Liquid leakage detection method, device and equipment of liquid cooling system and storage medium

Abstract

This invention discloses a method, apparatus, device, and storage medium for detecting leakage in a liquid cooling system. It involves collecting multi-node operational data reflecting the fluid transport state and constructing a baseline network characterizing the normal coupling relationship of the liquid cooling circuit based on this data. When a structural anomaly occurs in a local area of ​​the baseline network, the corresponding test area is identified. Under the condition of satisfying the thermal safety constraints of the energy storage battery system, a controlled perturbation input is applied to the test area, causing the liquid cooling circuit to form an identifiable dynamic response process. During the dynamic response process, fluid response data of the test area is collected, dynamic features are extracted, and the difference between these features and the baseline features is measured to output the leakage risk detection result. This invention can effectively identify leakage risks in liquid cooling systems under complex operating conditions, improving detection reliability and early warning capabilities.

Description

Technical Field

[0001] This invention relates to the field of liquid cooling system leakage detection technology, and in particular to a method, apparatus, equipment, and storage medium for detecting liquid cooling system leakage. Background Technology

[0002] Liquid cooling systems are a type of cooling method that uses liquid as a heat transfer medium to remove heat generated during equipment operation from heat-generating components and transfer it to external heat dissipation units. Compared to traditional air cooling, liquid cooling systems offer advantages such as high heat transfer efficiency, uniform temperature control, and low noise, making them widely used in high-power-density, high-energy-integration products, such as data center servers, power electronic equipment, and lithium battery systems. In lithium battery systems, liquid cooling systems typically include components such as cold plates, piping, manifolds, pumps, valves, and heat exchangers to continuously dissipate heat from battery modules or cells, ensuring the battery operates within a suitable temperature range, thereby improving battery safety, lifespan, and performance stability.

[0003] However, liquid cooling systems are typically closed-loop structures with multiple connection points and seals. Under long-term operation, vibration, shock, or thermal cycling, there is a risk of coolant leakage. For lithium battery systems, a leak in the liquid cooling system can lead not only to reduced cooling capacity and abnormal local temperatures, but also to coolant entering the battery pack and coming into contact with electrical components or high-voltage connections, causing reduced insulation performance, corrosion damage, and even safety hazards. Furthermore, leaks in liquid cooling systems are often insidious in their early stages, making them difficult to detect in time. Failure to effectively detect and assess leak risks in their early stages can lead to escalation of the fault or forced system shutdown. Therefore, it is necessary to perform leak detection on liquid cooling systems to achieve effective monitoring of their operational status and provide early warning of risks. Summary of the Invention

[0004] This application provides a method for detecting leakage in a liquid cooling system, aiming to solve the technical problems of existing liquid cooling systems having strong concealment of leakage during operation, being easily affected by changes in operating conditions, and being difficult to identify leakage risks in a timely and reliable manner.

[0005] This invention is implemented as follows: a method for detecting leakage in a liquid cooling system, comprising:

[0006] During the operation of the liquid cooling circuit, multi-node operating data reflecting the fluid transport status of the liquid cooling circuit are collected, and a baseline network characterizing the normal coupling relationship of the liquid cooling circuit is constructed based on the multi-node operating data. The multi-node operating data includes at least two types of data: pressure data, flow data, and temperature data.

[0007] When the baseline network exhibits structural anomalies in a local area, the test area corresponding to the local area is determined, and a preset controlled perturbation input is applied to the test area under the condition of satisfying the thermal safety constraints of the lithium battery, so that the liquid cooling circuit forms an identifiable dynamic response process, and at least one fluid response data of the test area is collected simultaneously.

[0008] Dynamic features characterizing the equivalent fluid properties of the liquid cooling circuit are extracted from the fluid response data. The difference between the dynamic features and the baseline features corresponding to the baseline network is measured, and the leakage risk detection result is output.

[0009] The present invention also provides a device for detecting leakage in a liquid cooling system, comprising:

[0010] The data acquisition module is used to collect multi-node operating data reflecting the fluid transport status of the liquid cooling circuit during operation, and to construct a baseline network characterizing the normal coupling relationship of the liquid cooling circuit based on the multi-node operating data. The multi-node operating data includes at least two types of data: pressure data, flow data, and temperature data.

[0011] The controlled perturbation module is used to determine the test area corresponding to the local area when the baseline network has a structural anomaly in a local area, and to apply a preset controlled perturbation input to the test area under the condition of meeting the thermal safety constraints of the lithium battery, so that the liquid cooling circuit forms an identifiable dynamic response process, and simultaneously collect at least one fluid response data of the test area.

[0012] The detection result output module is used to extract dynamic features characterizing the equivalent fluid properties of the liquid cooling circuit from the fluid response data, measure the difference between the dynamic features and the baseline features corresponding to the baseline network, and output the leakage risk detection result.

[0013] The present invention also provides a liquid cooling system leakage detection device, comprising: a memory and at least one processor, wherein the memory stores instructions, and the memory and the at least one processor are interconnected via a circuit; the at least one processor invokes the instructions in the memory to cause the liquid cooling system leakage detection device to perform the steps of the liquid cooling system leakage detection method described above.

[0014] The present invention also provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the steps of the above-described liquid cooling system leakage detection method.

[0015] It is understandable that when a liquid cooling circuit is in operation, the operating data of multiple nodes often exhibit stable coupling relationships within the same time period. For example, there is a linkage between pressure, flow rate, and supply and return liquid temperatures, which drift together with changes in operating conditions. By constructing node state vectors based on a sliding time window and calculating the correlation strength between nodes, this linkage can be separated from the absolute values ​​of each node, and the operating state can be characterized by the degree of coupling stability between node pairs. Furthermore, by using the statistics and uncertainties of the correlation strength together to construct an adaptive criterion, the judgment criteria can be automatically adjusted according to the current data distribution. When changes in the overall operating conditions cause synchronous changes in the values ​​of each node, the correlation structure remains consistent, and the baseline network is less prone to misjudgment. However, when a local sealing failure leads to leakage risk, the fluid transport state in the affected area will change asynchronously with other areas. For example, changes in local flow resistance and compressibility cause shifts in the pressure-flow relationship, and changes in local heat exchange capacity cause shifts in the supply and return liquid temperature difference relationship. The original coupling edges between nodes thus become structurally weakened, broken, or have increased uncertainty, which manifests as structural anomalies in local areas at the baseline network level. In this way, early-stage hidden anomalies no longer rely on a single threshold for triggering, but are exposed in advance through the disruption of coupling relationships, and have a stronger ability to distinguish overall operating condition drift.

[0016] When the baseline network exhibits structural anomalies in local areas, identifying the test area and applying controlled perturbation input under lithium battery thermal safety constraints allows for the construction of a more controllable observation window with manageable boundary conditions within a short time. This enables the differences caused by sealing failure to be presented more clearly in the form of dynamic response. Controlled perturbation input causes the loop to transition from an approximately steady state to a repeatable dynamic process. During this dynamic process, at least one fluid response data point is simultaneously acquired, and dynamic features characterizing the equivalent fluid properties are extracted. These features reflect the loop's response speed to disturbances, steady-state change amplitude, and hysteresis, among other dynamic attributes. Under normal sealing conditions, these dynamic attributes exhibit high repeatability under the same or similar operating conditions. However, when leakage risk occurs, factors such as local fluid loss and changes in gas-containing state alter the loop's equivalent impedance and equivalent cavity, causing a stable shift in the response curve in terms of rise time, steady-state gain, and phase hysteresis. By measuring the difference between dynamic features and baseline features corresponding to the baseline network, it is equivalent to measuring the degree of this offset within the same set of comparable feature spaces. This transforms early leakage risks that are difficult to observe directly into quantifiable and graded detection results, and outputs suspected areas and corresponding confidence levels or grades, enabling timely and reliable identification of leakage risks in lithium battery liquid cooling systems. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of one embodiment of the liquid cooling system leakage detection method according to the present invention;

[0019] Figure 2 This is a schematic diagram of one embodiment of the liquid cooling system leakage detection device according to the present invention;

[0020] Figure 3 This is a schematic diagram of one embodiment of a liquid cooling system leakage detection device according to the present invention;

[0021] Figure 4 This is a schematic diagram illustrating a specific application of the liquid cooling system of the present invention;

[0022] Figure 5 Comparison of fluid dynamic response in the measured region under controlled perturbation input.

