System and method for fast charging lithium-ion batteries

Abstract

A lithium-ion battery management system includes a lithium-ion battery cell, one or more sensors configured to sense one or more operating conditions of the lithium-ion battery cell, a controller having non-transitory memory for storing machine instructions to be executed by the controller and operably connected to the lithium-ion battery cell, the machine instructions, when executed by the controller, implement the following functions: receiving the one or more operating conditions and a trained machine learning (ML) model, and outputting an indicator value along a state of charge (SOC) trajectory in response to the one or more operating conditions and the trained ML model to control a fast state of charge of the lithium-ion battery cell from a current source.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to systems and methods for fast charging electrochemical cells, such as lithium-ion cells, while utilizing machine learning (ML) models. BACKGROUND

[0002] Lithium-ion cells have become ubiquitous and their use is widespread. Fast charging of lithium-ion cells is one of the technical challenges faced by companies providing various devices, such as consumer electronics, power tools, and electric vehicles, that use these cells. Consumers generally prefer minimal wait times for recharging their smartphones, power drills, electric vehicles, or other electronic devices. As a result, battery suppliers and device manufacturers compete to deliver devices that charge as fast as possible. However, fast charging can cause overpotential and mechanical stress in the cells, which can accelerate the aging process of the cells and can result in reduced life. SUMMARY

[0003] According to one embodiment, a lithium-ion battery management system is disclosed. The system includes a lithium-ion cell and one or more sensors configured to sense one or more operating conditions of the lithium-ion cell. The system further includes a controller having a non-transitory memory for storing machine instructions to be executed by the controller and operably connected to the lithium-ion cell, the machine instructions, when executed by the controller, implement the following functions: receiving the one or more operating conditions and a trained machine learning (ML) model; and outputting an indicator value along a state-of-charge (SOC) trajectory in response to the one or more operating conditions and the trained ML model to control a fast charging state of the lithium-ion cell from a current source. The SOC trajectory can include a constant current (CC) phase, a constant indicator (CI) phase, and / or a constant voltage (CV) charging phase. The indicator value can be a state of the cell along the SOC trajectory. The indicator value can be an overpotential value. The one or more operating conditions can include one or more of an internal cell temperature, a cell voltage, and a current. The machine instructions, when executed by the controller, can further implement the following function: determining a first indicator value at a time (t). The machine instructions, when executed by the controller, can further implement the following function: determining a second indicator value at a time (t+1). The trained ML model can include a training dataset comprising a plurality of individual charging trajectories, and each trajectory is generated for a particular combination of model parameters, initial conditions, and / or ambient temperature. The trained ML model can be a physics-based trained model.

[0004] In an alternative embodiment, a method of fast charging a lithium-ion battery cell is disclosed. The method can include sensing one or more operating conditions of the lithium-ion battery cell; receiving the one or more operating conditions and a trained machine learning (ML) model; and outputting, in response to the one or more operating conditions and the trained ML model, an indicator value along a state-of-charge (SOC) trajectory to control a fast charging state of the lithium-ion battery cell from a current source. The SOC trajectory can include a constant current (CC) phase, a constant indicator (CI) phase, and / or a constant voltage (CV) charging phase. The indicator value can be a state of the battery cell along the SOC trajectory. The indicator value can be an overpotential value. The one or more operating conditions can include one or more of an internal battery cell temperature, a battery cell voltage, and a current. The trained ML model can include a training set including a plurality of individual charging trajectories, and further including that each trajectory is generated for a particular combination of model parameters, initial conditions, and / or ambient temperature.

[0005] In another embodiment, a method of estimating a battery target indicator value for a lithium-ion battery cell is disclosed. The method can include training a machine learning (ML) model including a training data set including: a plurality of individual charging trajectories, charging phase variables, a measurement data set, and an indicator variable, wherein each trajectory is generated for a particular combination of model parameters, initial conditions, and / or ambient temperature. The method can further include sensing one or more operating conditions of the lithium-ion battery cell; receiving the one or more operating conditions and the trained ML model; and outputting, in response to the one or more operating conditions and the trained ML model, an indicator value along a state-of-charge (SOC) trajectory. The indicator value can be a state of the battery cell along the SOC trajectory. The method can further include supplying the training data set from a physics-based model. The measurement data set can include one or more operating conditions of the lithium-ion battery cell. The one or more operating conditions can include one or more of an internal battery cell temperature, a battery cell voltage, and a current. BRIEF DESCRIPTION OF DRAWINGS

[0006] Figure 1 A voltage profile during constant current constant voltage (CC-CV) charging of a lithium-ion battery is shown;

