System and method to accoustically detect early thermal runaways in lithium-ion batteries

Abstract

A thermal runaway detection system to detect early-stage lithium-ion battery thermal runaways at the safety valve breakage stage includes one or more microphones configured to transduce audio signals into electrical signals; and a controller configured to receive the electrical signals and, using a deep-learning detection model, determine when an electrical signal corresponding to a respective acoustic signal indicates a safety valve breakage in a lithium-ion battery. The deep-learning detection model was trained by acoustic data obtained from a plurality of incidents in which one or more lithium-ion batteries undergo thermal runaway.

Description

[0001] SYSTEM AND METHOD TO ACCOUSTICALLY DETECT EARLY THERMAL RUNAWAYS IN LITHIUM-ION BATTERIES

[0002] Related Applications

[0003] This application claims the benefit of U.S. Provisional Patent Application Serial No. 63 / 706,136 (filed October 11 , 2024), which is hereby incorporated herein by reference in its entirety.

[0004] Federally-Sponsored Research and Development

[0005] This invention was made with United States Government support from the National Institute of Standards and Technology (NIST), an agency of the United States Department of Commerce. The Government has certain rights in this invention.

[0006] Copyright Notice

[0007] This patent disclosure may contain material that is subject to copyright protection. The copyright owner has no objection to the facsimile reproduction by anyone of the patent document or the patent disclosure as it appears in the U.S. Patent and Trademark Office patent file or records but otherwise reserves any and all copyright rights.

[0008] Field of Invention

[0009] The present invention relates generally to acoustic detection, and more particularly to an acoustic method and system for detecting thermal runaway.

[0010] Background

[0011] The use of lithium-ion batteries (LiBs) has been increasingly prevalent in modem society. Today, LiBs are found in nearly every cell phone, laptop, and portable electronic device. The global demand for new LiBs is expected to increase from approximately 700 gigawatt-hours in 2022 to 4700 gigawatt-hours in 2030. The benefits of LiBs are profound as they facilitate the development of robust technologies, such as e-bikes, e-scooters, drones, electric vehicles, and energy storage systems.

[0012] However, LiBs can experience thermal runaway, a phenomenon in which the lithium-ion cell enters an uncontrollable, self-heating state which can result in life-threatening fire hazards that cannot be ignored. In 2023, in New York City alone (national statistics are not available), the fire department responded to 267 residential LiB fires which caused 150 injuries and 18 deaths. To illustrate this increasing trend, there were only 10 deaths associated with LiB fires for the combined years of 2021 and 2022. In commercial buildings such as battery warehouses and battery recycling plants, major fires were caused by thermal runaway in LiBs. In 2021 , the Superior Battery incident occurred in Morris, Illinois where nearly 4000 residents were forced to evacuate and millions of taxpayer dollars were spent for environmental clean-up.

[0013] Summary of Invention

[0014] Existing fire detection systems and battery management systems cannot detect early-stage thermal runaways. In residential and commercial buildings, the commonly used detection technologies are 1 ) smoke / gas, 2) heat, and 3) video detectors. The detection effectiveness of these technologies depends on the fire size and the relative distance between the fire and the detector. The amount of smoke and hot gases generated from the LiB in a thermal runaway is limited. Unless the detectors are placed directly above the LiB, they will not be triggered to alarm during this early-stage thermal runaway. In a commercial building such as a manufacturing facility or battery warehouse, the smoke / gas sampling devices are typically located in the return air ducts, or the smoke / gas is sampled near equipment / ceiling and then transported to the detectors. So, the detection of thermal runaway, even with the ignition of a LiB, can be significantly delayed because it takes time for the smoke / hot gases to travel, reach the detector, and arrive in the quantities needed to exceed the triggering threshold.

