A battery thermal runaway identification method based on pressure mutation

Abstract

The application relates to the technical field of battery thermal runaway identification, and specifically discloses a battery thermal runaway identification method based on pressure mutation, which comprises the following steps: S1, collecting pressure vibration signals and judging whether an abnormal event occurs in a battery pack based on the pressure vibration signals; S2, when the abnormal event occurs, synchronously collecting original waveform data of all pressure sensors in an event time window to form a signal data set; S3, conducting mode separation on the complete signal data set to obtain a first conduction mode signal subset and a second conduction mode signal subset; S4, calculating a first spatial coordinate solution set and a second spatial coordinate solution set based on the first conduction mode signal subset and the second conduction mode signal subset respectively; and S5, cross-verification of the first spatial coordinate solution set and the second spatial coordinate solution set to determine a thermal runaway physical source point. The application improves the safety and controllability of the whole operation process of a battery system.

Description

Technical Field

[0001] This invention relates to the field of thermal runaway identification technology, and more specifically to a battery thermal runaway identification method based on pressure mutation. Background Technology

[0002] During the operation of a power battery system, thermal runaway is usually accompanied by complex and violent physical changes. In its early stages, it often manifests as sudden changes in local structural stress and rapid disturbances in the internal pressure environment of the battery pack. Due to the highly integrated, spatially enclosed, and complex propagation paths of the internal structure of the battery pack, the pressure change signals caused by thermal runaway exhibit significantly different propagation characteristics in different media and structures, resulting in a high degree of coupling between related signals in time, space, and morphology.

[0003] In existing technologies, monitoring of battery thermal runaway largely relies on slowly changing information such as temperature, voltage, or gas concentration, making it difficult to reflect the physical initiation location of sudden thermal runaway events in a timely and accurate manner. Methods based on single pressure signals or simple threshold judgments are also insufficient for reliable identification and location of thermal runaway events under multi-source interference and complex propagation environments. Especially when solid structures and non-solid media coexist within the battery pack, pressure surge signals often propagate simultaneously along different paths, resulting in mixed signal arrival characteristics and unclear source points, thus leading to significant uncertainty in identifying the physical source of thermal runaway. Therefore, how to reliably identify battery thermal runaway events and accurately determine their physical location under complex propagation conditions without damaging the battery pack structure, utilizing the transient information carried by internal pressure surges, has become a critical technical problem urgently needing to be solved in the field of power battery safety monitoring. Summary of the Invention

[0004] The purpose of this invention is to provide a battery thermal runaway identification method based on pressure mutation, thereby solving the above-mentioned technical problems.

[0005] The objective of this invention can be achieved through the following technical solutions:

[0006] A method for identifying battery thermal runaway based on pressure mutation includes the following steps:

[0007] S1. Arrange multiple pressure sensors inside the battery pack to collect pressure vibration signals, and determine whether an abnormal event has occurred inside the battery pack based on the pressure vibration signals.

[0008] S2. When an abnormal event occurs, the raw waveform data of all pressure sensors within the event time window are collected synchronously to form a signal data set;

[0009] S3. Perform conduction mode separation on the complete signal data set to obtain a first conduction mode signal subset and a second conduction mode signal subset;

[0010] S4. Calculate the first spatial coordinate solution set and the second spatial coordinate solution set based on the first conduction mode signal subset and the second conduction mode signal subset, respectively;

[0011] S5. Cross-validate the solution sets of the first and second spatial coordinates to determine the physical source of thermal runaway.

[0012] As a further aspect of the present invention: the acquired signal data set includes:

[0013] The instantaneous energy of the signal output by each pressure sensor is continuously calculated. The instantaneous energy is the integral of the square of the signal amplitude over a short time window.

[0014] When the instantaneous energy continuously exceeds the preset background energy threshold for a preset duration, the moment when the preset duration is reached is extended forward by a predetermined duration to obtain the event start time.

[0015] The starting point of the event time window is set to the event start time, and the length of the event time window is based on a manual preset.

[0016] Within the event time window, raw waveform data from all pressure sensors are simultaneously acquired and stored to form a signal data set.

[0017] As a further aspect of the present invention: conduction mode separation includes:

[0018] Extract the signals of each pressure sensor in multiple dimensions from the signal data set, and combine the signals of all sensors to form a spatiotemporal signal matrix;

[0019] The spatiotemporal signal matrix is ​​processed using a mode decomposition method to obtain multiple global propagation mode components. The mode decomposition method includes multi-scale step mode decomposition.

[0020] For each global conduction mode component, its signal propagation characteristics among different pressure sensors are calculated;

[0021] The signal propagation characteristics are matched with preset solid-state structure conduction characteristics and non-solid medium conduction characteristics;

[0022] The global conduction mode components that match the conduction characteristics of solid structures and conduction characteristics of non-solid media are respectively merged into a first conduction mode signal subset and a second conduction mode signal subset.

