Monitoring batteries using low-frequency stress-waves

Abstract

A system includes a function generator configured to generate a plurality of signals that are transmitted from the function generator, through a first transducer, and to a battery. The system also includes a data acquisition system configured to measure a plurality of responses that are transmitted from the battery, through a second transducer, and to the data acquisition system. The responses include a plurality of first responses of the battery in response to first signals, and a second response of the battery in response to a second signal. The system also includes a computing system configured to train a machine learning model based upon the first responses and a plurality of corresponding actual SoC values to produce a trained machine learning model. The trained machine learning model is then configured to predict the SoC value of the battery based upon the second response.

Description

PATENT Attorney Docket No.: 2024-022 MONITORING BATTERIES USING LOW-FREQUENCY STRESS-WAVES Cross-Reference to Related Applications

[0001] This patent application claims priority to U.S. Provisional Patent Application No. 63 / 556,630, filed on February 22, 2024, which is incorporated by reference. Background

[0002] Most methods for measuring or estimating the state of charge (SoC) of a battery require the battery to be taken offline and are thus not suitable for inspection / testing in real-time while the battery is online and in operation. In addition to the requirement to take a battery offline from its load, electrical methods of a SoC measurement, such as electrical impedance spectroscopy (EIS), require adding to or modifying the electrical circuits powered by the battery to allow for the generation of an alternating current and measurement of the electrical impedance response, which is inconvenient or even impossible in many electrical applications. Electrical methods that attempt to correlate either the battery terminal voltage during online operation or the open-circuit voltage (OCV) with the battery SoC must deal with the non-linear nature of the voltage-SoC curve and its significant fluctuation with confounding variables such as temperature, recent charge / discharge history, current rest time or load, etc. In addition, certain battery chemistries such as LiFePO4 have a large flat region in that voltage-SoC relationship that leads to significant inaccuracy except when nearly empty or nearly full. Methods relying on the Ampere-Integral bookkeeping approach require a known state of capacity, and an accurate sensing of all currents in / out of the battery over time, and thus errors in measurement accumulate and lead to inaccurate estimations of charge state.

[0003] These shortcomings in accurately sensing online battery states cause many problems in operation, including limiting the ability for a battery management system (BMS) to maintain battery operation within safe parameters, or predict degraded or destabilized battery conditions. This lack of effective control can lead to damaging the battery thus shortening its lifespan or reliability. In the case of batteries with unstable chemistries such as lithium-ion batteries (LIB), not preventing operation outside of safe parameters can lead to thermal runaway, high-temperature difficult-to-extinguish fires, and even explosions, which have resulted in significant human injury, property damage, and pollution.PATENT Attorney Docket No.: 2024-022 Summary

[0004] A system for monitoring a state of charge (SoC) value of a battery is disclosed. The system includes a first transducer in contact with the battery, and a second transducer in contact with the battery. The system also includes a function generator configured to generate a plurality of signals that are transmitted from the function generator, through the first transducer, and to the battery. The signals include a plurality of first signals and a second signal. The system also includes a data acquisition system configured to measure a plurality of responses that are transmitted from the battery, through the second transducer, and to the data acquisition system. The responses include a plurality of first responses of the battery in response to the first signals, and a second response of the battery in response to the second signal. The system also includes a computing system configured to train a machine learning model based upon the first responses and a plurality of corresponding actual SoC values to produce a trained machine learning model. The trained machine learning model is then configured to predict the SoC value of the battery based upon the second response.

[0005] A method for monitoring a state of charge (SoC) value of a battery is also disclosed. The method includes generating a first signal to excite the battery. The first signal is transmitted through a first transducer. The method also includes measuring a first response of the battery in response to the first signal. The first response is received through a second transducer. The method also includes generating a plurality of second signals. The method also includes varying the SoC value of the battery to a plurality of different SoC values. The SoC value of the battery is varied simultaneously with generating the second signals. The method also includes measuring a plurality of second responses of the battery in response to the second signals. The method also includes training a machine learning model based upon the second responses and the different SoC values to produce a trained machine learning model. The method also includes generating a third signal. The method also includes measuring a third response of the battery in response to the third signal. The method also includes predicting the SoC value of the battery based upon the third response. The SoC value of the battery is predicted using the trained machine learning model while the battery is online.

[0006] A method for monitoring a state of charge (SoC) value of a battery while the battery is online. The method includes generating a first signal to excite the battery. The first signal isPATENT Attorney Docket No.: 2024-022 generated by a function generator. The first signal is transmitted through a first transducer that is in contact with a first side of the battery. The first signal has a frequency that varies within a frequency range between about 1 Hz and about 2 MHz. The method also includes measuring a first response of the battery in response to the first signal. The first response is measured by a data acquisition system. The first response is received through a second transducer that is in contact with a second, opposing side of the battery. The first response is received with a plurality of different frequencies within the frequency range. The method also includes determining peaks or local maxima in the first response. The method also includes determining a resonance frequency based upon the peaks or local maxima. The method also includes generating a plurality of second signals based upon the resonance frequency. The second signals are generated by the function generator. The second signals are transmitted through the first transducer. A central frequency of the second signals is the resonance frequency. The method also includes varying the SoC value of the battery to a plurality of different SoC values. The SoC value is varied using a battery charger / discharger. The SoC value of the battery is varied simultaneously with generating the second signals. The battery has a different SoC value when each of the second signals is generated. The method also includes measuring a plurality of second responses of the battery in response to the second signals. The second responses are measured by the data acquisition system. The second responses are received through the second transducer. The method also includes training a machine learning model based upon the second responses and the different SoC values to produce a trained machine learning model. The method also includes generating a third signal based upon the resonance frequency. The third signal is generated by the function generator. The third signal is transmitted through the first transducer. A central frequency of the third signal is the resonance frequency. The method also includes measuring a third response of the battery in response to the third signal. The third response is measured by the data acquisition system. The third response is received through the second transducer. The method also includes predicting the SoC value of the battery to produce a predicted SoC value. The SoC value is predicted based upon the third response. The SoC value of the battery is predicted using the trained machine learning model. The SoC value is predicted at a time that the third signal is generated.

