Adaptive optimization techniques for accelerated battery charging protocols

Abstract

Disclosed are methods and systems for implementing optimization techniques for accelerated battery charging protocols. A state estimator model configured as a synthetic data generator generates a synthetic dataset from an input data. The synthetic dataset comprises one or more states of a battery. A surrogate model generates a set of features comprising exotic parameters representing internal attributes of the battery by applying time-series analysis to the synthetic dataset. A reduced order ML model determines a minimum charging time based on an optimization control parameter and a subset of features derived from the set of features while maintaining the optimization control parameter in an acceptable range. One or more charging protocols are generated based on the minimum charging time and the optimization control parameter. In some embodiments, an optimal charging protocol is further selected from the one or more charging protocols.

Description

ADAPTIVE OPTIMIZATION TECHNIQUES FOR ACCELERATED BATTERY CHARGING PROTOCOLS CROSS-REFERENCE

[0001] The present application claims the benefit of US Serial No. 63 / 606,837, filed December 6, 2023, the entire content of which is incorporated herein by reference into this application.FIELD

[0002] This specification generally relates to charging protocols for rechargeable batteries (secondary batteries).BACKGROUND

[0003] Rechargeable batteries are desirable power sources for a wide range of applications. Rechargeable batteries are based on a wide range of battery technologies. For example, a rechargeable battery can be a lithium-ion battery (LIB), a lithium metal battery, liquid electrolyte based battery, a gel polymer electrolyte based battery, or an all-solid state battery (ASSB). In some embodiments, a secondary battery is anode-free (anode-less). The desirability of these rechargeable battery technologies stems in part from their relatively high energy density. However, with increased charge / discharge activity or ageing of the battery, rechargeable batteries such as LIBs undergo a decrease in performance, such as a decrease in the amount of energy available and an increased propensity for failure due to physical degradation.SUMMARY

[0004] The present disclosure describes optimization techniques for accelerated charging protocols for a rechargeable battery such as LIB. For example, a new or existing charging protocol can be optimized to accelerate a charging speed of the battery while also minimizing cell degradation from accumulated charge / discharge cycles and iterative usage of the battery. More specifically, the optimization techniques are adaptive such that a charging protocol can be optimized iteratively based on the age and chemistry of the battery.

[0005] The disclosed techniques can be implemented using a computing system configured to accelerate battery charging processes based on inferences computed by machine-learning (ML) models. More specifically, the computing system is configured to determine and / or select a particular charging protocol that is uniquely optimized for the underlying battery chemistry of the Li-ion battery. Additionally, the disclosed techniques are used in the design and optimization of fast charging algorithms to achieve rapid battery charging speeds while concurrently preserving battery health, cycling life, and overall safety. The proposed approachefficiently adapts (or optimizes) charging strategies based on comprehensive state values that accurately reflect charging levels and an overall condition of a battery.

[0006] The system includes a state estimator model that is a part of a digital twin of a particular battery. The state estimator model is an exemplary physics-based model configured as a synthetic data generator for the battery and is operable to generate a synthetic dataset comprising one or more states of the battery. The state estimator model generates the synthetic dataset by processing a set of input data. The input data can include data and information such as material properties, battery chemistry, current vs voltage profiles, and battery temperature measurements. In some embodiments, the input data is processed or formatted by a data processing algorithm into a pre-processed data, so that the pre-processed input data is in a format recognized or otherwise processable by a state estimator model. In some embodiments, the data processing may facilitate data generation and / or enhance accuracy thereof. In some implementations, the state estimator model is a pseudo-two-dimensional (P2D) model representing an electrochemical model that captures the kinetics, transport processes, and thermodynamics of a rechargeable battery such as LIB. In some embodiments, the input data comprises one or more measurable parameters.

[0007] The system further includes a surrogate ML model that generates a set of engineered features (alternatively features) from time-series analysis applied to the synthetic dataset, or to a combination of synthetic data and non-synthetic data such as real experimental data derived from direct physical testing of the battery. In some implementations, a set of engineered features includes a set of feature types and each feature type has a value or a range of value representing the feature type. In some implementations, a set of engineered features include a set of feature types, each feature type has a value or a range of value and has a weighting as determined by a surrogate ML model. The set of engineered features can include exotic parameters, which are certain non-measurable quantities that can represent a particular internal attribute of the battery. For example, a set of features can include a range of voltage, temperature, and electrode and interface stress values, as well as exotic parameters that indicate core temperature and internal spread of the battery. In some implementations, the surrogate model generates the set of features based on a respective weighting of different feature types, where the respective weighting is determined by the surrogate model. A surrogate model is an engineering method used when an outcome of interest cannot be easily measured or computed, so an approximate mathematical model of the outcome is used instead. In some embodiments, a surrogate model of the present disclosure is a transformer-based ML model.

[0008] To yield accurate battery performance predictions, the surrogate model such as transformer-based ML model cooperates with other data models of the computing system to capture sequential data and determine long-range dependencies among an observed set of input parameters. In some implementations, the surrogate model cooperates with a reduced order ML model to identify the charging protocols with a minimum charging time to charge a battery while simultaneously keeping the maximum temperature and stress during charge in an acceptable range for a battery with a specific state and / or condition. A minimum charging time in the present disclosure refers to an acceptable duration for an accelerated (fast) charging and may have a range of values as long as it meets the criteria for fast charging.

[0009] Based on the disclosed techniques, the computing system can generate a charging protocol (algorithm) for the battery at least based on the minimum charging time and an optimization control parameter, such as temperature, age, or stress. An optimization control parameter comprises one or more types of parameters for controlling the optimization. Each parameter comprises a parameter type and a value or a range of value corresponding to the parameter type. In some implementations, all the optimization control parameters can be user- defined. In some implementations, some of the optimization control parameters can be user- defined. In some embodiments, the optimization control parameters are generated and recommended by the system in terms of the importance in influencing the final results. For example, the system uses the reduced order ML model to compute or otherwise determine a minimum charging time for the battery based on one or more optimization control parameters and a subset of features that are derived from the set of features. The reduced order model interprets relationships between a set of features and the optimization control parameter and identifies a subset of features that are central to accelerating battery charging times while keeping values of the one or more optimization control parameters in an acceptable range, such as keeping the internal battery temperature and stress as minimum and in an acceptable range. In some cases, the set of features is an input as shown in Fig. 4A. The optimization control parameters are different from the set of features and used to optimize the charging protocol. In some implementations, the optimization control parameters are user-specified.

[0010] Other implementations of this and other aspects include corresponding systems, apparatus, and computer programs, configured to perform the actions of the methods, encoded on computer storage devices. A system of one or more computers can be so configured by virtue of software, firmware, hardware, or a combination of them installed on the system that in operation causes the system to perform the actions. One or more computer programs can beso configured by virtue of having instructions that, when executed by a data processing apparatus, cause the apparatus to perform the actions.

[0011] The subject matter described in this specification can be implemented in particular embodiments to realize one or more of the following advantages. The disclosed techniques can be used to enhance existing battery charging protocols and / or algorithms in a manner that balances accelerated charging speeds with battery health, cycling life, and overall battery safety. In addition to the enhancements, the optimization techniques allow for adapting to dynamic factors, such as the battery state of health and environmental conditions that affect or degrade battery performance.

[0012] One or more concepts of the disclosed techniques are inextricably tied to computer technology. For example, the subject matter of this specification enables efficient generation and distribution of data among various compute nodes of an exemplary computing system for optimized charging of a rechargeable battery. The system uses machine-learning logic, including predictive algorithms and iterative tuning of model weights, to refine its analytical approaches for optimizing a battery charging protocol. The disclosed techniques can integrate seamlessly into existing Battery Management Systems (BMS) and embedded technologies.

[0013] An optimization engine of the system executes specific computing rules for efficiently analyzing and detecting patterns, relationships, and dependencies among different parameters and / or latent variables of a dataset of real and synthetic values. The unique computational approaches of the optimization engine allow for optimizing fast charging protocols based on one or more states of battery, including without limitation, State-of-Health (SOH), State-of- Charge (SOC), State-of-Power (SOP), State-of-Energy (SOE) or any other State (SOX). As indicated above, the system leverages the unique approaches to efficiently adapt (or optimize) charging strategies based on one or more states of a battery, including without limitation that states that reflect charging levels and an overall condition. In some embodiments, each of one or more states includes a state type and a value representing the state type. In some embodiments, the charging protocol is adapted to the one or more states of the battery so that the battery can be charged within a minimum time while keeping the one or more optimization control parameters in an acceptable range.

[0014] The details of one or more implementations of the subject matter described in this specification are set forth in the accompanying drawings and the description below. Other potential features, aspects, and advantages of the subject matter will become apparent from the description, the drawings, and the claims.BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Fig. 1 is an exemplary computing system used in charging a battery.

[0016] Fig. 2A illustrates a block diagram of exemplary resources for optimizing aspects of charging a battery.

[0017] Figs. 2B-2D illustrate graphical examples of input data for processing by a multidimensional data model.

[0018] Fig. 3A illustrates a block diagram of a machine-learning (ML) model.

[0019] Figs. 3B-3D illustrate graphical examples of observed and predicted data associated with the ML model of Fig. 3 A.

[0020] Fig. 4A shows representative resources used to optimize charging protocols for a battery.