[0015] To overcome this detection bottleneck, sophisticated video smoke and / or heat detectors may be installed. However, there are still two major drawbacks: 1 ) the video detection system needs a clear line of sight of the targeted LiBs and 2) the systems are extremely expensive. Battery management systems (BSMs) are designed to prevent some types of battery faults, but current BSMs only provide oversight of the battery module for e-mobile devices (i.e. , e-bikes which cause most battery fires), and it is not capable of monitoring individual Li B cells, detecting early thermal runaway, and preventing the fires. Therefore, research efforts are needed to develop a robust, low-cost, and easy-to-use system for detecting early thermal runaways in LiBs which can be used to mitigate extreme fires.

[0016] According to an aspect of the invention, a thermal runaway detection system to detect early-stage lithium-ion battery thermal runaways at the safety valve breakage stage includes one or more microphones configured to transduce audio signals into electrical signals; and a controller configured to receive the electrical signals and, using a deep-learning detection model, determine when an electrical signal corresponding to a respective acoustic signal indicates a safety valve breakage in a lithium-ion battery; wherein the deep-learning detection model was trained by acoustic data obtained from a plurality of incidents in which one or more lithium-ion batteries undergo thermal runaway.

[0017] Optionally, the acoustic data used to train the deep-learning detection model includes data from a plurality of different states-of-charge.

[0018] Optionally, the acoustic data used to train the deep-learning detection model includes data from a plurality of battery orientations.

[0019] Optionally, the acoustic data used to train the deep-learning detection model includes data from a plurality of battery shapes.

[0020] Optionally, the acoustic data used to train the deep-learning detection model includes data from a plurality of internal chemistries.

[0021] Optionally, the acoustic data used to train the deep-learning detection model includes data having background noise configured to be representative of background noise found in a variety of indoor environments in which lithium-ion batteries are stored and used.

[0022] Optionally, the device is a mobile device.

[0023] Optionally, the acoustic data used to train the deep-learning detection model includes data that has been time stretched and data that has been pitch shifted. Optionally, the deep learning model includes a one-channel convolutional neural network having four convolutional layers, each layer having a different number of convolutional filters, and wherein a stride of 4 is used in a first convolutional layer and a stride of 1 is used for second, third, and fourth convolutional layers.

[0024] Optionally, the deep learning model includes one-dimensional maximum pooling.

[0025] Optionally, a maximum pooling of 4 is used after the first convolutional layer and a maximum pooling of 2 is used after the second and the third convolutional layers.

[0026] Optionally, the deep learning model includes global maximum pooling.

[0027] Optionally, the system includes an alert module configured to alert a user of a detected thermal runaway event.

[0028] Optionally, the alert module is configured to audibly alert a user of a detected thermal runaway event.

[0029] Optionally, the alert module includes a wireless communication device configured to send an electronic signal to another device and / or to emergency services of a detected thermal runaway event.

[0030] The foregoing and other features of the invention are hereinafter described in greater detail with reference to the accompanying drawings.

[0031] Brief Description of the Drawings

[0032] Fig. 1 shows a schematic of an experimental setup.

[0033] Fig. 2 shows acoustic data of an experiment and detail views of three selected events of the experiment.

[0034] Fig. 3 shows acoustic data of another experiment and detail views of three selected events.

[0035] Fig. 4 shows overall structure of an exemplary detection model.

[0036] Fig. 5 shows a validation loss and accuracy plot.

[0037] Fig. 6 shows a schematic of an exemplary detection device. Detailed Description

[0038] There are three crucial moments of a LiB undergoing a thermal runaway: the moment of i) safety valve breakage, ii) venting, and iii) ignition. The fire growth of a LiB is extremely fast. Once it is ignited, the fire intensity reaches its peak within about a second, generating a flame jet with temperatures of approximately 1100°C. These fire behaviors are dangerous because the thermal abuse from the ignited LiB will cause the adjacent LiBs to transition to thermal runaway. The fire typically spreads rapidly within a minute and becomes extremely difficult to extinguish. Therefore, early detection of thermal runaways, e.g. at the time of safety valve breakage, is vital because it provides extra time for users / occupants to leave or to mitigate fire hazards which helps to reduce potential injuries, fatalities, and property losses.

