System to predict pressure-composition-temperature curve for metal alloys used for solid-state hydrogen storage

Abstract

The present invention provides a technical solution in the form of a machine learning (ML) based system and a machine learning (ML) based method to predict a Pressure-Composition-Temperature (PCT) curve for a metal alloy to be used / implemented for solid-state hydrogen storage. The instant invention aims to solve the problems associated with determining the suitability of an alloy for solid state hydrogen storage, specifically by predicting the PCT curve that enables to determine the suitability of an alloy for solid state hydrogen storage, Accordingly, aspects of the present invention relate to a system and methods to analyze, predict and generate PCT (pressure-composition- temperature) curves for metal alloys w.r.t storage of H2 in solid-state, the system and methods provided herein predict such details using machine learning based mode.

Description

[0001] SYSTEM TO PREDICT PRESSURE-COMPOSITION-TEMPERATURE CURVE FOR METAL ALLOYS USED FOR SOLID-STATE HYDROGEN STORAGE

[0002] FIELD OF THE INVENTION

[0003] The present invention relates to a system for predicting pressure-composition-temperature (PCT) curve of metal alloys used to decide their suitability for solid-state hydrogen storage. In particular, the invention pertains to the field of hydrogen storage, and more particularly to a machine learning (ML) based system and a machine learning (ML) based method to predict a pressure-composition- temperature curve for a metal alloy to be used / implemented for solid-state hydrogen storage. The invention finds immense application in the renewable energy sector and alternative fuels, especially solid-state hydrogen storage. It shall help attain the 7thsustainable development goal of affordable and clean energy.

[0004] BACKGROUND OF THE INVENTION AND DESCRIPTION OF PRIOR ART

[0005] Solid-state hydrogen storage in metal alloys has emerged as a promising approach for achieving high-capacity, safe, and long-term hydrogen storage. Extensive research has been conducted on various metal alloys that can form hydrides through hydrogenation. However, the practical application of these alloys is hindered by their unattainable thermodynamic and kinetic hydrogen storage properties at ambient temperatures. The hydrogen storage properties of these alloys are highly dependent on their chemical composition. This presents a vast chemical space that needs to be explored to discover efficient solid-state hydrogen storage materials. Unfortunately, experimental analysis of the hydrogen storage properties of alloy compositions is resourceintensive and time-consuming, limiting the number of compositions that can be thoroughly investigated.

[0006] Understanding the thermodynamic properties is crucial for determining the operating conditions and hence feasibility of hydride systems for their practical applicability. These properties are typically measured through Pressure-Composition-Temperature (PCT) analysis. A typical PCT experiment involves analyzing response of the material to applied temperatures and hydrogen pressure. It provides valuable information such as hydrogen pressure for phase transition, maximum reversible hydrogen storage capacity, and the pressure at which hydrogen will be absorbed or released (i.e., plateau pressure) at a given temperature. However, conducting a PCT analysis for every available compound is practically impossible and time-consuming. Therefore, there is a need for a rapid characterization technique to expedite the search for new materials in this field.

[0007] Experimentally, PCT analysis generates a significant amount of data, which has been reported in the literature as PCT plots. One potential solution is the development of ML driven PCT analysis model. This model would be trained on existing PCT data and would learn the relationships and patterns between input parameters (such as temperature, pressure, and material composition) and the corresponding hydrogen content (PCT curve).

[0008] Reference may be made to the article by Jeong Min Kim et al., Metals and Materials International, volume 29, pages 861-869, published on 9thJuly 2022 which reports PCT curve predictions for some alloys using machine learning models, specifically based on three machine learning models such as random forest (RF), K-nearest neighbor (KNN) and deep neural network (DNN). These models are trained on the data specific to a particular family of alloys (AB2) consisting of 66 PCT curves and eight elements. Whereas, the present invention MH-PCTpro is trained on 300+ PCT curves of different types of alloys (AB5, AB2, AB, A2B, SS, and inter-metallics) and includes 30 elements. It can predict PCTs for alloys consisting of any of these 30 elements as constituents. Further, the study by Kim et al predicts PCT in a specific temperature window of 300-333K, contrary to the present model, which can predict in the temperature window of 298K- 700K. The ML model invented in the instant invention and features thereof are also distinctly different than the once used by Kim et. al. The present invention also include Active-PCTpro ML model which is trained on small datasets to predict PCTs. The Active-PCTpro utilize ML model uncertainty to improve model predictions in an iterative loop.

