Vanadium flow battery stack assembly leak detection method and system

Abstract

The application relates to the field of vanadium flow batteries, and discloses a vanadium flow battery stack assembly leak detection method and system, the method comprising the following steps: obtaining reference fluid and chemical parameter distribution in the stack by constructing a multi-level fluid mechanics model and a chemical reaction kinetics model; collecting and monitoring pressure, flow rate, vanadium ion concentration, oxidation-reduction potential, pH value and other chemical parameters in the stack in real time; performing multi-dimensional feature fusion on the collected data to generate a data matrix; and using a deep learning model of a convolutional neural network and a long short-term memory network to perform abnormal pattern recognition on the data matrix to accurately locate a leakage position. The application realizes comprehensive monitoring of fluid and chemical parameters in the stack, can quickly respond to the detection requirement of a small leakage, has high precision and real-time performance, and provides effective technical support for safe operation of the vanadium flow battery.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of vanadium flow battery, in particular to a vanadium flow battery stack assembly leak detection method and system. BACKGROUND

[0002] In the existing vanadium flow battery technology, the sealing performance of the stack assembly is crucial, because a small leakage can lead to electrolyte loss, stack performance degradation, and even safety hazards. The existing leak detection methods usually rely on single monitoring of fluid parameters or simple sensor detection, lacking deep analysis of the flow characteristics, pressure changes and abnormal fluctuations of chemical components of the fluid in different areas of the stack, resulting in low accuracy of leak detection. At the same time, the existing methods are difficult to effectively capture the changes of chemical parameters in the electrolyte of the vanadium flow battery, making it difficult to identify the changes of electrolyte composition in a timely manner, affecting the sensitivity of leak detection.

[0003] In addition, the existing technology has limited intelligence in data analysis, mostly relying on basic data comparison, lacking deep mining of time series characteristics and multi-dimensional characteristics of data, and being difficult to identify small leaks in complex working conditions in real time. Since the vanadium flow battery stack is affected by temperature, pressure and other factors in actual work, the traditional method is often difficult to adapt to the leak detection needs under different operating conditions, resulting in poor reliability of leak detection. SUMMARY

[0004] In view of the deficiencies of the prior art, the present application provides a vanadium flow battery stack assembly leak detection method and system, which solves the problems of low precision, slow response and inability to effectively monitor the changes of electrolyte chemical components in the vanadium flow battery stack leak detection of the prior art.

[0005] To achieve the above purpose, the present application realizes the following technical scheme: a vanadium flow battery stack assembly leak detection method, comprising the following steps:

[0006] A multi-level fluid mechanics model is established based on the overall fluid flow state of the vanadium flow battery stack, which includes fluid models of macroscopic, mesoscopic and microscopic layers;

[0007] The reference pressure, velocity and flow data of the electrolyte at different positions are obtained by numerical simulation;

[0008] The pressure and flow data of each key position of the stack are collected in real time, and the real-time data are compared with the reference data to detect whether there is abnormal pressure or flow fluctuation;

[0009] When abnormal fluctuation is detected, the chemical parameters in the electrolyte are monitored in real time, and the influence of leakage is analyzed by a chemical reaction kinetics model;

[0010] The collected fluid and chemical parameters are fused in multiple dimensions to form a multi-dimensional data matrix.

[0011] An abnormal pattern recognition is performed on the multi-dimensional data matrix using a deep learning model to determine a possible leakage position.

[0012] Preferably, the macroscopic fluid model describes the relationship between pressure, velocity and height of the fluid at different positions based on the principle of energy conservation, and determines the pressure and velocity distribution at different positions by Bernoulli's theorem.

[0013] Preferably, the mesoscopic fluid model is based on the permeability of the fluid in the porous medium, and models the permeation behavior at the seal and interface of the stack by Darcy's law to estimate the local permeation flow rate.

[0014] Preferably, the microscopic fluid model assumes that the leakage path is a fine capillary tube, and describes the flow rate and permeation flow rate of the electrolyte in the capillary or micro-crack based on the Poiseuille flow formula.