[0007] It will be appreciated that this summary is intended merely to introduce some aspects of the present methods, systems, and media, which are more fully described and / or claimed below. Accordingly, this summary is not intended to be limiting.PATENT Attorney Docket No.: 2024-022 Brief Description of the Drawings

[0008] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments of the present teachings and together with the description, serve to explain the principles of the present teachings. In the figures:

[0009] Figure 1 illustrates a lithium iron phosphate (LFP) battery with attached piezoceramic transducers, according to an embodiment.

[0010] Figure 2 illustrates a system for testing and / or monitoring a SoC of the battery, according to an embodiment.

[0011] Figure 3 illustrates a graph of a Gaussian-modulated sinusoidal impulse signal, according to an embodiment.

[0012] Figure 4A illustrates a graph showing a sine sweep being performed on the piezo-battery- piezo system to generate the signal, and Figure 4B illustrates a schematic flowchart of an excitation process for testing / monitoring the battery stress-wave response, according to an embodiment.

[0013] Figure 5 illustrates an architecture of the ML-based SoC estimation, according to an embodiment.

[0014] Figure 6 illustrates a schematic view of the (e.g., Multi-ROCKET) ML model, according to an embodiment.

[0015] Figures 7A-7C, 8A-8C, 9A-9C, and 10A-10C illustrate electrical data during the cycling, according to an embodiment.

[0016] Figures 11A-11C, 12A-12C, 13A-13C, and 14A-14C illustrate ultrasonic response signals under different SoCs, according to an embodiment.

[0017] Figure 15 illustrates SoC estimation errors in the four experiments under the conventional method and the proposed method, according to an embodiment.

[0018] Figures 16A, 16B, 17A, 17B, 18A, 18B, 19A, and 19B illustrate estimation results, according to an embodiment.

[0019] Figure 20 illustrates a flowchart of a method for measuring a SoC of the battery using low-frequency stress waves, according to an embodiment. Detailed DescriptionPATENT Attorney Docket No.: 2024-022

[0020] Reference will now be made in detail to embodiments, examples of which are illustrated in the accompanying drawings and figures. In the following detailed description, numerous specific details are set forth in order to provide a thorough understanding of the invention. However, it will be apparent to one of ordinary skill in the art that the invention may be practiced without these specific details. In other instances, well-known methods, procedures, components, circuits, and networks have not been described in detail so as not to unnecessarily obscure aspects of the embodiments.

[0021] It will also be understood that, although the terms first, second, etc. may be used herein to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element from another. For example, a first object or step could be termed a second object or step, and, similarly, a second object or step could be termed a first object or step, without departing from the scope of the present disclosure. The first object or step, and the second object or step, are both, objects or steps, respectively, but they are not to be considered the same object or step.

[0022] The terminology used in the description herein is for the purpose of describing particular embodiments and is not intended to be limiting. As used in this description and the appended claims, the singular forms “a,” “an” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will also be understood that the term “and / or” as used herein refers to and encompasses any possible combinations of one or more of the associated listed items. It will be further understood that the terms “includes,” “including,” “comprises” and / or “comprising,” when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. Further, as used herein, the term “if” may be construed to mean “when” or “upon” or “in response to determining” or “in response to detecting,” depending on the context.

[0023] Attention is now directed to processing procedures, methods, techniques, and workflows that are in accordance with some embodiments. Some operations in the processing procedures, methods, techniques, and workflows disclosed herein may be combined and / or the order of some operations may be changed.

[0024] A piezoceramic transducer using a low-frequency (e.g., sub-ultrasonic) impulse signal may be used to excite a battery with acoustic stress waves that interact with the constituent layersPATENT Attorney Docket No.: 2024-022 of the battery under investigation. These stress waves propagate through (and / or partially reflect from) each of the battery component layers in a non-uniform fashion because the battery is non- homogenous. In other words, the material properties and dimensions of the shell, current collector(s), anodes, cathodes, electrolyte, and / or separator membranes may be different.

[0025] The differences in stress wave propagation represent the changes of characteristic features of the state of the battery layers. Another (i.e., a second) receiving piezoceramic transducer affixed to the opposite side of the battery from the first piezoceramic transducer receives the resultant stress waves that have been modified through interaction with the battery’s layers and converts the stress wave into an equivalent electrical waveform. A data acquisition system records the received electrical signals. The differences between the received signals under different SoC values can be interpreted via one or more models using machine-learning (ML) methods to form a fingerprint for the conditions of the battery.

[0026] When the battery is subjected to controlled charge-discharge cycles, it causes the electrochemical and mechanical properties of the constituent layers of the battery to change. Collecting a series of pulse-response stress-wave excitation measurements over the course of these charge cycles creates a dataset of many fingerprints for the state of the battery. Using machine learning methods to develop a relationship between the vibrational response and the battery state of charge allows for the state of charge of the battery to be measured without the need for conventional electrical methods of determining state of charge that typically require the battery to be taken offline.