[0007] Figure 2 A comparison of a conventional CC-CV voltage profile to a physics-based estimator method voltage profile is shown;

[0008] Figure 3 is a schematic diagram of a non-limiting example of a system for fast charging a lithium-ion battery cell in accordance with one or more embodiments disclosed herein;

[0009] Figure 4schematic illustration of the training steps of the ML model disclosed herein; and

[0010] Figure 5 is a schematic illustration of a non-limiting example of a set of processing steps disclosed herein. DETAILED DESCRIPTION

[0011] Embodiments of the present disclosure are described herein. It should be understood, however, that the disclosed embodiments are merely examples and other embodiments can take various and alternative forms. The figures are not necessarily to scale; some features can be exaggerated or minimized for the purpose of clarity and illustration. Therefore, specific structural and functional details disclosed herein are not to be interpreted as limiting, but merely as a representative basis for teaching one skilled in the art to variously employ the embodiments. As those skilled in the art will appreciate, the various features described herein can be combined or integrated in various ways, and all such combinations and sub-combinations are intended to be within the scope of the disclosure. Combinations and sub-combinations of various features and steps can be made without departing from the scope of the disclosure.

[0012] Except where expressly stated to the contrary, all numerical quantities in description of the dimensions or material properties contained herein should be understood to be modified in all instances by the term "about." n

[0013] Reference will now be made in detail to compositions, embodiments and methods of embodiments known to the inventors. While the disclosed embodiments are amenable to various and alternative forms, specifics thereof have been set forth in this disclosure in order to provide an understanding of the present invention. It should be understood, however, that the intention is not to limit the invention to the particulars of the embodiments described.

[0014] The description of a group or class of materials suitable for a given purpose in connection with one or more embodiments means that a mixture of any two or more members of the group or class is also suitable. Description of components in chemical terms pertains to the composition of the components and can be made in terms of function of the components. Unless specifically set forth herein, notation representing components carried out sequential processes generally is not indicative of order except when the context requires otherwise. Description of components in chemical terms pertains to the composition of the components and, unless specifically stated to the contrary, can be made in terms of function of the components. The first definition of a first letter abbreviation or other abbreviation is applicable to all subsequent uses of the same abbreviation herein, and applies mutatis mutandis to normal grammatical variations of the initially-defined abbreviation. Unless explicitly stated otherwise, measurements of properties are determined by the same techniques as cited earlier or later for the same property.

[0015] Lithium-ion batteries have become the dominant product among rechargeable batteries. However, despite their popularity, lithium-ion batteries face various challenges. For example, in most applications, charging of lithium-ion batteries is a relatively demanding test that the batteries undergo. The charging process is an important factor in the aging and degradation of lithium-ion battery cells. The degradation of lithium-ion battery cells is mainly caused by side reactions, such as lithium plating, which consumes the recyclable lithium. Overpotential determines the occurrence of harmful side reactions. If the overpotential can be properly estimated, algorithms can be developed to minimize the side reactions inside the lithium-ion battery cells, thereby maintaining the expected life of the battery.

[0016] There have been many attempts to charge lithium-ion battery cells in the shortest time while trying to minimize their degradation. Generally, these methods can be divided into two groups: (1) methods that use only available real-time measurements from the battery cell, and (2) methods that model the internal processes of the battery cell and use estimates of internal variables to control the charging process.

[0017] One of the most common charging methods in the industry that uses current and voltage measurements is called the constant current-constant voltage method. Figure 1 The characteristic charging voltage profile of the CC-CV method is shown. As can be seen in Figure 1 The voltage profile includes two phases. The first phase corresponds to the part of the profile where the battery management system (BMS) maintains the charging current at a constant value. The second phase corresponds to the part where the charging current is adjusted (reduced) by the BMS to maintain the voltage of the battery cell at a constant value. The corresponding current and voltage thresholds used by the BMS to control the current allow the designer to adjust the BMS controller to be more aggressive or less aggressive depending on the application. Setting higher thresholds results in faster charging, as it allows higher current integration over the same time period. But this faster charging puts more stress on the lithium-ion battery cell, and if the thresholds are set too high, it can result in faster degradation. Therefore, this method allows a trade-off between minimizing the charging time and prolonging the life of the lithium-ion battery.