[0015] Preferably, the step of obtaining the reference pressure, velocity and flow rate data of the electrolyte at different positions by numerical simulation comprises:

[0016] The fluid mechanics model is solved by numerical simulation software. First, the calculation grid is divided in the vanadium flow battery stack, and the entire fluid region is discretized based on the density, viscosity, inlet and outlet boundary conditions of the fluid;

[0017] At the macroscopic level of the stack, the energy conservation equation based on the fluid density, velocity and height is solved, and the distribution relationship of pressure and flow rate is established by Bernoulli's theorem satisfied by the electrolyte density, flow rate and pressure at different positions of the stack, and the overall flow rate and pressure distribution of the electrolyte at different positions are calculated;

[0018] At the mesoscopic level, the local permeation flow rate is calculated at the seal or interface using the Darcy's law model, and the pressure and flow rate distribution of the seal or interface region is calculated based on the permeability coefficient, fluid viscosity and flow path length;

[0019] At the microscopic level, the leakage path is assumed to have capillary structure characteristics, and the flow rate and permeation flow rate in the micro-crack are calculated by the Poiseuille flow formula to establish the reference flow rate and flow rate distribution of the microscopic flow region;

[0020] The flow results of each level are integrated to obtain the reference fluid pressure, velocity and flow rate data distribution of the vanadium flow battery stack under normal working conditions, and the reference data is stored in a database for subsequent comparison and analysis with real-time monitoring data.

[0021] Preferably, the step of real-time monitoring the chemical parameters in the electrolyte when abnormal fluctuation is detected, and analyzing the leakage influence through a chemical reaction kinetics model comprises:

[0022] The concentration of vanadium ions in the electrolyte is measured using a vanadium ion selective electrode, and the initial reference concentration is recorded;

[0023] A differential equation of concentration change is established according to a vanadium ion redox reaction kinetics model, and the kinetics model includes oxidation and reduction reaction rate constants;

[0024] The rate constants in the redox kinetics model are used to calculate the rate of change of vanadium ion concentration at different temperatures, and the temperature dependence of the rate constants is described by the Arrhenius equation;

[0025] By real-time monitoring the trend of vanadium ion concentration change, abnormal values of concentration fluctuation are identified and the influence of leakage is judged;

[0026] The detected vanadium ion concentration fluctuation data is input into a computing system to judge the degree of concentration change, to identify the concentration change trend caused by leakage and determine the type of leakage component.

[0027] Preferably, the step of real-time monitoring the chemical parameters in the electrolyte when abnormal fluctuation is detected, and analyzing the leakage influence through a chemical reaction kinetics model further comprises:

[0028] Real-time acquisition of redox potential data of the electrolyte, which is collected by a redox potential sensor;

[0029] Based on the concentration ratio of vanadium ions, the redox potential of the electrolyte is calculated by the Nernst equation, and the potential calculation in the Nernst equation includes the standard electrode potential at room temperature and the logarithmic relationship of vanadium ion concentration;

[0030] When the redox potential fluctuation is detected to exceed a preset threshold, the fluctuation data is recorded as abnormal fluctuation, and it is judged whether the fluctuation indicates a leakage of vanadium electrolyte;

[0031] The redox potential data of abnormal fluctuation is collected and trend analyzed multiple times to confirm whether the fluctuation of redox potential is continuously and stably within the abnormal range, to further judge the chemical component change of the leakage position.

[0032] Preferably, the step of multi-dimensional feature fusion of the collected fluid, chemical parameters to form a multi-dimensional data matrix comprises:

[0033] The collected real-time pressure, flow data and electrolyte chemical parameter data are segmented according to time sequence to form a basic data set with time sequence characteristics, and the data set is normalized;

[0034] The normalized data is applied to convolution operation to extract a spatial feature matrix to obtain feature distribution under different positions and chemical states;

[0035] The time sequence variation of the feature matrix is processed by using a long short-term memory network to form a multi-dimensional data matrix that integrates time sequence characteristics.

