (such as strong hygroscopicity, high-temperature volatilization, and narrow sintering window), this system possesses unique adaptability in the following aspects: 1) The humidity control module can maintain the environment of the material transfer and storage area at RH≤15%; 2) The robotic arm and process area use components resistant to water vapor corrosion, and the gripper material is made of ceramic coating; 3) The sintering system supports high-precision temperature control (±1°C) and rapid heating and cooling strategies to prevent rare earth doping imbalance.
[0107] It should be noted that, compared with the prior art, the embodiments of this application have the following beneficial effects: 1. Closed-loop collaboration throughout the entire process: It constructs a complete closed loop of formula design, automated preparation, online characterization, and intelligent feedback, breaking through the limitations of existing technologies in optimizing single links or fragmented processes, realizing data linkage and seamless connection of each link, and greatly improving the continuity of R&D and the efficiency of implementation.
[0108] 2. Intelligent-driven precision and efficiency: Integrating multi-source data to build machine learning models, accurately recommending rare earth doping schemes and process parameters, and combining automated preparation and high-speed online characterization, it breaks through the dilemma of traditional manual experience screening and repeated experiments, significantly shortening the R&D cycle and improving parameter matching accuracy.
[0109] 3. Dedicated Adaptability: In response to the brittle and fragile nature of garnet-type lithium lanthanum zirconium oxide samples and the strong hygroscopicity of rare earth oxides, a dedicated solution including flexible grippers, humidity control, and high-precision temperature control is designed to solve problems such as sample breakage and doping imbalance caused by insufficient adaptability of existing technologies, and to ensure the consistency of material performance.
[0110] Please refer to Figure 2 , Figure 2 This paper illustrates a block diagram of a closed-loop design apparatus for a rare-earth-doped solid electrolyte provided in some embodiments of this application. It should be understood that this closed-loop design apparatus for the rare-earth-doped solid electrolyte is similar to that described above. Figure 1 Corresponding to the method embodiments, it is able to perform each step involved in the above method embodiments. The specific function of the closed-loop design device for the rare earth-doped solid electrolyte can be found in the description above. To avoid repetition, detailed descriptions are appropriately omitted here.
[0111] Figure 2 The closed-loop design device for the rare-earth-doped solid electrolyte includes at least one software function module that can be stored in a memory or embedded in the closed-loop design device of the rare-earth-doped solid electrolyte in the form of software or firmware. The closed-loop design device for the rare-earth-doped solid electrolyte includes: The performance prediction module 310 is used to input multiple solid electrolyte design schemes to be predicted into the performance prediction model to obtain the performance index prediction results corresponding to each solid electrolyte design scheme; wherein, each solid electrolyte design scheme to be predicted includes multi-dimensional parameter fields, including rare earth doping element type, doping concentration and preparation process parameters. The scheme screening module 320 is used to screen out the solid electrolyte design schemes whose performance index prediction results meet the preset target performance conditions as candidate design schemes, and determine the current design scheme based on the candidate design schemes. The sample preparation module 330 is used to prepare corresponding solid electrolyte samples based on the multidimensional parameter fields in the current design scheme. The performance testing module 340 is used to obtain the measured performance indicators of solid electrolyte samples. The scheme optimization module 350 is used to compare the measured performance indicators with the predicted performance indicators of the current design scheme, and to adaptively iterate and adjust the multi-dimensional parameter fields of the current design scheme according to the comparison deviation, until the comparison deviation between the measured performance indicators and the corresponding predicted performance indicators is less than a preset deviation threshold, thus obtaining the final solid electrolyte target design scheme.
[0112] It is understood that the above-described device embodiments correspond to the method embodiments of the present invention. The closed-loop design device for rare earth-doped solid electrolytes provided by the embodiments of the present invention can realize the closed-loop design method for rare earth-doped solid electrolytes provided by any one of the method embodiments of the present invention.
[0113] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the device described above can be referred to the corresponding process in the aforementioned method, and will not be elaborated further here.
