System and method for energy storage device generative design

Abstract

To provide a system and method for generative design of an energy storage device.SOLUTION: A method automatically builds at least one model of an energy storage device. The building is based on a design parameter space and employs a machine learning process. The method automatically performs a simulation of the energy storage device using the design parameter space, a design evaluation space, and the at least one model built. The performing produces at least one prediction. The method automatically evolves at least one of the design parameter space and the design evaluation space. In an event the at least one prediction indicates that a product design objective or model design objective has been achieved, the method automatically converges on the design parameter space evolved, thereby completing a generative design of the energy storage device and, otherwise, repeats the building, performing, and evolving.SELECTED DRAWING: Figure 1A

Description

Background Art

[0001] Generative design refers to a technology by which a model, such as a three-dimensional (3D) computer-aided design (CAD) model or another type of computer-based model, can be created and optimized by computer software. Designers / engineers / scientists can typically interact with the technology to create superior designs and drive product innovation more rapidly. Generative design can be considered a design exploration process that expands the designer's world of known effective solutions to their design challenges.

[0002] For example, as a non-limiting example, a designer may provide parameters such as materials, size, weight, strength, manufacturing methods, etc., and the generative design software may explore possible combinations of solutions and may rapidly generate hundreds or thousands of design options. From there, the designer, or the software, can filter and select the results that best meet the goals of the design process. Using generative design, the designer is not limited by their imagination or past experience.

[0003] The input to generative design tools may be similar to the input to many optimization tools, but generative design generates multiple valid (e.g., high-performance but cost-effective) designs or solutions instead of a single optimized version of a solution. In addition to creating completely new solutions, another area where generative design stands out is its ability to take manufacturability into account. Using generative design, simulations may be incorporated into the design process, and the generative design software can generate only designs that can be fabricated using a specified manufacturing method(s) or in accordance with the parameters and / or requirements of a specified manufacturing method(s).

Summary of the Invention

[0004] Various embodiments of the present disclosure generally relate to generative designs of energy storage devices. Exemplary embodiments enable, as non-limiting examples, compounds, components, additives, formulations, and recipes to be designed / specified via a generative design for use in an energy storage device. Exemplary embodiments enable such design / specification of compounds, components, additives, formulations, and recipes to affect, alone or in combination, as non-limiting examples, the performance of an energy storage device such as a battery. Performance may be based on, as non-limiting examples, capacity, temperature tolerance, maximum charge and discharge rates or profiles, cycle life, ease of manufacture, quality, safety, reprocessing or diversion at end-of-life, and attributes such as the shape, configuration, or form of the battery.

[0005] In connection with batteries, there are several issues that may be addressed via a generative design applied to the battery. Battery manufacturing is a complex, iterative, multi-element process. For example, battery manufacturing may include the production of electrodes. The production of electrodes may include, as non-limiting examples, the production of anodes, cathodes, active materials, electrolytes, binders, and separators. Battery manufacturing may include the production of cells, where a single unit that performs the function of a "battery" is produced. Battery manufacturing may include module assembly, where a plurality of cells enclosed in a metal case may be connected to an electronic management system. Battery manufacturing may further include performing pack assembly, where the connection of a plurality of modules, sensors, and controllers may be installed within a case.

[0006] As a non-limiting example, a generative design for a battery may take into account, as non-limiting examples, the state of charge, battery chemistry and technology, efficiency, safety, aging, temperature spread, durability, as well as sustainability and recyclability. In this way, the generative design may be optimized to enable the user of the battery to stop worrying about insufficient time to charge, and the generative design may find the most performant and safest chemistry for the battery. The generative design may make the battery efficient by generating sufficient power to increase functionality, and may evaluate the safety of the battery to prevent accidents. The generative design may evaluate the durability of the battery to prevent capacity loss and may also evaluate the effect of temperature spread on battery performance. The generative design may evaluate durability and enable the design of "green" batteries.

[0007] As a non-limiting example and in relation to a battery, the generative design may perform the identification of chemical components such as additives for electrolytes, coatings for electrodes, by-product scavengers, and decomposition inhibitors. The generative design may perform the identification of formulations such as mixture design (quantity) and specification design (grade, purity, and impurities). The generative design may determine the geometric shape of the battery such as size and form factor, electrode configuration and structure, as well as tab location and size. The generative design may determine the process design for the battery such as process sequence, process operating parameters, selection and identification of manufacturing equipment, as well as coating stage and operation. The generative design may further provide test and quality results for the battery at the end of the manufacturing line, on the line, at installation, during use / aging, and during recycling.

[0008] Generative design may advantageously combine advanced data science, machine learning, chemoinformatics - matinformatics, and structure - based modeling to explore and optimize chemical space, compositional space, and process space. Exemplary embodiments may automate virtual generation, testing, and selection of new components, amounts during their formulation, etc., and may incorporate process - level parameters. The goal of generative design may be not only to reduce the amount and cost of physical testing, but also, by way of non - limiting example, to perform trade - off studies between system variables.

[0009] According to an exemplary embodiment, a computer - implemented method for generative design of an energy storage device includes automatically constructing at least one model of the energy storage device. The constructing may be based on a design parameter space or may employ a machine - learning process. The computer - implemented method further includes automatically performing a simulation of the energy storage device. The simulation may employ the design parameter space, a design evaluation space, and the at least one constructed model. Performing may include generating at least one prediction of the energy storage device that achieves at least one product design objective or at least one model design objective for the at least one model constructed to achieve the at least one product design objective.

[0010] The computer - implemented method further includes automatically evolving at least one of (i) the design parameter space and (ii) the design evaluation space. Evolving may be based on the at least one generated prediction and may employ a machine - learning process. If the at least one prediction indicates that at least one product design objective has been achieved or at least one model design objective has been achieved, the computer - implemented method automatically converges on the evolved design parameter space, thereby completing the generative design of the energy storage device, and, otherwise, further includes repeating constructing, performing, and evolving.

[0011] The design parameter space and the design evaluation space may be associated with at least one of the energy storage device and at least one of the constructed models. The design parameter space may include variables. The variables may include material variables, system variables, or combinations thereof. The variables may be associated with the chemical space of the energy storage device, the formulation space of the energy storage device, the material space of the energy storage device, the configuration space of the energy storage device, the process space of the energy storage device, or combinations thereof.

[0012] The design evaluation space may include at least one test of the energy storage device, for example, a real-world test, and performing a simulation may include simulating at least one test using at least one of the constructed models. Non-limiting examples of such real-world tests may involve repeatedly measuring the state of charge over a number of charge cycles at different charge rates to identify the onset of cell failure in the energy storage device. For each design space, such measurements can be used for model validation and prediction of pre-production results. This enables determining better process and material specifications, as well as modifications to the design, physical configuration, or recipe of the cell components of the energy device.