[0023] Explanation of reference numerals in the attached figures:

[0024] 101. Data acquisition module; 102. Controlled perturbation module; 103. Detection result output module;

[0025] 200. Detection equipment for liquid cooling system leakage; 210. Processor; 220. Memory; 230. Storage medium; 231. Operating system; 232. Data; 233. Application program; 240. Power supply; 250. Wired or wireless network interface; 260. Input / output interface.

[0026] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0027] 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 a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0028] It should be noted that if the embodiments of the present invention involve directional indicators (such as up, down, left, right, front, back, etc.), the directional indicators are only used to explain the relative positional relationship and movement of the components in a specific posture. If the specific posture changes, the directional indicators will also change accordingly.

[0029] Furthermore, the use of terms such as "first" and "second" in this invention is for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the term "and / or" throughout the text includes three solutions; taking A and / or B as an example, it includes technical solution A, technical solution B, and a technical solution that simultaneously satisfies A and B. Furthermore, the technical solutions of various embodiments can be combined with each other, provided that they are feasible for those skilled in the art. When the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by this invention.

[0030] Figure 1 The implementation flow of the liquid cooling system leakage detection method provided in this embodiment is illustrated. For ease of explanation, only the parts relevant to this embodiment are shown, such as... Figure 1 As shown, this method includes the following steps:

[0031] S100. During the operation of the liquid cooling circuit, multi-node operating data reflecting the fluid transport status of the liquid cooling circuit are collected, and a baseline network characterizing the normal coupling relationship of the liquid cooling circuit is constructed based on the multi-node operating data. The multi-node operating data includes at least two types of data: pressure data, flow data, and temperature data.

[0032] In one embodiment of the present invention, the above-mentioned step S100 specifically includes:

[0033] S101. Multiple nodes are determined in multiple spatial regions of the liquid cooling circuit. The multiple spatial regions include at least a cooling region corresponding to the lithium battery module, a manifold region corresponding to the supply and return liquid collection structure, and a pump and valve region corresponding to the power components of the liquid cooling circuit.

[0034] Determining multiple nodes in a liquid cooling circuit essentially involves spatially discretizing the originally continuous and closed liquid cooling system according to its physical structure and functional characteristics, enabling the circuit's operating state to be perceived and analyzed in a regional and hierarchical manner. In lithium battery systems, the liquid cooling circuit is typically arranged around the battery pack, and its structure naturally exhibits distinct functional zoning characteristics. Therefore, the determination of nodes can be based on these objectively existing structural units.

[0035] Specifically, nodes can first be determined around the cooling area corresponding to the lithium battery module. Each lithium battery module typically exchanges heat with the coolant through a cold plate or internal flow channel. This area directly reflects the heat exchange state between the coolant and the battery. Therefore, each module cooling unit, or several module cooling units that are physically adjacent and have the same cooling structure, can be considered as a cooling area node. The data corresponding to this type of node can come from the module return temperature, module shell temperature, or module cold plate surface temperature, etc. These data can comprehensively reflect the flow and heat exchange characteristics of the coolant in this area. Second, nodes can be determined around the manifold area corresponding to the supply and return liquid collection structure. The manifold is a key location for the collection and distribution of multiple branches in the liquid cooling circuit. Its operating status can reflect whether the overall flow distribution is balanced and whether there are local flow resistance changes. Therefore, the supply manifold section, return manifold section, or a combination of the two can be considered as independent manifold area nodes. This type of node is usually associated with pressure, pressure difference, or collection point temperature to characterize the overall coupling state of multiple cooling branches during the collection or distribution process. Third, pump and valve area nodes can be determined around the power components in the liquid cooling circuit. The circulating pump, valves, and the pipe sections closely connected to them together determine the driving force and boundary conditions of the liquid cooling circuit. Their operating status directly affects the transport behavior of the coolant. Therefore, the locations near the pump outlet and upstream and downstream of the valve body can be regarded as pump-valve area nodes, and the operating status of this area can be characterized by pump speed, pump outlet pressure, valve position status, or related calculated parameters.

[0036] By organically combining the cooling zone, manifold zone, and pump / valve zone in the above manner, the liquid cooling circuit is mapped as a system structure composed of multiple spatial nodes with clear physical meanings. For example, in a liquid-cooled battery pack containing eight battery modules, the cooling unit corresponding to every two modules can be defined as a cooling zone node, the supply manifold and return manifold can each be defined as a manifold zone node, and the circulation pump and its adjacent pipe sections can be defined as a pump / valve zone node, thus forming a multi-node system covering the entire circuit. This provides a clear and feasible foundation for subsequent operational status analysis and leakage detection based on node relationships.

[0037] It should be noted that the same region can include multiple nodes, and is not limited to a single node.

[0038] S102. Within the sliding time window, at least two types of operating parameters with different physical meanings are collected for each node as node operating data, and the node operating data is processed for time alignment, outlier removal, and short-term fluctuation suppression.

[0039] After determining the nodes of each spatial region in the liquid cooling circuit, it is necessary to continuously collect node operation data reflecting the operating status of each node within a sliding time window to perform stable analysis of the coupling relationship between nodes. To ensure that the node state has sufficient expressive power, it is preferable to collect at least two types of operating parameters with different physical meanings for each node, thereby avoiding the problem of insufficient information caused by relying on a single parameter. Taking the liquid cooling circuit of a lithium battery system as an example, in the cooling area node corresponding to the battery module, the return liquid temperature and the battery module shell temperature of that area can be collected simultaneously. The former reflects the heat exchange result of the coolant in that area, and the latter reflects the thermal state of the battery itself. The combination of the two can more comprehensively characterize the coupling relationship between the coolant and the battery. In the manifold area node, pressure-related parameters and temperature-related parameters at the manifold can be collected simultaneously to reflect the fluid transport state when multiple branches converge or diverge. In the pump and valve area node, the pump speed or pump outlet pressure and the corresponding temperature parameters can be combined to characterize the driving characteristics of the power components on the entire circuit.

[0040] Since different operating parameters often originate from different sampling channels or sampling periods, directly using the raw data can easily introduce time bias. Therefore, it is necessary to perform time alignment processing on the operating data of each node within a sliding time window to pair different parameters under the same time reference. Common practices include timestamp-based interpolation or resampling to a unified sampling period. After completing time alignment, outliers in the node operating data also need to be removed to avoid interference from sensor momentary failures, communication jitter, or extreme noise in subsequent analysis. This can be achieved by limiting the physically reasonable range or using neighborhood statistical characteristics to identify isolated abrupt changes. Furthermore, to suppress high-frequency fluctuations caused by fluid turbulence, sampling noise, or short-term operating condition disturbances, short-term fluctuation suppression processing is also required on the node operating data. For example, moving averages or median filtering can be used to ensure that the node operating data retains the overall trend while possessing better stability. Through the above processing, a node operating data set with consistent timing, controlled noise, and clear physical meaning can be formed within the sliding time window, providing a reliable data foundation for subsequent construction of node state vectors and analysis of the correlation between nodes.

[0041] S103. Determine a valid data segment based on the operating condition information within the sliding time window. The valid data segment satisfies a preset operating condition consistency condition. The operating condition consistency condition includes at least the liquid cooling circuit being in a stable cycle state and the absence of a preset start-stop transient or valve switching transient.

[0042] Not all node operation data collected within the sliding time window is suitable for characterizing the normal coupling relationship of the liquid cooling circuit. Therefore, it is necessary to further filter the data by combining it with operating condition information to determine the effective data segments that can truly reflect the stable operating characteristics of the liquid cooling circuit. Operating condition information typically comes from a description of the overall operating status of the liquid cooling circuit, such as whether the circulating pump is continuously running, whether the pump speed or drive command remains relatively stable, whether there are any valve opening change commands, and whether the heat exchange unit has switched operating modes. This information can reflect whether the liquid cooling circuit is in a relatively stable and comparable operating phase.