[0021] Figs. 4B, 4C and 4D illustrate graphical examples generated using a multi -objective optimization process.

[0022] Fig. 5 is an exemplary process used to determine a charging protocol for charging a battery.

[0023] Fig. 6 is an exemplary process for optimizing parameters (aspects) of an ML model in the computing system of Fig. 1.

[0024] Like reference numbers and designations in the various drawings indicate like elements.DETAILED DESCRIPTION

[0025] Fig. 1 is an exemplary computing system 100 used in charging a battery 102. In general, the system 100 can represent a physical structure of a sophisticated, innovative system comprising distinct components for enhancing existing battery charging protocols.

[0026] For example, the components of system 100 cooperate in synergy to optimize and / or accelerate charging protocols for battery 102. The system 100 is configured to optimize battery charging protocols while ensuring safe operation of the battery 102 and efficient charging of the battery 102. The system 100 is further configured to optimize battery charging protocols and as a result enhance a life of battery 102. In some implementations, the system 100 can generate an optimized charging protocol to apply specific voltage and / or current values, including targeted charge duration values, that minimizes internal degradation to the battery 102, while enhancing a battery’s rate of charge.

[0027] For example, the system 100 can optimize a battery charging protocol such that, when the optimized protocol is used (e.g., repeatedly used), the battery 102 ages at a slower rate in comparison to the one charged using unoptimized protocols. Thus, the system 100 can be configured and leveraged to optimize battery charging protocols to extend the lifetime and / or ensure long-lasting performance of a battery 102.

[0028] In Fig. 1, the battery 102 is a rechargeable battery. More specifically, battery 102 is a rechargeable battery with a particular battery chemistry. In some embodiments, the battery chemistry is selected from a wide range of chemistries based on chemical compositions. For example, battery 102 can be a lithium-ion (Li-ion) battery, a lithium-sulfur battery (Li-S), a sodium-sulfur battery, a magnesium-ion battery, or other type of rechargeable battery. In general, battery 102 can be designed from a wide range of battery technologies. In some examples, battery 102 comprises liquid electrolytes, gel polymer electrolyte, semi-solid (quasisolid) electrolyte, inorganic solid electrolyte, such as sulfide-based electrolyte, or combinations thereof. For example, battery 102 is a lithium-ion battery, a lithium metal battery, all solid- state battery (ASSB), and an all solid-state lithium-ion battery. In some embodiments, battery 102 is an anode-free (anode-less) battery.

[0029] In some implementations, the battery 102 includes one or more battery cells 103. For example, the battery 102 can include n number of cells 103, where n is an integer greater than or equal to one. The number, / / , of cells 103 in battery 102 may vary based on the intended end-use application. As typically shown in Fig. 1, each of cells 103 includes a cathode, anode, and a separator 105. In some cases, at least one of cells 103 comprises an electrolyte between either electrode and the separator 105. In some examples, battery 102 may leverage alternate design methodologies (e.g., anode-free designs) using cell structures that differ from the example shown at Fig. 1. In some implementations, battery 102 is a rechargeable battery for an electronic vehicle, such as a car, bus, or aircraft, or for an electronic device, such as a smartphone or laptop.

[0030] The system 100 further includes a client device 104, a sensor system 106, a charging protocol optimization system 108, and a charge control module 110.

[0031] Although the client device 104 in Fig. 1 appears to be depicted as a desktop computer or console, the client device 104 can be any known computing device / system, such as a desktop computer, a laptop computer, a tablet device, a mobile device, a smartphone, or any other related computing device that receives user input and that can transmit, transfer, or otherwise provide data and input commands to another device of system 100. The client device 104 can be optionally coupled to a computing device / system 104-1. For example, the client device 104 can be coupled to computing device 104-1 via a wired or wireless interface connection. In some implementations, client device 104 is coupled to, or communicates with, a remote or nonlocal computing server 104-2 or cloud-based computing asset 104-2. For example, device 104- 2 can be a client-server that receives input from the client device 104, serves content to the client device 104, or both.

[0032] The sensor system 106 includes one or more groupings of sensors that are adapted to obtain a range of data, e.g., sensor data specific to the battery 102. For example, the sensor system 106 can include voltage sensors, current sensors, and thermal couples or related temperature sensors. In general, the sensor system 106 can include a variety of sensors or sensor types that are useful in evaluating or modeling parameters (aspects) of battery 102, which can include inferring or predicting boundary conditions that are central to safe operation of the battery 102.

[0033] The sensor system 106 collects and / or generates various types of data for characterizing performance of the battery 102. In some implementations, the sensor system 106 is configured to generate and / or analyze a broad range of sensor data and related battery chemistry information to determine State-of-Charge (SOC), State-of-Health (SOH), and State-of-Power (SOP) for the battery 102. The range of sensor values and related battery chemistry information analyzed and / or generated by the sensor system 106 can be characterized generally as battery performance metrics. The sensor system 106 is configured to generate and / or analyze battery performance metrics specific to charge and discharge activity of the battery 102.

[0034] The charging protocol optimization system 108 includes at least one model 122 such as a machine learning (ML) model and an optimization engine 124. The charging protocol optimization system 108 can be implemented in hardware, software, or both. In some embodiments, the model 122 employs one or more ML algorithms for generating a robust surrogate model. The optimization engine 124 performs its protocol optimization functions based on one or more algorithms. For example, at least one algorithm used by the optimization engine 124 can be a nondominated sorting genetic algorithm II (NSGA-II).

[0035] In general, machine learning relates to categories of algorithms in computer sciences that, when implemented, enable systems to exhibit abilities to detect and learn patterns, relational connections, and dependencies in data. Machine learning can sometimes come under the broader category of Artificial Intelligence. For example, learning relationships and patterns can correspond to the model 122 learning or interpreting charging protocol samples where there are differences from one sample to another as well as certain distinct underlying features of each step or task in a sample protocol. Learning the relationships and patterns can also include the model 122 learning or interpreting how adjusting details of certain steps or tasks in the sample protocol affects, for example, charging durations and battery performance for specific battery chemistries.

[0036] In other examples, machine learning can include mapping a set of inputs, such as a large set of performance data for batteries of varying chemistries, e.g., differing combinations ofelements, compounds, and chemical properties. Machine learning can also include mapping a set of output(s) which, in this battery example, can be groupings of favorable battery performance data detected among the varying battery chemistries. In this example battery context, supervised learning is a sub-category of ML algorithms that can be leveraged by the model 122 to determine the mappings. For example, the supervised learning approach can be marked by an availability of actual samples of specific battery charging protocols, voltage or current values, and battery chemistries, and corresponding correct annotations or labels of the samples verified by a human, another ML model, or both.

[0037] The charge control module 110 is a controller configured to generate and / or provide control signaling for controlling charge and discharge activity of the battery 102. The charge control module 110 includes a protocol logic 120 for identifying, selecting, generating, and / or determining a particular charging protocol. Example charging protocols include a multi-step constant current (MSCC) protocol, multi-step constant current constant voltage (MSCCCV) protocol, intermittent charging protocol, and a constant power (CP) protocol. Additional or related methods (or protocols) for charging a battery 102 are also within the scope of this specification.

[0038] The system 100 can use the charge control module 110 to determine one or more details of a charger that is, or will be, used to charge the battery 102. For example, the protocol logic 120 of charge control module 110 can be configured to determine a type of charger for use in charging battery 102. In some implementations, the charge control module 110 determines details of the charger based on details of the battery 102, such as the model number or chemistry of the battery 102. The system 100 can pass details of the charger to the charging protocol optimization system 108 and determine an optimized charging protocol based at least on a type or other details of the charger.

[0039] In some implementations, a particular charging protocol is implemented and / or augmented based on a corresponding charging algorithm. For example, the charge protocol model 110 can implement an MSCC protocol by iterating through the computational steps of a typical charging algorithm. At least one step of the algorithm involves the charge control module 110 applying a constant current with a selected value to battery 102 for a particular duration. An exemplary optimization scheme may include adjusting the duration of the constant current (CC), and / or other parameters, to achieve a particular charging speed. In some implementations, a initial (first) application of CC with a first value for a first duration may be followed by at least a second application of CC with a second value for a second duration. Thefirst value and the second value can be the same or different time durations. The first duration and the second duration can be the same or different time durations.

[0040] The charge control module 110 cooperates with the sensor system 106 to generate a comprehensive set(s) of battery performance metrics. At least one set of battery performance metrics can be provided as input data to the charging protocol optimization system 108 to generate a corresponding protocol optimization output. In some implementations, the system 100 analyzes and / or annotates certain data or parameters with discrete values in a set of battery performance metrics. The system 100 can generate one or more training datasets based on the analysis and / or annotation of values in the set of performance metrics. The training dataset is provided as input data to machine-learning models of the charging protocol optimization system 108, for example, during a training phase of system 108. This is described in more detail below with reference to Fig. 4A through 4D and Fig. 6.