[0013] At least one product design objective may include at least one user-specified criterion associated with the energy storage device, at least one machine-generated criterion associated with the energy storage device, or combinations thereof. At least one product design objective may include a target product profile (TPP). At least one model design objective may include at least one error threshold associated with the difference between the simulated measurement and the real-world measurement, the target size for the design parameter space, or combinations thereof.

[0014] Evolving may include (a) truncating the design parameter space, (b) truncating the design evaluation space, (c) expanding the design parameter space, (d) expanding the design evaluation space, or (e) a combination of (a) to (d).

[0015] Evolving may include maintaining diversity within the design parameter space. Maintaining may include adopting a clustering method, a Pareto method, or a combination thereof.

[0016] The computer-implemented method may further include adopting at least one monitored parameter in constructing, evolving, or a combination thereof. The at least one monitored parameter may be adopted in at least one iteration of constructing, evolving, or a combination thereof. The at least one monitored parameter may represent at least one real-world result generated via at least one real-world experiment adopting an energy storage device. The real-world result may be implemented based on the adoption of an evolved design parameter space in at least one real-world experiment.

[0017] The computer-implemented method may further include adopting an adversarial generative network (GAN), a deep neural network (DNN), a Bayesian optimization (BAO), a genetic function approximation method, or a combination thereof in a machine learning process.

[0018] The design parameter space may include semantically structured real-world evidence (RWE) data associated with experiments of an energy storage device.

[0019] The computer-implemented method may further include automatically storing in a database at least one constructed model associated with at least one generated prediction, at least one input of at least one constructed model, and at least one output from at least one constructed model. Performing a simulation may include inputting at least one input into at least one constructed model and generating at least one output from at least one constructed model in response to the at least one input.

[0020] The energy storage device may be a battery. Converging on an evolved design parameter space may include identifying at least one of a compound, component, additive, formulation, recipe, or combination thereof that enables achieving at least one product design objective of the energy storage device.

[0021] According to another exemplary embodiment, a computer-based system for generative design of an energy storage device may include at least one memory and at least one processor coupled to the at least one memory. The at least one processor may be configured to automatically construct at least one model of the energy storage device based on a design parameter space. Constructing may employ a machine learning process. The at least one processor may be further configured to automatically perform a simulation of the energy storage device. The simulation may employ the design parameter space, a design evaluation space, and at least one constructed model. Performing may include generating at least one prediction of the energy storage device that achieves at least one product design objective or at least one model design objective achieved by at least one constructed model.

[0022] At least one processor may be further configured to automatically evolve at least one of (i) a design parameter space and (ii) a design evaluation space. Evolving may be based on at least one generated prediction and may employ a machine learning process. If at least one prediction indicates that at least one product design objective has been achieved or at least one model design objective has been achieved, the at least one processor may automatically converge on the evolved design parameter space, thereby completing a generative design of the energy storage device, and otherwise may be further configured to repeat constructing, executing, and evolving.

[0023] Alternative computer-based system embodiments are similar to those described above in connection with the exemplary method embodiments.

[0024] According to yet another exemplary embodiment, a non-transitory computer-readable medium for a generative design of an energy storage device, when loaded and executed by at least one processor, may encode thereon a sequence of instructions that cause the at least one processor to automatically construct at least one model of the energy storage device based on a design parameter space. Constructing may employ a machine learning process.

[0025] The sequence of instructions may further cause the at least one processor to automatically execute a simulation of the energy storage device. The simulation may employ the design parameter space, the design evaluation space, and the at least one constructed model. Executing may include generating at least one prediction of the energy storage device that achieves at least one product design objective or at least one model design objective achieved by the at least one constructed model. The sequence of instructions may further cause the at least one processor to automatically evolve at least one of (i) the design parameter space and (ii) the design evaluation space.

[0026] Evolution may be based on at least one generated prediction, or a machine learning process may be employed. If at least one prediction indicates that at least one product design objective has been achieved or at least one model design objective has been achieved, the sequence of instructions may further cause at least one processor to automatically converge on an evolved design parameter space, thereby completing the generative design of the energy storage device, and, if not, may cause at least one processor to repeat constructing, executing, and evolving.

[0027] Alternative embodiments of the non-transitory computer-readable medium are similar to those described above in connection with the exemplary method embodiments.

[0028] Of course, the exemplary embodiments disclosed herein can be implemented in any combination and in such forms of a method, an apparatus, a system, or a non-transitory computer-readable medium having program code embodied thereon.

[0029] The foregoing will be apparent from the following more particular description of the exemplary embodiments, as illustrated in the accompanying drawings, wherein like reference numerals refer to the same parts throughout the different drawings. The drawings are not necessarily to scale, emphasis instead being placed upon illustrating embodiments.

Brief Description of the Drawings

[0030]

Figure 1A

Figure 1B

Figure 1C

Figure 1D

Figure 2

Figure 3

Figure 4

Figure 5

[0031] The description of the exemplary embodiments is set forth below.

[0032] The exemplary embodiments disclosed herein relate, by way of non-limiting example, to generative systems and methods therefor for designing compounds, components, additives, formulations, and recipes for use in energy storage devices and, by way of non-limiting example, for designing models of such energy storage devices such as batteries. The exemplary embodiments generally relate, by way of non-limiting example, to materials modeling and simulation of complex material-based systems such as batteries. The exemplary embodiments disclosed herein may be applicable to other areas where chemistry, materials, nano, and domain microstructures may be used to predict and understand the performance and behavior of products. The exemplary embodiments disclosed herein may be applicable, by way of non-limiting example, to general chemical reactivity systems such as fuel cells and chemical process effects.

[0033] Generative Material Design (GMD) is essentially the automation of model-based exploration of a variable space. The variables may be considered as chemistry (e.g., structure, isomers, toxicity, substitution, and sequence), materials (e.g., phase, polymorph, insertion, spacing), formulation (e.g., amount, purity, grade), and process (e.g., steps, actions, sequence, conditions, parameters). The variables may be specified at the design stage, and the system may automate the construction of samples. These constructions may then be evaluated using the next steps identified via model-based and evolutionary approaches.

[0034] Exemplary embodiments disclosed herein provide, as non-limiting examples, efficient and automated methods for determining optimal conditions in a property space from material and / or system variables. Exemplary embodiments of the computer-based systems disclosed herein may use, as non-limiting examples, virtual cycles by which the system can explore a property space for an energy storage device such as a battery by "learning" from the results of real-world experiments.