[0018] In an ideal battery, and without limiting the charging cell, all the charge needed to take the battery from one state of charge (SOC) to another SOC can be delivered instantaneously. However, the kinetic limitations in a real battery only allow a limited current to pass through the battery. Many internal processes of the battery have an impact on the charge transfer capability, for example, the limited diffusion rate of lithium ions in the electrolyte, the reduction / oxidation of materials other than the active material, the formation of a resistive film on the surface of the active particles, and the charge transfer limitations between the electrolyte and the active material. The faster the charge transfer is forced to occur, the stronger the impact of these processes on the health of the battery. Therefore, battery cell manufacturers always provide additional information about the utilization constraints on their battery cells. These constraints mainly involve limitations on the maximum charge or discharge current, lower and upper cut-off voltages, and operating temperature domain. Some manufacturers provide these limits at different operating ambient temperatures. All of these limits are suitable for the CC / CV charging method, and therefore are quite conservative, as these limits are specified for the entire lifetime of the battery.

[0019] Studies have shown that even though current and voltage thresholds can be used as a proper proxy for the amount of stress that induces degradation in the battery cell at low and high states of charge (SOC), at intermediate SOC, the current and voltage thresholds are too conservative and can lead to unnecessary prolongation of the charging process. Or, if the current and voltage thresholds are set too high, they can induce accelerated degradation by inducing high stress towards the end of their respective phases.

[0020] SOC can be defined as the percentage of the remaining charge inside the battery to the total charge, ranging between about 0% to 100%. SOC provides information about the performance of the battery and when the battery should be recharged. Furthermore, the BMS can use the SOC information for power management. Therefore, accurate SOC information is crucial, especially in some applications such as electric vehicles, where consumers rely on the SOC information to determine their driving range.

[0021] It has been shown that standard charging techniques, such as CC-CV, can cause damage to the battery due to large currents passing through the battery if used for fast charging. These large currents lead to dangerous overpotentials and mechanical stresses in the battery, which cause the battery to age rapidly, leading to reduced lifetime.

[0022] More optimal charging methods have been proposed, such as methods that utilize real-time estimation of internal battery states. Internal battery states are used as indicators of underlying aging mechanisms. Examples of internal states that can be used as indicators of battery cell aging can include the overpotential of the Li-plating reaction in the anode, the overpotential of the electrolyte degradation in the cathode, or both. When the BMS uses thresholds on such indicator variables in addition to current and voltage thresholds, the charging process can include more than two phases. Figure 2An example of a charging profile comprising three phases is shown. The first and last phases are CC and CV charging phases, respectively. The second phase is defined by the BMS controller, which regulates the charging current to maintain the indicator variable at the desired threshold. The corresponding phase CI is the constant indicator phase. Similar to the current and voltage thresholds, the threshold of the indicator variable provides a trade-off between charging speed or degradation.

[0023] The method with the indicator variable provides superior results over the traditional CC-CV method by allowing faster charging at the same rate of degradation or extending the battery life while maintaining the charging duration. The method utilizes a physics-based lithium-ion cell state estimator. The estimator provides a plurality of estimates to a feedback controller, which can adjust the charging current based on the received estimates. The feedback controller can switch from one phase to another, thereby selecting between various factors that limit the charging current based on the implemented control logic and inputs. The inputs can include at least one or a set of indicator variables and measured values.

[0024] As a result, the indicator variables can be actively controlled by the feedback controller such that their values do not fall below the threshold, thus preventing a charging regime with accelerated aging. Such an aging regime is visible in Figure 2 the conventional CC-CV charging, where the indicator variable falls below zero. Unlike the conventional CC-CV voltage profile, when using a physics-based estimator, the voltage profile does not fall below zero due to the limitation of the indicator variable to a certain range.

[0025] However, the estimator method is relatively computationally expensive, as it requires real-time estimation of the internal states of the battery cell with complex electrochemical models. The real-time estimation of the states can be provided by a state estimator, which represents a mathematical model of the physical processes occurring in the battery cell during charging. The mathematical model comprises a set of ordinary differential equations with algebraic constraints. A typical lithium-ion cell estimator based on such electrochemical models can have more than about 70 dynamic states, and since the charging controller typically uses only some of them as indicator variables during fast charging, the method has a certain computational inefficiency, especially when many battery cells are used simultaneously, for example, in an electric vehicle battery pack.

[0026] Therefore, there is a need for an optimized charging method, especially a fast charging method, which would be efficient, not computationally expensive, and provide fast charging while maintaining the expected or predetermined battery life.

[0027] In one or more embodiments, a system for fast charging of an electrochemical cell is disclosed. Without limiting the disclosure to a single definition, fast charging can be characterized as a process of supplying a charging current to a cell with the goal of increasing its SOC to a desired value in a minimum amount of time while taking into account the operational constraints of the cell with respect to voltage, current, temperature, and other variables. The charging time period can vary from a few seconds to a few minutes or hours. A non-limiting example charging time period can be about 10 to 20 minutes, 11 to 18 minutes, or 12 to 15 minutes, e.g., 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, or 20 minutes long.