[0023] As a further aspect of the present invention: calculating the first spatial coordinate solution set includes:

[0024] Based on the first transmission mode signal subset, the first peak moment of the signal arriving at each pressure sensor is extracted;

[0025] Based on the first peak time of all pressure sensors, calculate the difference of peak times between all non-repeating pressure sensor pairs, and sort the differences of peak times to obtain a first sequence.

[0026] Each node in the solid-state structure network topology model of the battery pack, except for the sensor nodes, is assumed to be the source point; wherein, the sensor nodes in the network topology model correspond one-to-one with the installation positions of the pressure sensors, and the edges represent continuous solid force transmission paths and are associated with theoretical wave velocity values.

[0027] For each of the hypothetical source points, based on the network topology model and the theoretical wave velocity values ​​of each edge, the theoretical propagation time of the vibration wave from the hypothetical source point to each sensor node is calculated.

[0028] Calculate the difference in theoretical propagation time between all non-repeating pressure sensor pairs, and sort the differences in theoretical propagation time to obtain the second sequence;

[0029] Calculate the root mean square error between the second sequence and the first sequence corresponding to each hypothetical source point, and select hypothetical source points whose root mean square error is less than a preset error tolerance. The three-dimensional spatial coordinates of all selected hypothetical source points constitute the first spatial coordinate solution set.

[0030] As a further aspect of the present invention: calculating the second conduction mode signal subset includes:

[0031] Based on the second transmission mode signal subset, the first peak moment of the signal arriving at each pressure sensor is extracted;

[0032] Based on the first peak time of all pressure sensors, the difference between the peak times of all non-repeating pressure sensor pairs is calculated, and the differences in peak times are sorted to obtain a third sequence.

[0033] Set the average propagation velocity v of the pressure vibration in the non-solid medium inside the battery pack;

[0034] For any two distinct pressure sensors i and j, with coordinates (xi, yi, zi) and (xj, yj, zj) respectively, the following equations are established:

[0035] ;

[0036] Where (x, y, z) represents the coordinates of an event source point to be solved, and ΔTij represents the difference between the peak times of pressure sensors i and j.

[0037] All the equations form a system of equations. A numerical optimization algorithm is used to solve the system of equations to find the coordinates of one or more event source points, such that the overall residual generated after substituting the coordinates of the event source points into all the equations is minimized.

[0038] The coordinates of all event source points constitute the second spatial coordinate solution set.

[0039] As a further aspect of the present invention: determining the physical source of thermal runaway includes:

[0040] Calculate the Euclidean distance between all pairs of coordinate points in the first spatial coordinate solution set and the second spatial coordinate solution set. If the Euclidean distance is less than the preset spatial distance threshold, mark the corresponding pair of coordinate points as the target pair.

[0041] If the proportion of the number of target pairs to the total number of coordinate point pairs exceeds a preset threshold, the geometric center point of the two coordinate points in the target pair is calculated, and the geometric center point with the highest frequency of occurrence is selected as the physical source point of thermal runaway.

[0042] As a further aspect of the present invention: calculating the first spatial coordinate solution set further includes:

[0043] After each determination of the physical source of thermal runaway, the coordinates of the physical source of thermal runaway corresponding to that abnormal event and the first sequence are stored;

[0044] When the number of stored abnormal events reaches a preset number, the coordinates of the thermal runaway physical source point of a single stored abnormal event are used as the vibration starting point. The corresponding second sequence is obtained by combining the network topology model and recorded as the verification sequence.

[0045] The error between the verification sequence and the corresponding first sequence is obtained and denoted as the verification error. The optimization problem is to minimize the total verification error. The gradient descent method is used to adjust the theoretical wave velocity values ​​of each edge in the network topology model.

[0046] Replace the original theoretical wave velocity values ​​of each edge in the network topology model with the adjusted theoretical wave velocity values.

[0047] The beneficial effects of this invention compared to the prior art are as follows:

[0048] This invention enables effective identification of battery thermal runaway events under conditions of sudden pressure changes within the battery pack, and reliably determines the physical location of thermal runaway in a complex propagation environment with both structural and media conditions. Through comprehensive temporal and spatial analysis of the pressure change signal, this invention avoids the uncertainties associated with single signal characteristics or single judgment criteria, making the identification process of thermal runaway events more reliable and reducing the impact of external interference or non-thermal runaway events on the judgment results.

[0049] This invention, without relying on battery pack disassembly or destructive testing, can cross-verify the location information reflected by different propagation paths within the battery pack, thereby improving the consistency and reliability of thermal runaway source identification results. Simultaneously, this invention can adapt to actual operating conditions with complex internal battery pack structures and variable propagation conditions, maintaining the stability of identification results through multiple abnormal events, which is beneficial for continuous monitoring and judgment during long-term operation. Furthermore, while identifying thermal runaway events, this invention can provide a clear physical location information basis for subsequent safety control or protection measures, thereby improving the overall safety and controllability of the battery system during operation, and has significant engineering application value. Attached Figure Description

[0050] The invention will now be further described with reference to the accompanying drawings.