[0036] Preferably, the step of using a deep learning model to perform abnormal pattern recognition on the multi-dimensional data matrix to determine the possible leakage position comprises:

[0037] The multi-dimensional data matrix is input into the convolution layer of the deep learning model, and the input data is convolved by multiple convolution kernels to generate an initial feature map, and a pooling operation is performed;

[0038] The long short-term memory network is used to perform time sequence processing on the pooled feature map to form a time sequence feature vector;

[0039] The time sequence feature vector is processed by using a classification layer to generate a leakage probability distribution for each position;

[0040] The most likely leakage position is determined according to the leakage probability distribution of each position, and a leakage warning signal is issued when the leakage probability exceeds a preset threshold.

[0041] The application also provides a vanadium flow battery stack assembly leak detection system, comprising:

[0042] A fluid mechanics modeling module is used to establish a multi-level fluid model of the vanadium flow battery stack to describe the flow characteristics of the electrolyte;

[0043] A numerical simulation module is used to simulate the normal state of the fluid inside the stack based on the fluid mechanics model and generate reference fluid data;

[0044] A sensor module includes multiple pressure, flow and chemical sensors for real-time collection of fluid, vanadium ion concentration, redox potential and pH data of the stack;

[0045] A data processing module is used to compare the real-time monitoring data and the reference data, and input the chemical data of the abnormal fluctuation position into a deep learning model;

[0046] A deep learning module is used to analyze the multi-dimensional feature data matrix, and perform leakage detection and positioning by using a convolutional neural network and a long short-term memory network.

[0047] This invention provides a method and system for leak detection during vanadium redox flow battery stack assembly. It offers the following advantages:

[0048] 1. By constructing multi-level fluid dynamics and chemical reaction kinetics models, this invention can comprehensively describe the fluid and chemical behavior inside the vanadium redox flow battery stack at the macroscopic, mesoscopic, and microscopic levels, ensuring sensitive detection of minute leaks inside the vanadium redox flow battery stack and significantly improving detection accuracy.

[0049] 2. The method of the present invention utilizes multi-point sensors to collect pressure, flow and chemical parameters in real time, and combines them with numerical simulation data for comparative analysis, thereby realizing real-time monitoring of leakage, ensuring timely identification and response in the early stage of leakage, and helping to prevent the further expansion of leakage risk.

[0050] 3. This invention generates a comprehensive multidimensional data matrix by fusing multidimensional features of fluid and chemical parameters, which helps deep learning models extract spatial and temporal features from complex data, achieving highly robust detection that is not affected by fluctuations or anomalies in single data.

[0051] 4. This invention employs a deep learning model combining convolutional neural networks and long short-term memory networks to perform joint analysis of the spatial and temporal features of the data matrix. This enables precise identification of possible leak locations, thereby improving the accuracy of location and providing a direct reference for leak repair.

[0052] 5. By detecting and accurately locating leaks in the early stages, this invention can effectively reduce the safety risks caused by leaks, extend the service life of the battery stack, improve the operational stability of the battery system, and provide technical support for the safe application of vanadium redox flow batteries. Attached Figure Description

[0053] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0054] Figure 2 This is a schematic diagram of the system structure of the present invention.

[0055] Among them, 100 is the fluid dynamics modeling module; 200 is the numerical simulation module; 300 is the sensor module; 400 is the data processing module; and 500 is the deep learning module. Detailed Implementation

[0056] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0057] The accompanying drawings are incorporated in and constitute a part of this specification and will be understood by those skilled in the art to be for illustration purposes only. Figure 1 The embodiment of the present application provides a vanadium flow battery stack assembly leak detection method, comprising the following steps:

[0058] S1, a multi-level fluid mechanics model is established based on the overall fluid flow state of the vanadium flow battery stack

[0059] Firstly, the vanadium flow battery stack assembly leak detection method of the present application is based on the internal electrolyte flow state of the vanadium flow battery stack, and a multi-level fluid mechanics model is established, so as to describe the flow characteristics of the electrolyte inside the stack from the macroscopic layer, the mesoscopic layer and the microscopic layer, and to provide accurate benchmark data for subsequent anomaly detection.