[0041] Fig. 2A illustrates a block diagram of resources 200 for optimizing charging protocols for a battery. The resources 200 includes a synthetic data generator 202 and the optimization engine 124. Each of the synthetic data generator 202 and the optimization engine 124 can be implemented in hardware, software, or both. A digital twin of a battery is a virtual representation of a physical battery system that integrates real-time sensor data, computational models, and analytics to simulate and predict the battery’s performance over time, under different operation conditions. In some implementations, the synthetic data generator 202 is a part of a digital twin of battery 102 comprising real-time data from the sensor system 106. In some cases, the digital twin comprises a pseudo two-dimensional (P2D) model 204. For example, the P2D model 204 can be configured for modeling galvanostatic charge and discharge of a battery cell, such as LIB and lithium metal battery comprising an electrolyte with or without a polymer. In other implementations, the digital twin comprises a single particle model, a Doyle-Fuller-Newman (DFN) model, an equivalent circuit model, or a combination thereof.

[0042] The digital twin may comprise a 3D model. For example, the digital twin (or model 204) may comprise a physics-based model, such as a porous electrode model, for predicting the performance and lifetime of LIBs. Some physics-based models can range from microscopic 3D models, which spatially resolve microstructural characteristics of all phases in porous electrodes, to reduced order and computationally effective models, which do not resolve the microstructure. In general, the digital twin of battery 102 may comprise any model suitable for modeling battery chemistry, including electrochemical parameters (aspects) for assessing overall performance of a rechargeable battery. In some implementations, the digital twin ofbattery 102 comprises real-time data from the sensor system 106, described above. For example, the digital twin of battery 102 can incorporate or integrate some (or all) real-time data from the sensor system 106.

[0043] A simulation model of the synthetic data generator 202 is calibrated in view of at least part of the real-time data of a digital twin of battery 102 as input data and / or calibration data . For example, a P2D model 204 can be calibrated using a combination of input data 206 and calibration data 208. The calibration data includes a set of experimental results as actually observed. In some implementations, Fig. 2B illustrates multiple graphical examples of calibration data (alternatively calibration input data) that include one or more selected from the group consisting of time-series voltage at a charge rate, time-series voltage at a discharge rate, normalized time-series voltage at a charge rate, normalized time-series voltage at a discharge rate, and derivatives thereof, such as difference and differential value. In some implementations, the P2D model 204 is calibrated by fine-tuning parameters (aspects) based on calibration data 208. As shown in Fig. 2B, the calibration data can include a set of experimental data related to various charge / discharge voltage curves at different charge / discharge rate. For example, the system 100 can perform a calibration based on Bayesian Optimization. The system 100 can perform a calibration process to ensure that alignment between a set of simulation results and observed experimental data satisfies a minimum accuracy threshold. The accuracy threshold may be set to achieve high-fidelity representations of battery behavior profiles for different fast charging protocols. In some embodiments, a state estimator model such as P2D model prior to calibration is referred to as a general state estimator model. In some embodiments, a state estimator model such as P2D model after calibration is referred to a calibrated model or specific model for a particular battery, which generates synthetic dataset with better accuracy.

[0044] The P2D model 204 is configured to generate a synthetic dataset 210 based on the input data 206. The synthetic dataset 210 can include multiple data corresponding to a state of battery 102. Additionally, the P2D model 204 can implement a particular sampling method to generate portions of synthetic dataset 210, a variety of fast charging protocol simulation results, or both. Based on the sampling methodology, the P2D model 204 can generate a synthetic dataset 210. As synthetic dataset 210 comprises a large volume of data, it would capture intricate relationships between charging parameters and battery responses. For example, when parameters associated with a charging current profile of a battery are adjusted, its relationship with the temperature of the cell is captured.

[0045] The intensity of a charge current governs the rate of electrochemical reactions in the anode and cathode of battery 102. A higher charge current usually causes a faster reaction and an increased heat generation. Heat is generated via Joule heating as the current flows from one electrode to the other via electrolyte by overcoming the internal resistance of the battery 102. The heat generation may be related to solid electrodes, electrolyte salt concentration, electrolyte salt concentration gradient and electrode / electrolyte interfacial resistance. In some implementations, the captured relationships reveal that higher currents lead to larger concentration gradients of lithium ions in the electrolyte and the electrode particles. These gradients can cause local overpotentials, which can result in additional heat generation.

[0046] As indicated at Fig. 2A, the input data 206 can encompass a comprehensive set of parameters. For example, the input data 206 can include material properties 214, pseudo-Open Circuit Voltage (OCV) 216, Galvanostatic Intermittent Titration Technique (GITT) experimental results 218, and core temperature measurements.

[0047] The material properties 214 can include sets of general physical properties and spatial properties. A representative set of general physical properties includes one or more selected from the group consisting of diffusion coefficients of components in electrolytes (such as lithium ions and solvent), diffusion coefficients of electrodes, ionic conductivity of electrolytes, electronic conductivity of electrodes, whereas an example set of spatial properties includes particle size distribution, microstructure porosity, microstructure tortuosity, active material surface area, or a combination of these.

[0048] Fig. 2C illustrates a graphical example of OCV input data 216 that includes voltage for processing by the P2D model 204. In Fig. 2C, the voltage relates to pseudo-open circuit potential obtained from C / 50 discharge activity. Fig. 2D illustrates a graphical example of GITT input data 218 that includes time-series potentials for processing by the P2D model 204.

[0049] The P2D model 204 can have a first calibration with a first accuracy and fidelity with respect to its outputs. In some embodiments, the P2D model 204 can have a first calibration based on a first measurement. In some embodiments, the P2D model 204 can have a first calibration based on a first measurement followed by a second calibration based on a second measurement, which leads to a second accuracy and fidelity. In some implementations, the first calibration is based on measurements of voltage. In some embodiments, the second calibration is based on measurements of core temperature of the battery. In some implementations, the P2D model 204 incorporates at least the core temperature measurements to obtain a second measure of accuracy and fidelity that exceeds the first measure. For example, the P2D model 204 incorporates the voltage as first measurement for first calibrationand the core temperature as second measurement for a second calibration to enhance the accuracy and fidelity of the digital twin corresponding to battery 102. This enables the P2D model 204 to provide a comprehensive representation of real battery behavior during a fast charging scenario. In some embodiments, the P2D model 204 is calibrated in view of three or more measurements such as voltage, core temperature, and stress.

[0050] In some implementations, a logical arrangement of resources 200 represents a workflow for implementing and / or executing multi-stage fast charging protocol optimization at system 100. As described in detail below, the workflow couples a digital twin of battery 102, a surrogate model 230 including its corresponding ML algorithms, and a charging protocol optimization block (240) for implementing a multi-objective optimization process. By coupling these resources in the manner disclosed, the example workflow represents a novel and efficient approach to addressing technical challenges for effective optimization of existing battery charging protocols (algorithms).

[0051] In Fig. 2A, outputs of the synthetic data generator 202, such as its experimental results and synthetic dataset 210, are provided as input data to the optimization engine 124 to generate an optimized charging protocol 250 (alternatively optimal charging protocol). The optimization engine 124 comprises a surrogate model 230 and a charging protocol optimization block (240) in operation with each other. In some implementations, the optimized charging protocol 250 is battery specific. For example, the optimization engine 124 cooperates with the surrogate model 230 to generate an optimized charging protocol 250 as an output, where the optimized charging protocol 250 can be one or more charging protocols that are uniquely optimized for the specific design details of battery 102.

[0052] The surrogate model 230 is an example of a model 122 included in the charging protocol optimization system 108 of Fig. 1. In some implementations, the surrogate model 230 is a transformer based machine learning model including a transformer architecture based on a self-attention mechanism. For example, the surrogate model 230 can include a Recurrent Neural Network (RNN) with one or more multi-head attention blocks that each use a scaled dot-product attention mechanism.

[0053] In general, a recurrent neural network is an artificial neural network instantiated in software, and that receives an input sequence and generates an output sequence from the input sequence. A recurrent neural network can use some or all of the internal state of the network from a previous time step in computing an output at a current time step. This allows the recurrent neural network to exhibit dynamic temporal behavior. In other words, the recurrentneural network summarizes all information it has received up to a current time step and is capable of capturing long-term dependencies in sequences of data.

[0054] Regarding the self-attention mechanism, this mechanism allows for each time-step data to refer to other data at other timesteps within the same sequence length. The surrogate model 230 such as transformer based ML model can leverage this approach to compute encoded representations of the sequence data, where the model 230 can perform these computations automatically and without requiring recurrence or convolution operations. In some implementations, the multi-head attention blocks allow the surrogate model 230 to jointly attend to information from different subspaces, such as a real-data subspace and a synthetic data subspace, at different positions in time.

[0055] Fig. 3A illustrates a block diagram of a machine-learning (ML) model. More specifically, the example of Fig. 3 A illustrates additional aspects of the surrogate model 230 described above with reference to Fig. 2A.

[0056] The surrogate model 230 is configured to process one or more datasets comprising the synthetic dataset 210, a non-synthetic dataset, or both. The non-synthetic dataset comprises a set of real experimental data values 302 derived from physical testing of the battery 102, such as data values generated from the sensor system 106 described above. The synthetic dataset 210 and non-synthetic dataset 302 serve as a training dataset for training the surrogate model 230.

[0057] In some examples, a training dataset can include historical battery performance data with predefined target attributes (e.g., annotated or labeled data with values), such as charge duration, voltage, current, constant current duration, constant voltage duration, battery temperature, or a stress parameter. The predefined target attributes or labels are defined for use during model training. An underlying algorithm employed during model training uses the labels to reliably and accurately identify those attributes, determine a corresponding weighting of those target attributes, and ultimately train the model to make predictions using the determined weightings.