[0039] Therefore, presented herein is an exemplary system and method to detect early-stage LiB thermal runaways at the safety valve breakage stage. The proposed system can provide instant detections in real-life settings where the existing technologies result in detection delays and there are limited line-of- sights between the LiB and the system. The system includes one or more microphones, an in-house deep-learning detection model, and a controller, e.g. a handheld mobile device. The deep-learning detection model uses acoustic data obtained from experiments in which LiBs undergo thermal runaway. Different state-of-charge conditions and battery orientations may be considered. An accurate and lightweight deep-learning detection model can be implemented on a mobile device to provide real-time actionable information to a user.

[0040] For practical engineering applications, acoustic-based detection has been widely used for detecting faulty machine parts, gas pipeline leakage, and impact damage. In the area of fire research, acoustic-based detection approaches have also been used to identify fire location, the heat release rate of fire, and the increasing trend of temperature within a compartment. There are two advantages to using acoustic-based detections. Firstly, the signals can be obtained at any moment despite challenging conditions with visual obstruction, limited spacing, and poor lighting. Secondly, because sound travels at the speed of sound, near real-time detection can be achieved. However, since previous studies utilize carefully produced acoustic signals (i.e. , the signals contain no background noise nor signals derived from any human activities), they will not work in real-life settings. This is because the detection problem becomes more complex if real-life acoustic conditions, such as acoustic signals from human activities and background noises, are considered. It is also worth noting that when the acoustic signals become more complex, conventional rule-based signal processing techniques are not reliable. The accuracy of a detection model using conventional rule-based techniques degrades significantly as noise levels increase. In order to overcome the technical challenge, the use of a deep learning paradigm is proposed. The primary advantage of using deep learning is that the model can learn the unique acoustic patterns of the safety valve breakage (an indicator of early-stage thermal runaway) and recognize the difference between other impulsive sounds, background noises, and acoustic signals from human activities. Exemplary systems and methods can provide extra time for users / occupants to leave or to mitigate fire hazards, which helps to reduce potential injuries, fatalities, and property losses due to LiB thermal runaway.

[0041] Figure 1 shows the schematic of an experimental setup for recording acoustic data from LiBs which includes an exhaust hood (approximately 2 m x 2 m), a laboratory desk with a heat shield, a battery holder, a lithium-ion battery, a heating element, a glass wall, and a video recorder.

[0042] Different fan speeds will yield different levels of background noise, providing more varied data for the deep-learning system. The laboratory desk may be positioned at the center of the exhaust hood and it may be covered by a stainless steel heat shield to avoid secondary ignition from the battery. The battery holder may be located on top of the heat shield. Additional weights may be placed on the base of the battery holder to avoid any unnecessary movement. Various battery sizes, capacities, and form factors may be used to record a variety of acoustic signals. Additionally, various internal structures and chemistries may be used. In the example discussed herein, and 18650 battery with a LiNiCoAIO2 cathode and graphite anode are used. The nominal capacity and voltage of the example LiBs are 3.2 Ah and 3.7 V, respectively. The LiB may be clamp-fastened by the battery holder. Various states-of-charge may be considered, however, in this example the state-of-charge is ranged from 0 % to 100 % with an increment of 25 %. In addition, five battery orientations, which include 0° (facing up), 45°, 90° (facing to the side to the page), 135°, and 180° (facing down), are accounted for. A total of 38 experiments were conducted in this example and duplicate experiments are carried out for certain battery configurations (i.e. , state-of-charge and orientation). A silicone heating pad powered by nominally DC 12 V and characterized by a 3.8 O resistance was used to facilitate the thermal abuse. It has the dimensions of approximately 63 mm x 30 mm. The thickness is approximately 0.25 mm. The heat pad provides a heating temperature range of approximately 175 °C to 200 °C and it was placed onto the lower half of the battery surface. A glass wall approximately 8 mm thick was located approximately 1.5 m away from the battery. It was used as a protective separation. There were gaps on two sides of the glass wall which allow sound to be recorded by a video recorder. The duration of each test varied with a mean video record duration of about 473 s (about 8 mins) with a standard deviation of about 126 s.