[0027] Previous research by others into stress-wave testing of lithium ion batteries used high- frequency (e.g., 1 MHz -5 MHz) ultrasonic stress waves, commercial ultrasonic testing equipment, and mostly tested pouch-type cells. The system and method described herein may monitor cylindrical battery cells using low-frequency (e.g., sub-ultrasound) stress-wave signals, which enables the use of inexpensive and simpler electrical circuits for producing and measuring the signal waveforms – at least one or two orders of magnitude less expensive than commercial ultrasonic testing equipment. These inexpensive active sensing methods can be integrated into / onto each battery in a battery pack and thus enable cost-effective real-time online monitoring of battery systems without having to modify the battery operating procedures or electrical circuits powered by the batteries. However, since lower frequency excitation signals have limited sensitivity to internal changes of batteries during charging and discharging cycles, which leads toPATENT Attorney Docket No.: 2024-022 a compromise in SoC estimation accuracy. To compensate for the effect introduced by the reduction of sensitivity, machine learning methods may be used to extract many features that correlate the vibrational response with desired measurements such as state of charge (SoC) or state of health (SoH).

[0028] Providing more accurate online monitoring of battery SoC and SoH enables battery management systems (BMS) to maintain battery operating conditions within recommended safe parameters (i.e., preventing overcharge, overdischarge, etc.) thus improving battery safety, reliability and lifespan. This technology enables accurately measuring the SoC of the battery while the battery remains online, without modifying or affecting the operation of the circuit connected to the battery, as the sensing principle is mechanical-vibrational rather than electrical signals. It may be suitable for monitoring spiral-wound cylindrical format batteries, as well as pouch or coin cell formats.

[0029] This technology is battery chemistry agnostic. For example, it is effective even with LiFePO4 batteries that are characterized by having a very wide flat region in their OCV-SoC curve compared to other batteries. This technology is also effective with other battery chemistries such as lithium-ion with Ni-Mn-Co (NMC) cathodes, and lithium-polymer (LiCoO2 + graphite).

[0030] The use of low-frequency stress-waves and small, inexpensive piezoceramic wafers as transducers enables this technology to be integrated into battery packs involving hundreds or thousands of cells and monitor each cell individually in real-time. This is something that is not feasible with commercially-available expensive and bulky high-frequency ultrasonic diagnostic imaging equipment.

[0031] Experimental Setup

[0032] In one embodiment, a rechargeable lithium-ion battery (LIB) in a cylindrical format such as the popular “18650” case may be selected for testing and / or monitoring. Currently nickel- manganese-cobalt (NMC) and lithium iron phosphate (LFP) battery chemistries have been tested using this process. The electrodes may be connected to a battery tester to simulate an electrical load and recharging current source that the battery would be subjected to in real-world conditions. In the current setup, a device may be used to control the battery cycling process using the common constant current-constant voltage (CC-CV) profiles used for charging, and constant-current (CC) for discharging lithium-ion batteries at C-rates commonly encountered by these batteries (e.g.,PATENT Attorney Docket No.: 2024-022 between 0.1C to 0.5C) while simultaneously logging the current and battery terminal voltage over time.

[0033] A pair of lead zirconate titanate (PZT) piezoceramic transducers may be affixed opposite each other and radially across the cylindrical battery. One may be selected as the transmitting or exciting piezoelectric element, while the other may be selected as the receiving or sensing element. The piezoceramic elements may be glued onto the battery case (e.g., using cyanoacrylate “super glue” type adhesive) to ensure effective mechanical coupling between the piezoceramic elements and the battery case.

[0034] Figure 1 illustrates a lithium iron phosphate (LFP) battery 100 with attached piezoceramic transducers 110A, 110B, according to an embodiment. More particularly, Figure 1 illustrates a LFP 18650 cell 100 with 3x13 mm PZT piezoceramic transducers 110A, 110B glued thereto with cyanoacrylate. An 18650 battery holder with female banana jacks is soldered to the battery electrode terminals.

[0035] Figure 2 illustrates a system 200 for testing and / or monitoring a SoC of the battery 100, according to an embodiment. The system 200 may include a programmable function generator 210 as the excitation signal source. In one example, a NI PXI-1042 function generator 210 may be used, populated with a PXI-5412 function generator card, and a PXI-5105 analog-digital converter (ADC) / oscilloscope capture card. The excitation signal produced using the function generator 210 may be amplified using an excitation amplifier 220 (e.g., an external Trek Piezo Amplifier), and the output of the amplifier 220 may be sent to the transmitting piezoelectric transducer 110A to convert the electrical signal into a stress wave that excites the layers of the battery 100.

[0036] The stress waves received by the sensing piezoelectric transducer 110B may induce a weak electrical waveform on the piezo electrodes, which are connected to a data acquisition system 240 on the function generator (e.g., NI PXI-5105 ADC) 210. A software program (e.g., written in the NI LabView programming language) produces a “Virtual Instrument” (VI) application that records the received response signals and / or triggers the output excitation signal at a regular interval of time during the charge / discharge cycles provided by a charger / discharger 230. The compilation of many excitation-response signal data results in the stress-wave dataset used to characterize the mechanical state or fingerprint of the battery 100 at periodic intervals as the electrochemical state changes.PATENT Attorney Docket No.: 2024-022

[0037] The excitation signal may be a Gaussian-modulated sinusoidal pulse (e.g., generated using MATLAB code before being copied to the NI PXI-1042 system for use in the experiments). However, the use of other excitation signal types is possible as well. Figure 3 illustrates a graph of a Gaussian-modulated sinusoidal impulse signal 300, according to an embodiment. Figure 4A illustrates a graph showing a sine / cosine sweep 400 being performed on the piezo-battery-piezo system to determine the center frequency of the signal 300, and Figure 4B illustrates a schematic flowchart of an excitation process for testing / monitoring the battery stress-wave response, according to an embodiment.