[0043] By analyzing the above operating condition information, different operating segments can be identified within the sliding time window, and it can be determined which segments meet the preset operating condition consistency conditions. The core of the operating condition consistency conditions is to ensure that the boundary conditions and drive conditions of the liquid cooling circuit are basically consistent within the selected data segment, so that the correlation between nodes is mainly determined by the physical state of the circuit itself, rather than by external control actions or transient disturbances. Among them, the liquid cooling circuit being in a stable circulation state is one of the most basic conditions. This usually means that the circulating pump has completed the startup process and entered the continuous operation stage, and the pump speed or equivalent driving force only fluctuates slightly within the data segment without obvious step changes. At the same time, it is also necessary to exclude start-up and shutdown transients or valve switching transients, because during pump startup, shutdown, or rapid opening and closing of valves, parameters such as pressure, flow rate, and temperature will change significantly in a short period of time. These changes mainly reflect the control actions themselves, rather than the inherent coupling characteristics of the liquid cooling circuit.

[0044] For example, in the liquid cooling circuit of a lithium battery system, when the vehicle is first powered on, the circulation pump accelerates from a standstill to the target speed. During this process, the pump outlet pressure and flow rate rise rapidly, while the temperature of each node has not yet established a stable heat transfer relationship. The data collected at this time does not meet the consistency condition requirements and should be excluded. Similarly, when the system executes valve switching commands to adjust the cooling capacity, the flow distribution of some branches changes instantaneously, causing non-stationary fluctuations in node parameters. Data during this period should not be considered valid data segments. By identifying and eliminating the above unstable segments within a sliding time window and combining operating condition information, only data intervals that meet the stable circulation state and have no obvious control transients can be retained to form valid data segments. This allows subsequent correlation analysis based on node state vectors to be established on consistent and comparable operating conditions, thereby improving the reliability and accuracy of the leak detection process.

[0045] S104. Based on the node operation data within the effective data segment, construct the node state vector.

[0046] After determining the valid data segment that meets the operating condition consistency condition, the construction of the node state vector needs to be based on the node operating data within this valid data segment that has undergone time alignment, outlier removal, and short-term fluctuation suppression processing. This ensures that each node is represented as a vector with clear physical meaning and statistical characteristics within the same time window. Since the node operating data in the liquid cooling loop itself has time-series characteristics, directly using instantaneous values ​​is often susceptible to residual noise or occasional disturbances. Therefore, when constructing the node state vector, it is usually necessary to statistically aggregate the operating data within the valid data segment so that the node state can reflect both the overall operating level within that time period and the necessary fluctuation characteristics. Specifically, for each node, at least two different types of operating parameters with different physical meanings collected within the valid data segment can be used as the basic data source. These parameters are then statistically processed within the valid data segment, such as calculating the mean, magnitude of change, or trend, thereby compressing the data that originally varied over time into a set of characteristic quantities that can represent the node's state. The advantage of this approach is that the node state vector no longer depends on a specific instant, but reflects the comprehensive characteristics within a stable operating range, which is more conducive to comparison and correlation analysis between different nodes.

[0047] In practical applications, the state vector composition of different types of nodes may vary depending on their physical functions, but the overall construction logic remains consistent. Taking the cooling area node corresponding to a lithium battery module as an example, this node may simultaneously possess return liquid temperature data and module shell temperature data within the effective data segment. Therefore, the node's state vector can be composed of the time average of the return liquid temperature, the fluctuation range of the return liquid temperature, and the time average of the module shell temperature. This vector reflects both the heat transfer result of the coolant in this area and the thermal state of the battery itself. For the manifold area node, the statistical characteristics of the pressure-related parameters and temperature-related parameters at the manifold within the effective data segment can be combined into a state vector to characterize the overall fluid transport state when multiple branches converge or diverge. For the pump and valve area node, the statistical characteristics of the pump speed or pump outlet pressure and the corresponding temperature parameters can be incorporated into the node's state vector to reflect the driving conditions of the liquid cooling circuit. In the above construction process, the dimensions and parameter meanings of each node's state vector remain consistent or comparable, making it possible to subsequently calculate the correlation strength between nodes based on the node state vectors. In this way, the complex liquid cooling circuit operation data is systematically mapped into a set of node state vectors with a clear structure and explicit physical meaning, providing a stable and reliable foundation for subsequent construction of the baseline network and identification of local anomalies.

[0048] In one embodiment of the present invention, step S100 further includes:

[0049] S105. Within the effective data segment, calculate the association strength between any two nodes based on the node state vector of each node. The association strength includes the correlation coefficient, mutual information, or a combination of the two.

[0050] Calculating the correlation strength between any two nodes within the valid data segment requires restoring the node state vectors to comparable time-series objects, or retaining the temporal components that support the correlation calculation when constructing the node state vectors, so that the coupling relationship between nodes can be quantified under the same time reference. More specifically, within the valid data segment, each node corresponds to a set of operating parameter sequences after time alignment and noise suppression. For example, a module cooling area node has a return liquid temperature sequence and a module housing temperature sequence, while a pump valve area node has a pump outlet pressure sequence and a pump speed sequence. To calculate the correlation strength between the two nodes, two sets of directly comparable sequence representations need to be formed at the same sampling time. A common approach is to select one or more representative sequences most relevant to the coupling relationship for each node within the valid data segment. For example, the cooling area node can be represented by a supply-return liquid temperature difference sequence or a return liquid temperature sequence, the manifold area node by a return liquid manifold temperature sequence or a manifold pressure sequence, and the pump valve area node by a pump outlet pressure sequence or a flow rate sequence derived from the pump speed. Then, the correlation strength is calculated for the representative sequences of any two nodes within the valid data segment.

[0051] When using the correlation coefficient as the correlation strength, the calculation process does not compare single-point values, but rather measures the degree of synchronous change between two sequences within the effective data segment. Therefore, the sequences need to be mean-removed and scale-normalized first to avoid bias in the results due to different dimensions or amplitude ranges. Then, the covariance of the two sequences within the window is calculated and divided by their respective standard deviations to obtain the correlation coefficient, which ranges from negative one to positive one. To make the correlation coefficient more closely reflect the actual coupling of the liquid cooling circuit, finite hysteresis alignment can be introduced within the same effective data segment. For example, considering the thermal inertia hysteresis of the return liquid temperature relative to the pump pressure change, the hysteresis quantity that maximizes the absolute value of the correlation coefficient can be searched within a small hysteresis range, and the corresponding maximum correlation coefficient is taken as the correlation strength between the two nodes. This avoids spurious low correlation caused by thermal hysteresis. For example, under stable circulation conditions, the increase in pump outlet pressure is often accompanied by an increase in flow rate, which in turn causes the return liquid temperature of some modules to drop or the temperature difference to decrease in a short period of time. If the correlation coefficient is calculated directly at the same moment, a weak correlation may be obtained. However, if a lag of several seconds is allowed, the correlation will be significantly enhanced, thus more accurately reflecting the physical coupling.

[0052] When mutual information is used as the correlation strength, the goal is to measure the statistical dependency between two sequences, covering nonlinear or piecewise relationships. Therefore, it is necessary to estimate the probability distribution of the two sequences within the effective data segment. This can be achieved using discretization and counting estimation: the range of values ​​for each sequence within the effective data segment is divided into several intervals according to quantiles or equal widths. Then, the joint frequency of the two sequences falling into each interval is counted and normalized to obtain the joint probability distribution. At the same time, the marginal probability distributions of each sequence are counted separately. Finally, the information gain between the joint distribution and the marginal distributions is calculated according to the definition of mutual information. The mutual information obtained in this way is particularly applicable to the piecewise characteristics caused by changes in valve opening or the nonlinear pump curve effect between pump speed and flow rate in liquid cooling systems. For example, within certain valve position ranges, changes in pump pressure have a significant impact on the branch flow rate, while the impact is weaker in other valve position ranges. This relationship is easily "averaged out" by a single correlation coefficient, but mutual information can still capture the dependency.