[0009] To summarize, in order to determine the suitability of an alloy for solid-state hydrogen storage, the Pressure-Composition-Temperature (PCT) measurements are extremely important. What is noted is that there is no generalized model available in literature to determine the suitability of an alloy for solid state hydrogen storage specifically by determining / predicting the PCT curve

[0010] To solve the above recited and associated problems of determining the suitability of an alloy for solid state hydrogen storage, specifically by predicting the PCT curve that enables to determine the suitability of an alloy for solid state hydrogen storage, the present invention provides a technical solution in the form of a machine learning (ML) based system and a machine learning (ML) based method to predict a Pressure-Composition-Temperature (PCT) curve for any metal alloy to be used / implemented for storage of a solid-state hydrogen.

[0011] OBJECTIVES OF THE PRESENT INVENTION

[0012] The main objective of the present invention is therefore to provide a simple, compact, and efficient system for predicting pressure-composition-temperature (PCT) curve for metal alloys to determine their suitability for a solid-state hydrogen storage.

[0013] Another object of the present invention is to determine / predict PCT curves i.e., hydrogen absorption in view of a variable temperature and a variable pressure based on multiple features for a metal alloy combination.

[0014] Yet another object of the present invention is to predict suitability of metal alloys for hydrogen storage.

[0015] Still another object of the present invention is to provide a system that trains on a smaller set of data and use uncertainties from each prediction to improve the next prediction.

[0016] Yet another object of the present invention is to provide a system that iteratively improves model training by incorporating the experimental setup in a loop.

[0017] Another object of the present invention is to provide a cost-effective method to predict suitable alloy compositions at a desired pressure and temperature ranges for the solid-state hydrogen storage.

[0018] SUMMARY OF THE INVENTION

[0019] Aspects of the present disclosure relate generally to systems for analysing metal alloys. In particular, the present disclosure relates to a system for predicting pressure-composition-temperature (PCT) curve for metal alloys used for solid-state hydrogen storage. Accordingly, a general aspect of the present invention relates to a system and methods to analyze, predict and generate PCT (pressure composition temperature) curves for metal alloys, the system and methods provided predict such details using machine learning based model.

[0020] In an aspect, a system to predict a Pressure-Composition Temperature (PCT) isotherm for a metal alloy to predict a suitability of the metal alloy for hydrogen storage. The system includes a processor. The processor is in communication with a first learning model and a second learning model. The first learning model and the second learning model are pre-stored in the system. A server is coupled to a memory storing instructions executable by the processor. The processor is configured toobtain pre-fed PCT isotherm data. The pre-fed PCT isotherm data is associated with a plurality of metal alloy combinations. The plurality of metal alloy combinations is available from the server. The processor is configured to generate a dataset. The dataset includes weighted averages, hydriding properties, and one or more experimental parameters such as a predefined absorption temperature and a predefined hydrogen pressure. The dataset further comprises the weighted average enthalpy of hydride formation and the volume difference between the element's bulk phase and its hydride phase.

[0021] The processor is configured to train the first learning model using a standard learning model with the PCT isotherm data from the server (106) and the generated dataset.

[0022] The processor is configured to improve prediction results generated by the trained first learning model iteratively by incorporating feedback from human experts using the second learning model.

[0023] The processor is configured to validate the improved predicted results generated by the trained first learning model. The validation is done by comparing the predicted results with the one or more experimental parameters to ensure the prediction accuracy of the trained first learning module.

[0024] The processor is configured to predict PCT isotherms for the plurality of metal alloy combinations. The PCT isotherms are predicted as a function of the predefined hydrogen pressure and the predefined absorption temperature using the trained first learning model.

[0025] The processor is configured to capture temperature-dependent variations in plateau pressure using the predicted PCT isotherms.

[0026] The processor is configured to determine the enthalpy and entropy of hydride formation for the plurality of metal alloy combinations through Van't Hoff plots. The Van't Hoff plots can be generated from the predicted PCT isotherms. The Van't Hoff plots determine the suitability of a metal alloy combination from the plurality of metal alloy combinations fit for storing hydrogen. In an embodiment, the processor using the first learning model may be configured to predict PCT isotherms for the metal alloy compositions. The PCT isotherms containing unseen elements not present in the pre-fed PCT isotherm data may be predicted.

[0027] In an embodiment, the dataset may include data with twenty-eight distinct features. The distinct features with collinearity values below 0.9 may be selected for the training the first learning model.

[0028] In an embodiment, the processor using the trained first learning model may be configured to generate an average Mean Absolute Error (MAE) of 0.17 ± 0.002 weight percentage (wt%) and an R2score of 0.96 when validated on the pre-fed PCT isotherm data from plurality of metal alloy combinations.

[0029] In an embodiment, the plurality of metal alloy combinations may be selected from a group comprising light metals, hydriding and non-hydriding elements, including transition metals, alkali metals, alkaline earth metals, non-metals, and lanthanides.

[0030] In an embodiment, the second learning model may be configured within the processor. The processor using the second learning model may be configured to select the most informative data points to be labeled by an expert. The second learning model may be configured to select one or more data points for which the first learning model may be least confident in its predictions. The second learning model may label the selected uncertain data points upon receiving feedback from human experts. The second learning model may train the first learning model using the labeled uncertain data points to improve the prediction results of the first learning model for metal alloy combinations in which it is previously uncertain.