[0028] The disclosed system and method include an ML model in place of or in addition to a physics-based estimator. The ML model allows for the function of the physics-based estimator to be streamlined. The ML model can be trained offline, thus reducing the computational requirements of the BMS, reducing cost, and improving the efficiency of the fast charging process.

[0029] Figure 3 A block diagram schematically illustrating a system 100 including a BMS 102 that controls a fast charging process of an electrochemical cell, in particular a lithium-ion cell 103, and including an ML model 104 is shown. The system 100 can also include one or more of a controller 108 having a memory 110, one or more sensors 112, a current source (not depicted), and a model 106 designed to train the ML model 104 offline, as discussed below. The BMS 102 is operably connected to the cell 103.

[0030] The ML model 104 can be a linear regression, a logistic regression, a K-Nearest Neighbor (KNN), a Support Vector Machine (SVM), a Decision Tree (DT), an ensemble, or an Artificial Neural Network (ANN). The ML model 104 can be embedded in the BMS 102.

[0031] During fast charging, the BMS 102 can use the ML model 104 to estimate a target variable to generate a control output. In particular, the BMS 102 uses the ML model 104 to estimate an optimal or ideal amount of current(s) to be supplied to the cell 103 during the entire duration of the fast charging. The optimal or ideal amount of current can vary over time during the charging process. The optimal or ideal amount of current is the amount of current that results in a minimum duration of fast charging or a minimum time period required for fast charging while achieving a minimum damage to the cell as a result of the charging process.

[0032] The ML model 104 is trained offline based on a set of input data as complete as possible. Thus, the ML model 104 is supplied with a training data set and trained based on this data set. The training data set can be generated by the model 106 parameterized and fitted to experimental data. Figure 4 Fig. 2 schematically depicts training the ML model 104 offline via the model 106.

[0033] The model 106 can be a physics-based model. The model 106 can be an analog model capable of simulating the internal state of the battery cell 103. The model 106 can be an electrochemical model providing an estimate of the state of the battery over its lifetime. The model 106 can be any model capable of simulating electrochemical processes of a lithium-ion battery cell. The model 106 can determine the state of the battery based on the level of degradation of the battery. The model 106 can be capable of predicting future states of the battery under possible future charging sequences and evaluating the charging performance over an extended period of time. The model 106 can approximate a nonlinear system of partial differential equations modeling electrochemical processes of the battery to a linearized system of algebraic equations. The model can be a dual foil model modeling electrochemical reactions in a lithium-ion battery cell using mass transport, diffusion, migration, and reaction kinetics. The model 106 can include a state filter algorithm, e.g., a Kalman filter for state estimation. The model 106 can be a simplified and computationally efficient model. The model 106 can be an equivalent circuit model. The model 106 can or can not be embedded in the BMS 102.

[0034] The training data set generated by the model 106 can include SOC trajectories corresponding to one or more (preferably, all) expected environmental conditions and / or initial states of the battery cell 103 from which charging is initiated. The training data set can also include SOC trajectories corresponding to various aging of the battery cell 103 and / or possible variations of battery cell parameters affecting the behavior of the battery cell during the charging process.

[0035] More specifically, the training data set can include a plurality of individual SOC trajectories or a collection thereof, where each trajectory is generated for a particular combination of model parameters, initial conditions, and environmental temperature.

[0036] As the chemistry of each lithium-ion cell 103 changes over time and usage, lithium-ion cells 103 can have different requirements for optimal charging throughout different phases of their life. Accordingly, the ML model 104 is trained on, or supplied with, data sets for different aging and life phases of the lithium-ion cell 103. Accordingly, the variation in model parameters results in charging trajectories corresponding to various ages of the lithium-ion cell 103 under consideration from the beginning of its life to the end of its expected life. The variation in model parameters also models the natural distribution of cell properties due to inaccuracies in the manufacturing process.

[0037] Varying the initial conditions at which charging begins (e.g., the SOC of the cell or the initial temperature) captures the dependence of the charging trajectory of the cell on the initial state. Varying ambient temperature results in capturing the effect of the environment on the charging trajectory. Ambient temperature, like internal temperature, has a significant impact on the performance of lithium-ion cells and limits their application at both low and high temperature ranges. At temperatures below the optimal range of about -20°C to 60°C, the chemical reaction activity and charge transfer speed can slow, which can result in reduced ionic conductivity in the electrolyte(s) and lithium ion diffusivity in the electrode(s). Accordingly, low temperatures can result in a reduction in the energy and power capacity of lithium-ion cells. Temperatures above the optimal range can likewise result in capacity loss due to lithium loss and a reduction in active material and power performance due to increased internal resistance. As temperature information can affect the quality of the cell 103 and its life, initial temperature as well as ambient temperature information is captured in the various charging trajectories of the training data.