[0038] The center frequency of the Gaussian pulse 300 can be selected in several ways. In one embodiment, the selection process includes exciting the battery 100 using a sweep or chirp cosine wave 400 with frequencies varying linearly (e.g., from 0 Hz – 2 MHz). The battery’s excitation response amplitude across the range of testing frequencies may then be analyzed for peaks or local maxima, indicating increased battery mechanical resonance at that frequency. Next, a frequency 410 (e.g., 123 kHz in this example) may be selected that demonstrates some degree of resonance on the lower end of the spectrum. This may be used to develop inexpensive active sensing circuits, and lower frequency operation enables the use of simpler circuits, less expensive components, and looser tolerances. Next, the selected frequency 410 of the sweep or chirp cosine wave 400 may be used as the center frequency input to the (e.g., MATLAB) code that generates the Gaussian- modulated sinusoidal pulse signal 300. Applicant has successfully used ultrasonic excitation signals 300 in the 50 kHz - 120 kHz range to accurately predict battery SoC. The system may also use signals 300 in the 20 kHz range at the lower threshold for ultrasonic signals as well as sonic signals at 15 kHz (and lower). For example, the center frequency of the signal 300 may be from about 1 kHz to about 1 MHz, about 5 kHz to about 100 kHz, or about 15 kHz to about 50 kHz. Previous research in testing the ultrasonic response characteristics of LIB has used much higher- frequency sensing, such as 2 MHz.

[0039] SoC Estimation Based on ML

[0040] Figure 5 illustrates an architecture of the ML-based SoC estimation, according to an embodiment. The collected response signals under different SOC states may be divided into a first (e.g., training) set and a second (e.g., test) set. Then, the training set may be used to train the machine learning / deep learning model, and the test set may be employed to test the estimationPATENT Attorney Docket No.: 2024-022 performance of the trained model. The machine learning model used in the experiment may be or include a multiple random Convolutional kernel transform (Multi-ROCKET) model.

[0041] Figure 6 illustrates a schematic view of the (e.g., Multi-ROCKET) ML model, according to an embodiment. The original Multi-ROCKET model is a fast time series classification (TSC) model that achieves high accuracy on the University of California Riverside TSC benchmark datasets. In present method, the Multi-Rocket model is modified to a regression model, and may include three steps which are convolution, feature extraction, and regression. In the convolutional operation, the Multi-Rocket model adopts a number of dilated kernels to convolve with the input (response signal), developing a set of feature maps. Subsequently, in the feature extraction, it employs four kinds of pooling operators (4 × number of feature maps) to extract four kinds of features (4 × number of feature maps) from each feature map and forms a final feature vector (1 by (4 × number of feature maps)). In the last step, a linear regressor (e.g., rigid regressor) may be utilized to predict the SoC with the final feature vector.

[0042] The four different pooling operators are named proportion of positive values (PPV), mean of positive values (MPV), mean of indices of positive values (MIPV), and longest stretch of positive values (LSPV), respectively. The PPV captures the proportion of positive values of each feature map and is defined as,^^^^^^ = ^^ ∑^^^^ ^^^^^^ (1)(2)map. Next, MPV computes the mean value of all positive values of the feature map,^^^^^^ = ^^ ∑^ ^^^^ ^^ ^^ ≤ ^^ (3)value of the feature map. In addition, MIPV captures information about the relative location of positive values of the feature map and is defined as, ^^ ^^^^^ = ∑^ ^^^^ ^ ^^ " > 0 ^^ ^^ (4)length of any subsequence of successive positive values in the feature map and is defined as,$%^^^^^ = "&'|) − ^|, ^∀^ ≤ , ≤ ), ^^,^ > 0^, ^^, ), , ≤ ^^ (5)PATENT Attorney Docket No.: 2024-022 where i, j, and k are all indices of positive values of the feature map. In addition to processing the raw time series, the Multi-Rocket model also applies four different pooling operators to their first order differences, which increases the diversity of the feature vector.

[0043] RESULTS AND DISCUSSION

[0044] In the experiments, two identical cylindrical batteries (e.g., lithium-ion NMC 18650) are used to demonstrate the performance of the proposed method. The settings of the galvanostatic charge-discharge cycles are listed in Table I, and the details of the ultrasonic detection are displayed in Table II. Table I. The settings of two experiments. Experiment Battery Voltage range Cycling method Current ID ID (V) constant current, 1 1 0.23C constant voltageconstant current, 2 20.5C constant voltage constant current, 3 3 2.75~4.2 0.5C constant voltage constant current, 4 4 2.75~4.2 0.05C constant voltage Table II. Details of ultrasonic detection. Central Experiment Battery frequency of the Deployment Detection ID ID excitation signal method interval (s) (kHz) 1 1 54 Pulser-Receiver 60 2 2 15 Pulser-Receiver 6 3 3 5 Pulser-Receiver 6 4 4 5 Pulser-Receiver 45

[0045] The electrical data during the cycling is presented in Figures 7A-7C, 8A-8C, 9A-9C, and 10A-10C. More particularly, Figures 7A-7C illustrate voltage versus time (Figure 7A), SoC versus time (Figure 7B), and current versus time (Figure 7C) of experiment 1, according to an embodiment. Figures 8A-8C illustrate voltage versus time (Figure 8A), SoC versus time (Figure 8B), and current versus time (Figure 8C) of experiment 2, according to an embodiment. FiguresPATENT Attorney Docket No.: 2024-022 9A-9C illustrate voltage versus time (Figure 9A), SoC versus time (Figure 9B), and current versus time (Figure 9C) of experiment 3, according to an embodiment. Figures 10A-10C illustrate voltage versus time (Figure 10A), SoC versus time (Figure 10B), and current versus time (Figure 10C) of experiment 4, according to an embodiment. As may be seen in Figures 9A-9C, the CC- CV charging profile is used from 0-100% SoC, while the CC discharging profile is used from 100%-0% SoC.