[0058] During an example training phase, the surrogate model 230 is iteratively trained by feeding labeled (or annotated) data of the training dataset into one or more transformer-based ML algorithms 304. In general, data labeling (or data annotation) is a process of curating a training dataset to include certain target attributes and labeling each input sample in accordance with a corresponding attribute, so that a machine learning model can learn what predictions it is expected to make. For example, an input sample can be a set of performance data for Battery type A and one or more corresponding attributes or labels can be battery age, voltage, current,and temperature for a specific charging duration. This process is one of the stages in preparing data for supervised machine learning.

[0059] The surrogate model 230 can apply a supervised ML algorithm 304 to the annotated inputs of one or more training datasets. In some implementations, the surrogate model 230 is trained to a threshold accuracy based on a particular underlying cost function. A representative algorithm 304 of the surrogate model 230 can be configured such that every single input data point in an input sample is ‘tokenized’ and / or encoded into embeddings that are used or processed by the model for all (or some) of its ML computations.

[0060] In some implementations, the surrogate model 230 is a transformed based ML model and its ML algorithm(s) 304 are uniquely adapted for encoding numerical data for the timeseries domain. Relatedly, the surrogate model 230 leverages its algorithm(s) 304 to generate predicted data consistent with observed data. For example, Fig. 3B illustrates a graphical example 310B of observed and predicted data 310 corresponding to voltages of battery 102. Likewise, Fig. 3C illustrates a graphical example 320C of observed and predicted data 320 corresponding to temperature values of battery 102; and Fig. 3D illustrates a graphical example 330D of observed and predicted data 330 corresponding to stress associated with battery 102.

[0061] The surrogate model 230 is configured to generate a set of features (engineered features) comprising a set of feature types with corresponding features 306 based on: i) time-series analysis applied to the synthetic dataset 210; ii) time-series analysis applied to the real experimental data 302; or iii) both. The features comprising feature types and feature values include one or more exotic parameters for representing one or more internal attributes of the battery 102. For example, the exotic parameters can indicate or be associated with one or more internal states of battery 102 that are not-directly measurable (e.g., non-measurable) through physical testing of battery 102. A set of engineered features from the transformer-based ML model may include without limitation slope, average value, standard deviation, kurtosis, entropy, skewness and derivatives thereof in relation to a time-series analysis.

[0062] In some implementations, the surrogate 230 is configured to capture intricate relationships between charging parameters (e.g., current, voltage, stress, temperature) and enable dynamic optimization with time-varying data. The surrogate model 230 can be also configurable as a pre-trained feature generator for generating or identifying a set of features from a set of user-specified features for a corresponding battery chemistry of the battery 102.

[0063] The surrogate model 230 is specifically designed and trained to capture sequential data and long-range dependencies, which makes it well-suited for generating realistic battery performance predictions. As an example, the surrogate model 230 is a transformer-based MLmodel and can be trained to effectively replicate behaviors of a corresponding digital twin of the battery 102. Such training can significantly enhance the model’s predictive capabilities.

[0064] Either P2D model or transformer-based ML model can be configured as a synthetic data generator for generating a synthetic dataset 210. More specifically, a trained transformerbased ML model can represent a digital twin of battery 102. The transformer-based ML model can provide advantages such as fast (or faster) computation times relative to a P2D model. In some implementations, a trained transformer-based ML model can be used to accelerate synthetic dataset generation by, for example, lOOx - lOOOx over the P2D model 204. Additionally, gains in speed or accelerated computations are also within the scope of this specification.

[0065] Fig. 4A shows resources to implement the charging protocol optimization block (240) for generating charging protocols. More specifically, it implements a multi -objective optimization (MOO) process 402 to generate one or more charging protocols 408 followed by identifying an optimal charging protocol (optimized charging protocol) 250 for a battery 102.

[0066] An objective of the multi-objective optimization process 402 is to generate an optimal charging protocol 250 with a minimum charging time. For example, the optimal charging protocol 250 has a minimum charging time, while simultaneously keeping a maximum temperature and stress for a fixed State-of-Charge (SOC) range in an acceptable range. The charging protocol can be configured as a multi-stage fast charging approach. In some implementations, the charging protocol can be configured to employ multi-stage constantcurrent (MSCC) each stage comprising charging the battery with a constant current (CC) for a duration. An overall configuration of the charging protocol can be particularly advantageous for maximizing battery cycle life.

[0067] In some implementations, the surrogate model 230 generates a set of features and determines corresponding feature values based on a single-head training, a multi-head training, or both (404). In some cases, the surrogate model is a transformer-based ML model and is trained following a single-head attention network (or model), a multi-head attention network, or both. The single-head training phase and / or the multi-head training is performed to determine a corresponding weighting for a particular feature type. In general, single-head and multi-head relate to attention networks that form the architecture of the transformer in the transformer-based ML model. Input samples are processed in accordance with the specific structure of a single-head attention network or multi-head attention network. In some implementations, a single-head attention network (or model) uses the same output layer foreach task, whereas in a multi-head attention network (or model) the output layer allocates a different set of output units (head) for each task.

[0068] To implement its multi -objective optimization process 402, the charging protocol optimization system 108 includes a reduced-order ML model that is configured to generate a condensed representation of time-series data derived from the time-series analysis applied to the synthetic dataset 210. The reduced-order ML model of system 100 can transform full charge profile data into parameterized versions for individual cycles. The reduced-order ML model is further configured to facilitate efficient handling of diverse charging scenarios and battery states of the battery 102. In some implementations and as shown in Fig. 4A, the engineered features, synthetic dataset and non-synthetic dataset are used as input for the charging protocol optimization 240 to generate an optimal charging protocol 250 as an output. The multi-objective optimization process 402 comprises a training 404 of a machine learning model and a subsequent multi -objective optimization (MOO) process 406. Fig. 4B is a graphical plot of 404B in Fig. 4A as an output from the training 404 according to one embodiment of the present disclosure. Fig. 4B shows the accuracy of the trained machine learning model in predicating the maximum temperature by comparing the predicated result with the true result. The comparison demonstrates that the training machine learning model exhibits high accuracy and consistency in prediction. Fig. 4C is a graphical plot of 406C in Fig. 4A as an output from the MOO process 406 according to one embodiment of the present disclosure. In Fig. 4C, Af represents the difference or improvement in the objective function values between successive generations in the optimization process. Fig. 4C demonstrates the convergence of the optimization. After iterative refinements, for example, after 50 to 80 generations of optimization, the system can obtain a stable solution.

[0069] Fig. 4D is a graphical plot of 408D in Fig. 4A according to one embodiment of the present disclosure. Fig. 4D demonstrates one or more optimal protocols projected to a 3D space defined by normalized time, normalized temperature and normalized stress.

[0070] The process 402 further includes: i) generating the subset of features based on the multiobjective optimization performed for each feature type (406); and ii) performing multiobjective optimization for each feature type using the condensed representation of the timeseries data. The process 402 includes an optimization block 406 indicating iterative optimization over time based on the multi-objective approach of the reduced-order ML model. Fig. 4D demonstrates a typical example of one or more optimal protocols in a 3D space defined by normalized time, normalized temperature and normalized stress. In the example of Fig. 4D, selection of an exemplary optimal protocol 408 can be defined using a collection of data pointsand associated features as illustrated herein. As also shown in Fig. 4A, one optimal charging protocol 250 is selected from the one or more charging protocols 408 in view of experimental validation. In some implementations, the system 100 employs ML algorithms to extract central and / or high-weighted features from an objective space 412 which is more specifically defined and much narrower than a general space. In some implementations, the extracted features are used to understand and / or profile underlying relationships between charging parameters and battery responses. In some implementations, the extracted features are used to accelerate the identification of the subset of features which are central to the design and optimization of the fast charging protocol.

[0071] Fig. 5 is an exemplary process for determining and / or selecting a charging protocol to charge a rechargeable battery. In some implementations, process 500 is implemented or executed at system 100 using at least the sensor system 106 and the charging protocol optimization system 108. In some examples, the steps or actions of process 500 are enabled by programmed software instructions, firmware instructions, or both. Each type of instruction may be stored in a non-transitory machine-readable storage device and is executable by one or more of the processors or other resources described in this specification.

[0072] In some embodiments, the rechargeable battery comprises an anode comparing an anode active material layer. In some embodiments, the anode active material layer comprises an anode active material such as lithium metal or lithium alloy. In some embodiments, the anode active material comprises at least one selected from the group consisting of lithium, sodium, magnesium, aluminum, silicon, calcium, titanium, manganese, iron, cobalt, nickel, zinc, molybdenum, silver, indium, tin, and tungsten.

[0073] In some embodiments, the anode further comprises an anode current collector. In some embodiments, an anode active material layer is assembled into the battery prior to the first charge. In some embodiments, an anode active material layer is formed after the first charge.

[0074] In some embodiments, the rechargeable battery comprises a liquid electrolyte, a polymer electrolyte or all-state electrolyte.

[0075] In some embodiments, the polymer electrolyte comprises an electrolyte salt, a solvent, and a polymer.

[0076] In some embodiments, the polymer is in situ polymerized after mixing the electrolyte salt, solvent and a polymer precursor (alternatively monomer).