[0038] The training data can also include a charging phase variable (a variable indicating which phase of the charging process (e.g., such as CC, CI, or CV) the cell 103 is in). The SOC trajectory can include a CC phase, a CI phase, and / or a CV phase.

[0039] In addition to the training data sets for different aging and life phases of the lithium-ion cell 103, the ML model 104 is also supplied with one or more operating conditions or measured data inputs as an offline measurement data set about the cell 103. The one or more operating conditions or offline measurement data sets can include cell voltage, current, internal temperature, capacity, ambient temperature, etc., or combinations thereof.

[0040] The model 106 also feeds the ML model 104, offline, with a target indicator variable range. The variable is not measurable. The indicator variable is an estimate of an internal battery state, such as a reaction overpotential or a state of a battery cell along an SOC trajectory. The indicator variable and the indicator value can vary based on model parameters, initial conditions, ambient temperature, etc., or a combination thereof.

[0041] After the training data set is provided to the ML model 104 offline by the physics-based model 106 and the ML model 104 is trained, the ML model 104 is made online or connected online. When online, the ML model 104 can be supplied with real-time input measurements or one or more operating conditions of the battery cell, such as a battery cell voltage, a current, an internal temperature, a capacity, an ambient temperature, etc., or a combination thereof. Based on the training data set and / or the measurements, the ML model 104 is able to estimate the indicator variable within the target range. The ML model 104 can accurately estimate the indicator along at least one or all possible charging trajectories. In other words, the ML model 104 can generate an indicator estimate along a charging trajectory of the battery cell.

[0042] In turn, the ML model 104 can provide, in real-time along the charging trajectory, an output to the controller 108 that is an estimate of the indicator value. The estimate of the indicator value can be used as an input to the controller 108. In response to receiving new inputs, such as one or more operating conditions or real-time measurements, the ML model 104 can generate and provide to the controller 108 a new estimated indicator value within the target. The generation and / or the supply can occur on a continuous, discontinuous, periodic, or random basis.

[0043] The variable can vary continuously, discontinuously, periodically, at regular or irregular intervals. The new estimate can be generated by the ML model 104 and provided to the controller 108 continuously, discontinuously, periodically, at regular or irregular intervals. Non-limiting examples of the intervals can be milliseconds, seconds, etc.

[0044] The trained ML model 104 can be included in the controller 108. The trained ML model 104 can be used as an input to the controller 108. The controller 108 can output the indicator value.

[0045] The controller 108 executes machine instructions stored in a memory 110 of the controller. The machine instructions can implement the following functions: receiving one or more operating conditions, receiving a trained ML model, outputting an indicator value along an SOC trajectory in response to the one or more operating conditions and the trained ML model to control a fast charging state of a lithium-ion battery cell from a current source, determining a first indicator value at a time (t), determining a second indicator value at a time (t+1), etc.

[0046] The controller 108 can output an indicator value along the SOC trajectory and / or an amount of current to be provided to the battery cell 103 during the charging process, particularly during one or more phases of the charging process. The control output can include an amount of current to be supplied to the lithium-ion battery cell 103 over a predetermined time period to achieve ideal or optimal charging or fast charging.

[0047] The ML model 104 is in communication with the controller 108, which can be embedded in the BMS 102 together with the ML model 104. The controller 108 can be operably connected to a current source (not shown) to which the controller 108 supplies a control output. The control output can include initiating or stopping the supply of current from the current source to the battery cell 103, increasing, decreasing, adjusting, or maintaining the amount of current to be provided to the battery cell 103 at any given instant during the charging time period or during the duration of fast charging or both.

[0048] The controller 108 includes one or more hardware and software components and can be implemented as a digital control device that executes stored program instructions in its non-transitory memory 110. The controller 108 can be implemented as a digital microcontroller, but in alternative embodiments, the controller 108 is, for example, a general-purpose microprocessor, a field-programmable gate array (FPGA), an application-specific integrated circuit (ASIC), a digital signal processor (DSP), or any other suitable digital processor that contains hardware and software components to implement the monitoring of the electrochemical battery cell 103 and the control of the level of input current applied to the electrochemical battery cell 103 during the fast charging process.