[0114] like Figure 4 As shown, some embodiments of this application provide an electronic device 400, which includes: a memory 410, a processor 420, and a computer program stored in the memory 410 and executable on the processor 420. When the processor 420 reads the program from the memory 410 via a bus 430 and executes the program, it can implement any of the methods included in the above-described closed-loop design method for rare earth-doped solid electrolytes.
[0115] Processor 420 can process digital signals and may include various computing architectures. For example, it may be a complex instruction set computer architecture, a reduced instruction set computer architecture, or an architecture that implements multiple instruction set combinations. In some examples, processor 420 may be a microprocessor.
[0116] Memory 410 can be used to store instructions executed by processor 420 or data related to the execution of instructions. These instructions and / or data may include code for implementing some or all of the functions of one or more modules described in the embodiments of this application. The processor 420 of this disclosure embodiment can be used to execute instructions in memory 410 to implement the methods shown above. Memory 410 includes dynamic random access memory, static random access memory, flash memory, optical memory, or other memories well known to those skilled in the art.
[0117] Some embodiments of this application also provide a computer-readable storage medium storing a computer program that, when executed by a processor, describes the method described in the method embodiments.
[0118] Some embodiments of this application also provide a computer program product that, when run on a computer, causes the computer to perform the method described in the method embodiments.
[0119] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For apparatus embodiments, since they are basically similar to method embodiments, the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.
[0120] It should be understood, in the several embodiments provided in this application, that the disclosed apparatus and methods can also be implemented in other ways. The apparatus embodiments described above are merely illustrative; for example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.
[0121] In addition, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.
[0122] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0123] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application. It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.
[0124] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
[0125] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
Claims
1. A closed-loop design method for rare-earth-doped solid electrolytes, characterized in that, include: Multiple solid electrolyte design schemes to be predicted are input into the performance prediction model to obtain the performance index prediction results corresponding to each solid electrolyte design scheme; wherein, each solid electrolyte design scheme to be predicted includes a multi-dimensional parameter field, which includes the type of rare earth doping element, doping concentration and preparation process parameters. The solid electrolyte design schemes whose performance index prediction results meet the preset target performance conditions are selected as candidate design schemes, and the current design scheme is determined based on the candidate design schemes. Prepare corresponding solid electrolyte samples based on the multidimensional parameter fields in the current design scheme; Obtain the measured performance indicators of the solid electrolyte sample; The measured performance indicators are compared with the predicted performance indicators of the current design scheme, and the multi-dimensional parameter fields of the current design scheme are adaptively and iteratively adjusted according to the comparison deviation until the comparison deviation between the measured performance indicators and the corresponding predicted performance indicators is less than a preset deviation threshold, thus obtaining the final solid electrolyte target design scheme.
2. The closed-loop design method for rare earth-doped solid electrolytes according to claim 1, characterized in that, The process of selecting solid electrolyte design schemes whose predicted performance indicators meet preset target performance conditions as candidate design schemes, and determining the current design scheme based on the candidate design schemes, includes: Solid electrolyte design schemes whose performance index prediction results meet the preset target performance conditions are selected as candidate design schemes, and the credibility score corresponding to each candidate design scheme is obtained; wherein, the credibility score is calculated by combining the model prediction uncertainty, cross-validation error, distance between candidate samples and training set distribution, consistency of multi-model output and number of historical similar cases; The current design scheme is selected from the candidate design schemes by combining the credibility score.
3. The closed-loop design method for rare earth-doped solid electrolytes according to claim 1, characterized in that, Also includes: When the number of iterations of the adaptive iterative adjustment exceeds a preset threshold, the backup design scheme is used as the current design scheme in the next round; wherein, the backup design scheme is the design scheme with the highest performance index prediction result among the solid electrolyte design schemes to be predicted, excluding the candidate design scheme.
4. The closed-loop design method for rare earth-doped solid electrolytes according to claim 1, characterized in that, Before inputting the multiple solid electrolyte design schemes to be predicted into the performance prediction model, the method further includes: The parameter field candidate set corresponding to the multidimensional parameter field is determined based on the preset parameter field range; wherein, the parameter field candidate set includes a rare earth doping element type candidate set, a doping concentration candidate set, and a preparation process parameter candidate set; The elements and parameter values in each of the candidate sets of the parameter fields are traversed and combined to generate the multiple solid electrolyte design schemes to be predicted.