[0053] Furthermore, to balance linear synchronicity and nonlinear dependence, correlation coefficients and mutual information can be combined with preset weights, or different correlation strength measures can be selected between different node types. For example, correlation coefficients can be used preferentially for pressure and flow series, while mutual information can be introduced for series such as temperature and pressure difference that may have lag and nonlinear coupling. This makes the correlation strength calculation based on effective data segments have a clear calculation path and can fit the real coupling characteristics of the liquid cooling loop.

[0054] S106. Determine a set of statistics for all association strengths within the effective data segment, wherein the set of statistics includes at least the mean and dispersion of the association strengths;

[0055] After calculating the correlation strength between any two nodes within the valid data segment, a set of correlation strengths describing the coupling state of the liquid cooling loop is obtained. This set can be understood as a correlation strength matrix or a correlation strength list, where each item corresponds to the coupling strength of a pair of nodes within the valid data segment. To enable subsequent judgment criteria to adaptively adjust to the current operating conditions, this entire set of correlation strengths needs to be statistically analyzed to extract a set of statistics representing the current coupling level. In practice, all available node pairs can be uniformly screened first, for example, excluding self-correlated terms and node pairs that cannot be reliably calculated due to missing data. Then, the correlation strength values ​​of the remaining node pairs are summarized into a numerical set. Based on this, the set of statistics should include at least the mean and dispersion. The mean is used to characterize the central level of the overall coupling strength within the current valid data segment, and the dispersion is used to characterize the degree of dispersion of coupling strength between different node pairs, thus reflecting the stability and differences of the coupling structure.

[0056] When calculating the mean, it's necessary to consider the range and meaning of the association strength metric. If the correlation coefficient is used, its value may be negative or positive. Directly calculating the arithmetic mean might result in the cancellation of positive and negative values, masking the coupling strength. Therefore, some implementations take the absolute value of the correlation coefficient before calculating the mean, or calculate positive and negative correlations separately to more accurately reflect the overall coupling degree. If mutual information is used, it is non-negative, and the mean can be calculated directly. Dispersion can usually be represented by standard deviation, variance, or interquartile range. Standard deviation directly reflects the fluctuation of the association strength relative to the mean, while interquartile range is more suitable for maintaining robustness when there are a few extreme correlation values.

[0057] S107. Determine the uncertainty quantity based on the node state vector, wherein the uncertainty quantity is used to characterize the volatility or uncertainty of the node state vector within the effective data segment;

[0058] The determination of uncertainty is used to characterize the volatility and instability of the node state vector within the effective data segment. Its core purpose is to distinguish between the two situations: "stable state but different levels" and "state itself is unstable". In implementation, the time series data on which the node state vector is based can be directly used to perform a brief statistical analysis of the node's change characteristics within the effective data segment, making the uncertainty both clear in its source and easy to understand and implement.

[0059] Specifically, for each node, statistical indicators reflecting the degree of fluctuation, such as standard deviation or amplitude of change, can be calculated for its operating parameter sequence within the effective data segment to describe the temporal dispersion of the node's state. When it is necessary to further distinguish fluctuations with strong randomness, the sequence can be simply discretized and the information entropy calculated to reflect the dispersion of the node's state value distribution. The above indicators can be normalized according to the number of parameters contained in the node and combined into a single node-level uncertainty, making different nodes comparable. Under normal and stable operation of the liquid cooling loop, the uncertainty of each node is usually at a low level, indicating that the node state changes smoothly and predictably within the effective data segment. However, when there are abnormal flows, gas content, or micro-leakage in a local area, even if the average level of the node's operating parameters does not change significantly, its time series often exhibits more frequent or more irregular fluctuations, leading to an increase in the corresponding uncertainty.

[0060] The uncertainty obtained in this way can intuitively reflect the reliability of the node state within the effective data segment, and can be used to modulate the subsequent correlation strength statistics and adaptive criteria, so that the system can automatically increase the prudence of the judgment when the node state is unstable, and gradually amplify it when continuous anomalies occur, thereby improving the robustness of liquid cooling system leakage detection.

[0061] S108. Construct an adaptive criterion threshold using the set of statistics and the uncertainty, wherein the adaptive criterion threshold includes at least a basic threshold term consisting of the mean and dispersion of the correlation strength, and a modulation term that modulates the basic threshold term using the uncertainty.

[0062] After obtaining the association strength of all node pairs within the effective data segment, the mean and dispersion of the association strength can reflect the central position and dispersion of the overall coupling level under this operating condition. Uncertainty, on the other hand, reflects whether the node state is stable within the effective data segment. Therefore, the construction of the adaptive criterion threshold can unify both onto the same logical chain. That is, under the current operating condition, a baseline threshold that is automatically updated with changes in the overall distribution is first given. Then, the threshold is tightened or loosened according to the stability of the data itself, so that the threshold does not become invalid due to operating condition drift, nor is it overly sensitive to short-term fluctuations. In specific implementation, a basic threshold item can be formed first using a set of statistics. For example, the mean and dispersion of the association strength can be combined with preset coefficients to obtain a basic threshold. The dispersion can be expressed as standard deviation or interquartile range, and the tolerance range for the dispersion can be controlled by coefficients, allowing the threshold to automatically shift upward or downward with the overall coupling strength. Taking the correlation coefficient as an example, if the absolute value of the correlation coefficients of all node pairs is taken to obtain a mean of 0.62 and a standard deviation of 0.10, the basic threshold term can be constructed as 0.62 plus 0.10 multiplied by a certain coefficient, thereby automatically excluding weakly correlated node pairs and retaining only coupling edges that are higher than the overall level by a certain margin. When the operating conditions change and the overall coupling weakens, for example, the mean drops to 0.48 and the standard deviation is 0.12, the basic threshold will decrease accordingly, so that the threshold still fits the current distribution rather than remaining fixed, thereby avoiding large-scale misjudgments caused by changes in external temperature or pump speed settings.

[0063] After obtaining the basic threshold term, an uncertainty is introduced to form a modulation term, which can make the threshold respond to data stability. This can be achieved by constructing a modulation factor using the uncertainty, which is multiplied or added to the basic threshold. This factor is used to raise the threshold when uncertainty is high, thereby reducing spurious strong correlations caused by random fluctuations, or to moderately lower the threshold when uncertainty is low, so as to more sensitively capture subtle changes in the coupling structure under stable data conditions. For example, the uncertainty can be normalized to between 0 and 1, and the modulation factor can be set to be larger as uncertainty increases. When the overall uncertainty of nodes in a certain effective data segment is high due to local gas content or flow fluctuations, the modulation term raises the basic threshold. Only when multiple node pairs show consistent and significant strong correlation changes will they be included in the baseline network, thereby suppressing structural disturbances caused by short-term fluctuations. Conversely, when the effective data segment is in a truly stable cyclic state and the uncertainty is low, the influence of the modulation term on the basic threshold is weakened, and the threshold is closer to the baseline level determined by the mean and dispersion, thereby improving the ability to distinguish between weakened or broken local coupling edges. By constructing an adaptive criterion threshold that includes a basic threshold term and a modulation term, the threshold update is constrained by both the overall coupling distribution and data stability, enabling it to maintain a consistent judgment scale under different operating conditions and avoid over-triggering when the node state is unstable. This provides a robust threshold for forming a reliable baseline network and subsequent structural anomaly identification.

[0064] S109. Determine the node pairs that satisfy the adaptive criterion threshold as network edges of the baseline network, or normalize and weight the association strength according to the adaptive criterion threshold and use it as the edge weight of the baseline network to form a baseline network.

[0065] Specifically, the baseline network is used to characterize the normal coupling relationships between different spatial regions of the liquid cooling loop under stable operating conditions. The baseline network is based on nodes corresponding to multiple spatial regions and the relationships between these nodes. Nodes represent different spatial regions within the liquid cooling loop, and relationships represent stable connections formed by fluid transport and heat transfer under stable operating conditions. It is important to note that the baseline network is not a communication network, control network, or neural network; it does not involve information transmission, control decisions, or learning processes. Rather, it is a structured representation of the operating state of the liquid cooling loop, used to describe the existence and changes in coupling relationships between different spatial regions. The construction of the baseline network is based on the statistical correlation characteristics of operating data. Its network structure and correlation strength are updated with changes in operating conditions and system state, thus providing a comparative reference for subsequent identification of abnormal local coupling relationships.