[0046] The ultrasonic response signals under different SoCs are shown in Figures 11A-11C, 12A-12C, 13A-13C, and 14A-14C. More particularly, Figures 11A-11C illustrate graphs showing low frequency stress wave response signals under different SoCs in the 1stcycle (Figure 11A), 5thcycle (Figure 11B), and 11thcycle (Figure 11C) of experiment 1, according to an embodiment. Figures 12A-12C illustrates graphs showing ultrasonic response signals under different SoCs in the 1stcycle (Figure 12A), 5thcycle (Figure 12B), and 11thcycle Figure 12C) of experiment 2, according to an embodiment. Figures 13A-13C illustrates graphs showing ultrasonic response signals under different SoCs in the 1stcycle (Figure 13A), 5thcycle (Figure 13B), and 11thcycle Figure 13C) of experiment 3, according to an embodiment. Figures 14A-14C illustrates graphs showing ultrasonic response signals under different SoCs in the 1stcycle (Figure 14A), 5thcycle (Figure 14B), and 11thcycle Figure 14C) of experiment 4, according to an embodiment.

[0047] Comparing Figures 11A-11C,12A-12C, 13A-13C, and 14A-14C, it may be seen that, even in the time domain, the variation in response signal amplitude (SA) and the time of flight (TOF) decreases from Figures 11A-11C to Figures 14A-14C. This decreased variation indicates that the lower frequency excitation signals in Figs 12A-12C and 13A-13C are less sensitive to the material property changes in battery material layers with respect to SoC. This decreased variation at lower frequencies presents more of a challenge for the ML model to correlate signal response with SoC

[0048] Additionally, at higher frequencies (e.g., Figures 11A-11C), there is more variation between the signal response at a given SoC at a different cycle counts, than the very little noticeable variation between signal responses at the same given SoC for batteries tested with lower frequency excitation signals (e.g., Figures 14A-14C).

[0049] Because battery aging effects accumulate (i.e., SoH decreases) with cycle counts, the decreased signal response sensitivity to cycle count aging at lower frequencies indicates that lowPATENT Attorney Docket No.: 2024-022 frequencies also present more of a challenge for the ML model to correlate signal response with SoH as well.

[0050] In experiment 1, the amplitude of response signal increases as the SoC grows between time index 2200 and 2900. Similarly, when the SoC increases, the time shifts of some peaks of the response signals rise. However, in experiment 2 and experiment 3, there are no obvious changes can be observed in the response signals under different SoCs. In addition, the collected response signals may be used to train and test the machine learning model. In each experiment, the response signals captured during the last several charging and discharging cycles may be used as the test set, and the remaining signals may be used as the training set.

[0051] The proposed method may be compared with conventional methods. They extract the time of flight (TOF), signal amplitude (SA), and battery voltage (vol) from the response signals, and employ a support vector machine (SVM) to estimate the SOC of the LIBs based on extracted features. Particularly, root mean square error may be adopted as the SOC estimation performance metric. At last, the estimation errors are displayed in Table III and Figure 15. More particularly, Figure 15 illustrates SoC estimation errors in three experiments under the conventional method and the proposed method, according to an embodiment.

[0052] Not only is acoustic response variation decreasing at lower excitation signal rates, but also acoustic response variation also decreases when using slower charge / discharge rates. Experiment 1 (e.g., 54 kHz) used 0.25 C, Experiment 2 (e.g., 15 kHz) & Experiment 3 (e.g., 5 kHz) used 0.5 C, and Experiment 4 also used 5 kHz excitation as in Experiment 3, however, with 1 / 10th of the charge rate: 0.05 C.

[0053] Conventional methods of acoustic SoC estimation lose accuracy greatly between Experiment 3 and Experiment 4, whereas the model described herein loses much less accuracy at both low excitation rates and low charge rates. As many battery usage scenarios involve varying loads, the model described herein is more effective at accurately estimating SoC when the charge / discharge rates are very low and the material property changes are much more subtle. For example, a cellphone or golfcart that can last 10-20 hours on a single charge indicates a low average discharge rate of 0.1 C-0.05 C, rather than the higher discharge rates (e.g., 0.5 C, 1 C) where other conventional models still have enough accuracy to be competitive.

[0054] The corresponding estimation results are displayed in Figures 16A, 16B, 17A, 17B, 18A, 18B, 19A, and 19B. More particularly, Figure 16A illustrates a graph showing estimation resultsPATENT Attorney Docket No.: 2024-022 on testing set in experiment 1 under (TOF+SA+vol)+SVM, and Figure 16B illustrates a graph showing estimation results on testing set in experiment 1 under the method proposed herein. Figure 17A illustrates a graph showing estimation results on testing set in experiment 2 under (TOF+SA+vol)+SVM, and Figure 17B illustrates a graph showing estimation results on testing set in experiment 2 under the method proposed herein. Figure 18A illustrates a graph showing estimation results on testing set in experiment 3 under (TOF+SA+vol)+SVM, and Figure 18B illustrates a graph showing estimation results on testing set in experiment 3 under the method proposed herein. Figure 19A illustrates a graph showing estimation results on testing set in experiment 3 under (TOF+SA+vol)+SVM, and Figure 19B illustrates a graph showing estimation results on testing set in experiment 4 under the method proposed herein.

[0055] As may be seen in Figures 17A, 17B, 18A, 18B, 19A, and 19B, not only is the present model more accurate in general, but its deviation is lower – Figures 18A vs 18B and 19A vs 19B show a few significant outlier predictions made by the conventional model that reach approximately 50% magnitude of the total scale. Those significant deviations in conventional models present more problems for a BMS than the present model.