[0051] Figure 1 This is a flowchart illustrating a battery thermal runaway identification method based on pressure mutation according to the present invention. Detailed Implementation

[0052] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0053] Please see Figure 1 As shown, this invention is a battery thermal runaway identification method based on pressure mutation, comprising the following steps:

[0054] S1. Arrange multiple pressure sensors inside the battery pack to collect pressure vibration signals, and determine whether an abnormal event has occurred inside the battery pack based on the pressure vibration signals.

[0055] It should be noted that the pressure sensor is not limited to measuring static or quasi-static pressure changes of free gas inside the battery pack, but can respond to transient pressure changes caused by structural vibration, stress wave propagation, and pressure disturbances in non-solid media.

[0056] The pressure sensor can be installed by direct or indirect coupling with the battery pack structural components, enabling it to sense local pressure changes induced at the structure-medium coupling interface by mechanical vibrations propagating in the solid structure; at the same time, the pressure sensor can also sense pressure disturbance signals propagating in the non-solid medium inside the battery pack.

[0057] Therefore, the pressure vibration signal collected by the pressure sensor can be regarded as a comprehensive response signal containing multiple propagation mechanisms superimposed. Subsequent steps distinguish and utilize the signal components corresponding to different propagation mechanisms through signal processing, without relying on the physical distinction of propagation mechanisms by the sensor itself.

[0058] In actual battery pack structures, solid-state structural vibration, stress wave propagation, and pressure disturbances in non-solid media can be mutually converted and superimposed in a local space through structure-medium coupling. The pressure sensor responds to the transient pressure changes caused by the above coupling effect, thereby equivalently reflecting the dynamic process related to vibration propagation without directly measuring structural displacement or strain.

[0059] In one optional embodiment, the pressure sensor can be a multi-axis pressure sensor or a pressure sensor with multi-directional response capability, used to acquire pressure vibration response information from different directions at the same sensor location. The multi-directional pressure response acquired by the multi-axis pressure sensor can enhance the characterization capability of the spatial features of pressure vibration, resulting in higher distinguishability of the spatial consistency characteristics presented by different conduction modes during signal propagation feature analysis. It should be noted that whether or not a multi-axis pressure sensor is used does not affect the basic principle of thermal runaway identification based on pressure mutation in this invention; a single-axis pressure sensor can also achieve equivalent spatial feature extraction through multi-sensor spatial deployment and signal processing methods.

[0060] S2. When an abnormal event occurs, the raw waveform data of all pressure sensors within the event time window are collected synchronously to form a signal data set;

[0061] In a preferred embodiment of the present invention, the acquired signal data set includes:

[0062] The raw pressure vibration signal output by each pressure sensor is processed for a short time. The signal amplitude change is integrated within a preset short time window to obtain instantaneous energy information that can characterize the change in pressure vibration intensity at that moment.

[0063] As time progresses, the instantaneous energy is continuously updated and compared with the corresponding background energy level. When the instantaneous energy is continuously higher than the background energy threshold for a period of time and this continuous state reaches a preset time condition, it is considered that an abnormal event has occurred inside the battery pack.

[0064] To ensure the integrity of early propagation information of abnormal events, after determining the moment when the persistence condition is met, this moment is extended forward by a predetermined duration on the timeline as the starting moment of the event time window. This starting moment is earlier than the anomaly detection trigger moment, thus covering the pressure vibration signals generated in the initial stage of the abnormal event. Subsequently, using this event starting moment as the starting point and combining it with the manually set time window length, the complete event time window range is determined. Within this event time window, all pressure sensors are synchronously acquired, and the raw waveform data output by each pressure sensor within this time range is completely stored, forming a signal data set containing pressure vibration information from multiple sensors and multiple time points. This provides the original data foundation for further analysis of the abnormal event.

[0065] S3. Perform conduction mode separation on the complete signal data set to obtain a first conduction mode signal subset and a second conduction mode signal subset;

[0066] In another preferred embodiment of the present invention, conduction mode separation includes:

[0067] After forming a set of signal data within the event time window, the signal data set is subjected to conduction mode separation processing. The signal data set includes raw pressure vibration waveform data synchronously collected by multiple pressure sensors within the same event time window. This pressure vibration waveform data reflects the comprehensive response formed by the combined action of multiple propagation mechanisms during the occurrence of the abnormal event.