[0066] The formation of the baseline network is not simply a matter of retaining or discarding relationships. Instead, it uses an adaptive criterion threshold to filter or weight the relationship strength between node pairs, ensuring that the network structure truly reflects the normal coupling state of the liquid cooling circuit under current operating conditions. Specifically, for each node pair, its relationship strength is compared with the adaptive criterion threshold. When the relationship strength reaches or exceeds the threshold, the node pair is considered to have a stable and significant coupling relationship within the current valid data segment, thus being identified as a network edge in the baseline network. Conversely, when the relationship strength is below the threshold, the coupling relationship is considered insufficient to represent a stable structure and can be excluded from the baseline network. In this way, the baseline network naturally presents itself as a sparse network composed of several nodes and their key coupling edges, rather than a dense network containing all weak relationships, thereby highlighting the truly physically meaningful coupling paths in the liquid cooling circuit.

[0067] In another implementation, to retain more hierarchical information about the strength of associations, instead of completely retaining or discarding them, the association strengths can be normalized and weighted according to an adaptive criterion threshold, and used as the edge weights of the corresponding network edges in the baseline network. For example, the association strength can be normalized after subtracting the adaptive criterion threshold, so that associations that just reach the threshold correspond to smaller weights, while associations that are significantly higher than the threshold correspond to larger weights, thus reflecting the relative importance of different coupling relationships in the network. The resulting baseline network not only structurally reflects which nodes have stable couplings, but also numerically reflects the differences in the strength of couplings, providing a more detailed basis for subsequent analysis of network structure changes.

[0068] Taking a lithium battery liquid cooling system as an example, under stable circulation conditions, there is usually a strong and stable correlation between the pump / valve region nodes and the supply / return manifold region nodes. Similarly, the module cooling region nodes and the return manifold nodes often maintain a moderate to high correlation strength. Therefore, after comparison with the adaptive criterion threshold, these node pairs are retained as key edges in the baseline network. However, due to differences in heat load and flow path distribution, the correlation strength between different module cooling regions may be lower than the threshold, thus they are not included in the baseline network or exist with only a low weight. The baseline network formed in this way can characterize the normal coupling skeleton of the liquid cooling circuit under the current operating conditions. When an anomaly occurs in a certain region during subsequent operation, the correlation strength of the relevant node pairs in that region will be difficult to continue to meet the adaptive criterion threshold, thus being reflected in changes in the network structure or edge weights, providing an intuitive and stable reference for identifying local structural anomalies.

[0069] S200. When the baseline network exhibits structural anomalies in a local area, the test area corresponding to the local area is determined, and a preset controlled perturbation input is applied to the test area under the condition of satisfying the thermal safety constraints of the lithium battery, so that the liquid cooling circuit forms an identifiable dynamic response process, and at least one fluid response data of the test area is collected simultaneously.

[0070] In one embodiment of the present invention, determining the test region corresponding to the local region when the baseline network exhibits structural anomalies in a local area specifically includes:

[0071] S201. Obtain the network structure parameters of the baseline network within the sliding time window, and calculate the regional risk measure of each spatial region based on the network structure parameters; wherein, the network structure parameters include at least one of the following: the change in the number of associated edges of a node, the change in the weight of the associated edges, or the increment of the uncertainty of a node;

[0072] After forming a baseline network within a sliding time window, structural changes in this baseline network can be tracked within continuously updated windows to quantify the spatial risk distribution of the liquid cooling circuit. Specifically, a corresponding baseline network is generated for each sliding time window. By comparing the changes in network structural parameters within adjacent time windows, feature information reflecting local anomalies can be extracted. Network structural parameters can include changes in the number of network edges connected to nodes, the magnitude of changes in network edge weights relative to the historical baseline, and the increment of node uncertainty between adjacent windows. These parameters can reflect whether the coupling relationship between nodes and their surroundings has been disrupted from different perspectives. To map changes at the node level to physically meaningful spatial regions, the network structural parameters corresponding to nodes belonging to the same spatial region can be aggregated or weighted averaged to obtain a regional risk measure for that spatial region. For example, when the number of effective associated edges between a node in a module's cooling area and nodes in the manifold area significantly decreases and the corresponding edge weights continuously decline within a continuous sliding time window, while the uncertainty of that node gradually increases, the aggregated regional risk measure will be significantly higher than that of other regions, indicating that the coupling relationship in that region is undergoing structural weakening. Conversely, if the network structure parameters of a region fluctuate only slightly within a single window and recover rapidly in subsequent windows, the regional risk measure remains at a low level, thus avoiding misjudging short-term disturbances as persistent risks. In this way, regional risk measures based on network structure parameters can reflect the spatial concentration and persistence of anomalies within a sliding time window, providing a clear and stable basis for subsequent determination of the measured region.

[0073] S202. Compare the regional risk measurement of each spatial region with the corresponding regional baseline value. When the regional risk measurement of at least one spatial region meets the preset triggering condition, determine that spatial region as the measured region.

[0074] Regional baseline values ​​can be obtained from risk metrics statistics of the spatial region during historical stable operation phases or previous sliding time windows, and are used to characterize the normal fluctuation level of the region under normal conditions. By comparing the regional risk metric calculated within the current sliding time window with the corresponding regional baseline value, it can be determined whether the current risk exceeds the acceptable range for the region under normal operating conditions.

[0075] To avoid frequent triggering due to occasional fluctuations, the preset triggering conditions typically focus not only on single instances of exceeding limits but also on the magnitude or persistence of the exceedance. For example, they may require the regional risk metric to exceed the regional baseline value plus a certain allowable offset, or to remain above the baseline value for multiple consecutive sliding time windows. When the preset triggering conditions are met, it indicates that the network coupling structure change in that spatial region has become stable and concentrated, and is no longer caused by short-term disturbances. Therefore, this spatial region can be identified as the test region. For instance, in a lithium battery liquid cooling system, if the number of associated edges and the edge weights of a module's cooling area continuously decrease and significantly decline over several consecutive sliding time windows, causing the regional risk metric to remain consistently higher than the historical baseline level of that region, while the risks in other regions remain stable, then this module's cooling area will be identified as the test region for subsequent, more refined testing.

[0076] In one embodiment of the present invention, the process of applying a preset controlled perturbation input to the tested area under the condition of satisfying the thermal safety constraints of the lithium battery, so that the liquid cooling circuit forms an identifiable dynamic response process, specifically includes:

[0077] S203. Obtain the operating condition information of the liquid cooling circuit, and determine the disturbance amplitude and duration of the controlled perturbation input based on the operating condition information, so that the controlled perturbation input meets the preset thermal safety constraint conditions during the disturbance period, wherein the thermal safety constraint conditions include at least one of the target temperature range constraint or the temperature rise rate constraint of the lithium battery system.

[0078] Operating condition information typically includes the current pump speed or equivalent flow rate of the liquid cooling circuit, the supply and return liquid temperatures, the temperature distribution of the battery module, and the system's load status. This information reflects the thermal balance between the liquid cooling system and the lithium battery. Based on this operating condition information, the system's tolerance margin for changes in cooling capacity at the current moment can be determined, thereby defining the safe range for perturbation inputs. For example, when the battery module temperature is close to the upper limit of the target temperature, the allowable perturbation amplitude and duration need to be reduced accordingly to avoid a temporary drop in cooling capacity during the perturbation, which could lead to temperature exceeding the limit. Conversely, when the battery temperature is in the middle of the target range and changes gradually, a slightly larger or longer perturbation can be applied, while ensuring safety, to improve the identifiability of the dynamic response. Thermal safety constraints can be achieved by limiting the target temperature range or the rate of temperature rise, i.e., continuously monitoring the battery module temperature change during the perturbation process to ensure that it remains within the preset safe range or that its rate of change does not exceed the allowable upper limit.

[0079] For example, in a lithium battery liquid cooling system, when the module temperature is stable near the midpoint of the target range and the temperature rise rate is low, a small change in the circulation pump speed lasting for several seconds can be applied as a perturbation; while when the system is under high load and the module temperature is close to the upper limit, only a shorter time or a smaller amplitude perturbation is allowed, so as to complete the controlled detection under the condition of meeting thermal safety constraints.