[0035] Exemplary embodiments of the computer-based systems disclosed herein may, as non-limiting examples, use a combination of machine learning models and structure-based modeling and simulation methods to virtually screen and optimize candidate compounds. Multi-objective optimization methods may be employed to balance competing objectives and enable product designers to generate compounds that improve towards a designed or target product profile (TPP).

[0036] In the "actual" cycle stage, such product designers may synthesize and test the most promising virtual compounds in a laboratory setting. New data generated through such tests may be input into a computer-based system to improve the prediction model and, as a non-limiting example, to refine the exploration of chemical space. Such execution of real-world experiments may be useful for improving the model and generating new data used to perform optimization. According to an exemplary embodiment, such virtual (V) and real (R), i.e., V+R, active learning cycles may continue until the product designer identifies, as a non-limiting example, a compound that meets the TPP.

[0037] Exemplary embodiments disclosed herein may include, as a non-limiting example, a computer-implemented method that uses an adversarial deep neural computing approach that employs an iterative and probabilistic process to obtain a realistic computer model of a battery. As a non-limiting example, exemplary embodiments of a computer-based system or computer-implemented method may link the following four areas of science and modeling. 1. Quantitative Structure-Activity Relationship Modeling (QSAR) - QSAR modeling of empirical data related to quality, chemical, and material systems can be performed using established and validated machine learning methods. These provide highly predictive and extensible estimates of performance / activity for proposed chemicals and mixtures. 2. TPP - Early attention to balancing all the attributes required for a chemical system to progress from the innovation stage of a new product to the pilot scale and manufacturing stages may employ multi-parameter optimization at all stages (e.g., safety, manufacturability, novelty, etc.). 3. Active Learning (AL) - This special case of machine learning closely couples the real world with virtual activities, enabling the learning method to query the user (or another information source) to label new data points with the desired output. In GMD, this means intentionally synthesizing mixtures and recipes that expand the domain of applicability of the machine learning model. 4. Generative Chemistry - As a non-limiting example, it is impossible with reasonable resources to enumerate all the chemical substances that could potentially have a significant impact on the development of battery electrode coatings or electrolytes. However, chemists can explore large regions of chemical space through iterative modification of the starting chemistry or formulation sequences derived by a property-based multi-dimensional fitness function.

[0038] The GMD disclosed herein may then use an evolutionary approach to drive a material-based system, such as a battery, not only to maximum desirability but also to maximum synthesizability / processability. Such a process may generate millions of options over multiple cycles, which, according to an exemplary embodiment, can be automatically trimmed against user-defined criteria to ultimately obtain a small optimal set of options for the solution set. According to an exemplary embodiment, a clustering method or Pareto method may be employed to maintain diversity and ensure that the set of solutions is not a nearly identical group of materials with only minor variations from each other. After optimization, i.e., "virtual experiments" or simulations are complete, the materials scientist can look for convergence within the chart(s) to determine whether to synthesize and test a portion of the resulting structure. Alternatively, the materials scientist can adjust the optimization parameters and perform another run, i.e., another virtual experiment / simulation. Exemplary embodiments of computer-based systems that may be used by such materials scientists are further disclosed below with respect to FIGS. 1A and 2.

[0039] Exemplary embodiments of a computer-based system may, as non-limiting examples, enable a user, such as a materials scientist, to advantageously do the following. · Configure each optimization objective using a desirability profile. · Graphically adjust each desirability profile based on probabilities determined by applying a machine learning model to known compounds. · Increase the “weight” of the most important objectives (e.g., activity) to help ensure achievement of such objectives during optimization. · Specify the overall goals of the optimization, such as, as non-limiting examples, whether immediate improvement of a compound or exploration of chemical space should be used for model improvement, whether a compound improvement function should be applied to improve product production performance and quality metrics, and whether a series of low-scale models should be employed to enable better, more accurate, or more robust model development.

[0040] Exemplary embodiments of the computer-based system or computer-implemented method disclosed herein may, as non-limiting examples, at two levels, first, in a closed-loop approach, enrich the form used in feature generation, and second, in a high-level subsequent model, use an abstraction of the form of an energy storage device of the model that has been specified. The exemplary embodiments disclosed herein may be used automatically without user input or with minimal user input, which enables rapid exploration of the design parameter space and self-governing model development as new data accumulates from a test or production system, resulting in an improved user experience due to the availability of multiple models. An advantage of this approach is incorporating the synthetic accessibility of compounds. The enumeration may also, as non-limiting examples, be based on a Markush structure having a core molecule with a reaction-based (R) group position and a set of linked fragments. Such an approach may be employed in a computer-based system, such as the computer-based system 104 of FIG. 1A disclosed below.

[0041] FIG. 1A is a block diagram of an exemplary embodiment of a computing environment 100 in which a materials scientist 102 is using a computer-based system 104 for a generative design of an energy storage device 106. A materials scientist is a person who studies the structural and chemical properties of various materials in order to develop new products or enhance existing ones. Of course, the computer-based system 104 is not limited to being used by a materials scientist. Further, in the exemplary embodiment of FIG. 1A, by way of non-limiting example, the energy storage device 106 is a battery. The exemplary embodiment of FIG. 1A may describe the energy storage device 106 as a battery, and more specifically, as a lithium-ion battery. Of course, the exemplary embodiments disclosed herein are not limited to batteries or lithium-ion batteries.

[0042] In the exemplary embodiment of FIG. 1A, the materials scientist 102 is interacting with two user interfaces provided by the computer-based system 104, namely, a first user interface (UI) 103 and a second UI 105. In one computer operating system, the user interfaces 103, 105 are implemented as so-called windows. In a global network (Internet) server-based computer system, the user interfaces 103, 105 may be implemented as different tabs, screen views, or the like. The first UI 103 and the second UI 105 are presented on a display screen 107 of the computer-based system 104 for user interaction. Of course, the computer-based system 104 is not limited to providing two user interfaces (UIs).

[0043] In an exemplary embodiment, the energy storage device 106 is a lithium-ion battery. Lithium-ion batteries can last for many years, but sometimes exhibit rapid non-linear degradation that significantly limits the battery's lifespan. In the exemplary embodiment of FIG. 1A, the first UI 103 includes a first simulated representation 109 at the start of the lifespan of a slice of the "jelly roll" of the lithium-ion battery, and the second UI 105 includes a second simulated representation 111 after the cycle life simulation of the slice of the jelly roll. In the second simulated representation 111 of the slice, a deformation 113 is formed such that the jelly roll is deformed inwards towards the core 115 (i.e., the center) of the jelly roll. Such a deformation 113 may be attributed to a simulated increase in the internal pressure of the jelly roll during the cycle life simulation. The materials scientist 102 may employ a computer-based system 104 to perform GMD of such lithium-ion batteries and, as a non-limiting example, generate a design to avoid such a deformation 113.