[0031] In an embodiment, the standard learning model may be an Extra Trees Regression (ETR) algorithm.

[0032] In an embodiment, the predefined absorption temperature may be selected in the range of 200K to 700K. The predefined hydrogen pressure may be selected between 0.001 MPa to 100 MPa.

[0033] In an embodiment, a method for predicting Pressure-Composition-Temperature (PCT) isotherms for metal compositions to determine their suitability for hydrogen storage includes a step of obtaining a pre-fed PCT isotherm data associated with a plurality of metal alloy combinations from a server using a processor. The method includes a step of generating with a first learning model a dataset that includes weighted averages, hydriding properties, and one or more experimental parameters such as a predefined absorption temperature and a predefined hydrogen pressure using the processor. The dataset may further include the weighted average enthalpy of hydride formation and the volume difference between the element's bulk phase and its hydride phase. The method includes a step of training the first learning model with the PCT isotherm data from the server and the generated dataset using the processor with a standard learning model.

[0034] In an embodiment, the method includes a step of improving prediction results generated by the trained first learning model iteratively by incorporating feedback from human experts using a second learning model. The method includes a step of validating the improved predicted results generated by the trained first learning model by comparing the predicted results with the one or more experimental parameters to ensure the prediction accuracy of the trained first learning model using the processor.

[0035] The method includes a step of predicting with the trained first learning model the PCT isotherms for the plurality of metal alloy combinations as a function of the predefined hydrogen pressure and the predefined absorption temperature using the processor. The method include a step of capturing with the trained first learning model temperature-dependent variations in plateau pressure using the predicted PCT isotherms using the processor.

[0036] The method includes a step of determining with the trained first learning model (104) the enthalpy and entropy of hydride formation for the plurality of metal alloy combinations through Van't Hoff plots generated from the predicted PCT isotherms using the processor. The suitability of a metal alloy combination from the plurality of metal alloy combinations for storing hydrogen is determined.

[0037] Various objects, features, aspects, and advantages of the inventive subject matter will become more apparent from the following detailed description of preferred embodiments, along with the accompanying drawing figures in which like numerals represent like components.

[0038] BRIEF DESCRIPTION OF THE ACCOMPANYING DRAWINGS

[0039] Preferred representations of the non-invasive system of the present invention are described in detail below with reference to the drawings wherein: FIG. 1A illustrates a block diagram describing a system for predicting pressure-composition- temperature (PCT) curve for metal alloys used for solid-state hydrogen storage, in accordance with the present invention.

[0040] FIG. IB illustrates a flow diagram that describes working of the proposed system, in accordance with the present invention.

[0041] FIG. 2A illustrates a block diagram describing a method for predicting pressure-composition- temperature (PCT) curve for metal alloys used for solid-state hydrogen storage, in accordance with the present invention.

[0042] FIG. 2B illustrates an active learning framework of the proposed system for a used case, in accordance with the present invention.

[0043] FIG. 3A illustrates Comparison between ML-predicted (MH-PCTpro) and experimental PCT diagrams of LaMg3.70Ni1.i8 composition at 573 K in accordance with the present invention.

[0044] FIG. 3B illustrates Comparison between ML-predicted (PCTpro) and experimental PCT diagrams of La0.7Mg0.3Ni3.2Co0.35 composition at 298 K in accordance with the present invention.

[0045] FIG. 4A illustrates comparison between ML-predicted (Active-PCTpro) and experimental PCT diagrams of La5Mg9oNi5at 633K in accordance with the present invention.

[0046] FIG. 4B illustrates comparison between ML-predicted (Active-PCTpro) and experimental PCT diagrams of LaMg2Ni5at 514K in accordance with the present invention.

[0047] DETAILED DESCRIPTION OF THE INVENTION

[0048] The word "exemplary" or "embodiment" is used herein to mean "serving as an example, instance, or illustration." Any implementation or aspect described herein as "exemplary" or as an "embodiment" is not necessarily to be construed as preferred or advantageous over other aspects of the disclosure. Likewise, the term "aspects" does not require that all aspects of the disclosure include the discussed feature, advantage, or mode of operation.

[0049] Detailed embodiments of the claimed structures and methods are disclosed herein; however, it can be understood that the disclosed embodiments are merely illustrative of the claimed structures and methods that may be embodied in various forms. This invention may, however, be embodied in many different forms and should not be construed as limited to the exemplary embodiments set forth herein. Rather, these exemplary embodiments are provided so that this disclosure will be thorough and complete and will fully convey the scope of this invention to a person skilled in the art. In the description, details of well-known features and techniques may be omitted to avoid unnecessarily obscuring the presented embodiments.