[0060] In this embodiment, a macroscopic layer fluid model is first constructed to describe the overall flow state of the electrolyte inside the vanadium flow battery stack. The macroscopic layer model is based on the principle of energy conservation, and the flow of the electrolyte is regarded as a steady incompressible fluid flow, and it is assumed that the electrolyte in the stack follows Bernoulli's theorem. Therefore, in this macroscopic layer model, there is a constant energy conservation relationship between the fluid pressure, velocity and height at different positions. Specifically, according to Bernoulli's theorem, the total energy of the fluid remains unchanged, which can be described as:

[0061]

[0062] Where g is the fluid pressure, p is the electrolyte density, v is the flow rate, p is the acceleration of gravity, and h is the height. In the specific calculation, the fluid pressure, velocity and height at different positions inside the vanadium flow battery stack are substituted into the Bernoulli equation to obtain the fluid parameter distribution at each position. As an option, numerical simulation software (such as ANSYS Fluent or COMSOL Multiphysics) is used to solve the macroscopic layer model to obtain the overall flow rate and pressure distribution of the electrolyte in the stack.

[0063] In this embodiment, a mesoscopic layer fluid model is further established to describe the local permeation behavior at the sealing and interface of the stack. Generally, the flow at the sealing and interface is closely related to the permeability of the fluid, so Darcy's law is used in the mesoscopic layer to model the permeation behavior at the sealing and interface. Specifically, it is assumed that there may be small-scale leakage near the sealing, and Darcy's law is used to evaluate the fluid distribution at this place by calculating the permeation flow rate of the sealing area. Darcy's law is as follows:

[0064]

[0065] Where Q is the permeation flow rate, j is the permeability coefficient, A is the flow cross-sectional area, μ is the fluid viscosity, Δp is the pressure difference between the two ends, and L is the flow path length. Using Darcy's law, a fluid model is established at the vanadium redox flow battery stack seals and interfaces to obtain the fluid pressure and flow rate distribution in these areas. Alternatively, appropriate boundary conditions are applied in the numerical simulation to simulate potential leakage at the seals and accurately obtain flow rate and pressure data for these areas.

[0066] At the microscopic level in this embodiment, it is assumed that there are tiny cracks or capillary structures within the fuel cell stack to describe the flow within these capillaries or microcracks. The fluid model at the microscopic level is based on Poiseuille's flow equations, assuming that the leakage path is a slender capillary structure in which the electrolyte exhibits laminar flow, suitable for description by Poiseuille's law. Specifically, the Poiseuille flow formula is as follows:

[0067]

[0068] Among them, Q capillary Let denoted by , r be the flow rate in the capillary, Δp be the capillary radius, Δp be the pressure difference between the two ends, μ be the fluid viscosity, and L be the capillary length. Poiseuille's flow formula is used to calculate the flow distribution in microcracks or capillary structures and to obtain the permeation velocity in the microlayer. Generally, the flow velocity in a capillary structure is constrained by the channel radius and the fluid viscosity. After determining the geometric parameters of the capillary channel, the permeation flow rate can be obtained using Poiseuille's equation.

[0069] Alternatively, in some embodiments, the three-layer model is integrated to systematically analyze the fluid flow within the entire fuel cell stack. Specifically, the overall results of the three-layer fluid model can be summarized using numerical simulation software to obtain the overall flow velocity, pressure, and permeation flow distribution data of the vanadium redox flow battery stack, which serves as baseline data in leak detection methods.

[0070] S2. Obtain reference pressure, velocity, and flow rate data of the electrolyte at different locations through numerical simulation.

[0071] To accurately obtain the fluid state of a vanadium redox flow battery stack under normal operating conditions, this invention constructs a multi-level fluid dynamics model and then calculates the baseline pressure, velocity, and flow rate data at each location using numerical simulation methods. These baseline data will serve as the flow characteristics of the electrolyte during normal operation, providing a reference for comparing subsequent real-time data and thus identifying abnormal pressure or flow rate fluctuations.

[0072] In this embodiment, the grid division and parameter initialization are first performed in the fluid region inside the stack, aiming to discretize the fluid space of the electrolyte and facilitate the subsequent numerical solution. Specifically, a plurality of calculation grids are divided in the flow region of the electrolyte, wherein the grid density varies according to the region characteristics. Generally, the flow characteristics of the electrolyte are relatively complex around the seal, the interface region and the capillary structure, so denser grid division is adopted at these positions to ensure the calculation accuracy. As an option, for the overall flow region of the stack, a moderately dense grid division scheme can be selected to balance between the accuracy and the calculation efficiency.