[0043] In general, the battery goes through three different stages during the thermal runaway. Firstly, the internal temperature of the battery cell increases. The pressure inside the battery also increases. The safety valve of the battery cell breaks to relieve pressure and a very limited amount of smoke is generated (denoted as Stage 1 ). It should be noted that about 200 s to 350 s are needed for the battery to reach Stage 1. The gas and smoke disappear in less than 0.5s.

[0044] Secondly, the internal battery temperature continues to increase. The rate of smoke generation also continues to increase. Visually, smoke gradually appears and this is denoted as the venting stage (Stage 2). Based on previous experimental experience, the venting stage will last for about 84 s to 346 s.

[0045] Thirdly, the battery enters the ignition stage (Stage 3) and it typically lasts for about 5 s to 10 s. It should be noted that if the battery has less than about 30% SOC, the battery is less likely to be ignited and this criterion is important for battery storage and handling.

[0046] Acoustic signals of a selected experiment is presented in Fig. 2 for illustration. The safety valve breakage of the Li B happens at around 258 s and the battery ignition takes place at around 504 s. Three important observations are noticed in Fig. 2. Firstly, background noise is relatively negligible. This is because the fan speed from the exhaust fan is low. Secondly, there are three types of sounds with relatively low amplitude due to various human activities and they are a) whispering, b) adjusting the camera, and c) door closing. Thirdly, there are also four types of acoustic signals with relatively large amplitude, and they include i) flipping switch, ii) safety valve breakage (SVB), iii) hammering, and iv) battery ignition. The zoom-in plots of Fig. 2 show that the SVB signals have unique acoustic shapes and characteristics (i.e. , a longer oscillation duration and a larger peak-to-peak amplitude as compared to signals from flipping switch and hammering).

[0047] Fig. 3 shows the acoustic signals for another experiment. The safety valve breakage (SVB) and battery ignition happen at around 261 s and 394 s, respectively. As compared to Fig. 2, there are three major differences. Firstly, the background noise is larger due to an increase in the fan speed from the exhaust hood. Secondly, there is a wide range of human activities which include loud discussion, laughing, walking, preparation for the next experiment, and grinding metal. In addition, Fig. 3 also shows other acoustic signals, such as clipping objects and battery ignition. Finally, and most importantly, the SVB acoustic signals are significantly different from each other. Fig. 3 shows the zoom-in plots for force-closing a window, SVB, and door closing. It can be seen that the acoustic signals of SVB have much lower peak amplitude. Also, the signals begin with maximum oscillations and then decay relatively linearly. In contrast, the SVB signals from Fig. 2 first increase and then decrease exponentially after reaching their peak with a peak amplitude of nearly 0.75. The differences in the signals are likely due to different LiB state-of-charge and its orientation. It is also worth noting that the overall characteristics of acoustic signals for force-closing a window from Fig. 3 are very similar to the SVB shown in Fig. 2. Since the SVB signals can vary significantly due to different state-of- change and its relative orientation and the fact that the signals from other events can be similar to SVB, the development of a robust early-stage thermal runaway detection model that can be used in real-life settings may be challenging.