[0056] As may be seen, the proposed method performs better than the conventional (e.g., ((TOF+SA+vol)+SVM) method in the SOC estimation of the cylindrical batteries with low frequency excitation signals. Table III. Root mean square errors under two different methods in two experiments. Method RMSE Source of method ID Experiment 1 Experiment 2 Experiment 3 Experiment 4 The proposed 1 0.0072 0.0141 0.0346 0.0531 method (TOF+SA+vol)+S 2 0.0266 0.0476 0.0851 0.0778 VM

[0057] Related configurations:

[0058] Other configurations of excitation transducers / actuators can be used to generate the stress-waves in the battery under testing / monitoring used for the active sensing principle. This may take the form of other piezoelectric materials such as the use of PVDF film instead of PZT ceramics, using one or more transducers for excitation, and having the transducers eitherPATENT Attorney Docket No.: 2024-022 permanently attached or removable or integrated into the battery during manufacturing for example.

[0059] Similarly, to receive the stress-wave response signal from the battery, other methods besides the use of a PZT piezoceramic transducer may be possible, such as using fiberoptic FBG sensors which may simultaneously monitor many batteries simultaneously and thus could be suitable for use in battery pack or battery module monitoring. It is also possible to capture a response signal from the same transducer used to excite the battery, by using echo mode, in which the transducer first excites the battery with an impulse signal and then detects the echoes of that signal as portions of the stress-wave are reflected backwards as the wave propagates through the battery’s constituent layers.

[0060] Figure 20 illustrates a flowchart of a method 2000 for measuring a SoC of the battery 100 using low-frequency stress waves, according to an embodiment. The method 2000 may be performed while the battery 100 is offline or while the battery 100 remains online. An illustrative order of the method 2000 is provided below; however, one or more portions of the method 2000 may be performed in a different order, simultaneously, repeated, or omitted.

[0061] The method 2000 may include generating a first signal to excite the battery, as at 2005. The first signal is generated by the function generator 210. The first signal is transmitted through the first transducer 110A that is in contact with the first side of the battery 100. The first transducer 110A may be adhered or clamped to the first side of the battery 100. The first transducer 110A may be or include a piezoceramic transducer. The first signal may be or include a sweep cosine wave or a chirp cosine wave. The first signal has a frequency that varies linearly. The frequency varies within a frequency range between about 1 Hz and about 2 MHz, about 10 Hz and about 1.75 MHz, or about 100 Hz and about 1.5 MHz.

[0062] The method 2000 may also include measuring a first response of the battery 100 in response to the first signal, as at 2010. The first response is measured by the data acquisition system 240. The first response is received through the second transducer 110B that is in contact with a second, opposing side of the battery 100. The second transducer 110B may be adhered or clamped to the second, opposing side of the battery 100. The second transducer 110B may be or include a piezoceramic transducer. The first response may be received at a plurality of different frequencies within the frequency range.PATENT Attorney Docket No.: 2024-022

[0063] The method 2000 may also include determining peaks or local maxima in the first response, as at 2015. The peaks or local maxima may indicate an increased mechanical resonance in the battery 100 at the frequencies of the peaks or local maxima.

[0064] The method 2000 may also include determining a resonance frequency based upon the peaks or local maxima, as at 2020. The resonance frequency demonstrates greater than a predetermined degree of resonance on a lower end of a spectrum that includes the peaks or local maxima.

[0065] The method 2000 may also include generating a plurality of second signals based upon the resonance frequency, as at 2025. The second signals may be generated by the function generator 210. The second signals may be transmitted through the first transducer 110A. A central frequency of the second signals may be the resonance frequency. The second signals may be Gaussian-modulated sinusoidal pulse signals. A frequency of the second signals may be from about 1 kHz to about 200 kHz or about 5 kHz to about 100 kHz.

[0066] The method 2000 may also include amplifying the second signals to produce amplified second signals, as at 2030. The second signals may be amplified by the amplifier 220.

[0067] The method 2000 may also include varying the SoC of the battery to a plurality of different SoCs, as at 2035. The SoC may be varied using the battery charger / discharger 230. The SoC of the battery may be varied simultaneously with generating the second signals. The battery 100 has a different SoC when each of the second signals is generated.

[0068] The method 2000 may also include measuring a plurality of second responses of the battery 100 in response to the second signals, as at 2040. The second responses may be measured by the data acquisition system 240. The second response may be in response to the amplified second signals. The second responses may be received through the second transducer 110B.

[0069] The method 2000 may also include training a machine learning model based upon the different SoC values (from step 2035) and the second responses (from step 2040) to produce a trained machine learning model, as at 2045. More particularly, the machine learning model may be trained based upon the second responses under the different SoC values (e.g., dataset) and the different SoC values (e.g., label). The different SoC values are actual SoC values from a known source (i.e., the potentiostat used to accurately charge / discharge the battery 100), as opposed to a “predicted SoC” or “estimated SoC,” which is the output of this method 2000. As used herein, each second response corresponds to an actual SoC value in Figure 7B / 8B / 9B / 10B. Based onPATENT Attorney Docket No.: 2024-022 collected electrical data (Figures 7C, 8C, 9C, 10C), the different (e.g., actual) SoC values (Figures 7B, 8B, 9B, 10B) are computed through Ampere-hour integral method.

[0070] The method 2000 may also include generating a third signal based upon the resonance frequency, as at 2050. The third signal may be generated by the function generator. The third signal may be transmitted through the first transducer 110A. A central frequency of the third signal may be the resonance frequency. The third signal may be or include a Gaussian-modulated sinusoidal pulse signal. A frequency of the third signal may be from about 1 kHz to about 200 kHz or about 5 kHz to about 100 kHz.