[0068] Based on the aforementioned signal data set, multi-dimensional signal information (such as time-domain characteristic signals to characterize the trend of signal strength changes, frequency band signals or time-frequency representation signals to reflect the distribution of different frequency components, and characteristic signals to characterize the arrival time sequence, phase change relationship, or envelope change characteristics of the signal) is extracted from the original waveform data corresponding to each pressure sensor. This multi-dimensional signal information can simultaneously characterize the changes in signal characteristics during time evolution and the changes in propagation characteristics reflected after frequency band processing or time-frequency processing. The signal information of all pressure sensors in the aforementioned multi-dimensional dimensions is uniformly organized and aligned according to the sensor spatial layout relationship and time sampling order, and then combined to form a spatiotemporal signal matrix that can simultaneously reflect the sensor spatial position relationship and the signal time evolution relationship. The spatiotemporal signal matrix serves as the input data for subsequent mode decomposition processing.

[0069] Based on the spatiotemporal signal matrix, a mode decomposition method is used to process it, so that the spatiotemporal signal matrix is ​​decomposed into multiple global transmission mode components with a consistent expression form across all pressure sensors. The mode decomposition method includes multi-scale step mode decomposition, so as to extract the propagation components related to the abnormal event at different time scales, and so that signal components with different propagation characteristics can be reflected in the global transmission mode components.

[0070] For each obtained global conduction mode component, based on the signal response relationship of the global conduction mode component at different pressure sensors, the signal propagation characteristics of the global conduction mode component among the pressure sensors are calculated. These characteristics characterize the propagation relationship of the mode component within the sensor array. The signal propagation characteristics include at least a phase difference sequence formed by the phase difference changes between the signals from different pressure sensors, and the spatial distribution of the principal vibration direction, reflecting the spatial consistency and dominant propagation directionality of the mode component. The spatial distribution of the principal vibration direction describes the consistent propagation trend of the global conduction mode component in the sensor array space. It does not represent a strictly physical vector direction, but rather an equivalent spatial directivity description based on the comprehensive characteristics of the temporal consistency, amplitude variation, and phase relationship of the signals from different pressure sensors.

[0071] It should be noted that signal propagation characteristics are used to characterize the differences in different propagation properties and do not require them to correspond to completely independent physical propagation channels. In actual battery pack structures, there may be a coupling relationship between solid structure vibration and pressure disturbance in non-solid media. Propagation characteristics are used to reflect signal components that exhibit differences in propagation speed, arrival time, and spatial consistency.

[0072] Subsequently, the signal propagation characteristics corresponding to the global propagation mode components are matched with preset solid-state structure propagation characteristics and non-solid medium propagation characteristics. These preset characteristics describe the typical features exhibited by different propagation mechanisms in terms of propagation delay distribution, spatial consistency, and phase difference variation. The matching process determines the most suitable propagation type for each global propagation mode component. Based on the matching results, global propagation mode components determined to conform to solid-state structure propagation characteristics are merged to form a first propagation mode signal subset, while those determined to conform to non-solid medium propagation characteristics are merged to form a second propagation mode signal subset. This allows for the effective differentiation of pressure vibration signal components with different propagation characteristics without requiring strict physical isolation between different physical propagation mechanisms, providing a basic signal input for subsequent source location calculations based on different propagation characteristics.

[0073] S4. Calculate the first spatial coordinate solution set and the second spatial coordinate solution set based on the first conduction mode signal subset and the second conduction mode signal subset, respectively;

[0074] In a preferred embodiment of the present invention, calculating the first spatial coordinate solution set includes:

[0075] Based on the previously obtained first transmission mode signal subset, time series analysis is performed on the signals corresponding to each pressure sensor in the signal subset. Time feature points that can characterize the first significant arrival of the signal are extracted from the signals, and these time feature points are recorded as the first peak moment when the signal arrives at the pressure sensor.

[0076] Understandably, the first peak moment is not limited to the peak value with the largest amplitude in the signal, but is used to characterize the time feature point when the signal first arrives significantly.

[0077] In practical implementation, the timing of the first peak can be determined using any of the following methods or a combination thereof, based on the actual signal characteristics:

[0078] (1) The moment when the signal envelope first exceeds the preset amplitude threshold;

[0079] (2) The moment when the signal energy or short-term energy undergoes its first sudden change;

[0080] (3) The first local extremum point in the signal waveform corresponding to the first arrival propagation;

[0081] (4) After bandpass filtering or time-frequency analysis, it is used to characterize the characteristic moment of the first arrival propagation.

[0082] After obtaining the first peak time corresponding to all pressure sensors, the pressure sensors are paired up, and the time difference between the first peak times of any two different pressure sensors is calculated. After removing duplicate combinations, a complete set of time difference values ​​is formed. The set of time difference values ​​is sorted according to their numerical values ​​to obtain the first sequence used to characterize the actual propagation arrival time relationship.