[0048] Additional acoustic signals are required to teach the deep-learning detection model to distinguish events other than SVB. To facilitate the learning, three publicly available acoustic datasets are considered, and they are SubstiTUTion (TUT), Acoustic Event (AE), and Environmental Sound Classification (ESC). Together, these datasets have approximately 11 ,900 distinct samples covering a wide range of activities and events. In total, they are about 11 ,700 mins in length and more than 13 GB in size. In an exemplary model, manual extraction may be conducted to obtain a selective dataset. Since the detection model is intended to be used in indoor residential-like settings, only acoustic data of related events need be utilized. These events include a) data subsets of Home, Office, and Library from TUT and b) data subsets of pet, speech, non-speech, interior, and exterior noise sounds from AE and ESC. Table 1 shows the summary of the events from TUT, AE, and ESC datasets. Together, 1 ,128 data samples for other events (non-SVB samples) are gathered. These data have a sampling rate of 44,100 Hz with a resolution of 24 bit. The duration of these data ranges from approximately 4 s to more than 60 s and the data size is about 5 GB. It should be noted that a more robust deep learning detection model can be developed when the acoustic data from other events are considered. This is because since the detection model has seen these events, the model will be less likely to misclassify these events as SVB.

[0049] Table 1. A summary of events used in TUT

[0022] , AE

[0023] , and ESC

[0024] datasets.

[0050] Data augmentation may be carried out to address the data imbalance problem. As mentioned above, acoustic data from other events are vital to train a robust model. However, there are 1 ,128 samples for the acoustic data from other events. In contrast, there may be only tens of samples for SVB. Training the model with such a biased dataset will be problematic because 1 ) the model does not have enough SVB sample to learn from and 2) the model is being “forced” to learn the data pattern from other events. To resolve this problem, the easiest solution is to reduce the number of other-event samples. Although this will help, and the model will be able to recognize SVB events, it will not be able to distinguish non-SVB / other events, such as those from human activities. Therefore, more SVB samples are needed.

[0051] Two data augmentation methods may be utilized to generate synthetic SVB samples. The two augmentation methods are described below:

[0052] 1 . Time Stretching (TS): To slow down or speed up the acoustic sample.

[0053] Each sample may be time-stretched by, e.g., five different factors: [e.g., 0.81 , 0.93, 1.00, 1.07, and 1.19],

[0054] 2. Pitch Shifting (PS): To raise or lower the pitch of the acoustic sample. Each sample may be pitch-shifted by, e.g., seven different values (in semitones): [e.g., -3.5, -2.5, -1 , 0, 1 , 2.5, and 3.5],

[0055] It should be noted that the original acoustic signals are preserved if TS and PS are equal to 1 and 0, respectively. By applying five TS factors and seven PS values to an original number of 38 SVB samples, an additional 1 ,292 acoustic signals are obtained. It is also worth mentioning that the augmented data can simulate various SVB events for different types of LiBs with different capacities and state-of-charges. In total, there are 1 ,330 samples for SVB and 1 ,128 samples for other others. The entire dataset is now more balanced.

[0056] Four data preprocessing processes may be carried out to construct the data subsets that can be used for model training. Firstly, a tumbling window may be utilized to extract 10-s samples from the dataset. The use of a tumbling window (i.e. , 0 s to 10 s, 11 s to 20 s, etc.) enables the acoustic data of interest to be located anywhere within the 10-s sample. Within the 10-s samples, the acoustic data of interest might also be chopped off and it is believed that the consideration of these data characteristics facilitates the development of a more robust model that can be used in real-life settings. Secondly, padding may be used to obtain consistent samples in length. Typical background noises from the TUT dataset may be used to pad the samples that have fewer than 10-s of data. Thirdly, z-score normalization may be applied to the entire dataset. Finally, a ratio of 60 / 20 / 20 may be utilized to construct the training, validation, and testing subsets. A convolutional neural network (CNN) may be used as the backbone of the model structure in an exemplary embodiment. The model may use a singlechannel 10-s acoustic sample as the input. The model output may be the corresponding class of the input sample (i.e., non-SVB or SVB).