[0049] The controller 108 can receive the following inputs: one or more operating conditions or real-time measurements from one or more sensors 112, the trained ML model 104, estimates of indicators from the ML model 104, additional inputs such as predetermined constraint parameters for limiting the input current to the battery cell 103, data corresponding to predetermined physical, chemical, and electrochemical properties of the battery cell 103, additional data provided from one or more external sources, and the like, or combinations thereof. In response to receiving new inputs (e.g., new estimated indicator values or one or more operating conditions), the controller 108 can output new indicator values, match the estimated indicator values to a target, generate new control outputs, supply the new control outputs to the current source, the battery cell 103, or both, and the like. The outputting, matching, generating, and / or supplying can occur on a continuous, discontinuous, or periodic basis.

[0050] The memory 110 includes one or more digital data storage devices including, but not limited to, random access memory (RAM), solid state storage memory including NAND and NOR flash memory or EEPROM memory, magnetic and optical data storage media, and the like. The memory 110 can also store data corresponding to predetermined physical, chemical, and electrochemical properties of the electrochemical cell 103, measured values of current, voltage, or temperature of the cell 103, or ambient temperature, and the like, or combinations thereof. The memory 110 can be a non-transitory memory. The memory 110 can also store ML model 104 data corresponding to estimated target indicator variables, predetermined constraint parameters for limiting input current to the electrochemical cell 103, sensor data received from one or more sensors 112 in the system 100, current source data, machine instructions to be executed by the controller 108.

[0051] The system 100 can include one or more sensors 112, such as a cell 103 current sensor, a voltage sensor, an ambient temperature sensor, a cell internal temperature sensor, one or more additional sensors, or combinations thereof. The one or more sensors 112 can be located within the cell 103, on an outer surface of the cell 103, or both. The one or more sensors 112 can sense one or more operating conditions, measured data, or data sets, provide the one or more operating conditions or measured data or data sets as inputs to the controller 108, the ML model 104, and / or the BMS, in real-time, offline, or both. Data from the one or more sensors 112 can be used by the controller 108 and / or the ML model 104 to generate target indicator variable values.

[0052] The cell 103 includes two electrodes that are electrically connected to a current source to enable the current source to deliver an input current that charges the cell 103. The cell 103 can include a plurality of electrochemical cells. The cell 103 can include one or more electrochemical cells that are integrated into a single physical package with two electrical terminals that receive current from an external current source during a fast charge operation. The battery package can optionally include one or more of the controller 108 and the sensors 112 to control the charging process of the battery.

[0053] In one or more embodiments, a fast charging process of an electrochemical cell 103 is described herein. The method can include generating training data offline and training the ML model 104. The trained ML model 104 can be connected online to accurately estimate indicator values along at least one charging trajectory. The method can include generating a dataset offline that covers all possible charging trajectories as described above.

[0054] The method can include the disclosed model development, including selection of machine learning model architecture and model inputs. As Figure 4 As shown, the method can include generating a training dataset offline by the physics-based model 106 that is capable of producing experimentally verified estimates of the desired indicator variable values in response to provided inputs. The training dataset can include charging trajectories, charging phase data, and an offline measurement dataset, including voltage, current, and temperature of the battery cell 103. The method can include training the ML model 104 offline with the training dataset.

[0055] The method can include setting a target, predetermined target, or predetermined value for a variable. The variable can be an internal state of the battery cell 103 along the SOC trajectory, such as a reaction overpotential during charging. The method can also include setting an indicator for the target. The method can also include providing the indicator variable to the ML model 104.

[0056] The method can also include connecting or causing the ML model 104 online once the ML model 104 is trained offline with the generated training dataset. The method can include providing the trained ML model as input to the controller 108. The method can also include providing the ML model with one or more operating conditions or real-time measurement values of the battery cell 103 once the ML model 104 is connected online, such as Figure 3 The generated training dataset and / or real-time measurement values or one or more operating conditions provide sufficient information to the ML model 104 or controller 108 to accurately estimate the indicator values along the charging trajectory online. Once the indicator is estimated by the ML model 104, the method can include supplying the estimated indicator by the ML model 104 to the controller 108.

[0057] The method can include executing machine instructions by the controller 108 that implement the following functions: receiving the one or more operating conditions and the trained ML model, outputting the indicator values along the SOC trajectory in response to the one or more operating conditions and the trained ML model to control the fast charging state of the lithium ion battery from the current source, determining a first indicator value at time (t), determining a second indicator value at time (t+1), and so on.