[0077] In some embodiments, the monomer contains one or more polymerizable groups. In some embodiments, non-limiting specific polymerizable groups include vinyl (-CH=CH2), substituted vinyl (-CRI=CR2R3) and a combination thereof, wherein Ri, R2 and R3 areindependently hydrogen, halogen, -CN, -NO2, C1-6 alkyl, Ci-ehaloalkyl, C1-6 hydroxyalkyl, C1-6 aminoalkyl, C2-6 alkenyl, C2-e alkynyl, Ce-w aryl or any combination thereof.

[0078] In some embodiments, the monomer comprises at least one selected from the group consisting of 2,2,3,3-tetrafluorobutane-l,4-diacrylate, 2,2,3,3,4,4,5,5-octafhiorohexane-l,6- diyl diacrylate, 2,2,3,3,4,4,5,5-octafluorohexane-l,6-diyl bis(2-methylacrylate), poly(ethylene glycol) diacrylate (Mn=500-5000), triethylene glycol dimethacrylate (TEGDMA), diurethane dimethacrylate, tetraallyl silane (TAS), 2,4,6,8-tetramethyl-2,4,6,8- tetravinylcyclotetrasiloxane, triethoxyvinylsilane, allyltriethoxysilane, pentaerythritol tetraacrylate (PETA), pentaerythritol tetramethacrylate (PETMA), tris[2-(acryloyloxy)ethyl] isocyanurate (TAEI), di(trimethylolpropane) tetraacrylate (Di-TMPTA), trimethylolpropane propoxylate triacrylate, trimethylolpropane trimethacrylate, pentaerythritol triacrylate, and dipentaerythritol hexaacrylate.

[0079] In some embodiments, the polymer in the polymer electrolyte is a crosslinked polymer. In some embodiments, the precursor (or monomer) for a crosslinked polymer includes at least two or more polymerizable groups. In some embodiments, the precursor (or monomer) for a crosslinked polymer includes at least three or more polymerizable groups

[0080] In some embodiments, the electrolyte salt is selected from the group consisting of lithium perchlorate (LiCICh), lithium nitrate (LiNCh), lithium hexafluorophosphate (LiPFe), lithium borofluoride (LiBF4), lithium hexafluoroarsenide (LiAsFe), lithium trifluoromethanesulfonate (LiCFsSCh), lithium bis(perfluoroethanesulfonyl)imide (LiBETI), lithium bis(fluorosulfonyl)imide (LiFSI), lithium bis(trifluoromethanesulfonyl)imide (LiN(CF3SO2)2, LiTFSI), lithium bis(oxalato)borate (LiBOB), lithium difluoro(oxalato)b orate (LiDFOB), lithium fluoroalkylphosphates (Li[PFx(CyF2y+i-zHz)6-x]) (l<x<5, l<y<8, and 0<z<2y-l), lithium fluorophosphate (Li2PO3F), lithium difluorophosphate (LiDFP), lithium difluoro(bisoxalato)phosphate (LiC4PO8F2), lithium tetrafluoro oxalato phosphate (LiC2PO4F4), lithium tris(trifluoromethanesulfonyl)methide (LiC(CF3SO2)3), LiF, LiCl, LiBr, Lil, Li2SO4, LisPCh, Li2CO3, lithium acetate, lithium trifluoromethyl acetate, lithium oxalate, and a mixture thereof.

[0081] In some embodiments, a process comprises:1) obtaining input data;2) generating, by a state estimator model configured as synthetic data generator, one or more synthetic datasets based on the input data;3) generating, by a surrogate model, such as transformer based ML model, in operation with an optimization engine, a set of features based on time-series analysis applied to the synthetic dataset;4) determining, using a reduced order ML model, a minimum charging time for the battery based on one or more optimization control parameters and a subset of features that are derived from the set of features, wherein the reduced order ML identifies the subset of features for an accelerated charging while maintaining the optimization control parameter with a value in an acceptable range, and5) generating one or more charging protocols for the battery at least based on the minimum charging time and the optimization control parameter.

[0082] In some embodiments, the input data is pre-processed by a data processing algorithm. As shown in Fig. 5, the process 500 comprises a step of obtaining input data (502) by the system 100. In some embodiments, the input data comprises one or more sets of sensing data from the sensor system 106 embedded or integrated at the battery 102. In some implementations, the process comprises a step of pre-pressing the input data into pre-processed input data. In some implementations, the process further comprises a step of generating one or more synthetic datasets (504) based on the input data and / or pre-processed input data by a state estimator model such as P2D model, equivalent circuit model (ECM) and transformer based ML model. For example, a state estimator model that is configured as a synthetic data generator for a particular battery 102 can generate the synthetic dataset based on the input data and / or pre-processed input data. In some embodiments, the state estimator model is a physics-based model. In some embodiments, the process 500 comprises a step of generating a synthetic dataset based on part of or all the pre-processed input data. In some embodiments, the process 500 uses the P2D model to generate a synthetic dataset based on some (or all) of the input data 506 and / or pre-processed data. The synthetic dataset comprises one or more states of the battery 102.

[0083] The system 100 generates a set of features based on time-series analysis (508). For example, the transformer-based ML model generates a set of features based on time-series analysis applied to discrete data of the synthetic dataset. The set of features include one or more exotic parameters for representing one or more internal attributes of the battery 102. The exotic parameters can represent non-measurable quantities specific to battery 102, including one or more of its cells, without breaking the battery. For example, the exotic parameters can represent non-measurable quantities such as an internal stress of the battery 102, a core temperature of battery 102, individual resistance of one or more electrodes of the battery 102,an electrolyte concentration at a particular cell of battery 102, or a combination thereof. Additional exotic parameters that represent internal attributes of battery 102, or non- measurable quantities specific to battery 102, can include individual resistance of one or more electrodes of the battery, lithium concentration in a solid active material, interface kinetic exchange current density, temperature gradient in a cell (e.g., a large format cell), voltage gradient in a cell (e.g., a large format cell), overpotential, or a combination of these.

[0084] The system 100 uses a reduced order ML model to determine a minimum charging time for the battery (510). In some implementations, the minimum charging time is determined for the particular battery 102 based on an optimization control parameter with a value in an acceptable range and a subset of features derived from the set of features. Exemplary optimization control parameters can include age, temperature, stress, voltage, or some other battery parameter. The system 100 is configured to dynamically change or adjust the minimum charging time based on a particular battery parameter or optimization control parameter. For example, the system 100 can dynamically adjust the minimum charging time based on the age , a state of battery, or both. The minimum charging time can correspond to a desired charging duration provided as an input to the client device 104. In some implementations, the desired charging duration is one of a variety of optimization control parameters that can be provided or received as an input (e.g., a user input) to the protocol optimization system 108. The system 100 generates one or more charging protocols for the battery at least based on the minimum charging time and the optimization control parameter (512). In some embodiments, the system 100 further selects an optimal charging protocol from the one or more charging protocols in view of experimental validation, or an analysis of accuracy or fidelity.

[0085] Fig. 6 is an exemplary process 600 for optimizing parameters (aspects) of an ML model in the computing system 100 of Fig. 1. In some implementations, process 600 is implemented or executed at the charging protocol optimization system 108using input data comprising the real-time data from the sensor system 106 In some examples, the steps or actions of process 600 are enabled by programmed software instructions, firmware instructions, or both. Each type of instruction may be stored in a non-transitory machine-readable storage device and is executable by one or more of the processors or other resources described herein.

[0086] As shown in Fig. 6, process 600 comprises a step of training for a multi-head optimization technique associated with each feature type (602). The process 600 further comprises a step of applying a supervised ML algorithm based on a training dataset comprising annotated inputs (604). The process 600 further comprises a step of determining respective weights of the model for each feature type (606). In some implementations, the process 600comprises a step of iteratively tuning one or more of its ML models to enhance predictive accuracy for a particular battery chemistry (608). For example, the system 100 can iteratively tune the model based on operations executed to perform at least steps 604 and 606 of process 600.

[0087] In some implementations, one or more steps of process 500 and process 600, or other processes described here, are performed at a hardware integrated circuit(s) as part of a larger compute operation to generate a machine-learning (ML) outputs, including an output for a neural network layer of a neural network that implements one or more ML models. For example, the output can be a portion of a computation for a ML task or inference workload to generate outputs for a feature engineering process for optimizing a charging protocol for a rechargeable battery such as LIB. As indicated above, the integrated circuit(s) can include special-purpose processors, such as a neural network processor or a hardware ML accelerator configured to accelerate computations for generating different types of data processing outputs.

[0088] In some implementations, the fast charging protocol comprises charging the battery from a first threshold value of state of charge (SOCa) to a second threshold value (SOCb) following a multi-step constant current (MSCC). A fast charging refers to a charging protocol capable of charging a battery from SOCa to SOCb with a duration of no greater than 20 min, no greater than 15 min or no greater than 10 min

[0089] In some embodiments, a multi-step constant current (MSCC) is part of a battery charging protocol depending on the initial state of charge (SOCi) of the battery. In some embodiments, a method of charging a battery comprises:1) if SOCi is less than a first threshold value of SOC (SOCa), charging the battery with a low constant current (la) until SOC reaches SOCa, and then charging the battery with a multi-step constant current (MSCC) charging protocol comprising multiple charging steps,2) if SOCi is equal to or greater than a second threshold value of SOC (SOCb), charging the battery with a fixed constant current (lb), and3) when if SOCi is equal to or higher than SOCa but less than SOCb, charging the battery following an MSCC charging protocol comprising multiple charging steps.