[0073] After the grid division and parameter initialization are completed, the specific boundary conditions are set according to the fluid properties of the electrolyte and the operating conditions of the stack. Generally, the inlet is set as the known pressure and flow rate boundary conditions, and the outlet is set as the known pressure boundary. Therefore, in this embodiment, the inlet pressure and flow rate are defined as constant values, and the constant pressure is set as the boundary condition at the outlet. When setting the boundary conditions, the permeability and the viscosity of the fluid also need to be considered for the seal and the interface region of the stack to ensure the accurate simulation of the permeation flow.

[0074] In this embodiment, the multi-level fluid mechanics model is numerically solved by the numerical simulation software to calculate the pressure, velocity and flow distribution of the electrolyte at different positions in the stack. At the macro level, the fluid model is based on the energy conservation principle of Bernoulli's theorem, assuming that the electrolyte is a steady incompressible fluid and satisfies the energy conservation relationship under normal operating conditions. Specifically, in the numerical simulation, the pressure, velocity and height of the electrolyte at each position are substituted into the Bernoulli equation to obtain the overall flow rate and pressure distribution. As an option, the finite volume method or the finite element method is used to discretize and solve the model to improve the convergence and accuracy of the simulation results.

[0075] At the mesoscopic level, the fluid model is based on Darcy's law, which is applicable to the permeation fluid behavior at the seal and the interface. Generally, the permeation behavior of the seal region can be described by the permeation flow, which depends on the permeation coefficient, the flow cross-sectional area and the viscosity of the fluid. Therefore, in the numerical simulation, the permeation parameters of the seal are set according to Darcy's law to accurately solve the permeation flow and pressure distribution. Specifically, the permeation coefficient in Darcy's law is a material characteristic parameter, which can be set according to the flow characteristics of the electrolyte near the seal to ensure that the calculation accurately reflects the fluid permeation state of the seal region.

[0076] At the micro level, the fluid model assumes that the flow of electrolyte in microcracks or capillaries is laminar, which meets the Poiseuille flow condition. Specifically, the flow rate in the microcracks or capillaries is calculated according to the Poiseuille flow equation, and the radius of the capillary channel and the viscosity parameter of the fluid are set in the simulation to ensure that the calculated flow rate distribution is consistent with the actual situation. The Poiseuille flow formula is suitable for describing the laminar flow behavior of fluid in capillaries or microcracks, and can accurately simulate the flow characteristics at this level.

[0077] In some embodiments, in order to improve the accuracy of the simulation data, the boundary constraint conditions of different fluid characteristic regions are considered in the numerical simulation. Specifically, the no-slip boundary condition is adopted for the macroscopic layer model to ensure that the flow rate of the electrolyte at the boundary is zero; for the mesoscopic layer permeation model, a suitable pressure gradient is applied to reflect the fluid permeation behavior at the seal and interface; for the microcosmic layer capillary flow model, the pressure difference boundary condition of the capillary is set to ensure that the fluid produces laminar permeation flow along the direction of the microcrack.

[0078] As an option, the pressure, flow rate and velocity data obtained by numerical simulation can be stored in a database to establish a baseline fluid parameter data set of the vanadium flow battery stack, so as to compare the real-time data and detect abnormalities in the subsequent leak detection process.

[0079] S3, real-time acquisition of pressure and flow rate data at each key position of the stack, and comparison of real-time data with baseline data to detect abnormal pressure or flow rate fluctuations

[0080] In order to realize real-time monitoring of the operating state of the vanadium flow battery stack, the method of the present application compares the real-time data with the baseline data on the basis of the baseline fluid parameter data set obtained by numerical simulation. This step aims to quickly identify whether there is an abnormal pressure or flow rate fluctuation in the stack, so as to determine whether there is a leak. By real-time monitoring and comparison with the baseline data, abnormal conditions of the fluid state can be detected in time, providing preliminary judgment basis for subsequent chemical parameter monitoring.