[0050] The present invention may be a system, a method, and / or a device at any possible technical detail level of integration. The device may include a computer readable storage medium (or media) having computer readable program instructions thereon for causing a processor to carry out aspects of the present invention. Aspects of the present invention are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer readable program instructions.

[0051] The flowchart and block diagrams in the Figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of instructions, which comprises one or more executable instructions for implementing the specified logical function(s). In some alternative implementations, the functions noted in the blocks may occur out of the order noted in the Figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and / or flowchart illustration, and combinations of blocks in the block diagrams and / or flowchart illustration, can be implemented by special purpose hardware-based systems and smartphone devices that perform the specified functions or acts or carry out combinations of special purpose hardware, smartphone devices and computer instructions.

[0052] The present disclosure may be a system, a method, and / or a device at any possible technical detail level of integration. The device may include a computer readable storage medium (or media) having computer readable program instructions thereon for causing a processor to carry out aspects of the present invention. Embodiments explained herein relate to a system for predicting pressure- composition-temperature (PCT) curve for metal alloys used for solid-state hydrogen storage and a method thereof.

[0053] In an aspect, the proposed system determines the suitability of an alloy for solid-state hydrogen storage upon predicting Pressure-Composition-Temperature (PCT) data of plurality of metal alloy combinations owing to a fact that scanning PCT characteristics of every metal alloy combination from an available infinite number of metal alloy combinations to predict the suitability of that metal alloy combination suitable for a solid-state hydrogen storage is not possible. The system includes two machine learning models for predicting the Pressure-Composition-Temperature (PCT) curve for the metal alloy to be used / implemented for storage of the solid-state hydrogen.

[0054] The first model, MH-PCTpro is trained on complete database of the said metal alloy combinations and a second model, active-PCTpro is trained on a category of metal alloys or single family of alloys. The system and method predicts PCT curves as a function of pressure at a temperature range and a pressure range for different metal-alloy combinations. The MH-PCTpro model is used for faster and more efficient screening of plurality of metal-alloy combinations, while the active-PCTpro model acts as a tool to learn from real-time experiments performed on a plurality of metal-alloy combinations and train the MH-PCTpro model for efficient screening of the plurality of metal-alloy combinations to predict metal alloy combinations that is suitable for storing solid-state hydrogen. Furthermore, the method describes steps to predict the suitable metal-alloy combinations at the desired pressure and temperature ranges.

[0055] As noted above, to determine the suitability of an alloy for solid-state hydrogen storage, the Pressure-Composition-Temperature (PCT) measurements are extremely important. Further, scanning of the infinite chemical space experimentally, is not possible.

[0056] In an embodiment, the present disclosure relates to a system to predict a Pressure-Composition Temperature (PCT) isotherm for a metal alloy to predict a suitability of the metal alloy for hydrogen storage, the system (100) comprising: a processor (102) is in communication with a first learning model (104) and a second learning model (110), the first learning model (104) and the second learning model (110) are prestored in the system (100), and a server (106) coupled to a memory (108) storing instructions executable by the processor (102) is configured to: obtaining pre-fed PCT isotherm data associated with a plurality of metal alloy combinations available from the server (106); generate a dataset that includes weighted averages, hydriding properties, and one or more experimental parameters such as a predefined absorption temperature and a predefined hydrogen pressure, wherein the dataset further comprises the weighted average enthalpy of hydride formation and the volume difference between the element's bulk phase and its hydride phase; training the first learning model (104) using a standard learning model with the PCT isotherm data from the server (106) and the generated dataset; improving prediction results generated by the trained first learning model (104) iteratively by incorporating feedback from human experts using the second learning model (110); validating the improved predicted results generated by the trained first learning model (104) by comparing the predicted results with the one or more experimental parameters to ensure the prediction accuracy of the trained first learning module; predicting PCT isotherms for the plurality of metal alloy combinations as a function of the predefined hydrogen pressure and the predefined absorption temperature using the trained first learning model (104); capturing temperature-dependent variations in plateau pressure using the predicted PCT isotherms; and determining the enthalpy and entropy of hydride formation for the plurality of metal alloy combinations through Van't Hoff plots generated from the predicted PCT isotherms, thereby determining the suitability of a metal alloy combination from the plurality of metal alloy combinations fit for storing hydrogen.

[0057] In another embodiment, the processor (102) using the first learning model (104) is configured to predict PCT isotherms for the metal alloy compositions containing unseen elements not present in the pre-fed PCT isotherm data. In another embodiment, the dataset comprises data with twenty-eight distinct features, wherein the distinct features with collinearity values below 0.9 are selected for the training the first learning model (104).

[0058] In another embodiment, the processor (102) using the trained first learning model (104) is configured to generate an average Mean Absolute Error (MAE) of 0.17 ± 0.002 weight percentage (wt%) and an R2score of 0.96 when validated on the pre-fed PCT isotherm data from plurality of metal alloy combinations.