[0080] S204. During the application of the controlled perturbation input, at least one boundary condition of the liquid cooling circuit is kept constant or in a preset stable state; wherein the controlled perturbation input is a preset perturbation sequence applied to at least one operating parameter of the liquid cooling circuit.

[0081] During a controlled perturbation input, to ensure that the dynamic response of the liquid cooling circuit primarily reflects the changes in the fluid characteristics of the measured area without being masked by other control actions or external factors, the boundary conditions of the liquid cooling circuit need to be constrained. This means that at least one key boundary condition must remain constant or in a preset stable state during the perturbation. Boundary conditions typically refer to factors that have a decisive impact on the overall operating state of the liquid cooling circuit, such as the operating mode of the heat exchange unit, the cooling capacity of the secondary circuit, the opening status of major valves, or the inlet and outlet temperature settings of the coolant.

[0082] Before implementing perturbations, a set of boundary conditions can be selected based on the current operating conditions, and relevant control commands can be frozen during the perturbation period. This avoids valve switching, heat exchange mode switching, or other large-scale adjustment behaviors occurring simultaneously, thus ensuring that the loop responds to the controlled perturbation input under relatively fixed external constraints. The controlled perturbation input manifests as a preset perturbation sequence applied to at least one operating parameter of the liquid cooling loop. This perturbation sequence can be a step change with limited amplitude, a pulse change that recovers after a short period of time, or a small fluctuation with a fixed rhythm. Their common characteristic is that the perturbation form and duration are determined before application. For example, when detecting the cooling area of ​​a module, a short-term, small-amplitude perturbation can be applied to the circulating pump speed while keeping the heat exchanger operating conditions and branch valve positions unchanged. This causes identifiable changes in the loop pressure and flow rate. Since other boundary conditions remain stable, this change can more directly reflect the fluid response characteristics of the liquid cooling loop itself, thus providing a clear and reliable data foundation for subsequent dynamic feature extraction and risk assessment.

[0083] In one embodiment of the present invention, during the dynamic response process, at least one fluid response data of the measured region is simultaneously acquired, specifically including:

[0084] S205. During the period of applying the controlled perturbation input and within the preset observation period after the perturbation ends, the measured area is synchronously observed according to the sampling period corresponding to the controlled perturbation input, and fluid response data related to the fluid transport state is collected; wherein, the fluid response data includes at least one of pressure-related response data, flow-related response data, and temperature difference-related response data, and the fluid response data is temporally aligned with the perturbation sequence of the controlled perturbation input.

[0085] During the application of controlled perturbation input, in order to fully and accurately characterize the dynamic response of the liquid cooling circuit, it is necessary to synchronously observe the measured area and collect fluid response data related to the fluid transport state during the perturbation period and within a preset observation period after the perturbation ends. To achieve this, the sampling period must first be determined based on the type of controlled perturbation input used, ensuring that the sampling frequency covers the main dynamic changes caused by the perturbation. Since controlled perturbations typically have a defined time structure, such as a known step moment or duration, the sampling clock can be unified before the perturbation begins, and the sampling rhythm can be kept consistent throughout the perturbation and subsequent recovery phases, thus ensuring the continuity and comparability of the collected data in the time dimension. Based on this, parameters directly related to fluid transport within the measured area are synchronously collected. These parameters may include at least one of pressure-related response data, flow-related response data, and temperature difference-related response data reflecting heat transfer results. The specific selection depends on the existing monitoring conditions of the system but does not affect the overall implementation logic of the method.

[0086] To ensure that subsequent analysis accurately corresponds to disturbance behavior, the collected fluid response data needs to be aligned temporally with the disturbance sequence of the controlled micro-perturbation input. This means that the response data at each moment can clearly correspond to the stage of the disturbance input, such as the pre-disturbance steady state, the disturbance rise phase, the disturbance hold phase, and the recovery phase after the disturbance ends. In practice, this can be achieved by unifying timestamps or recording the trigger time and duration of the disturbance input in the control system, thus mapping the response data sequence and the disturbance sequence on the same time axis. The preset observation period after the disturbance ends is also important because liquid cooling circuits often undergo a recovery process after the disturbance is removed. The changes in pressure, flow rate, or temperature difference during this process can reflect the energy storage and dissipation characteristics of the circuit, which is particularly useful for distinguishing between normal states and states with leakage risks.

[0087] For example, in a lithium battery liquid cooling system, when a small, controlled perturbation lasting several seconds is applied to the circulating pump speed, the pump outlet pressure and return manifold temperature can be synchronously collected at a fixed sampling period. These data are continuously recorded during the start of the perturbation, its maintenance, and the recovery phase after the perturbation ends. Because the sampling period matches the perturbation rhythm, the resulting pressure and temperature sequences clearly reflect the dynamic response pattern caused by the perturbation and correspond one-to-one with the perturbation sequence in time. This provides a reliable data foundation for subsequent extraction of dynamic features and assessment of fluid characteristic changes in the measured area.

[0088] S300. Extract dynamic features characterizing the equivalent fluid properties of the liquid cooling circuit from the fluid response data, measure the difference between the dynamic features and the baseline features corresponding to the baseline network, and output the leakage risk detection result.

[0089] In one embodiment of the present invention, dynamic features characterizing the equivalent fluid properties of the liquid cooling circuit are extracted from the fluid response data, and the difference between the dynamic features and the baseline features corresponding to the baseline network is measured, specifically including:

[0090] S301. Based on the fluid response data and the perturbation sequence of the controlled perturbation input, extract at least one type of feature quantity reflecting dynamic response speed, at least one type of feature quantity reflecting steady-state change, and at least one type of feature quantity reflecting dynamic hysteresis relationship as dynamic features.

[0091] Because controlled perturbations have definite start times and change patterns, fluid response data can be divided into three phases according to the perturbation sequence: before the perturbation, during the perturbation, and the recovery phase after the perturbation ends. The response morphology of each phase can then be analyzed separately. Characteristic quantities reflecting the dynamic response speed are typically obtained by measuring the time or rate of change required for the fluid response to reach a new level from its initial state. Examples include the time required for pressure or flow rate to reach a certain percentage change after the perturbation begins, or the slope of change per unit time. These characteristics reflect how quickly the loop responds to a perturbation.

[0092] Characteristic quantities used to reflect steady-state changes focus on the level of fluid response after the disturbance stabilizes, such as the average pressure change, flow rate change, or temperature difference change during the disturbance holding phase or before the disturbance ends, to characterize the response amplitude of the loop in the new equilibrium state.

[0093] The characteristic quantity used to reflect the dynamic hysteresis relationship is obtained by comparing the time offset between the moment when the controlled perturbation input changes and the moment when the fluid response changes significantly, or by calculating the maximum correlation hysteresis between the input and output sequences to describe the system's delay characteristics.

[0094] For example, when a short-term disturbance is applied to the speed of a circulating pump, the pressure response in a normal liquid cooling circuit usually changes significantly within a short period of time and quickly stabilizes. However, when there is a risk of leakage in the measured area, the pressure response may rise more slowly, the steady-state change amplitude may decrease, and there may be a more significant lag relative to the change in pump speed. These differences can be quantified into dynamic characteristics through dynamic response speed characteristics, steady-state change characteristics, and dynamic lag characteristics, which can be used for subsequent risk assessment.

[0095] S302. Normalize the dynamic features and calculate the difference metric value of the dynamic features relative to the baseline features. The difference metric value is a weighted difference value, a distance metric value, or a combination of the two.

[0096] The normalized benchmark can be derived from a historical stable operating phase or a set of baseline features that matches the current operating conditions. This maps each dynamic feature to a uniform numerical scale, for example, by converting feature values ​​into deviations from the baseline mean or by linearly scaling them according to the range of variation of the baseline features. In this way, features reflecting response speed, steady-state changes, and dynamic hysteresis, even if they differ in physical sense, can be comprehensively evaluated within the same feature space.