5. The closed-loop design method for rare earth-doped solid electrolytes according to claim 1, characterized in that, The preparation of the corresponding solid electrolyte sample based on the multidimensional parameter fields in the current design scheme includes: Using an automated preparation system, the ingredients are sequentially prepared, mixed, shaped, and sintered based on the multidimensional parameter fields in the current design scheme to obtain a solid electrolyte sample corresponding to the current design scheme.
6. The closed-loop design method for rare earth-doped solid electrolytes according to claim 1, characterized in that, The performance prediction model is constructed in the following ways: A sample dataset of solid electrolyte materials is obtained from a multi-source database; wherein, each sample data in the sample dataset contains the multidimensional parameter field and the corresponding performance index parameter; The pre-constructed initial machine learning model is trained and cross-validated using a sample dataset to obtain a trained performance prediction model; wherein the initial machine learning model includes at least one of the following: random forest model, gradient boosting regression model, support vector regression model, and Bayesian neural network model.
7. The closed-loop design method for rare-earth-doped solid electrolytes according to claim 6, characterized in that, The categories of performance index parameters include ionic conductivity, activation energy, phase type, and phase purity; The process of training and cross-validating a pre-built initial machine learning model using a sample dataset to obtain a trained performance prediction model includes: For different categories of performance index parameters, corresponding candidate model sets are established; among them, a corresponding regression model set is established for continuous performance indexes, and a corresponding classification model set is established for categorical performance indexes; each candidate model set includes multiple models such as random forest model, gradient boosting regression model, support vector regression model, and Bayesian neural network model. For each set of candidate models, the candidate model with the best performance is selected as the final prediction model corresponding to that performance index parameter based on at least one of the following evaluation metrics: validation set error, classification accuracy, stability, and generalization ability.
8. The closed-loop design method for rare-earth-doped solid electrolytes according to any one of claims 1 to 7, characterized in that, The adaptive iterative adjustment of the multidimensional parameter fields of the current design scheme based on the comparison deviation includes: If the comparison deviation is greater than a preset deviation threshold, the adjustment strategy will be executed in the following priority order: First priority: Adjust the doping concentration; Second priority: Change the type of rare earth doping element; Third priority: Adjust the sintering temperature and / or holding time in the preparation process parameters.
9. A closed-loop design device for a rare-earth-doped solid electrolyte, characterized in that, include: The performance prediction module is used to input multiple solid electrolyte design schemes to be predicted into the performance prediction model to obtain the performance index prediction results corresponding to each solid electrolyte design scheme; wherein, each solid electrolyte design scheme to be predicted includes a multi-dimensional parameter field, which includes the type of rare earth doping element, the doping concentration, and the preparation process parameters. The scheme screening module is used to screen out the solid electrolyte design schemes whose performance index prediction results meet the preset target performance conditions as candidate design schemes, and determine the current design scheme based on the candidate design schemes. The sample preparation module is used to prepare corresponding solid electrolyte samples based on the multidimensional parameter fields in the current design scheme. The performance testing module is used to obtain the measured performance indicators of the solid electrolyte sample. The scheme optimization module is used to compare the measured performance indicators with the predicted performance indicators of the current design scheme, and to adaptively iteratively adjust the multi-dimensional parameter fields of the current design scheme according to the comparison deviation, until the comparison deviation between the measured performance indicators and the corresponding predicted performance indicators is less than a preset deviation threshold, thereby obtaining the final solid electrolyte target design scheme.
10. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, it can implement the closed-loop design method of the rare earth-doped solid electrolyte according to any one of claims 1-8.
11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, performs the closed-loop design method for rare-earth-doped solid electrolytes as described in any one of claims 1-8.
12. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the closed-loop design method for rare-earth-doped solid electrolytes according to any one of claims 1-8.