[0083] Subsequently, a pre-established solid-state structure network topology model of the battery pack is introduced. This network topology model includes at least a set of nodes, a set of edges, and propagation attribute information associated with the edges. The node set represents key spatial locations within the battery pack related to vibration propagation. Nodes include sensor nodes corresponding to the installation locations of each pressure sensor, as well as structural nodes characterizing the internal structural connections of the battery pack. Structural nodes can correspond to connection points of structural components, intersections of components, or characteristic locations in continuous structural paths within the battery pack, describing the spatial locations that vibration propagation may traverse. The edge set describes the continuous force transmission relationships formed between nodes in the solid structure. Each edge corresponds to an actual solid structural force transmission path within the battery pack, characterizing the propagation of vibration waves along this path between adjacent nodes. The connection relationships of the edges reflect the topological connectivity characteristics of the internal structure of the battery pack. The propagation attribute information associated with each edge describes the vibration propagation characteristics along the solid force transmission path, including at least a theoretical wave velocity value characterizing the speed of vibration propagation. This theoretical wave velocity value serves as an equivalent parameter to describe the average characteristics of vibration wave propagation along the corresponding solid path.

[0084] In one alternative implementation, the propagation speed of pressure vibration in the non-solid medium is an equivalent propagation speed parameter used to characterize the average characteristics of pressure disturbance propagation within the event time window in the non-solid medium environment inside the battery pack.

[0085] The equivalent propagation velocity can be estimated based on historical experimental data, environmental parameters, or through numerical optimization along with the coordinates of the event source point; its value is not limited to a strictly constant physical constant, but is allowed to reflect the influence of changes in the medium state on the propagation characteristics within a reasonable range.

[0086] By introducing an equivalent propagation velocity parameter, a robust estimation of the location of the event source point can be achieved without relying on accurate modeling of the instantaneous state of the medium.

[0087] The nodes in the network topology model are traversed. Except for the sensor nodes that correspond one-to-one with the installation positions of the pressure sensors, the remaining structural nodes are successively used as hypothetical source points for analysis. For each hypothetical source point, based on the connection relationship between the nodes in the network topology model and the theoretical wave velocity value associated with each edge, the theoretical propagation time required for the vibration wave to propagate from the hypothetical source point along the solid structure path to each sensor node is calculated, and the theoretical arrival time sequence corresponding to each sensor is obtained accordingly.

[0088] After obtaining the theoretical propagation time, the sensor nodes are paired up in pairs, and the difference in theoretical propagation time between any two different sensor nodes is calculated. The theoretical propagation time differences are then sorted to form a second sequence that structurally corresponds to the first sequence.

[0089] Based on the degree of difference between the first and second sequences, the root mean square error between the corresponding second and first sequences is calculated for each hypothetical source point. This root mean square error is used as an evaluation index to measure the consistency between the hypothetical source point and the actual observation time sequence under the solid structure propagation model. Hypothetical source points with root mean square errors less than the preset error tolerance are selected. Finally, the three-dimensional spatial coordinates corresponding to all hypothetical source points that meet the conditions are gathered to form the first spatial coordinate solution set, thereby obtaining a set of candidate source point positions that match the actual pressure vibration propagation characteristics under the solid structure propagation constraint.

[0090] From a physics perspective, the first sequence reflects the "overall distribution characteristics of arrival time differences" formed after an abnormal event propagates in a real structure, while the second sequence reflects the "overall distribution characteristics of arrival time differences" that can be formed under the assumption of a source point and a given structural model. If the source point location is assumed to be correct, even if there are uncertainties in the local path, these two distributions should still be highly similar in overall ordering. Conversely, if the source point location is incorrect, even if the time differences of some sensor pairs are accidentally close, the overall ordering distribution will be difficult to match. Therefore, comparing the ordered first and second sequences essentially uses the overall consistency of the propagation time distribution to constrain the source point location, which is a more robust criterion design under complex structural conditions.

[0091] In another preferred embodiment of the present invention, calculating the second conduction mode signal subset includes:

[0092] Based on the second conduction mode signal subset, the time sequence analysis of the signal waveforms corresponding to each pressure sensor is performed. The time feature point that can characterize the first significant arrival of the mode component at the sensor is extracted from each signal. This time feature point is recorded as the first peak moment of the signal arriving at the pressure sensor. The first peak moment is used to equivalently describe the arrival time sequence information under the propagation condition of non-solid medium, and is not limited to the peak with the largest amplitude (refer to the previous text).

[0093] After obtaining the first peak time of all pressure sensors, the pressure sensors are paired up and duplicate pairs are removed. The time difference of the first peak time of each pair of pressure sensors is calculated to form a time difference set. The time difference set is then sorted by numerical value to obtain the third sequence. The third sequence is used as the observation time sequence input for subsequent solving of event source points.

[0094] By combining the three-dimensional coordinate information of each pressure sensor in the battery pack, the spatial geometric relationship between any two sensors is associated with the difference in their corresponding peak times. Based on the average propagation velocity parameter of the non-solid medium, a set of equations is established to constrain the coordinates of the event source point to be determined, where the average propagation velocity parameter is the equivalent propagation velocity (see the previous explanation).