[0097] After normalization, the difference between the dynamic features detected by the current perturbation and the corresponding baseline features can be measured. This difference measure quantifies the deviation between the current response and the normal state. The difference measure can be implemented using a weighted difference value, which involves multiplying each normalized feature difference by a preset weight and then summing the results, giving features with higher leakage sensitivity a greater weight in the overall evaluation. Alternatively, a distance metric can be used, where the distance between the current dynamic feature vector and the baseline feature vector is used as the difference measure value, reflecting the overall shift in the feature space. If necessary, the weighted difference value and the distance metric can be combined to account for changes in individual key features and alterations in the overall feature structure.

[0098] For example, in a normal liquid cooling circuit, the pressure response speed, steady-state change amplitude, and lag time after pump speed disturbance are close to the baseline characteristics, and the normalized difference metric is small. However, when there is a risk of leakage in the measured area, the response speed slows down, the steady-state change weakens, and the lag time prolongs. These characteristics deviate significantly from the baseline characteristics after normalization, making the calculated difference metric significantly larger, thus providing a quantitative basis for the determination of leakage risk.

[0099] After calculating the difference between dynamic characteristics and baseline characteristics, the difference measurement result can serve as a direct quantitative basis for determining whether there is a risk of leakage in the tested area, thus outputting the leakage risk detection result. In implementation, the currently obtained difference measurement value needs to be compared with the corresponding judgment reference. This reference can be determined by the statistical results of difference measurements during historical stable operation phases or by a pre-set allowable deviation range, ensuring the judgment process is based on repeatability and interpretability. When the difference measurement value is within the reference range, it indicates that the dynamic response of the tested area under controlled perturbation remains consistent with the normal state, and the equivalent fluid characteristics of the liquid cooling circuit have not changed significantly. Therefore, the current leakage risk can be determined to be at a low level. When the difference measurement value exceeds the reference range and remains at a high level during continuous or repeated detection, it indicates that the tested area has a stable deviation from the baseline state in terms of response speed, steady-state changes, or dynamic hysteresis. This deviation is difficult to explain solely by random fluctuations or short-term operating condition changes, thus indicating a leakage risk in the area. To improve the reliability of the results, the output of leakage risk detection results typically includes not only a judgment on the existence of risk, but also a classification of risk level based on the magnitude of the difference metric, such as distinguishing between a warning level and a high-risk level, and correlating the results with the corresponding spatial region. For example, in a lithium battery liquid cooling system, if a module's cooling area shows a significantly increased difference metric in multiple perturbation detections, the system can output a detection result indicating a high leakage risk in that module area, which can be used for subsequent early warning or maintenance decisions.

[0100] Please refer to the preceding description and other relevant materials. Figure 4 and Figure 5 , Figure 4 This is a physical diagram illustrating an application scenario of the liquid cooling system in this solution. Figure 5 This is a comparison of the fluid dynamic response in the measured region under controlled perturbation input. Specifically, Figure 5 This illustration shows a comparison of the fluid dynamic response of the liquid cooling circuit when a controlled perturbation input is applied to the test area in one embodiment.

[0101] like Figure 5 As shown, the horizontal axis represents time (seconds), indicating the entire observation process from before the application of the controlled perturbation input, during the application, until the end of the perturbation; the vertical axis represents the normalized values, used to represent the relative changes of the controlled perturbation input and fluid response data on a uniform scale. The "normalized values" refer to the dimensionless representation of the actual physical quantities (such as pump speed, pressure, or flow rate) after scaling them relative to the reference steady-state values, so that different physical quantities can be compared in the same coordinate system.

[0102] The curve represented by line 1 (i.e., the dashed line) in the figure represents the controlled perturbation input, which indicates a preset perturbation sequence applied to at least one operating parameter of the liquid cooling circuit (e.g., the speed of the circulating pump) under the condition of satisfying the thermal safety constraints of the lithium battery system. The perturbation exhibits a limited change in amplitude within a preset time period and returns to the original steady-state level after the perturbation ends.

[0103] The two curves represented by solid lines in the figure correspond to the fluid response of the measured region under different conditions, where:

[0104] Line 2 (i.e., the orange line) represents the fluid response of the tested area to the controlled perturbation input when the liquid cooling circuit is properly sealed.

[0105] Line 3 (the green line) represents the fluid response of the measured area to the same controlled perturbation input when there is a risk of leakage in the measured area.

[0106] The fluid response data can be any one of pressure-related response data, flow-related response data, or temperature difference-related response data, or a combination thereof. In this embodiment, a single type of fluid response is used as an example for illustration.

[0107] from Figure 5 It can be seen that, under the same controlled perturbation input conditions, the fluid response with leakage risk exhibits at least one of the following differences compared to the normal state: a slower response rise rate, a smaller steady-state change amplitude, and a prolonged recovery process after the perturbation ends. These differences reflect that leakage risk alters the equivalent fluid characteristics of the liquid cooling circuit, thereby causing a stable deviation in the dynamic response pattern from the normal state.

[0108] By measuring the difference between this dynamic response feature and the baseline feature corresponding to the baseline network, the risk of leakage in the tested area can be effectively identified.

[0109] The above describes the method for detecting leakage in a liquid cooling system according to embodiments of the present invention. The following describes the device for detecting leakage in a liquid cooling system according to embodiments of the present invention. Please refer to [link / reference]. Figure 2 One embodiment of the liquid cooling system leakage detection device of the present invention includes:

[0110] The data acquisition module 101 is used to acquire multi-node operating data reflecting the fluid transport status of the liquid cooling circuit during the operation of the liquid cooling circuit, and to construct a baseline network characterizing the normal coupling relationship of the liquid cooling circuit based on the multi-node operating data. The multi-node operating data includes at least two types of data: pressure data, flow data, and temperature data.

[0111] The controlled perturbation module 102 is used to determine the test area corresponding to the local area when the baseline network has a structural anomaly in a local area, and to apply a preset controlled perturbation input to the test area under the condition of meeting the thermal safety constraints of the lithium battery, so that the liquid cooling circuit forms an identifiable dynamic response process, and simultaneously collect at least one fluid response data of the test area.

[0112] The detection result output module 103 is used to extract dynamic features characterizing the equivalent fluid properties of the liquid cooling circuit from the fluid response data, measure the difference between the dynamic features and the baseline features corresponding to the baseline network, and output the leakage risk detection result.

[0113] above Figure 2 The liquid cooling system leakage detection device in this embodiment of the invention is described in detail from the perspective of modular functional entities. The liquid cooling system leakage detection device in this embodiment of the invention is described in detail from the perspective of hardware processing.

[0114] Figure 3This is a schematic diagram of a liquid cooling system leakage detection device 200 provided in an embodiment of the present invention. The liquid cooling system leakage detection device 200 can vary significantly due to different configurations or performance characteristics. It may include one or more processors 210 (e.g., one or more processors) and a memory 220, and one or more storage media 230 (e.g., one or more mass storage devices) storing application programs 233 or data 232. The memory 220 and storage media 230 can be temporary or persistent storage. The program stored in the storage media 230 may include one or more modules (not shown in the diagram), each module including a series of instruction operations on the liquid cooling system leakage detection device 200. Furthermore, the processor 210 may be configured to communicate with the storage media 230 and execute the series of instruction operations in the storage media 230 on the liquid cooling system leakage detection device 200 to implement the steps of the above-described liquid cooling system leakage detection method.

[0115] The liquid cooling system leakage detection device 200 may also include one or more power supplies 240, one or more wired or wireless network interfaces 250, one or more input / output interfaces 260, and / or one or more operating systems 231, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will understand that... Figure 3 The illustrated liquid cooling system leakage detection device structure does not constitute a limitation on the liquid cooling system leakage detection device provided by the present invention. It may include more or fewer components than illustrated, or combine certain components, or have different component arrangements.

[0116] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when the instructions are executed on a computer, cause the computer to perform the steps of the liquid cooling system leakage detection method.

[0117] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system, device, or unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0118] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0119] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural transformations made using the contents of the present invention's specification and drawings under the inventive concept of the present invention, or direct / indirect applications in other related technical fields, are included within the patent protection scope of the present invention.