[0057] Fig. 4 shows a one-channel convolutional neural network for an exemplary detection model. There are four convolutional layers 410, 420, 430, 440 and each layer consists of a different number of convolutional filters. A stride of 4 is used in the first convolutional layer 410 and a stride of 1 is used for the rest of the convolutional layers. Table 2 describes the configuration details for each layer. In general, the use of convolutional filters offers temporal locality to capture the dynamics of the signals. For the early layers, the filter describes basic shape features, such as a peak, a valley, or a certain degree of slope. In the later layers, where the filter has a larger receptive field, it is more likely to contain shape semantics, such as a large magnitude peak or an impulse with multiple peaks.

[0058] Table 2. Configuration details for each layer of the detection model.

[0059] Two regularization techniques may be used to facilitate the training.

[0060] The first technique is one-dimensional maximum pooling and there are three benefits to this technique: 1 ) to retain the most significant features that can be crucial to determining non-SVB and SVB samples, 2) to neglect small distortions such as those due to noise, and 3) to reduce the number of elements in the sample going to the next layer. As shown in Table 2, a maximum pooling (maxpool) of 4 is used after the first convolutional layer and a maximum pooling of 2 is used after the second and the third convolutional layers. The * symbol indicates the dimension of the input size.

[0061] The second technique is global maximum pooling (gmaxpool). In contrast to 1-D maximum pooling, the global maximum pooling takes the maximum value from the feature map across all positions. It means that only the most prominent feature is preserved. The advantage is that the exact location of the important feature within the acoustic sample does not matter and so a more robust CNN can be obtained. The exact locations of maxpool and gmaxpool are shown in Fig. 4.

[0062] As shown in Fig. 4, there are two fully connected (fc) layers and they are used to combine the higher lever features to form a classification. The last layer is SoftMax which provides the predicted probability of the two classes (non-SVB and SVB). Fig. 5 presents the loss and accuracy plot for the validation set. The loss function is taken to be the binary cross-entropy. An Adam optimizer with an initial learning rate of 1e-4 was used to optimize the model. Early-stopping with a patience of 50 was used to avoid overfitting. With this configuration, if the loss from the validation subset does not improve for 50 consecutive epochs, the training stops. As seen in Fig. 5, early-stopping happens in Epoch 42 and the model is saved for testing.

[0063] Shown in Fig. 6 is a schematic representation of an exemplary acoustic sensing system for early detection of thermal runaway at 600. The system 600 includes one or more acoustic detectors 610 and a controller 620. The controller 620 includes an exemplary deep-learning system for identifying a thermal runaway event based on acoustic data gathered by the acoustic detector 610. The system 600 also includes an alert module 630 for alerting a user of a thermal runaway event. Preferably the alert module 630 includes an audible alarm. In some embodiments, the alert module can include a wireless communication device for sending an electronic signal to another device and / or to emergency services. Such an alert module 630 may send additional data such as location of the system 600 and audio and / or video data from the system 600. In some exemplary embodiments, the system 600 is a mobile phone. In other exemplary embodiments, the system 600 is a dedicated special-purpose device. In other exemplary embodiments, the system 600 is a dual-purpose consumer device, such as a wireless router, a television, or a kitchen appliance.

[0064] The processes described herein may be embodied in, and fully automated via, software code modules executed by a computing system that includes one or more general purpose computers or processors. The code modules may be stored in any type of non-transitory computer-readable medium or other computer storage device. Some or all the methods may alternatively be embodied in specialized computer hardware. In addition, the components referred to herein may be implemented in hardware, software, firmware, or a combination thereof.

[0065] Many other variations than those described herein will be apparent from this disclosure. For example, depending on the embodiment, certain acts, events, or functions of any of the algorithms described herein can be performed in a different sequence, can be added, merged, or left out altogether (e.g., not all described acts or events are necessary for the practice of the algorithms). Moreover, in certain embodiments, acts or events can be performed concurrently, e.g., through multi-threaded processing, interrupt processing, or multiple processors or processor cores or on other parallel architectures, rather than sequentially. In addition, different tasks or processes can be performed by different machines and / or computing systems that can function together.