[0058] The method can further include matching, by the controller 108, the estimated metric to a predetermined target. The method can further include receiving, by the controller 108, one or more controller inputs. The one or more controller inputs can include one or more operating conditions or one or more real-time measurements from one or more sensors 112, a trained ML model, a metric estimate from the ML model 104, and the like, or a combination thereof. The method can further include outputting, from the controller 108, a control output in the form of a metric value along a state of charge (SOC) trajectory or an amount of current to be supplied to the battery cell 103 from a current source to complete the fast charge in a shortest amount of time with a minimum amount of damage to the battery cell 103.

[0059] The method can further include measuring one or more operating conditions, such as a voltage, a current, a capacity, an internal temperature, an ambient temperature, and / or other properties of the battery cell 103 and / or the system 100. The measurements can be taken one, at least one, one or more times continuously, discontinuously, randomly, regularly, during, before, and / or after the fast charging process, in real-time, online, offline, or a combination thereof. The method can further include collecting the measurements from the one or more sensors 112 at least one or more times at regular or irregular intervals, continuously, or discontinuously. The method can include providing or supplying the measurements to the model 106. The method can include providing the real-time measurements as real-time inputs to the ML model 104, the BMS 102, the controller 108, or a combination thereof.

[0060] The method can include generating the estimated metric value one, at least one, one or more times continuously, discontinuously, randomly, regularly, during, before, and / or after the fast charging process. The generation can be in response to the ML model 104 and / or the controller 108 receiving a new set of inputs, such as the one or more operating conditions, measurements of the voltage, current, and / or temperature of the battery cell 103, before learning the ML model 104, or both.

[0061] The method can include further learning of the ML model 104 from real-time or offline inputs from the particular battery cell 103, external data, additional training data sets, or a combination thereof, before, during, and / or after the fast charging process.

[0062] The method can include initiating and / or stopping the output of the control output to the current source. The method can include initiating and / or stopping the flow of the current from the current source to the battery cell 103. The method can include adjusting, increasing, decreasing, changing, maintaining, or regulating the amount of current supplied to the battery cell 103 during the charging process.

[0063] The method can include rapidly charging the battery cell 103 to a predetermined value. The predetermined value can be about 40, 50, 55, 60, 65, 70, 75, 80, 85, 90, 91, 92, 93, 94, 95, 96, 97, 98, 99, or 100% SOC or full capacity. Capacity refers to the total amount of charge that can be drawn from the battery cell until the battery cell is depleted. A fully charged battery has a 100% SOC.

[0064] Figure 5 A non-limiting example of a sequence of method steps is depicted. Figure 5 The process 500 is schematically illustrated, including ML training process steps 502 and 504, estimation of metrics in steps 506-514, and a rapid charging process in steps 516-524.

[0065] The processes, methods, or algorithms disclosed herein can be deliverable to, or implemented by, a processing device, controller, or computer that can include any existing programmable electronic control unit or specialized electronic control units. Similarly, the processes, methods, or algorithms can be stored as data and instructions on a non-transitory storage medium included in or accessible to the controller or computer, including a read-only memory (ROM) device, a magnetic or optical disk, or any other non-transitory storage medium known in the art. The processes, methods, or algorithms can also be embodied in software executable on the controller or computer. Alternatively, the processes, methods, or algorithms can be implemented entirely or in part with a specialized control device, such as a dedicated integrated circuit (ASIC), field-programmable gate array (FPGA), state machine, controller, or other hardware components or devices, or a combination of hardware, software, and firmware components.

[0066] While the example embodiments have been described above, it is not meant that the embodiments describe all possible forms of the claims. The words used in the specification are words of description rather than limitation, and it is understood that various changes can be made without departing from the spirit and scope of the disclosure. As previously described, features of various embodiments can be combined to form further embodiments of the present application that can not be expressly described or illustrated. While various embodiments can have been described as providing advantages or being preferred over other embodiments or prior to the application with respect to one or more desired characteristics, one of ordinary skill in the art will recognize that one or more features or characteristics can be deemed unimportant under the circumstances, and that some features or characteristics can be deemed important for certain applications. As such, it should be understood that it is intended to cover all such modifications and variations of various embodiments included within the scope of the disclosed concepts. It is intended that each of the claims cover all such modifications and variations of this application.

Claims

1. A lithium-ion battery management system, comprising: a lithium-ion battery cell; one or more sensors configured to sense one or more operating conditions of the lithium-ion battery cell; a controller having non-transitory memory for storing machine instructions to be executed by the controller and operably connected to the lithium-ion battery cell, the machine instructions, when executed by the controller, implement the following functions: receiving first values of the one or more operating conditions at time (t) and second values of the one or more operating conditions at time (t+1), and a trained machine learning (ML) model including a first state of charge (SOC) trajectory at a first phase of a fast charging process and a second SOC trajectory at a second phase of the fast charging process; and outputting a first indicator value along the first SOC trajectory in response to the first values of the one or more operating conditions and the trained ML model, and outputting a second indicator value along the second SOC trajectory in response to the second values of the one or more operating conditions and the trained ML model, to control the fast charging process of the lithium-ion battery cell from a current source.