[0059] In another embodiment, the plurality of metal alloy combinations are selected from a group comprising light metals, hydriding and non-hydriding elements, including transition metals, alkali metals, alkaline earth metals, non-metals, and lanthanides.

[0060] In another embodiment, the second learning model (110) is configured within the processor (102), wherein the processor (102) using the second learning model is configured to select the most informative data points to be labeled by an expert, the second learning model is configured to: selecting one or more data points for which the first learning model is least confident in its predictions; labeling the selected uncertain data points upon receiving feedback from human experts; and training the first learning model using the labeled uncertain data points to improve the prediction results of the first learning model for metal alloy combinations in which it is previously uncertain.

[0061] In another embodiment, the standard learning model is an Extra Trees Regression (ETR) algorithm.

[0062] In another embodiment, the predefined absorption temperature is selected between a temperature range of 200K to 700K, and wherein the predefined hydrogen pressure is selected between 0.001 MPa to 100 MPa. In another embodiment, the present disclosure provides a method for predicting Pressure-Composition-Temperature (PCT) isotherms for metal compositions to determine their suitability for hydrogen storage, the method (200) comprising a step of: obtaining (210), using a processor (102) a pre-fed PCT isotherm data associated with a plurality of metal alloy combinations from a server; generating (220), using the processor (102) with a first learning model (104) a dataset that includes weighted averages, hydriding properties, and one or more experimental parameters such as a predefined absorption temperature and a predefined hydrogen pressure, wherein the dataset further comprises the weighted average enthalpy of hydride formation and the volume difference between the element's bulk phase and its hydride phase; training (230), the processor (102) with a standard learning model, the first learning model (104) with the PCT isotherm data from the server and the generated dataset; improving (240), using a second learning model (110), prediction results generated by the trained first learning model (104) iteratively by incorporating feedback from human experts; validating (250), using the processor (102) the improved predicted results generated by the trained first learning model by comparing the predicted results with the one or more experimental parameters to ensure the prediction accuracy of the trained first learning model; predicting (260), using the processor (102) with the trained first learning model (104) the PCT isotherms for the plurality of metal alloy combinations as a function of the predefined hydrogen pressure and the predefined absorption temperature; capturing (270), using the processor (102) with the trained first learning model (104) temperature-dependent variations in plateau pressure using the predicted PCT isotherms; and determining (280), using the processor (102) with the trained first learning model (104) the enthalpy and entropy of hydride formation for the plurality of metal alloy combinations through Van't Hoff plots generated from the predicted PCT isotherms such that the suitability of a metal alloy combination from the plurality of metal alloy combinations for storing hydrogen is determined. Accordingly, embodiments of the present invention, provides a machine learning (ML) based system and a machine learning (ML) based methods to predict a Pressure-Composition- Temperature (PCT) curve for a metal alloy to be used / implemented for solid-state hydrogen storage. The present invention provides two different approaches that assists in predicting the PCT curve for a given alloy.

[0063] In the first approach ML-based system is trained on 300 to 350 PCT plots extracted from the literature. This model is trained on compositions consisting of more than 30 elements and their PCTs reported at wide range of pressure and temperature. This model can predict PCTs for alloys consisting of any of these 30 elements as constituents. The model is named as MH-PCTpro. The MH-PCTpro can be used for high-throughput screening of alloys for solid state hydrogen storage.

[0064] In an exemplary embodiment, the present invention provides a machine learning (ML) based system (MH-PCTpro) to predict a Pressure-Composition-Temperature (PCT) curve for determining suitability of a metal alloy used / implemented for storage of a solid-state hydrogen.

[0065] The machine learning system includes a data collection / retrieval module that is configured to collect PCT data of metal alloy compositions; a data analysis module that is configured to analyse the collected PCT data, the data is analyzed in terms of composition, pressure and temperature; and a training module that is configured to utilize extra trees regression (ETR) algorithm to train PCTpro model.

[0066] An alternate model, named as Active-PCTpro is based on active machine learning that assists in predicting the PCT curve for a specific alloy family with better performance in case of small dataset.

[0067] Active-PCTpro offers a solution by providing an efficient on-the-fly learning approach to build model using minimal amount of labeled data. The process typically begins with an initial small labeled data-set. A model is trained on this initial labeled data, and then it is used to predict the labels of the remaining unlabeled instances. The active learning algorithm analyses the model's uncertainty or confidence in its predictions and selects the instances where it is most uncertain or least confident. These instances are then labeled by human oracle. The labeled instances are incorporated into the training set, and the model is retrained. The updated model is then used to make predictions on the remaining unlabeled instances, and the process repeats iteratively. The Active-PCTpro can serve as a tool to guide real-time experiments. In an exemplary embodiment, the present invention provides an active-machine learning (ML) based system (Active-PCTpro) to predict a Pressure-Composition-Temperature (PCT) curve for determining suitability of a metal alloy used / implemented for storage of a solid-state hydrogen.