[0066] Any logical blocks, modules, and algorithm elements described or used in connection with the embodiments disclosed herein can be implemented as electronic hardware, computer software, or combinations of both. To clearly illustrate this interchangeability of hardware and software, various illustrative components, blocks, modules, and elements have been described above generally in terms of their functionality. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the overall system. The described functionality can be implemented in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the disclosure.

[0067] The various illustrative logical blocks and modules described or used in connection with the embodiments disclosed herein can be implemented or performed by a machine, such as a processing unit or processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof designed to perform the functions described herein. A processor can be a microprocessor, but in the alternative, the processor can be a controller, microcontroller, or state machine, combinations of the same, or the like. A processor can include electrical circuitry configured to process computerexecutable instructions. In another embodiment, a processor includes an FPGA or other programmable device that performs logic operations without processing computer-executable instructions. A processor can also be implemented as a combination of computing devices, e.g., a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors in conjunction with a DSP core, or any other such configuration. Although described herein primarily with respect to digital technology, a processor may also include primarily analog components. For example, some or all of the signal processing algorithms described herein may be implemented in analog circuitry or mixed analog and digital circuitry. A computing environment can include any type of computer system, including, but not limited to, a computer system based on a microprocessor, a mainframe computer, a digital signal processor, a portable computing device, a device controller, or a computational engine within an appliance, to name a few.

[0068] The elements of a method, process, or algorithm described in connection with the embodiments disclosed herein can be embodied directly in hardware, in a software module stored in one or more memory devices and executed by one or more processors, or in a combination of the two. A software module can reside in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disk, a removable disk, a CD-ROM, or any other form of non-transitory computer-readable storage medium, media, or physical computer storage known in the art. An example storage medium can be coupled to the processor such that the processor can read information from, and write information to, the storage medium. In the alternative, the storage medium can be integral to the processor. The storage medium can be volatile or nonvolatile.

[0069] While one or more embodiments have been shown and described, modifications and substitutions may be made thereto without departing from the spirit and scope of the invention. Accordingly, it is to be understood that the present invention has been described by way of illustrations and not limitation. Embodiments herein can be used independently or can be combined.

[0070] All ranges disclosed herein are inclusive of the endpoints, and the endpoints are independently combinable with each other. The ranges are continuous and thus contain every value and subset thereof in the range. Unless otherwise stated or contextually inapplicable, all percentages, when expressing a quantity, are weight percentages. The suffix (s) as used herein is intended to include both the singular and the plural of the term that it modifies, thereby including at least one of that term (e.g., the colorant(s) includes at least one colorants). Option, optional, or optionally means that the subsequently described event or circumstance can or cannot occur, and that the description includes instances where the event occurs and instances where it does not. As used herein, combination is inclusive of blends, mixtures, alloys, reaction products, collection of elements, and the like.

[0071] As used herein, a combination thereof refers to a combination comprising at least one of the named constituents, components, compounds, or elements, optionally together with one or more of the same class of constituents, components, compounds, or elements.

[0072] All references are incorporated herein by reference.

[0073] The use of the terms “a,” “an,” and “the” and similar referents in the context of describing the invention (especially in the context of the following claims) are to be construed to cover both the singular and the plural, unless otherwise indicated herein or clearly contradicted by context. It can further be noted that the terms first, second, primary, secondary, and the like herein do not denote any order, quantity, or importance, but rather are used to distinguish one element from another. It will also be understood that, although the terms first, second, etc. are, in some instances, used herein to describe various elements, these elements should not be limited by these terms. For example, a first current could be termed a second current, and, similarly, a second current could be termed a first current, without departing from the scope of the various described embodiments. The first current and the second current are both currents, but they are not the same condition unless explicitly stated as such.