[0071] The method 2000 may also include amplifying the third signal to produce an amplified third signal, as at 2055. The third signal may be amplified by the amplifier 220.

[0072] The method 2000 may also include measuring a third response of the battery 100 in response to the third signal, as at 2060. The third response may be measured by the data acquisition system 240. The third response may be in response to the amplified third signal. The third response may be received through the second transducer 110B.

[0073] The method 2000 may also include predicting / estimating / determining the SoC of the battery 100 based upon the third response, as at 2065. The SoC of the battery 100 may be predicted using the trained machine learning model. More particularly, the trained machine learning model may adopt a plurality of dilated kernels to convolve with the third response to develop a set of feature maps. The trained machine learning model may then extract a plurality of features from each feature map using a plurality of (4) different pooling operators to produce a final feature vector. The trained machine learning model may then predict the SoC value with the final feature vector using a linear regressor.

[0074] The method 2000 may also include charging or replacing the battery 100 in response to the determined SoC, as at 2070. More particularly, the battery 100 may be charged or replaced in response to the SoC being less than a predetermined threshold.

[0075] Clauses

[0076] Clause 1. A system for monitoring a state of charge (SoC) value of a battery, the system comprising: a first transducer in contact with the battery; a second transducer in contact with the battery; a function generator configured to generate a plurality of signals that are transmitted from the function generator, through the first transducer, and to the battery, wherein the signals include: a plurality of first signals; and a second signal; a data acquisition system configured to measure aPATENT Attorney Docket No.: 2024-022 plurality of responses that are transmitted from the battery, through the second transducer, and to the data acquisition system, wherein the responses include: a plurality of first responses of the battery in response to the first signals; and a second response of the battery in response to the second signal; and a computing system configured to train a machine learning model based upon the first responses and a plurality of corresponding actual SoC values to produce a trained machine learning model, wherein the trained machine learning model is then configured to predict the SoC value of the battery based upon the second response.

[0077] Clause 2. The system of clause 1, wherein the first transducer is in contact with a first side of the battery, and wherein the second transducer is in contact with a second, opposing side of the battery.

[0078] Clause 3. The system of clause 1, further comprising a battery charger / discharger configured to vary the corresponding actual SoC values of the battery.

[0079] Clause 4. The system of clause 3, wherein the corresponding actual SoC values of the battery are varied simultaneously with generating the first signals to produce the first responses.

[0080] Clause 5. A method for monitoring a state of charge (SoC) value of a battery, the method comprising: generating a first signal to excite the battery, wherein the first signal is transmitted through a first transducer; measuring a first response of the battery in response to the first signal, wherein the first response is received through a second transducer; generating a plurality of second signals; varying the SoC value of the battery to a plurality of different SoC values, wherein the SoC value of the battery is varied simultaneously with generating the second signals; measuring a plurality of second responses of the battery in response to the second signals; training a machine learning model based upon the second responses and the different SoC values to produce a trained machine learning model; generating a third signal; measuring a third response of the battery in response to the third signal; and predicting the SoC value of the battery based upon the third response, wherein the SoC value of the battery is predicted using the trained machine learning model while the battery is online.

[0081] Clause 6. The method of clause 5, wherein the first transducer is in contact with a first side of the battery, and wherein the second transducer is in contact with a second, opposing side of the battery.

[0082] Clause 7. The method of clause 5, wherein the first signal has a frequency that varies within a frequency range between about 1 Hz and about 2 MHz.PATENT Attorney Docket No.: 2024-022

[0083] Clause 8. The method of clause 5, wherein the first response is received at a plurality of different frequencies within the frequency range.

[0084] Clause 9. The method of clause 5, further comprising: determining peaks or local maxima in the first response; and determining a resonance frequency based upon the peaks or local maxima.

[0085] Clause 10. The method of clause 9, wherein the second signals are transmitted through the first transducer, and wherein the second signals are generated based upon the resonance frequency.

[0086] Clause 11. The method of clause 10, wherein a central frequency of the second signals is the resonance frequency.

[0087] Clause 12. The method of clause 9, wherein the third signal is transmitted through the first transducer, and wherein the third signal is generated based upon the resonance frequency.

[0088] Clause 13. The method of clause 12, wherein a central frequency of the third signal is the resonance frequency.

[0089] Clause 14. The method of clause 5, wherein the battery has a different SoC value when each of the second signals is generated.

[0090] Clause 15. The method of clause 5, wherein the second responses and the third response are received through the second transducer.

[0091] Clause 16. A method for monitoring a state of charge (SoC) value of a battery while the battery is online, the method comprising: generating a first signal to excite the battery, wherein the first signal is generated by a function generator, wherein the first signal is transmitted through a first transducer that is in contact with a first side of the battery, and wherein the first signal has a frequency that varies within a frequency range between about 1 Hz and about 2 MHz; measuring a first response of the battery in response to the first signal, wherein the first response is measured by a data acquisition system, wherein the first response is received through a second transducer that is in contact with a second, opposing side of the battery, and wherein the first response is received with a plurality of different frequencies within the frequency range; determining peaks or local maxima in the first response; determining a resonance frequency based upon the peaks or local maxima; generating a plurality of second signals based upon the resonance frequency, wherein the second signals are generated by the function generator, wherein the second signals are transmitted through the first transducer, and wherein a central frequency of the second signals isPATENT Attorney Docket No.: 2024-022 the resonance frequency; varying the SoC value of the battery to a plurality of different SoC values, wherein the SoC value is varied using a battery charger / discharger, wherein the SoC value of the battery is varied simultaneously with generating the second signals, and wherein the battery has a different SoC value when each of the second signals is generated; measuring a plurality of second responses of the battery in response to the second signals, wherein the second responses are measured by the data acquisition system, and wherein the second responses are received through the second transducer; training a machine learning model based upon the second responses and the different SoC values to produce a trained machine learning model; generating a third signal based upon the resonance frequency, wherein the third signal is generated by the function generator, wherein the third signal is transmitted through the first transducer, and wherein a central frequency of the third signal is the resonance frequency; measuring a third response of the battery in response to the third signal, wherein the third response is measured by the data acquisition system, wherein the third response is received through the second transducer; and predicting the SoC value of the battery to produce a predicted SoC value, wherein the SoC value is predicted based upon the third response, wherein the SoC value of the battery is predicted using the trained machine learning model, and wherein the SoC value is predicted at a time that the third signal is generated.