[0095] The equations formed by all non-repeating sensor pairs constitute the constraints to be solved. The equations are solved by numerical optimization algorithm. The consistency between the observation time difference of each sensor pair and the geometric constraints is used as the objective to minimize the overall residual generated after substituting the coordinates of the event source points into the equations. This results in the output of one or more event source point coordinates that meet the condition of minimum residual. The three-dimensional coordinates of all event source points are then collected to form a second spatial coordinate solution set, which is used to characterize the candidate source point positions obtained by inverting the signal time series information under the propagation conditions of the second conduction mode.

[0096] The above formula is based on the principle of time difference positioning. Its core idea is to use the time difference of arrival of the same abnormal event observed at different spatial locations to deduce the spatial location of the event. When a pressure surge related to thermal runaway occurs inside the battery pack, the pressure disturbance will propagate outwards in the non-solid medium at a certain propagation speed. Pressure sensors at different locations will receive this disturbance signal at different times due to their different spatial distances from the event source. The time difference between the first peak observed between any two pressure sensors essentially reflects the time difference relationship corresponding to the difference in path length from the event source to these two sensors under the constraint of propagation speed. By treating the coordinates of the event source as unknowns and establishing a geometric distance relationship between them and the known spatial coordinates of each pressure sensor, a corresponding relationship can be established between the spatial distance difference and the observed time difference, thus forming an equation to constrain the location of the event source. The combined effect of the time difference information provided by multiple different sensors ensures that the coordinates of the event source must simultaneously satisfy multiple sets of geometric and temporal constraints. This process is physically equivalent to finding a spatial location where the difference in propagation distance to each sensor is consistent with the actual observed time difference. Due to factors such as noise, multipath propagation, and non-uniform propagation speed in the real environment, a single sensor pair often cannot uniquely determine the source location. Therefore, by constructing a set of equations composed of multiple sensor pairs and using numerical optimization to find the solution with the minimum overall residual, the event source location that best matches the observation data can be obtained in a statistical sense, thereby realizing the spatial inversion of the physical source of thermal runaway.

[0097] It should be noted that the first spatial coordinate solution set and the second spatial coordinate solution set are calculated using different methods because the physical propagation mechanisms followed by the two types of pressure vibration signals inside the battery pack are fundamentally different. Therefore, the constraints and modeling assumptions relied upon when performing source point inversion are also significantly different.

[0098] The signal corresponding to the first spatial coordinate solution set mainly manifests as vibration components propagating along the solid structure of the battery pack. This type of vibration is significantly affected by the internal structural connection relationship, force transmission path, and structural continuity of the battery pack. Its propagation process is more in line with the characteristic of transmission segment by segment along a given structural path. The propagation time is not only related to the spatial distance but also to the specific structural path traversed. Therefore, it is necessary to use a solid structure network topology model to explicitly introduce the internal structural connection relationship of the battery pack. By assuming source points on the topology nodes and calculating the theoretical propagation time sequence along the structural path, and matching it with the actual observed arrival time sequence, reasonable candidate source point positions under structural propagation constraints can be selected.

[0099] The signal corresponding to the second spatial coordinate solution set mainly reflects the pressure disturbance component propagating in the non-solid medium. This type of propagation is physically closer to the process of diffusion in a continuous medium. Its propagation characteristics are mainly determined by spatial geometric relationships and equivalent propagation speed, and its correlation with specific structural paths is relatively weak. Therefore, geometric constraints can be established directly based on the spatial coordinate relationships and arrival time differences between sensors, and the location of the event source point can be inverted through time difference positioning. Although the two calculation methods are different in form, they have inherent consistency in physical logic. Both are based on the time sequence differences formed by pressure mutations starting from the same physical source point and arriving at each sensor under different propagation conditions. They simply perform location inversion for the same thermal runaway event from the perspectives of structural constraints and spatial geometric constraints, respectively. By adopting calculation methods that are more consistent with actual physical characteristics under different propagation mechanisms, the credibility of each solution set is improved, and a complementary information basis is provided for subsequent cross-validation.

[0100] S5. Cross-validate the solution sets of the first and second spatial coordinates to determine the physical source of thermal runaway.

[0101] In a preferred embodiment of the present invention, determining the physical source of thermal runaway includes:

[0102] After obtaining the first spatial coordinate solution set and the second spatial coordinate solution set respectively, the candidate coordinates in the two spatial coordinate solution sets are used as input data for subsequent consistency determination. Each coordinate point in the first spatial coordinate solution set and each coordinate point in the second spatial coordinate solution set are combined in pairs. For each combination, the spatial distance between the corresponding coordinate points is calculated to characterize the degree of proximity of the candidate source points obtained under different propagation characteristic constraints in spatial location.