[0074] The modifier about used in connection with a quantity is inclusive of the stated value and has the meaning dictated by the context (e.g., it includes the degree of error associated with measurement of the particular quantity). The conjunction or is used to link objects of a list or alternatives and is not disjunctive; rather the elements can be used separately or can be combined together under appropriate circumstances.

[0075] Although the invention has been shown and described with respect to a certain embodiment or embodiments, it is obvious that equivalent alterations and modifications will occur to others skilled in the art upon the reading and understanding of this specification and the annexed drawings. In particular regard to the various functions performed by the above described elements (components, assemblies, devices, compositions, etc.), the terms (including a reference to a "means") used to describe such elements are intended to correspond, unless otherwise indicated, to any element which performs the specified function of the described element (i.e., that is functionally equivalent), even though not structurally equivalent to the disclosed structure which performs the function in the herein illustrated exemplary embodiment or embodiments of the invention. In addition, while a particular feature of the invention may have been described above with respect to only one or more of several illustrated embodiments, such feature may be combined with one or more other features of the other embodiments, as may be desired and advantageous for any given or particular application.

Claims

ClaimsWhat is claimed is:1 . A thermal runaway detection system to detect early-stage lithium- ion battery thermal runaways at the safety valve breakage stage, the system comprising: one or more microphones configured to transduce audio signals into electrical signals; and a controller configured to receive the electrical signals and, using a deeplearning detection model, determine when an electrical signal corresponding to a respective acoustic signal indicates a safety valve breakage in a lithium-ion battery; wherein the deep-learning detection model is trained by acoustic data obtained from a plurality of incidents in which one or more lithium-ion batteries undergo thermal runaway.

2. The thermal runaway detection system of claim 1 , wherein the acoustic data used to train the deep-learning detection model includes data from a plurality of different states-of-charge.

3. The thermal runaway detection system of any one of claims 1 -2, wherein the acoustic data used to train the deep-learning detection model includes data from a plurality of battery orientations.

4. The thermal runaway detection system of any one of claims 1 -3, wherein the acoustic data used to train the deep-learning detection model includes data from a plurality of battery shapes.

5. The thermal runaway detection system of any one of claims 1 -4, wherein the acoustic data used to train the deep-learning detection model includes data from a plurality of internal chemistries.

6. The thermal runaway detection system of any one of claims 1 -5, wherein the acoustic data used to train the deep-learning detection model includes data having background noise configured to be representative of background noise found in a variety of indoor environments in which lithium-ion batteries are stored and used.

7. The thermal runaway detection system of any one of claims 1 -6, wherein the device is a mobile device.

8. The thermal runaway detection system of any one of claims 1 -7, wherein the acoustic data used to train the deep-learning detection model includes data that has been time stretched and data that has been pitch shifted.

9. The thermal runaway detection system of any one of claims 1 -8, wherein the deep learning model includes a one-channel convolutional neural network having four convolutional layers, each layer having a different number of convolutional filters, and wherein a stride of 4 is used in a first convolutional layer and a stride of 1 is used for second, third, and fourth convolutional layers.

10. The thermal runaway detection system of claim 9, wherein the deep learning model includes one-dimensional maximum pooling.11 . The thermal runaway detection system of claim 10, wherein a maximum pooling of 4 is used after the first convolutional layer and a maximum pooling of 2 is used after the second and the third convolutional layers.

12. The thermal runaway detection system of claim 9, wherein the deep learning model includes global maximum pooling.

13. The thermal runaway detection system of any one of claims 1 -12, further comprising an alert module configured to alert a user of a detected thermal runaway event.

14. The thermal runaway detection system of any one of claims 1 -13, wherein the alert module is configured to audibly alert a user of a detected thermal runaway event.

15. The thermal runaway detection system of any one of claims 1 -14, wherein the alert module includes a wireless communication device configured to send an electronic signal to another device and / or to emergency services of a detected thermal runaway event.

Images