[0090] In some embodiments, the MSCC charging protocol in 1) and 3) above is the same or different.

[0091] In some embodiments, an MSCC charging protocol comprises at least three steps, at least four steps, or at least five steps. The battery charged by the MSCC charging protocol exhibits an improved cycle life. The battery charged by the MSCC charging protocol exhibitsa cycle life of at least 10% longer than a battery charged by a charging protocol with a single constant current.

[0092] In some embodiments, the MSCC charging protocol comprises a first, second and third charging steps in the order, wherein a first constant current (Ii), a second constant current (I2), and a third constant current (I3) are used in the first, second and third charging steps, respectively, and wherein the constant currents in the first, second and third steps exhibit a pattern of Ii < I2 > I3 or Ii > I2 < I3.

[0093] In some implementations, the MSCC charging protocol comprises a first, second, third and fourth charging steps in the order, wherein a first constant current (Ii), a second constant current (I2), a third constant current (I3) and a fourth constant current (I4) are used in the first, second, third and fourth charging steps, respectively, and wherein the constant currents in the first through fourth steps exhibit a pattern of Ii < I2 > I3 < I4 or Ii > I2 < I3 > I4.

[0094] In some embodiments, SOCa and SOCb are each independently defined by a user or a program.

[0095] In some embodiments, SOCa has a value in a range from 5% to 75%, from 5% to 70%, from 5% to 65%, from 5% to 60%, from 5% to 55%, from 5% to 55%, from 5% to 50%, from 5% to 45%, from 5% to 40%, from 5% to 35%, from 5% to 30%, from 5% to 25%, from 5% to 20%, from 5% to 15%, or from 5% to 10%.

[0096] In some embodiments, SOCb is greater than SOCa. In some embodiments, SOCb has a value in a range from 70% to 95%, from 70% to 90%, from 70% to 85%, from 70% to 80%, or from 70% to 75%.

[0097] Embodiments of the subject matter and the functional operations described in this specification can be implemented in digital electronic circuitry, in tangibly-embodied computer software or firmware, in computer hardware, including the structures disclosed in this specification and their structural equivalents, or in combinations of one or more of them. Embodiments of the subject matter described in this specification can be implemented as one or more computer programs, i.e., one or more modules of computer program instructions encoded on a tangible non transitory program carrier for execution by, or to control the operation of, data processing apparatus.

[0098] Alternatively or in addition, the program instructions can be encoded on an artificially generated propagated signal, e.g., a machine-generated electrical, optical, or electromagnetic signal, that is generated to encode information for transmission to suitable receiver apparatus for execution by a data processing apparatus. The computer storage medium can be a machine-readable storage device, a machine-readable storage substrate, a random or serial access memory device, or a combination of one or more of them.

[0099] The term “computing system” encompasses all kinds of apparatus, devices, and machines for processing data, including by way of example a programmable processor, a computer, or multiple processors or computers. The apparatus can include special purpose logic circuitry, e.g., an FPGA (field programmable gate array) or an ASIC (application specific integrated circuit). The apparatus can also include, in addition to hardware, code that creates an execution environment for the computer program in question, e.g., code that constitutes processor firmware, a protocol stack, a database management system, an operating system, or a combination of one or more of them.

[0100] A computer program (which may also be referred to or described as a program, software, a software application, a module, a software module, a script, or code) can be written in any form of programming language, including compiled or interpreted languages, or declarative or procedural languages, and it can be deployed in any form, including as a standalone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.

[0101] A computer program may, but need not, correspond to a file in a file system. A program can be stored in a portion of a file that holds other programs or data, e.g., one or more scripts stored in a markup language document, in a single file dedicated to the program in question, or in multiple coordinated files, e.g., files that store one or more modules, sub programs, or portions of code. A computer program can be deployed to be executed on one computer or on multiple computers that are located at one site or distributed across multiple sites and interconnected by a communication network.

[0102] The processes and logic flows described in this specification can be performed by one or more programmable computers executing one or more computer programs to perform functions by operating on input data and generating output. The processes and logic flows can also be performed by, and apparatus can also be implemented as, special purpose logic circuitry, e.g., an FPGA (field programmable gate array), an ASIC (application specific integrated circuit), or a GPGPU (General purpose graphics processing unit). Computers suitable for the execution of a computer program include, by way of example, general or special purpose microprocessors or both, or any other kind of central processing unit.

[0103] Generally, a central processing unit will receive instructions and data from a read only memory or a random-access memory or both. Some elements of a computer are a central processing unit for performing or executing instructions and one or more memory devices forstoring instructions and data. Generally, a computer will also include, or be operatively coupled to receive data from or transfer data to, or both, one or more mass storage devices for storing data, e.g., magnetic, magneto optical disks, or optical disks. However, a computer need not have such devices. Moreover, a computer can be embedded in another device, e.g., a mobile telephone, a personal digital assistant (PDA), a mobile audio or video player, a game console, a Global Positioning System (GPS) receiver, or a portable storage device, e.g., a universal serial bus (USB) flash drive, to name just a few.

[0104] Computer readable media suitable for storing computer program instructions and data include all forms of nonvolatile memory, media and memory devices, including by way of example semiconductor memory devices, e.g., EPROM, EEPROM, and flash memory devices; magnetic disks, e.g., internal hard disks or removable disks; magneto optical disks; and CD ROM and DVD-ROM disks. The processor and the memory can be supplemented by, or incorporated in, special purpose logic circuitry.

[0105] To provide for interaction with a user, embodiments of the subject matter described in this specification can be implemented on a computer having a display device such as client device 104 for displaying information to the user and a keyboard and a pointing device, e.g., a mouse or a trackball, by which the user can provide input to the computer. Other kinds of devices can be used to provide for interaction with a user as well; for example, feedback provided to the user can be any form of sensory feedback, e.g., visual feedback, auditory feedback, or tactile feedback; and input from the user can be received in any form, including acoustic, speech, or tactile input. In addition, a computer can interact with a user by sending documents to and receiving documents from a device that is used by the user; for example, by sending web pages to a web browser on a user’s client device in response to requests received from the web browser.

[0106] Embodiments of the subject matter described in this specification can be implemented in a computing system that includes a back end component, e.g., as a data server, or that includes a middleware component, e.g., an application server, or that includes a front end component, e.g., a client computer having a graphical user interface or a Web browser through which a user can interact with an implementation of the subject matter described in this specification, or any combination of one or more such back end, middleware, or front end components. The components of the system can be interconnected by any form or medium of digital data communication, e.g., a communication network. Examples of communication networks include a local area network (“LAN”) and a wide area network (“WAN”), e.g., the Internet.

[0107] The computing system can include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other.

[0108] The disclosure will be better understood by reference to the Experimental Details which follow, but those skilled in the art will readily appreciate that the specific experiments detailed are only illustrative, and are not meant to limit the disclosure as described herein, as numerous variations and modifications of these exemplary embodiments are possible without undue experimentation. All such variations and modifications are within the scope of the teachings of this disclosure. It will be appreciated that the foregoing description and following examples, no matter how detailed they may appear in text, the disclosure may be practiced in many ways, and the disclosure should be construed in accordance with the appended claims and equivalents thereof.Example 1

[0109] Input data including material properties, pseudo-Open Circuit Voltage (OCV) with C / 50, Galvanostatic Intermittent Titration Technique (GITT) results, and temperature measurements were collected and provided into a state estimator model such as P2D model configurated as a synthetic data generator. The state estimator model was calibrated based on a calibration data comprising a set of experimental data for a battery, thus obtaining a calibrated state estimator model specific for the battery. The battery was a pouch cell (0.75 Ah) comprising Li metal as anode, microporous membrane as separator, NMC811 as cathode, and a polymer electrolyte as electrolyte.

[0110] The calibrated state estimator model then generated a synthetic dataset based on the input data and / or pre-processed input data, wherein the synthetic dataset comprises one or more states of the battery. The synthetic dataset captures intricate relationships between charging parameters and battery responses. The generation of synthetic dataset could be performed using one or more sampling methods.[OHl] A transformer-based ML model was used as a typical surrogate model for generating a set of features based on time-series analysis applied to the synthetic dataset, wherein the set of features comprises exotic parameters for representing one or more internal attributes of the battery. In some implementations, the one or more internal attributes are not directly measurable through physical testing of the battery, particularly during operation of the battery. In some implementations the one or more internal attributes include without limitation at leastone selected from the group consisting of internal stress of the battery, a core temperature of the battery, an electrolyte salt concentration, a solvent concentration in electrolyte, individual resistance of one or more electrodes, lithium concentration in a solid electrode active material, interface kinetic exchange current density, temperature gradient in a cell, voltage gradient in a cell, overpotential at either electrode, and electrode utilization. The transformer-based ML model was adapted to one or more states of the battery with a range of value aligned with the detected state from one or more sensors. In some implementations, the transformer-based ML model was adapted by generating and / or identifying a set of features so that the charging protocol parameters are well aligned with the actual state and condition of the battery.