[0044] According to an exemplary embodiment, the computer-based system 104 may be configured as follows, as a non-limiting example. · Explore chemical space, · Include reaction-based / R-group-based enumeration, · Filter virtual compounds against undesirable substructures based on universal, company, and project-specific criteria, · Incorporate design preferences based on a battery component profile customized for the materials functional area, and · Prioritize based on considerations of synthesizability.

[0045] GMD provides various methods for generating virtual compounds. Evolutionary methods may use various molecular transformations, including matched molecular pairs (MMPs), substitution of ring assemblies, molecular scaffolds, molecular variations (fragmentation, trimming, elongation, rearrangement, crossover), and molecular mutations (changes in atom and bond types, addition and deletion of atoms, ring opening and closing), to generate molecules for each input molecule. There is an option to bias generative methods to generate compounds similar to those containing a target structure or similar to a particular scaffold.

[0046] One of the challenges related to lead compound search is to understand the trade - offs made when optimizing multiple design criteria. GMD enables materials scientists, chemists, etc. to design improved lead compounds by maximizing an overall desirability function based on individual desirability profiles based on the TPP. For each objective in the TPP, materials scientist 102 may use the profile to specify how "good enough" is "good enough" to achieve the project's goals. GMD may be used to assist in this process, for example, via display screen 107, to display model predictive distributions and empirical probability values (e.g., the probability that a compound is active if the model score exceeds a certain value). Alternatively, GMD may be used to provide a reasonable default desirability profile based on the TPP.

[0047] Immediate access to both positive and negative data enables scientists, such as materials scientist 102, to enhance predictive models and track the lineage of compounds from concept to development. This "digital continuity" is useful for tracking intellectual property (IP) generated within a discovery organization. An exemplary embodiment of the generative process enabling this is described below with respect to Figure 1B.

[0048] FIG. 1B is a block diagram of an exemplary embodiment of a generative process 150. The generative process 150 may include a data input stage 152, an ideation stage 154, a prototype stage 156, a testing stage 158, and an evaluation stage 160. The generative process 150 may perform an iterative design 162 (returning from the evaluation stage 160 to the ideation stage 154), which ultimately leads to a product design 164 based on the results of the evaluation stage 160. The generative process 150 may be employed as part of an automated molecular design environment, such as the environment 170 disclosed below with respect to FIG. 1C.

[0049] FIG. 1C is a block diagram of an exemplary embodiment of an automated molecular design environment 170. The automated molecular design environment 170 includes data 172, tools 174, and a client 176. By way of non-limiting example, the data 172 may include an assay 172-1, a candidate 172-2, an input 172-3, and / or a history 172-4. The tools 174 may include data access 174-1, data preparation and search 174-2, machine learning methods 174-3, and / or delivery, integration, and deployment 174-5.

[0050] Machine learning method 174-3 may include, by way of non-limiting example, performance and scalability method 175-1, automation method 175-2, active learning method 175-3, and / or multi-parameter optimization method 175-4. Client 176 may include, by way of non-limiting example, optimization workbench 176-1, project management 176-2, model management 176-3, collaboration and analysis 176-4, and / or visual programming environment 176-5. Such an automated molecular design environment 170 may be included within the computing environment 100 of FIG. 1A, disclosed above and further detailed below. Chemical and materials design approaches may include the design of anti-aging additives in battery electrolytes. Given the numerous competing design requirements, the ability to screen a large chemical space down to a human-understandable set and then prioritize that set against either performance goals or model development goals can be achieved by artificial intelligence. Additionally, competing requirements can be evaluated using such models to achieve a balance between competing needs.

[0051] Referring to FIGS. 1A and 1C, the automated molecular design environment 170 represents the components of a computer-based system 104 according to an exemplary embodiment. As shown in FIG. 1C, such components include data sources and real-world evidence, i.e., data 172, including assays 172-1, candidates 172-2, inputs 172-3, and histories 172-4. The components further include tools 174 that are used to access, transform, instantiate, and analyze these various data sources 172. The components further include a consuming system, referred to as client technology or simply client 176. The client 176 can interact with the tools 174, their derivatives, models thereof, and native data 172. Such a framework, i.e., the automated molecular design environment 170, is much more than the mere sum of its parts. The environment 170 implicitly contains not only the life cycle of entities but also state and message frameworks within the multi-user / multi-role environment specified by the data 172, tools 174, and client 176.

[0052] Returning to FIGS. 1A and 1B, the generative process 150 may be implemented by a computer-based system 104, or may implement a generative method, such as disclosed below with respect to FIG. 1D.

[0053] FIG. 1D is a flowchart of an exemplary embodiment of a computer-implemented generative method (180). The generative method (180) begins (182) and (1) collects and semantically aligns data, (2) constructs a model of properties of interest, such as, by way of non-limiting example, open circuit voltage, flammability, capacity, maximum cycle life, maximum charge current (C-rate) during discharge and / or charge, if the product design 164 is, by way of non-limiting example, for a battery, (3) uses the model as a predictor to perform design of experiments for the design space over a set of variables, (4) stores it in a database, (5) filters on user or defined machine criteria, (6) selects the best candidates and validates such selections, and (7) may iterate from (2) if such selections (6) do not meet at least one criterion, or upon convergence (e.g., the selections meet the at least one criterion), the method then ends (184) in the exemplary embodiment. The at least one criterion may be associated with, by way of non-limiting example, at least one product design objective or at least one model design objective, further disclosed below with respect to FIG. 4, for example. In one embodiment, the at least one criterion may be set by a user or automatically set based on requirements.

[0054] Continuing to refer to FIG. 1D, such alignment of data in (1), model construction based on real-world evidence in (2), and optimization of the design space for either the goals of characteristics / performance or improvement of the model construction in (3) lead to better virtual evaluation in the evaluation stage 160. The storage action taken in (4) above provides storage that encourages or enables iteration to new ideas and limitation to a manageable set of tests and evaluations of the design space in the ideation stage 154. The prototype stage 156 and the test stage 158 may be actions performed in the real world that provide results input to the evaluation stage 160 executed by the computer-based system 104, and by returning to the ideation stage 154, cause the computer-based system 104 to execute an iterative design 162. Further details regarding the actions (1)-(7) disclosed above are provided below. Such actions may be referred to as generative actions and may be executed by the computer-based system 104 and the computer-based system 204 of FIG. 2 further disclosed below.