[0103] When the spatial distance between a set of coordinate points is less than a pre-set spatial distance threshold, the set of coordinate points is considered to be spatially consistent. This set of coordinate points is then marked as a target pair and recorded. This process is used to filter out a subset of coordinate point pairs that are spatially close to each other from all coordinate point pairs. Based on the quantitative relationship between the target pairs and all coordinate point pairs, the proportion of target pairs in the overall set of coordinate point pairs is assessed. When the proportion of target pairs to all coordinate point pairs meets a preset condition, the first spatial coordinate solution set and the second spatial coordinate solution set are considered to have sufficient consistency in spatial positioning results. If the preset condition is not met, a fault is directly reported.

[0104] Based on this, geometric processing is performed on the two corresponding coordinate points of each target pair to calculate the geometric center position of these two coordinate points in three-dimensional space. This geometric center point is used to characterize the spatial position supported by both types of propagation characteristic constraints. Statistical analysis is performed on the geometric center points corresponding to all target pairs. Based on the frequency of occurrence of the geometric center points in space, the geometric center point with the highest frequency in the set is selected as the final consistent result obtained after cross-validation of the first and second spatial coordinate solution sets. This geometric center point is then identified as the physical source point of thermal runaway. Thus, based on the comprehensive inversion results of different propagation characteristics, the actual location of the thermal runaway event is determined and reported.

[0105] In each preferred embodiment of the present invention, calculating the first spatial coordinate solution set further includes:

[0106] After determining the physical source of thermal runaway in an abnormal event, the three-dimensional spatial coordinates of the physical source of thermal runaway corresponding to the abnormal event and the first sequence obtained from the first conduction mode signal subset in the abnormal event are stored together, so that the stored content can fully reflect the correspondence between the spatial location and propagation timing characteristics of the abnormal event.

[0107] As abnormal events accumulate, the number of stored abnormal events is statistically analyzed. When the number of stored abnormal events reaches a preset threshold, the coordinates of the thermal runaway physical source point corresponding to each individual abnormal event are selected from the stored events. These coordinates are then input into the network topology model as the starting position for vibration propagation. Combining the connection relationships between nodes in the network topology model and the theoretical wave velocity values ​​associated with each edge, the theoretical propagation time sequence of vibration from the thermal runaway physical source point along the solid structure path to each sensor node is calculated. Based on this theoretical propagation time sequence, a second sequence corresponding to the first sequence in terms of structural form is generated. This second sequence is used as the verification sequence for validation.

[0108] Subsequently, each set of verification sequences is compared and analyzed with its corresponding stored first sequence to obtain the degree of difference in their temporal ordering relationship, and this difference is recorded as the verification error. After obtaining the verification errors corresponding to multiple abnormal events, minimizing the verification errors of all abnormal events is taken as the overall optimization objective. Under the premise of satisfying the physical constraints of the network topology model, the theoretical wave velocity values ​​associated with each edge in the network topology model are iteratively adjusted through gradient descent, so that the adjusted theoretical wave velocity values ​​can reduce the verification error as a whole and better reflect the equivalent propagation characteristics in the solid-state structure of the battery pack.

[0109] After optimization, the original theoretical wave velocity value in the network topology model is replaced with the adjusted theoretical wave velocity value, thereby achieving continuous correction of the propagation parameters of the network topology model, so that the source localization process of subsequent abnormal events can be calculated based on the updated propagation parameters.

[0110] When adjusting the theoretical wave velocity values ​​of each edge in the network topology model, the adjustment process is carried out under preset physical constraints. The physical constraints include at least a reasonable range of wave velocity values ​​and a smoothness constraint on the wave velocity changes of adjacent structural units.

[0111] By introducing the above constraints, parameter overfitting caused by abnormal data or noise can be avoided, thus ensuring that the adjusted theoretical wave velocity value still has a clear physical meaning and can stably reflect the overall changing trend of the propagation characteristics of the battery pack structure.

[0112] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the present invention should still fall within the scope of the present invention.