[0081] In this embodiment, high-precision pressure and flow rate sensors are arranged at each key position of the stack to real-time acquire the flow parameters of the electrolyte. Specifically, the sensors are arranged at the inlet, outlet, seal and other nodes that may be at risk of leakage of the electrolyte. Generally, arranging sensors at key areas of the fluid flow path helps to more accurately acquire real-time change data of the fluid, so in this embodiment, nodes with significant flow rate changes and susceptible to leakage are arranged as the focus to ensure the integrity and representativeness of the real-time data.

[0082] In this embodiment, the real-time data collected by the sensors include pressure and flow information at each position. Generally, the sensors need to have sufficient sampling accuracy and response speed to ensure a quick response to changes in fluid parameters, so high-precision and high-sensitivity pressure and flow sensors are used in the selection of sensors. As an option, the collected real-time data is compared with the reference data stored in the database after data preprocessing to identify whether there is abnormal fluctuation.

[0083] When comparing data, the real-time collected fluid parameters are compared with the upper and lower limit ranges of the reference data one by one. If the pressure or flow at a certain position exceeds the preset tolerance range, it is determined that there is abnormal fluctuation at that position. Specifically, the tolerance range of pressure fluctuation is set according to the standard deviation of the pressure distribution and flow of the normal electrolyte flow to ensure tolerance for slight fluctuations in the normal state, but sensitivity to significant fluctuations beyond the reference state. As an option, the real-time pressure signal is subjected to frequency spectrum analysis by data analysis methods such as fast Fourier transform (FFT) or wavelet transform to extract fluctuation characteristics and further verify the pressure abnormal fluctuation at the position.

[0084] In this embodiment, when significant fluctuations in pressure or flow are detected, it is preliminarily determined that the position is a possible leakage area, and the corresponding monitoring data is recorded. Generally, the comparison results of real-time data and reference data will indicate the possible position of the leakage, so the real-time data of the position is stored each time an abnormal fluctuation is detected for subsequent chemical parameter monitoring and detailed analysis.

[0085] S4, when an abnormal fluctuation is detected, real-time monitoring of chemical parameters in the electrolyte and analysis of the impact of leakage by chemical reaction kinetics model

[0086] To further confirm the leakage in the vanadium flow battery stack, the present application real-time monitors the chemical parameters in the electrolyte through chemical sensors after detecting abnormal pressure or flow fluctuations, and analyzes the impact of leakage on the electrolyte composition by combining the chemical reaction kinetics model. Through this step, the dynamic changes of key chemical parameters such as vanadium ion concentration, redox potential and pH value in the electrolyte can be obtained, which provides an important basis for subsequent leakage component analysis and position determination.

[0087] In this embodiment, the real-time monitoring of the key chemical parameters of the electrolyte is first achieved by the chemical sensor arrangement. Specifically, the concentration of vanadium ions, the oxidation-reduction potential (ORP), and the pH value in the electrolyte are collected by the vanadium ion selective electrode, the ORP sensor, and the pH sensor, respectively. Generally, these sensors are arranged near the inlet and outlet of the electrolyte and the sealing element, so as to monitor the changes of the chemical parameters in the area where abnormal fluid fluctuations are detected. Therefore, in this embodiment, the chemical sensors are arranged at the key nodes on the electrolyte flow path to ensure comprehensive monitoring of the leakage impact.

[0088] In this embodiment, the changes of the monitored chemical parameters are analyzed by a chemical reaction kinetics model, and in particular, the concentration change trend of vanadium ions is calculated in real time. Specifically, the oxidation-reduction reaction between vanadium ions in different valence states is regarded as a first-order reaction, and the concentration change rate thereof satisfies the chemical kinetics equation. It is assumed that the concentration change between vanadium divalent and trivalent ions conforms to the following reaction rate equation:

[0089]

[0090] where k ox is the oxidation reaction rate constant, and k red is the reduction reaction rate constant. In some embodiments, the reaction rate constants also need to be corrected according to the temperature to ensure the accuracy of the concentration change under different temperature conditions. Generally, the relationship between the rate constant and the temperature conforms to the Arrhenius equation, so in this model, the rate constant is corrected according to the temperature change to obtain more accurate kinetic model parameters.