[0055] Generative method

[0056] (1) Collect and semantically align data

[0057] Data for devices such as batteries may come from many different levels. By way of non-limiting example, data may come from the sub-cell chemistry level (e.g., atomic structure, quantum electronic properties) and the materials level (e.g., not only purity, grade, structure, and / or phase, but also thermodynamic state and dynamic or reactive behavior, chemical decomposition pathways, and / or other information). Data may further come from mesoscopic information such as particle size, particle distribution, structure, or layer arrangement in the electrodes, and the distribution structure of the materials, and porosity, flexibility, and permeability on the electrodes (i.e., separators, anodes, and / or cathodes). In addition, physical characteristics of many materials such as pore size distribution on the electrodes and the effects of processing (e.g., calendar processing winding and compression or stacking) can directly affect the performance of the cell.

[0058] At the cell level, information comes not only from the geometric shape of the cell, e.g., cylindrical vs. pouch vs. prismatic vs. blade, but by way of non-limiting example, also from their sizes (e.g., in the case of cylindrical, there are different dimensions such as the diameter and height of the cell (e.g., 18×65, 23×70, or 46×80)). Such cell and sub-cell characteristics are manufactured, and thus, by way of non-limiting example, there are secondary attributes around the ambient conditions of manufacturing, and the processes or equipment used for coating the electrodes, the geometric shape of the press for calendar processing, and the stacking configuration of the “jelly roll” or pouch. When the battery is assembled, in its first “formation” step, it is tested to create a solid electrolyte interface and then tested for cycle life performance at different charge rates and different test temperatures.

[0059] Such semantic alignment of the data is for the purpose of specifying not only the situation (e.g., which tests are relevant to which cells), but also which processes the cells passed through and from which batches of materials those cells were constructed. Such alignment provides a connection of material characteristics at the raw material, sub-cell, cell, and test levels.

[0060] The complexity of what is a batch vs. sample vs. test vs. experiment vs. quality gate depends not only on simple linguistic semantics, but also on the mutual conversion of scales of time and length (from nanoseconds to hours, or cycles), the connection of the design to models and raw data (master data alignment), and the persistence of requirement and verification implementation. Exemplary embodiments enable, as non-limiting examples, the alignment of chemical, material, test metrics for both virtual and real data within a unified data model incorporating logical, functional, physical, and result requirements.

[0061] In the above action (2), the generative method 180 constructs, as non-limiting examples, models of properties of interest such as open circuit voltage, flammability, capacity, maximum cycle life, maximum charge current (C-rate) during discharge and / or charge. The generative method 180 may construct models, as non-limiting examples, estimated from a hybrid of atomic / material and in-use data such as battery telemetry, life, and aging behavior. The generative method 180 may construct models representing the degree of expected variance in battery production quality metrics in volume and scale-up production.

[0062] Once data from tests is integrated into the system, models can be constructed using full three-dimensional fidelity, enabling high-fidelity monitoring of processes and changes occurring in different parts of the storage device (e.g., battery) life cycle. Such models may be pyramids constructed of physical property-based features, geometry-based features, process-based features, and test-based data. Further, as additional data is generated, these models may be versioned and then incremented relative to their preceding models and evaluated with respect to the accuracy of their values and error prediction.

[0063] In the above action (3), the computer-based system 104 may use a model as a predictor and execute a design space experiment design over variables using the model as a simulation for physical testing, and may use Monte Carlo and evolutionary approaches to identify an optimal solution to the design space requirements.

[0064] In the above action (4), the computer-based system 104 may store data into a database (not shown). As a non-limiting example, the computer-based system 104 may store models, their versions, their input / output / parameters / components into the database for both activity tracking and tracing and for building increased storage of knowledge and prediction. The simulation may be fast and may perform screening and verification of complex solutions, so the ability to perform approximate trade-offs at a lower resolution is useful for identifying interesting domains of the design space (e.g., chemistry, recipes, and processes).

[0065] In the above action (5), the computer-based system 104 may filter user or defined machine criteria. By filtering and sorting the output of the models within the system, the user is provided with the ability for Pareto-based optimization within the loop, and additional generation of values and sub-scenario analysis of the generated data becomes possible. This is useful at a stage after the base model construction of a hybrid or adversarial training process.

[0066] In action (6), the computer-based system 104 may select the best candidate and confirm the design. Real-world evidence of the prediction may be created and linked to the experimental data. The deviation in the model prediction may be linked to the model version, and the re-learning process may be automatically initiated to improve and optimize the model in a cyclic form.

[0067] In Action (7), the computer-based system 104 may be iterative. In this approach, useful elements are integrated data and the cyclic nature of the process. The data is generated from both experiments (reality) and models (virtual). As the computer-based system 104 constructs models and data within its framework, the balance between reality and virtual shifts. Value is that the specifications and limits of materials, processes, and innovations in chemistry can be virtually screened and optimized without the need for continuous testing and can be verified at the end of the cycle. FIG. 2 disclosed below may be employed as the computer-based system 104 disclosed above and may also be employed to implement Actions (1)-(7), and is a block diagram of an exemplary embodiment of a computer-based system 204.

[0068] FIG. 2 is a block diagram of an exemplary embodiment of a computer-based system 204 for the generative design of an energy storage device 206. The computer-based system 204 may be employed as the computer-based system 104 of FIG. 1A disclosed above. The computer-based system 204 includes, by way of non-limiting example, at least one memory 208 and at least one processor 218 coupled to the at least one memory 208, such as a central processing unit 518 coupled to the memory 508, further disclosed below with respect to FIG. 5.

[0069] Continuing to refer to FIG. 2, at least one processor 218 of the computer-based system 204 is configured to automatically construct at least one model 212 of the energy storage device 206 based on the design parameter space 210. Such construction may employ a machine learning process 214. The at least one processor 218 may be further configured to automatically execute a simulation 216 of the energy storage device 206. The simulation 216 may employ the design parameter space 210, the design evaluation space 220, and the at least one constructed model 212. The execution 216 may include generating at least one prediction 222 of the energy storage device 206 that achieves at least one product design objective (not shown) or at least one model design objective (not shown) achieved by the at least one constructed model 212.

[0070] As described above, building at least one model 212 may employ a machine learning process 214. Machine learning can be used at several stages of model building and evaluation. The model is effectively mapped from a parameter space to a characteristic space and includes feature generation. Features may include real-world measurements and, by way of non-limiting example, may include aggregates, derivatives, or transforms of such real-world experiments. By way of non-limiting example, a signal measured per second may be averaged over days or weeks for a long-term trend model. The augmentation of real-world evidence with derived or generated features can be performed using the model. Further, when a large descriptor space is so generated, the selection of the important variables that are actually mapped to the characteristics of interest can be performed using the model. Of course, the evaluation of the model, i.e., the evaluation of its construction, can be performed by mechanical means. Some of these evaluations may include model form, the functions used in the model, the modeled domain of applicability, the model error function, and other features or metrics. Still further, the evaluation of the model for optimal conditions or trade-off regions can be performed using any of a plurality of models, consensus models, adversarial models, or even a system of system models.