[0068] The active machine learning system includes a data collection / retrieval module that is configured to collect PCT data of one or more metal alloy compositions; a data analysis module that is configured to analyse the collected PCT data, the data is analysed in terms of composition, pressure and temperature; a selection module that is configured to select compositions from the available data as training data (Mg-Ni); and a training module that is configured to employ ML algorithm to train an Active-PCTpro for PCT curve prediction of compositions in test data (Mg-Ni-La). Uncertainty measurement module is configured to compute the standard deviation in model's predicted values of PCT curve. The selection module is further configured to select one or more compositions from the test set (Mg-Ni-La), compositions having highest standard deviation as feedback for re-training the Active-PCTpro.

[0069] It may be appreciated that, Mg-Ni as a training set and Mg-Ni-La as test set is used as an example for demonstration of Active-PCTpro's working. The formalism is valid for any set. The training can be started with PCT curves of any family of binary alloys and then introduce (any) third metal and so on. Further, it could be trained on a PCT data of ternary alloys and tested for binary or quaternary alloys.

[0070] The invention leverages a processor using MH-PCTpro model and active PCTpro model to provide accurate and efficient predictions, thereby accelerating the discovery of new materials suitable for solid-state hydrogen storage.

[0071] FIG. 1(A and B) illustrates a flowchart demonstrating an automated method of the present invention (PCTpro / Active-PCTpro).

[0072] Referring to FIG.1A, a system (hereinafter referred to as "system 100") to predict a Pressure- Composition Temperature (PCT) curve for a metal alloy to predict a suitability of the metal alloy for hydrogen storage is described. The system 100 comprises a processor 102 in communication with a first learning model 104 which is interchangeably referred as MH-PCTpro model and a second learning model 110 which is interchangeably referred as Active-PTCpro. Both learning models are pre-stored in the system 100. The system also includes a server 106 coupled to a memory 108 that stores instructions executable by the processor 102. The processor 102 is configured to perform the following steps:

[0073] The processor 102 obtains pre-fed PCT isotherm data associated with a plurality of metal alloy combinations from the server 106. This data forms the basis for training the machine learning models.

[0074] The processor 102 generates a dataset that includes weighted averages, hydriding properties, and one or more experimental parameters such as a predefined absorption temperature and a predefined hydrogen pressure. The dataset further comprises the weighted average enthalpy of hydride formation and the volume difference between the element's bulk phase and its hydride phase. This comprehensive feature set ensures that the model captures all relevant factors influencing PCT isotherms.

[0075] The first learning model 104 is trained using a standard learning model, specifically an Extra Trees Regression (ETR) algorithm, with the PCT isotherm data from the server 106 and the generated dataset. This training process enables the model to learn the relationships between the features and the PCT isotherms.

[0076] The prediction results generated by the trained first learning model 104 are iteratively improved by incorporating feedback from human experts using the second learning model 110. The second learning model 110 is configured to select the most informative data points to be labeled by an expert. It selects one or more data points for which the first learning model is least confident in its predictions, labels the selected uncertain data points upon receiving feedback from human experts, and retrains the first learning model using the labeled uncertain data points to improve the prediction results for metal alloy combinations in which it was previously uncertain.

[0077] The improved predicted results generated by the trained first learning model 104 are validated by comparing the predicted results with the one or more experimental parameters to ensure the prediction accuracy of the trained first learning model. This validation step ensures that the model's predictions are reliable and accurate.

[0078] The processor 102 uses the trained first learning model 104 to predict PCT isotherms for the plurality of metal alloy combinations as a function of the predefined hydrogen pressure and the predefined absorption temperature. This step provides the core functionality of the system, enabling the prediction of PCT isotherms for various metal compositions.

[0079] The processor 102 captures temperature-dependent variations in plateau pressure using the predicted PCT isotherms. This capability is crucial for understanding the thermodynamic properties of the metal alloys.

[0080] The processor 102 determines the enthalpy and entropy of hydride formation for the plurality of metal alloy combinations through Van't Hoff plots generated from the predicted PCT isotherms. This step provides valuable thermodynamic insights, thereby determining the suitability of metal alloy combinations from the plurality of metal alloy combinations for storing hydrogen.

[0081] FIG. 2 (A and B) illustrates a method 200 for predicting PCT isotherms for metal compositions to determine their suitability for hydrogen storage comprises the following steps:

[0082] Step:-1 Obtaining 210 a pre-fed PCT isotherm data associated with a plurality of metal alloy combinations from a server using a processor 102.