[0092] Clause 17. The method of clause 16, wherein the peaks or local maxima indicate an increased mechanical resonance in the battery at the frequencies of the peaks or local maxima, and wherein the resonance frequency demonstrates greater than a predetermined degree of resonance on a lower end of a spectrum that includes the peaks or local maxima.

[0093] Clause 18. The method of clause 16, wherein the first signal comprises a sweep cosine wave or a chirp cosine wave, wherein the second signals are Gaussian-modulated sinusoidal pulse signals, and wherein the third signal is a Gaussian-modulated sinusoidal pulse signal.

[0094] Clause 19. The method of clause 16, wherein a frequency of the second signals is from about 5 kHz to about 100 kHz, and wherein a frequency of the third signal is from about 5 kHz to about 100 kHz.

[0095] Clause 20. The method of clause 16, wherein the trained machine learning model: adopts a plurality of dilated kernels to convolve with the third response to develop a set of feature maps; extracts a plurality of features from each feature map using a plurality of different pooling operators to produce a final feature vector; and predicts the SoC value with the final feature vector using a linear regressor.PATENT Attorney Docket No.: 2024-022

[0096] The foregoing description, for purpose of explanation, has been described with reference to specific embodiments. However, the illustrative discussions above are not intended to be exhaustive or limiting to the precise forms disclosed. Many modifications and variations are possible in view of the above teachings. Moreover, the order in which the elements of the methods described herein are illustrate and described may be re-arranged, and / or two or more elements may occur simultaneously. The embodiments were chosen and described in order to best explain the principles of the disclosure and its practical applications, to thereby enable others skilled in the art to best utilize the disclosed embodiments and various embodiments with various modifications as are suited to the particular use contemplated.

Claims

PATENT Attorney Docket No.: 2024-022 CLAIMS What is claimed is:

1. A system for monitoring a state of charge (SoC) value of a battery, the system comprising: a first transducer in contact with the battery; a second transducer in contact with the battery; a function generator configured to generate a plurality of signals that are transmitted from the function generator, through the first transducer, and to the battery, wherein the signals include: a plurality of first signals; and a second signal; a data acquisition system configured to measure a plurality of responses that are transmitted from the battery, through the second transducer, and to the data acquisition system, wherein the responses include: a plurality of first responses of the battery in response to the first signals; and a second response of the battery in response to the second signal; and a computing system configured to train a machine learning model based upon the first responses and a plurality of corresponding actual SoC values to produce a trained machine learning model, wherein the trained machine learning model is then configured to predict the SoC value of the battery based upon the second response.

2. The system of claim 1, wherein the first transducer is in contact with a first side of the battery, and wherein the second transducer is in contact with a second, opposing side of the battery.

3. The system of claim 1, further comprising a battery charger / discharger configured to vary the corresponding actual SoC values of the battery.

4. The system of claim 3, wherein the corresponding actual SoC values of the battery are varied simultaneously with generating the first signals to produce the first responses.

5. A method for monitoring a state of charge (SoC) value of a battery, the method comprising: generating a first signal to excite the battery, wherein the first signal is transmitted throughPATENT Attorney Docket No.: 2024-022 a first transducer; measuring a first response of the battery in response to the first signal, wherein the first response is received through a second transducer; generating a plurality of second signals; varying the SoC value of the battery to a plurality of different SoC values, wherein the SoC value of the battery is varied simultaneously with generating the second signals; measuring a plurality of second responses of the battery in response to the second signals; training a machine learning model based upon the second responses and the different SoC values to produce a trained machine learning model; generating a third signal; measuring a third response of the battery in response to the third signal; and predicting the SoC value of the battery based upon the third response, wherein the SoC value of the battery is predicted using the trained machine learning model while the battery is online.

6. The method of claim 5, wherein the first transducer is in contact with a first side of the battery, and wherein the second transducer is in contact with a second, opposing side of the battery.

7. The method of claim 5, wherein the first signal has a frequency that varies within a frequency range between about 1 Hz and about 2 MHz.

8. The method of claim 5, wherein the first response is received at a plurality of different frequencies within the frequency range.

9. The method of claim 5, further comprising: determining peaks or local maxima in the first response; and determining a resonance frequency based upon the peaks or local maxima.

10. The method of claim 9, wherein the second signals are transmitted through the first transducer, and wherein the second signals are generated based upon the resonance frequency.PATENT Attorney Docket No.: 2024-022 11. The method of claim 10, wherein a central frequency of the second signals is the resonance frequency.

12. The method of claim 9, wherein the third signal is transmitted through the first transducer, and wherein the third signal is generated based upon the resonance frequency.

13. The method of claim 12, wherein a central frequency of the third signal is the resonance frequency.

14. The method of claim 5, wherein the battery has a different SoC value when each of the second signals is generated.

15. The method of claim 5, wherein the second responses and the third response are received through the second transducer.

Images