[0071] At least one processor 218 may be further configured to automatically evolve at least one of (i) a design parameter space 210 and (ii) a design evaluation space 220. The evolution may be based on at least one generated prediction 222 and may employ a machine learning process 214. If at least one prediction 222 indicates that at least one product design objective (e.g., longer lifespan, lower weight, lower cost, etc.) has been achieved or at least one model design objective (e.g., accuracy, range of applicability, etc.) has been achieved, at least one processor 218 may be further configured to automatically converge on the evolved design parameter space 210, thereby completing the generative design 224 of the energy storage device 206 and, otherwise, repeating constructing, executing, and evolving.

[0072] In many real-world cases, what affects the accuracy and applicability of a model is the constraint of the design space with respect to a manageable set of axes or the choice of forms and functions used within the model. As a non-limiting example, to model a performance gradient, a simple straight line linking two points can be used, which can then evolve such that more information is presented in a polynomial fit and then further evolve such that even more information is presented in a discontinuous surface function.

[0073] As another method, the number of similar terms may increase. For example, a first-order model y = f(x) may evolve into a second-order model y2 = f(x) + f(y) + f(x*y) using a certain level of cross-variable interaction effects while remaining linear. As the number of variables increases to four or more variables, the number of combinations of x, y, z, and w increases to 14, and an evolutionary process (i.e., evolution) may be used to identify which set or combination of these variables is best for the known data, i.e., which fits the known data best. In the evolutionary process, variables and their functions, polynomials, or non-linearities may be described by encoded patterns. The patterns may be ranked against the data, and then pair-wise crossovers may be randomly performed to form a second generation. The second generation may be ranked and compared against the first generation, and the best set may proceed for the next complete iteration of function evolution. The advantages of such evolution may be realized from both the perspective of the range and spread of fitting functions that the system can handle as well as the non-obvious discoveries that it can reveal. This identification of the primary variables and their combinations serves to reduce the dimensionality of the design space and enable its prioritization for virtual and physical testing.

[0074] At least one model 212 may include, by way of non-limiting example, a material model, a mixture and recipe model, a configuration and design model, a process and production model, a use and end-of-life model, or combinations thereof.

[0075] At least one processor 218 implements a machine learning process 214, and in the machine learning process 214 may be further configured to employ a generative adversarial network (GAN), a deep neural network (DNN), a Bayesian optimization (BAO), a genetic function approximation method, or combinations thereof.

[0076] The design evaluation space 220 may include at least one test (not shown) of the energy storage device 206. To perform the simulation 216, at least one processor 218 may be further configured to simulate at least one test using the at least one constructed model 212.

[0077] At least one product design objective may include at least one user-specified criterion (not shown) associated with the energy storage device 206, at least one machine-generated criterion (not shown) associated with the energy storage device 206, or a combination thereof. At least one product design objective may further include a TPP.

[0078] At least one model design objective may include at least one error threshold associated with the difference between the simulated measurements and the real-world measurements, the target size for the design parameter space 210, or a combination thereof.

[0079] The design parameter space 210 may include semantically structured real-world evidence (RWE) data (not shown) associated with experiments of the energy storage device 206. The design parameter space 210 and the design evaluation space 220 may be associated with at least one of the energy storage device 206 and the at least one constructed model 212. The design parameter space 210 may include variables, for example, as disclosed with respect to FIG. 3 below.

[0080] FIG. 3 is a block diagram of an exemplary embodiment of a design parameter space 310 that may be employed as the design parameter space 210 disclosed above in FIG. 2. In the exemplary embodiment of FIG. 3, the design parameter space 310 includes a variable 330. Referring to FIGS. 2 and 3, the variable 330 may include, by way of non-limiting example, a material variable (not shown), a system variable (not shown), or a combination thereof. The variable 330 may be associated with a chemical space 332 of the energy storage device 206, a formulation space 334 of the energy storage device 206, a material space 336 of the energy storage device 206, a configuration space 338 of the energy storage device 206, a process space 340 of the energy storage device 206, or a combination thereof.

[0081] Referring back to FIG. 2, at least one processor 218 of the computer-based system 204 may be further configured to (a) clip the design parameter space 210, (b) clip the design evaluation space 220, (c) expand the design parameter space 210, (d) expand the design evaluation space 220, or (e) perform a combination of (a) through (d) to automatically evolve the design parameter space 210, the design evaluation space 220, or a combination thereof. The at least one processor 218 may be further configured to maintain diversity within the design parameter space 210 by employing, by way of non-limiting example, a clustering method, a Pareto method, or a combination thereof.

[0082] At least one processor 218 may be further configured to employ at least one monitored parameter (not shown) in construction, evolution, or a combination thereof. The at least one monitored parameter may be employed in at least one iteration of construction, evolution, or a combination thereof. The at least one monitored parameter may represent at least one real-world result generated via at least one real-world experiment that employs the energy storage device 206. The real-world result may be implemented via the use of the evolved design parameter space 210 in at least one real-world experiment.

[0083] At least one processor 218 may be further configured to automatically store in at least one memory 208 a constructed at least one model 212 associated with at least one generated prediction 222, at least one input (not shown) of the constructed at least one model 212, and at least one output (not shown) from the constructed at least one model 212. To execute the simulation 216, at least one processor 218 may be further configured to input at least one input to the constructed at least one model 212. The constructed at least one model 212 may be configured to generate at least one output in response to the at least one input.

[0084] According to a non-limiting example, the energy storage device 206 may be a battery, as disclosed above with respect to FIG. 1A. To converge on the evolved design parameter space 210, at least one processor 218 may be further configured to identify at least one of a compound, component, additive, formulation, recipe, or combination thereof that enables at least one product design objective of the energy storage device 206 to be achieved. Referring back to FIG. 1A, as a non-limiting example, at least one product design objective may be that no deformation 113 is formed at the center 115 of the jelly roll of the energy storage device 106 over a specified life cycle. The generative design of the energy storage device 106 may be implemented via a computer-implemented method, as disclosed below with respect to FIG. 4, for example.