[0083] Step:-2 Generating 220 a dataset that includes weighted averages, hydriding properties, and one or more experimental parameters such as a predefined absorption temperature and a predefined hydrogen pressure using the processor 102 with a first learning model 104. The dataset further comprises the weighted average enthalpy of hydride formation and the volume difference between the element's bulk phase and its hydride phase.

[0084] Step:-3 Training 230 the first learning model 104 with the PCT isotherm data from the server and the generated dataset the processor 102 with a standard learning model.

[0085] Step:-4 Improving 240 prediction results generated by the trained first learning model 104 iteratively by incorporating feedback from human experts using a second learning model 110.

[0086] Step:-5 Validating 250 the improved predicted results generated by the trained first learning model by comparing the predicted results with the one or more experimental parameters to ensure the prediction accuracy of the trained first learning model using the processor 102.

[0087] Step:-5 Predicting 260 with the trained first learning model (104), the PCT isotherms for the plurality of metal alloy combinations as a function of the predefined hydrogen pressure and the predefined absorption temperature using the processor 102. Step:-6 Capturing 270 temperature-dependent variations in plateau pressure using the predicted PCT isotherms using the processor 102 with the trained first learning model 104.

[0088] Step:-7 Determining 280 with the trained first learning model 104, the enthalpy and entropy of hydride formation for the plurality of metal alloy combinations through Van't Hoff plots generated from the predicted PCT isotherms, thereby determining the suitability ofa metal alloy combination from the plurality of metal alloy combinations for storing hydrogen using the processor 102.

[0089] Referring to FIG. 3A describes a comparison between the first learning model predicted and experimental PCT diagrams of LaMg3.70Nil.18 composition at 573 K. The round-dotted curve "0thiteration" represents PCT curve predicted by the first learning model 104 of the system 100 and compared with the experimentally reported PCT.

[0090] Referring to FIG. 3B describes a comparison between the first learning model predicted and experimental PCT diagrams of La0.7Mg0.3Ni3.2Co0.35 composition at 298 K. The round-dotted curve "0thiteration" represents PCT curve predicted by the first learning model 104 of the system 100 and compared with the experimentally reported PCT curve represented with cubes.

[0091] FIG. 4A illustrates comparison between predicted and experimental PCT diagrams of La5Mg90Ni5 at 633K through Active-PCTpro in accordance with the present invention.

[0092] FIG. 4B illustrates comparison between predicted and experimental PCT diagrams of LaMg2Ni5 at 514 K through Active-PCTpro in accordance with the present invention.

[0093] To summarize, a PCT database of more than 300 PCTs from published literature was built. For training a model, a feature set consisting of elemental properties, absorption pressure, and temperature as feature were built. ETR, a tree-based model was trained using the dataset and features. Two different models, MH-PCTpro (an ML-based model trained on diverse database) and Active-PCTpro (trained for predicting compositions from specific family of alloy), were builtfor predicting alloy PCT curves. The validated model performance was checked by predicting PCT curves of experimentally reported compositions (not seen by model) as function of hydrogen pressure and temperature. The model predicts PCTs of new multi-component (binary, ternary, quaternary, etc.) metal alloys at different temperatures. It is understood in advance that although this disclosure includes a detailed description on cloud computing, implementation of the teachings recited herein are not limited to a cloud computing environment. Rather, embodiments of the present invention are capable of being implemented in conjunction with any other type of computing environment now known or later developed.

[0094] ADVANTAGES OF THE INVENTION

[0095] • The present system and method are advantageous to determine / predict PCT curves i.e., hydrogen absorption in view of variable temperature and pressure, for different alloys composition.

[0096] • The MH-PCTpro can be used for high-throughput screening of alloys, while the Active-PCTpro can serve as a tool to guide real-time experiments.

[0097] • This study is cost effective and preliminary analysis method to know the suitable alloy compositions at desired pressure and temperature ranges for solid state hydrogen storage.

[0098] • The present invention provides a simple and efficient system for predicting pressure- composition-temperature (PCT) curve for metal alloys used for a solid-state hydrogen storage.

[0099] • The present invention predicts the suitability of a metal alloy combination for hydrogen storage.

[0100] • The present invention provides for a system that trains a MH-PCTpro and active-PCTpro for calculating one or more uncertainties from each prediction based on a pre-fed standard deviation.

[0101] • The present invention provides acost-effective method to predict suitable alloy compositions at a desired pressure and temperature ranges for the solid-state hydrogen storage.