2. The lithium-ion battery management system of claim 1, wherein, the SOC trajectory includes a constant current (CC) phase, a constant indicator (CI) phase, and / or a constant voltage (CV) charging phase.

3. The lithium-ion battery management system of claim 1, wherein, the first and second indicator values are first and second states of the lithium-ion battery cell along the first and second SOC trajectories.

4. The lithium-ion battery management system of claim 1, wherein, the first and second indicator values are overpotential values.

5. The lithium-ion battery management system of claim 1, wherein, the one or more operating conditions include one or more of an internal battery cell temperature, a battery cell voltage, and a current.

6. The lithium-ion battery management system of claim 1, wherein, the first SOC trajectory is different from the second SOC trajectory.

7. The lithium-ion battery management system of claim 6, wherein, the first phase of the fast charging process is earlier than the second phase of the fast charging process.

8. The lithium-ion battery management system of claim 1, wherein, the trained ML model includes a training data set including a plurality of individual charging trajectories, and each trajectory is generated by the trained ML model for a particular combination of model parameters, initial conditions, and / or ambient temperature.

9. The lithium-ion battery management system of claim 1, wherein, the trained ML model is a physics-based trained model.

10. A method of fast charging a lithium-ion battery cell, the method comprising: sensing one or more operating conditions of the lithium-ion battery cell; receiving first values of the one or more operating conditions at time (t) and second values of the one or more operating conditions at time (t+1), and a trained machine learning (ML) model including a first state of charge (SOC) trajectory at a first phase of a fast charging process and a second SOC trajectory at a second phase of the fast charging process; and outputting a first indicator value along the first SOC trajectory in response to the first values of the one or more operating conditions and the trained ML model, and outputting a second indicator value along the second SOC trajectory in response to the second values of the one or more operating conditions and the trained ML model, to control the fast charging process of the lithium-ion battery cell from a current source.

11. The method of claim 10, wherein, the SOC trajectory includes a constant current (CC) phase, a constant indicator (CI) phase, and / or a constant voltage (CV) charging phase.

12. The method of claim 10, wherein, The first and second indicator values are first and second states of the lithium-ion battery cell along the first and second SOC trajectories.

13. The method of claim 10, wherein, The first and second indicator values are overpotential values.

14. The method of claim 10, wherein, The one or more operating conditions include one or more of an internal battery cell temperature, a battery cell voltage, and a current.

15. The method of claim 10, wherein, The trained ML model includes a training set that includes a plurality of individual charge trajectories and further includes each trajectory generated by the trained ML model for a particular combination of model parameters, initial conditions, and / or ambient temperature.

16. A method of estimating a battery target indicator value for a lithium-ion battery cell, the method comprising: training a machine learning (ML) model including a training data set that includes: a plurality of individual charge trajectories, a charge phase variable, a measurement data set, and an indicator variable, wherein the plurality of individual charge trajectories includes a first state of charge (SOC) trajectory at a first phase of a fast charge process and a second SOC trajectory at a second phase of the fast charge process, each trajectory generated by the trained ML model for a particular combination of model parameters, initial conditions, and / or ambient temperature; sensing one or more operating conditions of the lithium-ion battery cell; receiving a first value of the one or more operating conditions at a time (t) and a second value of the one or more operating conditions at a time (t+1), and a trained ML model including first and second SOC trajectories; and outputting, in response to the first value of the one or more operating conditions and the trained ML model, a first indicator value along the first SOC trajectory, and in response to the second value of the one or more operating conditions and the trained ML model, a second indicator value along the second SOC trajectory, to control a fast charge process of the lithium-ion battery cell from a current source.

17. The method of claim 16, wherein, The first and second indicator values are first and second states of the lithium-ion battery cell along the first and second SOC trajectories.

18. The method of claim 16, further comprising supplying the training data set from a physics-based model.

19. The method of claim 16, wherein, The measurement data set includes one or more operating conditions of the lithium-ion battery cell.

20. The method of claim 16, wherein, The one or more operating conditions include one or more of an internal battery cell temperature, a battery cell voltage, and a current. The one or more operating conditions include one or more of an internal battery cell temperature, a battery cell voltage, and a current.

Images