[0085] FIG. 4 is a flow diagram of an exemplary embodiment of a computer-implemented method (400) for a generative design of an energy storage device. The method begins (402) and automatically constructs (404) at least one model of the energy storage device. The construction (404) may be based on a design parameter space or may employ a machine learning process. The computer-implemented method automatically performs a simulation (406) of the energy storage device. The simulation may employ the design parameter space, a design evaluation space, and the at least one constructed model. The performance (406) may include generating at least one prediction of the energy storage device that achieves at least one product design objective or at least one model design objective achieved by the at least one constructed model. The computer-implemented method automatically evolves (408) at least one of (i) the design parameter space and (ii) the design evaluation space. The evolution (408) may be based on the at least one generated prediction or may employ a machine learning process. The computer-implemented method checks (410) whether at least one prediction indicates that at least one product design objective has been achieved or at least one model design objective has been achieved. If at least one prediction indicates that at least one product design objective has been achieved or at least one model design objective has been achieved, the computer-implemented method automatically converges (412) on the evolved design parameter space, thereby completing the generative design of the energy storage device, and in the exemplary embodiment, the computer-implemented method then ends (414). Otherwise, the computer-implemented method repeats the construction (404), performance (406), and evolution (408) disclosed above.

[0086] Evolution (408) may include (a) truncating a design parameter space, (b) truncating a design evaluation space, (c) expanding a design parameter space, (d) expanding a design evaluation space, or (e) a combination of (a) to (d). Evolution (408) may include maintaining diversity within the design parameter space. Maintenance may include adopting a clustering method, a Pareto method, or a combination thereof.

[0087] The computer-implemented method (400) may further include adopting at least one monitored parameter in construction, evolution, or a combination thereof. The at least one monitored parameter may be adopted in at least one iteration of construction (404), evolution (408), or a combination thereof. The at least one monitored parameter may represent at least one real-world result generated via at least one real-world experiment adopting an energy storage device. The real-world result may be implemented based on the adoption of an evolved design parameter space in at least one real-world experiment.

[0088] The computer-implemented method (400) may further include adopting a GAN, DNN, BAO, genetic function approximation method, or a combination thereof in a machine learning process.

[0089] The computer-implemented method (400) may further include automatically storing at least one constructed model in a database, associated with at least one generated prediction, at least one input of the at least one constructed model, and at least one output from the at least one constructed model. Executing a simulation may include inputting at least one input into the at least one constructed model and generating at least one output from the at least one constructed model in response to the at least one input.

[0090] The energy storage device may be a battery. The convergence (412) on the evolved design parameter space may include identifying at least one of a compound, a component, an additive, a formulation, a recipe, or a combination thereof that enables at least one product design objective of the energy storage device to be achieved. The computer-implemented method (400) may be implemented via a computer having an internal structure, as a non-limiting example, for example as disclosed below.

[0091] FIG. 5 is a block diagram of an example of the internal structure of a computer 500 in which various embodiments of the present disclosure may be implemented. The computer 500 includes a system bus 502, which is a set of hardware lines used for data transfer between components of a computer or digital processing system. The system bus 502 is essentially a shared conduit that connects different elements of a computer system (e.g., a processor, disk storage, memory, input / output ports, network ports, etc.) to enable information transfer between the elements. An I / O device interface 504 is coupled to the system bus 502 to connect various input and output devices (e.g., a keyboard, a mouse, a display monitor, a printer, a speaker, a microphone, etc.) to the computer 500. A network interface 506 enables the computer 500 to be connected to various other devices attached to a network (e.g., a global computer network, a wide area network, a local area network, etc.). The memory 508 provides volatile or non-volatile storage for computer software instructions 510 and data 512 that may be used to implement embodiments of the present disclosure (e.g., methods 150, 400), where the volatile and non-volatile memories are examples of non-transitory media. The disk storage 513 also provides non-volatile storage for computer software instructions 510 and data 512 that may be used to implement embodiments of the present disclosure (e.g., method 400). The central processing unit 518 is also coupled to the system bus 502 and provides for the execution of computer instructions.

[0092] Further exemplary embodiments disclosed herein may be configured using a computer program product; for example, the control may be programmed in software to implement the exemplary embodiments. Further exemplary embodiments may include a non-transitory computer-readable medium containing instructions executable by a processor, which, when loaded and executed, cause the processor to complete the methods and techniques described herein. Of course, the elements of the block diagrams and flowcharts may be implemented in software or hardware, or their equivalents, firmware, combinations thereof, or other similar implementations to be determined in the future, for example, via one or more arrangements of the circuits of FIG. 5 disclosed above.

[0093] In addition, the elements of the block diagrams and flowcharts described herein may be combined or divided in any manner in software, hardware, or firmware. When implemented in software, the software may be written in any language capable of supporting the exemplary embodiments disclosed herein. The software may be stored in any form of computer-readable medium, such as random access memory (RAM), read-only memory (ROM), compact disc read-only memory (CD-ROM), etc. In operation, a general-purpose or special-purpose processor or processing core loads and executes the software in a manner well understood in the art. Of course, the block diagrams and flowcharts may include more or fewer elements, be arranged or oriented differently, or be represented differently. Of course, the implementation may also indicate block diagrams, flowcharts, and / or network diagrams showing the execution of the embodiments disclosed herein, as well as the numbers of the block diagrams and flowcharts.

[0094] Although the exemplary embodiments have been specifically shown and described, it will be understood by those skilled in the art that various changes in form and detail may be made therein without departing from the scope of the embodiments encompassed by the appended claims.

Claims

Claim 1 A method executed by a computer for a generative design of an energy storage device, comprising: automatically constructing at least one model of the energy storage device, said constructing being based on a design parameter space and employing a machine learning process; automatically performing a simulation of the energy storage device, said simulation employing the design parameter space, a design evaluation space, and the at least one constructed model, said performing including generating at least one prediction of the energy storage device to achieve at least one product design objective or at least one model design objective achieved by the at least one constructed model; automatically evolving at least one of (i) the design parameter space and (ii) the design evaluation space, said evolving being based on the at least one generated prediction and employing the machine learning process; employing at least one monitored parameter in said constructing, evolving, or a combination thereof, said at least one monitored parameter representing at least one real-world result generated through at least one real-world experiment employing the energy storage device, said real-world result being implemented based on employing the evolved design parameter space in the at least one real-world experiment; completing the generative design of the energy storage device by automatically converging in the evolved design parameter space when the at least one prediction indicates that the at least one product design objective has been achieved or the at least one model design objective has been achieved, and repeating said constructing, performing, and evolving in response to the at least one prediction indicating that the at least one product design objective has not been achieved and the at least one model design objective has not been achieved; A method comprising the above steps. Claim 2 The method according to claim 1, wherein the design parameter space and the design evaluation space are associated with at least one of the energy storage device and the at least one constructed model, the design parameter space includes variables, the variables include material variables, system variables, or combinations thereof, and the variables are associated with the chemical space of the energy storage device, the formulation space of the energy storage device, the material space of the energy storage device, the configuration space of the energy storage device, the process space of the energy storage device, or combinations thereof.