Claims

1. A method for detecting a liquid leakage of a liquid cooling system, characterized by, Includes the following steps: During the operation of the liquid cooling circuit, multi-node operational data reflecting the fluid transport status of the liquid cooling circuit are collected. A baseline network characterizing the normal coupling relationship of the liquid cooling circuit is constructed based on this multi-node operational data. The multi-node operational data includes at least two of the following: pressure data, flow data, and temperature data; specifically: Multiple nodes are defined in multiple spatial regions of the liquid cooling circuit. The multiple spatial regions include at least a cooling region corresponding to the lithium battery module, a manifold region corresponding to the supply and return liquid collection structure, and a pump and valve region corresponding to the power components of the liquid cooling circuit. Within the sliding time window, at least two types of operating parameters with different physical meanings are collected for each node as node operating data, and the node operating data is processed for time alignment, outlier removal, and short-term fluctuation suppression. The effective data segment is determined based on the operating condition information within the sliding time window. The effective data segment satisfies the preset operating condition consistency condition. The operating condition consistency condition includes at least that the liquid cooling circuit is in a stable cycle state and there are no preset start-stop transients or valve switching transients. Based on the node operation data within the valid data segment, construct a node state vector; Within the effective data segment, the association strength between any two nodes is calculated based on the node state vector of each node. The association strength includes the correlation coefficient, mutual information, or a combination of both. A set of statistics is determined for all association strengths within the effective data segment, and the set of statistics includes at least the mean and dispersion of the association strengths; An uncertainty is determined based on the node state vector, and the uncertainty is used to characterize the volatility or uncertainty of the node state vector within the effective data segment. An adaptive criterion threshold is constructed using the set of statistics and the uncertainty. The adaptive criterion threshold includes at least a basic threshold term consisting of the mean and dispersion of the correlation strength, and a modulation term that modulates the basic threshold term using the uncertainty. The nodes that satisfy the adaptive criterion threshold are determined as the network edges of the baseline network, or the association strength is normalized and weighted according to the adaptive criterion threshold and then used as the edge weights of the baseline network to form the baseline network. When the baseline network exhibits structural anomalies in a local area, the test area corresponding to the local area is determined, and a preset controlled perturbation input is applied to the test area under the condition of satisfying the thermal safety constraints of the lithium battery, so that the liquid cooling circuit forms an identifiable dynamic response process, and at least one fluid response data of the test area is collected simultaneously. Dynamic features characterizing the equivalent fluid properties of the liquid cooling circuit are extracted from the fluid response data. The difference between the dynamic features and the baseline features corresponding to the baseline network is measured, and the leakage risk detection result is output.

2. The method of claim 1, wherein, When the baseline network exhibits structural anomalies in a local region, determining the test region corresponding to the local region specifically includes: Within the sliding time window, network structure parameters of the baseline network are acquired, and regional risk metrics for each spatial region are calculated based on the network structure parameters; wherein, the network structure parameters include at least one of the following: changes in the number of network edges, changes in edge weights, or increases in node uncertainty. The regional risk measurement of each spatial region is compared with the corresponding regional baseline value. When the regional risk measurement of at least one spatial region meets the preset triggering condition, the spatial region is determined to be the measured region.

3. The method of claim 2, wherein the step of detecting a liquid leak in the liquid cooling system comprises: Under the condition of meeting the thermal safety constraints of lithium batteries, a preset controlled perturbation input is applied to the tested area to enable the liquid cooling circuit to form a identifiable dynamic response process, specifically including: The system acquires the operating condition information of the liquid cooling circuit and determines the disturbance amplitude and duration of the controlled perturbation input based on the operating condition information, so that the controlled perturbation input meets the preset thermal safety constraints during the disturbance period. The thermal safety constraints include at least one of the target temperature range constraint or the temperature rise rate constraint of the lithium battery system. During the application of the controlled perturbation input, at least one boundary condition of the liquid cooling circuit is kept constant or in a preset stable state; wherein the controlled perturbation input is a preset perturbation sequence applied to at least one operating parameter of the liquid cooling circuit.

4. The method of claim 1, wherein, The synchronous acquisition of at least one fluid response data of the measured area specifically includes: During the period of the applied controlled perturbation input and within a preset observation period after the perturbation ends, the measured area is synchronously observed according to the sampling period corresponding to the controlled perturbation input, and fluid response data related to the fluid transport state is collected; wherein, the fluid response data includes at least one of pressure-related response data, flow-related response data, and temperature difference-related response data, and the fluid response data is temporally aligned with the perturbation sequence of the controlled perturbation input.

5. The method of claim 4, wherein the step of detecting a liquid leak in the liquid cooling system comprises: Dynamic features characterizing the equivalent fluid properties of the liquid cooling circuit are extracted from the fluid response data. The difference between these dynamic features and the baseline features corresponding to the baseline network is measured, specifically including: Based on the fluid response data and the perturbation sequence of the controlled perturbation input, at least one type of feature quantity reflecting dynamic response speed, at least one type of feature quantity reflecting steady-state change, and at least one type of feature quantity reflecting dynamic hysteresis relationship are extracted as dynamic features. The dynamic features are normalized, and a difference metric is calculated between the dynamic features and the baseline features. The difference metric is a weighted difference metric, a distance metric, or a combination of both.

6. A device for detecting leakage in a liquid cooling system, characterized in that, include: The data acquisition module is used to collect multi-node operating data reflecting the fluid transport status of the liquid cooling circuit during operation, and to construct a baseline network characterizing the normal coupling relationship of the liquid cooling circuit based on the multi-node operating data. The multi-node operating data includes at least two types of data: pressure data, flow data, and temperature data. Specifically, it includes: identifying multiple nodes in multiple spatial regions of the liquid cooling circuit, including at least a cooling region corresponding to the lithium battery module, a manifold region corresponding to the supply and return fluid collection structure, and a pump and valve region corresponding to the power components of the liquid cooling circuit; collecting at least two types of operating parameters with different physical meanings as node operating data for each node within a sliding time window, and performing time alignment, outlier removal, and short-term fluctuation suppression processing on the node operating data; determining valid data segments based on the operating condition information within the sliding time window, wherein the valid data segments meet preset operating condition consistency conditions, which at least include the liquid cooling circuit being in a stable circulation state and the absence of preset start-stop transients or valve switching. Transient; Based on the node operation data within the effective data segment, construct node state vectors; Within the effective data segment, calculate the correlation strength between any two nodes based on the node state vectors of each node, the correlation strength including correlation coefficient, mutual information, or a combination of both; Determine a set of statistics for all correlation strengths within the effective data segment, the set of statistics including at least the mean and dispersion of the correlation strength; Determine an uncertainty based on the node state vectors, the uncertainty being used to characterize the volatility or uncertainty of the node state vectors within the effective data segment; Construct an adaptive criterion threshold using the set of statistics and the uncertainty, the adaptive criterion threshold including at least a basic threshold term composed of the mean and dispersion of the correlation strength, and a modulation term modulating the basic threshold term using the uncertainty; Determine the node pairs that satisfy the adaptive criterion threshold as network edges of the baseline network, or normalize and weight the correlation strength according to the adaptive criterion threshold as the edge weights of the baseline network to form the baseline network; The controlled perturbation module is used to determine the test area corresponding to the local area when the baseline network has a structural anomaly in a local area, and to apply a preset controlled perturbation input to the test area under the condition of meeting the thermal safety constraints of the lithium battery, so that the liquid cooling circuit forms an identifiable dynamic response process, and simultaneously collect at least one fluid response data of the test area. The detection result output module is used to extract dynamic features characterizing the equivalent fluid properties of the liquid cooling circuit from the fluid response data, measure the difference between the dynamic features and the baseline features corresponding to the baseline network, and output the leakage risk detection result.

7. A device for detecting leakage in a liquid cooling system, characterized in that, The liquid cooling system leakage detection device includes: a memory and at least one processor, wherein the memory stores instructions; The at least one processor invokes the instructions in the memory to cause the liquid cooling system leakage detection device to perform the steps of the liquid cooling system leakage detection method as described in any one of claims 1 to 5.

8. A computer-readable storage medium having stored thereon instructions, the computer-readable storage medium comprising: When the instruction is executed by the processor, it implements the steps of the liquid cooling system leakage detection method as described in any one of claims 1 to 5.

Images