[0112] A reduced order ML model was then conducted in conjunction with a typical multiobjective optimization (MOO) process to determine a minimum charging time for the battery based on one or more optimization control parameters and a subset of features that are derived from the set of features. The minimum charging time was determined in a way that the charging duration is in an acceptable range while keeping values of the one or more optimization control parameters in an acceptable range, such as keeping the internal battery temperature and stress in an acceptable range, i.e., not beyond a threshold value. In some implementations, the reduced order ML model identified one or more optimization control parameters affecting the degradation and lifetime of the battery. In some implementations, the multi-objective optimization (MOO) model determined the minimum charging time in a way that the charging duration is acceptable minimum while the one or more optimization control parameters such as degradation of the battery, maximum temperature and maximum stress are also controlled to an acceptable extent.

[0113] Based on the minimum charging time and / or the one or more optimization control parameters, one or more charging protocols were then generated. Table 1 summarizes one of the charging protocols.Table 1 MSCC charging protocol of example 1

[0114] The battery as prepared above was cycled between 2.8V to 4.25 V at 25 °C under an external pressure. In each cycle, the pouch cell was charged by following the MSCC chargingprotocol in Table 1 and discharged at 1C at 25 °C. In some embodiments, the external pressure is in a range from 0.5MPa to 5.0 MPa.

[0115] A pouch cell was similarly prepared and cycled except that the cell was charged at a single constant current of 2.8C.

[0116] Fig. 7A shows a voltage profile of a battery charged with a conventional protocol, wherein a single constant current of 2.8C was used and the starting SOC and ending SOC were 10% and 80%, respectively. Fig. 7B shows a voltage profile of a battery charged with an exemplary MSCC charging protocol as summarized in Table 1.

[0117] Cycle life is the number of cycles for the battery to reach a threshold value such as 80% of its original capacity or a threshold value such as 98.0% of its columbic efficiency (CE), whichever is earlier. Cycle life is usually used to measure the cycling performance of a secondary battery. As shown in Figs. 8A, 8B and Table 2, the cell charged with the MSCC charging protocol exhibited a cycle life of more than 90 cycles, which is around 260% longer than that of the comparative example with a cycle life of 25 cycles.Table 2 Cycle life of cell charged with MSCC and single constant currentAspects

[0118] In a first aspect, the present disclosure provides a computer-implemented method for generating a charging protocol of a battery. The method comprises:1) processing, by a data processing algorithm, a plurality of input data of a battery into a plurality of pre-processed input data;2) providing a plurality of pre-processed input data to a state estimator model configured as a synthetic data generator;3) generating, by the state estimator model, a synthetic dataset based on the plurality of pre- processed input data, the synthetic dataset comprising one or more states of the battery;4) generating, by a surrogate model in operation with an optimization engine, a set of features based on time-series analysis applied to the synthetic dataset, the set of features comprising exotic parameters for representing one or more internal attributes of the battery;5) determining, using a reduced order ML model, a minimum charging time for the battery based on one or more optimization control parameters and a subset of features that are derived from the set of features, and6) generating one or more charging protocols for the battery at least based on the minimum charging time and the one or more optimization control parameters.

[0119] In a second aspect according to the first aspect, the method further comprises selecting an optimal charging protocol from the one or more charging protocols.

[0120] In a third aspect according to the first or second aspect, the state estimator model is calibrated by a calibration data comprising a set of experimental data for the battery.

[0121] In a fourth aspect according to the first aspect, the state estimator model is a pseudo- two-dimensional (P2D), equivalent circuit model (ECM) or a transformer-based ML model.

[0122] In a fifth aspect according to the first aspect, the surrogate model, in operation with the optimization engine, adapts the set of features and / or the subset of features in view of the one or more states with a value or within a range of value that is aligned with one or more sets of sensing data retrieved from one or more sensors.

[0123] In a sixth aspect according to the fifth aspect, the one or more states comprise one or more selected from the group consisting of State-of-Health (SOH), State-of-Charge (SOC), State-of-Power (SOP), and State-of-Energy (SOE)

[0124] In a seventh aspect according to the first aspect, the reduced order ML model adapts the minimum charging time and / or the optimization control parameter in view of the one or more states of the battery and one or more sets of sensing data retrieved from one or more sensors.

[0125] In an eighth aspect according to first aspect, the surrogate model generates the set features based on time-series analysis applied to the synthetic dataset and time-series analysis applied to non-synthetic data comprising real experimental data derived from physical testing of the battery. In some embodiments, the surrogate model is pre-trained based on the set of features so that the subset of features is adapted to a corresponding battery chemistry of the battery.

[0126] In a nineth aspect according to the eighth aspect, the one or more internal attributes are not-directly measurable through physical testing of the battery.

[0127] In a tenth aspect according to the nineth aspect, the one or more internal attributes comprise at least one selected from the group consisting of at least one selected from the group consisting of internal stress of the battery, a core temperature of the battery, an electrolyte salt concentration, a solvent concentration in electrolyte, individual resistance of one or more electrodes, lithium concentration in a solid electrode active material, interface kinetic exchange current density, temperature gradient in a cell, voltage gradient in a cell, overpotential at either electrode, and electrode utilization.

[0128] In an eleventh aspect, the surrogate model is pre-trained with a supervised machinelearning algorithm, a single-head attention network or multi-head attention network using a plurality of annotated inputs as one or more training datasets. In some embodiments, the surrogate model is a transformer-based machine learning (ML) model. In some embodiments, the surrogate model is a transformer-based machine learning (ML) model and is trained by a single-head attention network for generating or identifying one of the set of features and determining a corresponding weighting thereof. In some embodiments, the surrogate model is a transformer-based ML model and is trained by a multi-head attention network for generating or identifying two or more features of the set of features and determining a respective corresponding weighting thereof.

[0129] In a twelfth aspect according to the first aspect, the reduced order ML model generates the subset of feature values by generating a condensed representation of time-series data derived from the time-series analysis applied to the synthetic dataset followed by a multiobjective optimization for each feature type using the condensed representation of the timeseries data.

[0130] In a thirteenth aspect according to the first aspect, the surrogate model is part of a digital twin of the battery.

[0131] In a fourteenth aspect according to the thirteenth aspect, the plurality of input data comprises a set of data in the digital twin of the battery, and the plurality of input data is updated once the set of data in the digital twin is updated.

[0132] In a fifteenth aspect according to the first aspect, at least one of the one or more charging protocols is a multi-step constant current (MSCC) charging protocol. In some embodiments, the MSCC charging protocol comprises a first, second and third charging steps in the order, wherein a first constant current (Ii), a second constant current (I2), and a third constant current (I3) are used in the first, second and third charging steps, respectively, and wherein the constant currents in the first, second and third steps exhibit a pattern of Ii < I2 > I3 or Ii > I2 < I3. In some embodiments, the MSCC charging protocol comprises a first, second, third and fourth charging steps in the order, wherein a first constant current (Ii), a second constant current (I2), a third constant current (I3) and a fourth constant current (I4) are used in the first, second, third and fourth charging steps, respectively, and wherein the constant currents in the first through fourth steps exhibit a pattern of Ii < I2 > I3 < I4 or Ii > I2 < I3 > I4.

[0133] In a sixteenth aspect, the present disclosure provides a method for charging a battery, comprising:1) processing, by a data processing algorithm, a plurality of input data of a battery into a plurality of pre-processed input data;2) providing a plurality of pre-processed input data to a state estimator model configured as a synthetic data generator;3) generating, by the state estimator model, a synthetic dataset based on the plurality of pre-processed input data, the synthetic dataset comprising one or more states of the battery;4) generating, by a surrogate model in operation with an optimization engine, a set of features based on time-series analysis applied to the synthetic dataset, the set of features comprising exotic parameters representing one or more internal attributes of the battery;5) determining, using a reduced order ML model, a minimum charging time for the battery based on one or more optimization control parameters and a subset of features that are derived from the set of features;6) generating one or more charging protocols for the battery at least based on the minimum charging time and the one or more optimization control parameters;7) selecting an optimal charging protocol from the one or more charging protocols; and8) charging the battery according to the optimal charging protocol.

[0134] In a seventeenth aspect, the present disclosure provides a system comprising: a processing device; and a non-transitory machine-readable storage medium storing instructions that are executable by the processing device to cause performance of operations comprising:1) processing, by a data processing algorithm, a plurality of input data of a battery into a plurality of pre-processed input data;2) providing a plurality of pre-processed input data to a state estimator model configured as a synthetic data generator;3) generating, by the state estimator model, a synthetic dataset based on the plurality of pre-processed input data, the synthetic dataset comprising one or more states of the battery;4) generating, by a surrogate model in operation with an optimization engine, a set of features based on time-series analysis applied to the synthetic dataset, the set of features comprising exotic parameters representing one or more internal attributes of the battery;5) determining, using a reduced order ML model, a minimum charging time for the battery based on one or more optimization control parameters and a subset of features that are derived from the set of features; and6) generating one or more charging protocols for the battery at least based on the minimum charging time and the one or more optimization control parameters.

[0135] In an eighteenth aspect according to the seventeenth aspect, the surrogate model generates the set features based on time-series analysis applied to the synthetic dataset; and time-series analysis applied to non-synthetic data comprising real experimental data derived from physical testing of the battery.