3. The method according to claim 1, wherein the design evaluation space includes at least one test of the energy storage device, and performing the simulation includes simulating the at least one test using the at least one constructed model.

4. The method according to claim 1, wherein the at least one product design objective includes at least one user-specified criterion associated with the energy storage device, at least one machine-generated criterion associated with the energy storage device, or combinations thereof, the at least one product design objective includes a target product profile (TPP), and the at least one model design objective includes at least one error threshold associated with the difference between the simulated measurement and the real-world measurement, a target size for the design parameter space, or combinations thereof.

5. The method according to claim 1, wherein evolving includes (a) truncating the design parameter space, (b) truncating the design evaluation space, (c) expanding the design parameter space, (d) expanding the design evaluation space, or (e) combinations of (a) to (d).

6. The method according to claim 1, wherein evolving includes maintaining diversity within the design parameter space, and maintaining includes employing a clustering method, a Pareto method, or combinations thereof.

7. The method according to claim 1, further including employing an adversarial generative network (GAN), a deep neural network (DNN), a Bayesian optimization (BAO), a genetic function approximation method, or combinations thereof in the machine learning process.

8. The method of claim 1, wherein the design parameter space includes semantically structured real-world evidence (RWE) data associated with experiments on the energy storage device.

9. The method of claim 1, further comprising automatically storing the at least one constructed model in a database, associated with the at least one generated prediction, at least one input of the at least one constructed model, and at least one output from the at least one constructed model, wherein performing the simulation further includes inputting the at least one input into the at least one constructed model and generating the at least one output from the at least one constructed model in response to the at least one input.

10. The method of claim 1, wherein the energy storage device is a battery, and identifying at least one of a compound, component, additive, formulation, recipe, or combination thereof that converges in the evolved design parameter space and enables achieving the at least one product design objective of the energy storage device.

11. A computer-based system for generative design of an energy storage device, comprising: at least one memory; at least one processor coupled to the at least one memory, automatically constructing at least one model of the energy storage device based on a design parameter space, the constructing employing a machine learning process; and automatically performing a simulation of the energy storage device, the simulation employing the design parameter space, a design evaluation space, and the at least one constructed model, the performing including generating at least one prediction of the energy storage device that achieves at least one product design objective or at least one model design objective and achieving the at least one constructed model. Automatically evolving at least one of (i) the design parameter space and (ii) the design evaluation space, wherein the evolving is based on the at least one generated prediction and employs the machine learning process, In the constructing, evolving, or combinations thereof, employing at least one monitored parameter, wherein the at least one monitored parameter represents at least one real-world result generated via at least one real-world experiment employing the energy storage device, and the real-world result is implemented in the at least one real-world experiment based on employing the evolved design parameter space, Completing the generative design of the energy storage device by automatically converging in the evolved design parameter space when the at least one prediction indicates that the at least one product design objective has been achieved or the at least one model design objective has been achieved, and repeating the constructing, performing, and evolving in response to the at least one prediction indicating that the at least one product design objective has not been achieved and the at least one model design objective has not been achieved, At least one processor configured to perform, A computer-based system comprising.

12. The design parameter space and the design evaluation space are associated with at least one of the energy storage device and the at least one constructed model, The design parameter space includes variables, The variables include material variables, system variables, or combinations thereof, The variables are associated with the chemical space of the energy storage device, the formulation space of the energy storage device, the material space of the energy storage device, the configuration space of the energy storage device, the process space of the energy storage device, or combinations thereof, The design evaluation space includes at least one test of the energy storage device, and the at least one processor for performing the simulation is further configured to simulate the at least one test using the at least one constructed model. The at least one product design objective includes at least one user-specified criterion associated with the energy storage device, at least one mechanically generated criterion associated with the energy storage device, or a combination thereof. The at least one product design objective includes a target product profile (TPP), and The at least one model design objective includes at least one error threshold associated with a difference between a simulated measurement and a real-world measurement, a target size for the design parameter space, or a combination thereof. The computer-based system according to claim 11.

13. The at least one processor is configured to (a) clip the design parameter space, (b) clip the design evaluation space, (c) expand the design parameter space, (d) expand the design evaluation space, or (e) perform a combination of (a) to (d) maintain diversity within the design parameter space by employing a clustering method, a Pareto method, or a combination thereof. The computer-based system according to claim 11, further configured to perform the above.

14. The at least one processor is further configured to implement the machine learning process and to employ a generative adversarial network (GAN), a deep neural network (DNN), a Bayesian optimization (BAO), a genetic function approximation method, or a combination thereof in the machine learning process. The computer-based system according to claim 11.

15. The design parameter space includes semantically structured real-world evidence (RWE) data associated with experiments of the energy storage device. The computer-based system according to claim 11.

16. The at least one processor is further configured to automatically store the at least one constructed model, associated with the at least one generated prediction, at least one input of the at least one constructed model, and at least one output from the at least one constructed model, in the at least one memory, the at least one processor is further configured to input the at least one input to the at least one constructed model to perform the simulation, and the at least one constructed model is configured to generate the at least one output in response to the at least one input, the computer-based system of claim 11.

17. The energy storage device is a battery, and the at least one processor is further configured to identify at least one of a compound, component, additive, formulation, recipe, or combination thereof that enables the at least one product design objective of the energy storage device to be achieved in order to converge in the evolved design parameter space, the computer-based system of claim 11.

18. A non-transitory computer-readable medium for generative design of an energy storage device, which, when loaded and executed by at least one processor, causes the at least one processor to automatically construct at least one model of the energy storage device based on a design parameter space, the constructing employing a machine learning process, automatically perform a simulation of the energy storage device, the simulation employing the design parameter space, a design evaluation space, and the at least one constructed model, the performing including generating at least one prediction of the energy storage device that achieves at least one product design objective or at least one model design objective achieved by the at least one constructed model. Automatically evolving at least one of (i) the design parameter space and (ii) the design evaluation space, wherein the evolving is based on the generated at least one prediction and employs the machine learning process; In the constructing, evolving, or combination thereof, employing at least one monitored parameter, wherein the at least one monitored parameter represents at least one real-world result generated via at least one real-world experiment employing the energy storage device, and the real-world result is implemented in the at least one real-world experiment based on employing the evolved design parameter space; Completing the generative design of the energy storage device by automatically converging in the evolved design parameter space when the at least one prediction indicates that the at least one product design objective has been achieved or the at least one model design objective has been achieved, and repeating the constructing, executing, and evolving in response to the at least one prediction indicating that the at least one product design objective has not been achieved and the at least one model design objective has not been achieved; A non-transitory computer-readable medium having encoded thereon a sequence of instructions to cause the above.

Images