Claims

1. A method for identifying battery thermal runaway based on pressure mutation, characterized in that, Includes the following steps: S1. Arrange multiple pressure sensors inside the battery pack to collect pressure vibration signals, and determine whether an abnormal event has occurred inside the battery pack based on the pressure vibration signals. S2. When an abnormal event occurs, the raw waveform data of all pressure sensors within the event time window are collected synchronously to form a signal data set; S3. Perform conduction mode separation on the complete signal data set to obtain a first conduction mode signal subset and a second conduction mode signal subset; S4. Calculate the first spatial coordinate solution set and the second spatial coordinate solution set based on the first conduction mode signal subset and the second conduction mode signal subset, respectively; S5. Cross-validate the solution sets of the first and second spatial coordinates to determine the physical source of thermal runaway. Conduction mode separation includes: Extract the signals of each pressure sensor in multiple dimensions from the signal data set, and combine the signals of all sensors to form a spatiotemporal signal matrix; The spatiotemporal signal matrix is ​​processed using a mode decomposition method to obtain multiple global propagation mode components. The mode decomposition method includes multi-scale step mode decomposition. For each global conduction mode component, its signal propagation characteristics among different pressure sensors are calculated; The signal propagation characteristics are matched with preset solid-state structure conduction characteristics and non-solid medium conduction characteristics; The global conduction mode components that match the conduction characteristics of solid structures and conduction characteristics of non-solid media are respectively merged into a first conduction mode signal subset and a second conduction mode signal subset; The calculation of the solution set of the first spatial coordinates includes: Based on the first transmission mode signal subset, the first peak moment of the signal arriving at each pressure sensor is extracted; Based on the first peak time of all pressure sensors, calculate the difference of peak times between all non-repeating pressure sensor pairs, and sort the differences of peak times to obtain a first sequence. Each node in the solid-state structure network topology model of the battery pack, except for the sensor nodes, is assumed to be the source point; wherein, the sensor nodes in the network topology model correspond one-to-one with the installation positions of the pressure sensors, and the edges represent continuous solid force transmission paths and are associated with theoretical wave velocity values. For each of the hypothetical source points, based on the network topology model and the theoretical wave velocity values ​​of each edge, the theoretical propagation time of the vibration wave from the hypothetical source point to each sensor node is calculated. Calculate the difference in theoretical propagation time between all non-repeating pressure sensor pairs, and sort the differences in theoretical propagation time to obtain the second sequence; Calculate the root mean square error between the second sequence and the first sequence corresponding to each hypothetical source point, and select hypothetical source points whose root mean square error is less than a preset error tolerance. The three-dimensional spatial coordinates of all selected hypothetical source points constitute the first spatial coordinate solution set.

2. The battery thermal runaway identification method based on pressure mutation according to claim 1, characterized in that, The acquired signal data set includes: The instantaneous energy of the signal output by each pressure sensor is continuously calculated. The instantaneous energy is the integral of the square of the signal amplitude over a short time window. When the instantaneous energy continuously exceeds the preset background energy threshold for a preset duration, the moment when the preset duration is reached is extended forward by a predetermined duration to obtain the event start time. The starting point of the event time window is set to the event start time, and the length of the event time window is based on a manual preset. Within the event time window, raw waveform data from all pressure sensors are simultaneously acquired and stored to form a signal data set.

3. The battery thermal runaway identification method based on pressure mutation according to claim 2, characterized in that, Calculating the second conduction mode signal subset includes: Based on the second transmission mode signal subset, the first peak moment of the signal arriving at each pressure sensor is extracted; Based on the first peak time of all pressure sensors, the difference between the peak times of all non-repeating pressure sensor pairs is calculated, and the differences in peak times are sorted to obtain a third sequence. Set the average propagation velocity v of the pressure vibration in the non-solid medium inside the battery pack; For any two distinct pressure sensors i and j, with coordinates (xi, yi, zi) and (xj, yj, zj) respectively, the following equations are established: ； Where (x, y, z) represents the coordinates of an event source point to be solved, and ΔTij represents the difference between the peak times of pressure sensors i and j. All the equations form a system of equations. A numerical optimization algorithm is used to solve the system of equations to find the coordinates of one or more event source points, such that the overall residual generated after substituting the coordinates of the event source points into all the equations is minimized. The coordinates of all event source points constitute the second spatial coordinate solution set.

4. The battery thermal runaway identification method based on pressure mutation according to claim 3, characterized in that, Determining the physical source of thermal runaway includes: Calculate the Euclidean distance between all pairs of coordinate points in the first spatial coordinate solution set and the second spatial coordinate solution set. If the Euclidean distance is less than the preset spatial distance threshold, mark the corresponding pair of coordinate points as the target pair. If the proportion of the number of target pairs to the total number of coordinate point pairs exceeds a preset threshold, the geometric center point of the two coordinate points in the target pair is calculated, and the geometric center point with the highest frequency of occurrence is selected as the physical source point of thermal runaway.

5. The battery thermal runaway identification method based on pressure mutation according to claim 4, characterized in that, Calculating the solution set of the first spatial coordinates also includes: After each determination of the physical source of thermal runaway, the coordinates of the physical source of thermal runaway corresponding to that abnormal event and the first sequence are stored; When the number of stored abnormal events reaches a preset number, the coordinates of the thermal runaway physical source point of a single stored abnormal event are used as the vibration starting point. The corresponding second sequence is obtained by combining the network topology model and recorded as the verification sequence. The error between the verification sequence and the corresponding first sequence is obtained and denoted as the verification error. The optimization problem is to minimize the total verification error. The gradient descent method is used to adjust the theoretical wave velocity values ​​of each edge in the network topology model. Replace the original theoretical wave velocity values ​​of each edge in the network topology model with the adjusted theoretical wave velocity values.

Images