Claims

We Claim:

1. A system to predict a Pressure-Composition-Temperature (PCT) isotherm for a metal alloy to predict its suitability for hydrogen storage, wherein the system (100) comprising a processor (102) in communication with a first learning model (104) and a second learning model (110), wherein the first learning model (104) and the second learning model (110) being pre-stored in the system (100), connected to a server (106) coupled to a memory (108) storing instructions executable by the processor (102) being configured to: a) obtaining pre-fed PCT isotherm data associated with a plurality of metal alloy combinations available from the server (106); b) generating a dataset that includes weighted averages, hydriding properties, and one or more experimental parameters such as a predefined absorption temperature and a predefined hydrogen pressure, wherein the dataset further comprises the weighted average enthalpy of hydride formation and the volume difference between the element's bulk phase and its hydride phase; c) training the first learning model (104) using a standard learning model with the PCT isotherm data from the server (106) and the generated dataset; d) improving prediction results generated by the trained first learning model (104) iteratively by incorporating feedback from human experts using the second learning model (110); e) validating the improved predicted results generated by the trained first learning model (104) by comparing the predicted results with the one or more experimental parameters to ensure the prediction accuracy of the trained first learning module; f) predicting PCT isotherms for the plurality of metal alloy combinations as a function of the predefined hydrogen pressure and the predefined absorption temperature using the trained first learning model (104); g) capturing temperature-dependent variations in plateau pressure using the predicted PCT isotherms; andh) determining the enthalpy and entropy of hydride formation for the plurality of metal alloy combinations through Van't Hoff plots generated from the predicted PCT isotherms, thereby determining the suitability of a metal alloy combination from the plurality of metal alloy combinations fit for storing hydrogen.

2. The system (100) as claimed in claim 1, wherein the processor (102) using the first learning model (104) is configured to predict PCT isotherms for the metal alloy compositions containing unseen elements not present in the pre-fed PCT isotherm data.

3. The system (100) as claimed in claim 1, wherein the dataset comprises data with twentyeight distinct features, wherein the distinct features with collinearity values below 0.9 are selected for the training the first learning model (104).

4. The system as claimed in claim 1, wherein the processor (102) using the trained first learning model (104) is configured to generate an average Mean Absolute Error (MAE) of 0.17 ± 0.002 weight percentage (wt%) and an R2score of 0.96 when validated on the pre-fed PCT isotherm data from plurality of metal alloy combinations.

5. The system (100) as claimed in claim 1, wherein the plurality of metal alloy combinations are selected from the group comprising light metals, hydriding and non-hydriding elements, including transition metals, alkali metals, alkaline earth metals, non-metals, and lanthanides.

6. The system as claimed in claim 1, wherein the second learning model (110) is configured within the processor (102), wherein the processor (102) using the second learning model is configured to select the most informative data points to be labeled by an expert, the second learning model is configured to:(i) selecting one or more data points for which the first learning model is least confident in its predictions;(ii) labeling the selected uncertain data points upon receiving feedback from human experts; and(iii) training the first learning model using the labeled uncertain data points to improve the prediction results of the first learning model for metal alloy combinations in which it is previously uncertain.

7. The system (100) as claimed in claim 5, wherein the standard learning model is an Extra Trees Regression (ETR) algorithm.

8. The system (100) as claimed in claim 1, wherein the predefined absorption temperature is selected between a temperature range of 200K to 700K, and wherein the predefined hydrogen pressure is selected between 0.001 MPa to 100 MPa.

9. A method for predicting Pressure-Composition-Temperature (PCT) isotherms for metal compositions to determine their suitability for hydrogen storage, the said method (200) comprising the step of:(i) obtaining (210), using a processor (102) a pre-fed PCT isotherm data associated with a plurality of metal alloy combinations from a server;(ii) generating (220), using the processor (102) with a first learning model (104) a dataset that includes weighted averages, hydriding properties, and one or more experimental parameters such as a predefined absorption temperature and a predefined hydrogen pressure, wherein the dataset further comprises the weighted average enthalpy of hydride formation and the volume difference between the element's bulk phase and its hydride phase;(iii) training (230), the processor (102) with a standard learning model, the first learning model (104) with the PCT isotherm data from the server and the generated dataset;(iv) improving (240), using a second learning model (110), prediction results generated by the trained first learning model (104) iteratively by incorporating feedback from human experts;(v) validating (250), using the processor (102) the improved predicted results generated by the trained first learning model by comparing the predicted results with the one or more experimental parameters to ensure the prediction accuracy of the trained first learning model;(vi) predicting (260), using the processor (102) with the trained first learning model (104) the PCT isotherms for the plurality of metal alloy combinations as a function of the predefined hydrogen pressure and the predefined absorption temperature;(vii) capturing (270), using the processor (102) with the trained first learning model (104) temperature-dependent variations in plateau pressure using the predicted PCT isotherms; and(viii) determining (280), using the processor (102) with the trained first learning model (104) the enthalpy and entropy of hydride formation for the plurality of metal alloy combinations through Van't Hoff plots generated from the predicted PCT isotherms such that the suitability of a metal alloy combination from the plurality of metal alloy combinations for storing hydrogen is determined.

Images