[0136] In a nineteenth aspect, the present disclosure provides a non-transitory machine- readable storage device storing instructions that are executable by a processing device to cause performance of operations comprising:1) processing, by a data processing algorithm, a plurality of input data of a battery into a plurality of pre-processed input data;2) providing a plurality of pre-processed input data to a state estimator model configured as a synthetic data generator;3) generating, by the state estimator model, a synthetic dataset based on the plurality of pre-processed input data, the synthetic dataset comprising one or more states of the battery;4) generating, by a surrogate model in operation with an optimization engine, a set of features based on time-series analysis applied to the synthetic dataset, the set of features comprising exotic parameters representing one or more internal attributes of the battery;5) determining, using a reduced order ML model, a minimum charging time for the battery based on one or more optimization control parameters and a subset of features that are derived from the set of features;6) generating one or more charging protocols for the battery at least based on the minimum charging time and the one or more optimization control parameters; and7) selecting an optimal charging protocol from the one or more charging protocols;

[0137] In a twentieth aspect, the present disclosure provides a vehicle comprising the non- transitory machine-readable storage device according to the nineteenth aspect.

[0138] While this specification contains many specific implementation details, these should not be construed as limitations on the scope of any invention or of what may be claimed, but rather as descriptions of features that may be specific to particular embodiments of particularinventions. Certain features that are described in this specification in the context of separate embodiments can also be implemented in combination in a single embodiment. Conversely, various features that are described in the context of a single embodiment can also be implemented in multiple embodiments separately or in any suitable subcombination. Moreover, although features may be described above as acting in certain combinations and even initially claimed as such, one or more features from a claimed combination can in some cases be excised from the combination, and the claimed combination may be directed to a subcombination or variation of a subcombination.

[0139] All transitional phrases such as “comprising,” “including,” “carrying,” “having,” “containing,” “involving,” “holding,” “composed of,” and the like are to be understood to be open-ended, i.e., to mean including but not limited to. Only the transitional phrases “consisting of’ and “consisting essentially of’ shall be closed or semi-closed transitional phrases, respectively. Phrases such as “based on”, “at least based on”, “on the basis of’, “considering”, “factoring”, “in view of’ and the like are to be understood to be open-ended, i.e., to mean considering at least one factor including but not limited to.

[0140] While operations are depicted in the drawings in a particular order, this should not be understood as requiring that such operations be performed in the particular order shown or in sequential order, or that all illustrated operations be performed, to achieve desirable results. In certain circumstances, multitasking and parallel processing may be advantageous. Moreover, the separation of various system modules and components in the embodiments described above should not be understood as requiring such separation in all embodiments, and it should be understood that the described program components and systems can generally be integrated together in a single software product or packaged into multiple software products.

[0141] Various embodiments of the features of this disclosure are described herein. However, it should be understood that such embodiments are provided merely by way of example, and numerous variations, changes, and substitutions can occur to those skilled in the art without departing from the scope of this disclosure. It should also be understood that various alternatives to the specific embodiments described herein are also within the scope of this disclosure.

Claims

What is claimed is:

1. A computer-implemented method for generating a charging protocol of a battery, comprising:1) processing, by a data processing algorithm, a plurality of input data of a battery into a plurality of pre-processed input data;2) providing a plurality of pre-processed input data to a state estimator model configured as a synthetic data generator;3) generating, by the state estimator model, a synthetic dataset based on the plurality of pre-processed input data, the synthetic dataset comprising one or more states of the battery;4) generating, by a surrogate model in operation with an optimization engine, a set of features based on time-series analysis applied to the synthetic dataset, the set of features comprising exotic parameters representing one or more internal attributes of the battery;5) determining, using a reduced order ML model, a minimum charging time for the battery based on one or more optimization control parameters and a subset of features that are derived from the set of features, and6) generating one or more charging protocols for the battery at least based on the minimum charging time and the one or more optimization control parameters.

2. The computer-implemented method of claim 1, further comprising: selecting an optimal charging protocol from the one or more charging protocols.

3. The computer-implemented method of claim 1, wherein the state estimator model is calibrated by a calibration data comprising a set of experimental data for the battery.

4. The computer-implemented method of claim 1, wherein the state estimator model is a pseudo-two-dimensional (P2D) or a transformer-based ML model.

5. The computer-implemented method of claim 1, wherein the surrogate model, in operation with the optimization engine, adapts the set of features and / or the subset of features in view of the one or more states with a value or within a range of value that is aligned with one or more sets of sensing data retrieved from one or more sensors.

6. The computer-implemented method of claim 5, wherein the one or more states comprise one or more selected from the group consisting of State-of-Health (SOH), State-of-Charge (SOC), State-of-Power (SOP), and State-of-Energy (SOE)7. The computer-implemented method of claim 1, wherein the reduced order ML model adapts the minimum charging time and / or the optimization control parameter in view of the one or more states of the battery and one or more sets of sensing data retrieved from one or more sensors.

8. The computer-implemented method of claim 1, wherein the surrogate model generates the set features based on time-series analysis applied to the synthetic dataset and time-series analysis applied to non-synthetic data comprising real experimental data derived from physical testing of the battery.

9. The computer-implemented method of claim 1, wherein the one or more internal attributes are not-directly measurable through physical testing of the battery.

10. The computer-implemented method of claim 9, wherein the one or more internal attributes comprise at least one selected from the group consisting of: internal stress of the battery; a core temperature of the battery; an electrolyte salt concentration of the battery; a solvent concentration in electrolyte of the battery individual resistance of one or more electrodes of the battery; lithium concentration in a solid electrode active material; interface kinetic exchange current density; temperature gradient in a cell; voltage gradient in a cell; overpotential at either electrode; and electrode utilization.

11. The computer-implemented method of claim 11, wherein the surrogate model is pre-trained with a supervised machine-learning algorithm, a single-head attention network or multihead attention network using a plurality of annotated inputs as one or more training datasets.

12. The computer-implemented method of claim 1, wherein the reduced order ML model generates the subset of feature values by generating a condensed representation of timeseries data derived from the time-series analysis applied to the synthetic dataset followed by a multi-objective optimization for each feature type using the condensed representation of the time-series data.

13. The computer-implemented method of claim 1, wherein the surrogate model is part of a digital twin of the battery.

14. The computer-implemented method of claim 13, wherein the plurality of input data comprises a set of data in the digital twin of the battery, and the plurality of input data is updated once the set of data in the digital twin is updated.

15. The computer-implemented method of claim 1, wherein at least one of the one or more charging protocols is a multi-step constant current (MSCC) charging protocol.

16. A method for charging a battery, comprising:1) processing, by a data processing algorithm, a plurality of input data of a battery into a plurality of pre-processed input data;2) providing a plurality of pre-processed input data to a state estimator model configured as a synthetic data generator;3) generating, by the state estimator model, a synthetic dataset based on the plurality of pre-processed input data, the synthetic dataset comprising one or more states of the battery;4) generating, by a surrogate model in operation with an optimization engine, a set of features based on time-series analysis applied to the synthetic dataset, the set of features comprising exotic parameters representing one or more internal attributes of the battery;5) determining, using a reduced order ML model, a minimum charging time for the battery based on one or more optimization control parameters and a subset of features that are derived from the set of features;6) generating one or more charging protocols for the battery at least based on the minimum charging time and the one or more optimization control parameters;7) selecting an optimal charging protocol from the one or more charging protocols; and8) charging the battery according to the optimal charging protocol.

17. A system comprising: a processing device; and a non-transitory machine-readable storage medium storing instructions that are executable by the processing device to cause performance of operations comprising:1) processing, by a data processing algorithm, a plurality of input data of a battery into a plurality of pre-processed input data;2) providing a plurality of pre-processed input data to a state estimator model configured as a synthetic data generator;3) generating, by the state estimator model, a synthetic dataset based on the plurality of pre-processed input data, the synthetic dataset comprising one or more states of the battery;4) generating, by a surrogate model in operation with an optimization engine, a set of features based on time-series analysis applied to the synthetic dataset, the set of features comprising exotic parameters representing one or more internal attributes of the battery;5) determining, using a reduced order ML model, a minimum charging time for the battery based on one or more optimization control parameters and a subset of features that are derived from the set of features; and6) generating one or more charging protocols for the battery at least based on the minimum charging time and the one or more optimization control parameters.

18. The system of claim 17, wherein the surrogate model generates the set features based on time-series analysis applied to the synthetic dataset; and time-series analysis applied to non-synthetic data comprising real experimental data values derived from physical testing of the battery.

19. A non-transitory machine-readable storage device storing instructions that are executable by a processing device to cause performance of operations comprising:1) processing, by a data processing algorithm, a plurality of input data of a battery into a plurality of pre-processed input data;2) providing a plurality of pre-processed input data to a state estimator model configured as a synthetic data generator;3) generating, by the state estimator model, a synthetic dataset based on the plurality of pre-processed input data, the synthetic dataset comprising one or more states of the battery;4) generating, by a surrogate model in operation with an optimization engine, a set of features based on time-series analysis applied to the synthetic dataset, the set of features comprising exotic parameters representing one or more internal attributes of the battery;5) determining, using a reduced order ML model, a minimum charging time for the battery based on one or more optimization control parameters and a subset of features that are derived from the set of features;6) generating one or more charging protocols for the battery at least based on the minimum charging time and the one or more optimization control parameters; and7) selecting an optimal charging protocol from the one or more charging protocols;20. A vehicle comprising the non-transitory machine-readable storage device of claim 19.

Images