[0091] As an option, this embodiment also calculates the oxidation-reduction potential of the electrolyte based on the Nernst equation, and then determines whether the concentration ratio of vanadium ions fluctuates significantly. Specifically, the Nernst equation describes the relationship between the concentration ratio of vanadium divalent and trivalent ions and the oxidation-reduction potential:

[0092]

[0093] where E is the actual oxidation-reduction potential, E 0 is the standard electrode potential, R is the gas constant, T is the temperature, n is the number of electron transfer, and F is the Faraday constant. Generally, when the concentration ratio of vanadium ions fluctuates due to leakage, the oxidation-reduction potential will change significantly. Therefore, by monitoring the change trend of the oxidation-reduction potential in real time, this embodiment determines whether the proportion of vanadium ions in the electrolyte deviates from the normal range.

[0094] In some embodiments, to improve detection accuracy, the present embodiment also monitors the pH value as a supplementary chemical parameter. The pH value change of the vanadium flow battery can reflect the acid-base balance state of the electrolyte, and when the leakage affects the composition of the electrolyte, the pH value may abnormally fluctuate. Specifically, the change of the pH value can be associated with the oxidation-reduction reaction of vanadium ions, and by comprehensively analyzing the change trend of the vanadium ion concentration, the oxidation-reduction potential and the pH value, the chemical composition of the leakage is further confirmed.

[0095] In summary, through real-time monitoring of chemical parameters and analysis of kinetic models, the present embodiment can effectively track the change of the leakage composition, and provide detailed chemical characteristic data for subsequent leakage location determination.

[0096] S5, multi-dimensional feature fusion of the collected fluid and chemical parameters to form a multi-dimensional data matrix

[0097] To improve the accuracy of vanadium flow battery stack leakage detection, the present method generates a multi-dimensional data matrix through feature data fusion based on the obtained fluid and chemical parameters, to provide complete data input for the deep learning model. This step provides high-quality data support for subsequent leakage anomaly pattern recognition by multi-dimensional fusion of the spatial and temporal features of the fluid and chemical parameters.

[0098] In the present embodiment, the real-time collected pressure, flow, vanadium ion concentration, oxidation-reduction potential and pH value and other parameter data are first preprocessed to ensure the numerical consistency between the various dimensions. Generally, the parameter data is standardized to eliminate the influence of different dimensions, so the present embodiment adopts the standardization method of subtracting the mean value and dividing by the standard deviation to ensure that the parameter data is analyzed on the same scale.

[0099] In the present embodiment, the spatial features of the standardized multi-dimensional data matrix are extracted. Specifically, the preprocessed data is input into the convolution layer of the convolutional neural network (CNN) model, and different convolution kernels are used for convolution operation to capture the spatial features of each node position. The size of the convolution kernel is set according to the distribution characteristics of different parameters, so as to ensure the accuracy of feature extraction. For example, for flow and pressure data with large variation amplitude, larger convolution kernels can be used to extract significant variation features in these data. The initial feature map after convolution contains spatial feature information of each key position.

[0100] After obtaining the initial feature map, the embodiment adopts a pooling operation to reduce the dimension of the feature map. Specifically, the pooling operation extracts representative data points within the convolution region to retain key features and reduce data redundancy. Generally, the pooling operation can enhance the robustness of the feature map, so in this embodiment, the MaxPooling method is adopted to improve the recognition ability of the significant features.

[0101] After completing the time series processing, the time series feature vector is fused with the spatial features to form a multi-dimensional data matrix. Specifically, the fused multi-dimensional data matrix contains the spatial and temporal feature information of each position, and the matrix form is X = {p, Q, [V 2+ ],[V 3+ ], pH, E}, where p represents pressure, Q represents flow rate, [V 2+ ] and [V 3+ ] represent the concentrations of divalent and trivalent vanadium ions, respectively, pH represents the pH value, and E represents the redox potential. Generally, through this multi-dimensional feature fusion, the model can comprehensively grasp the fluid and chemical state of different positions.

[0102] In some embodiments, the generated multi-dimensional data matrix can be further stored as reference data for long-term monitoring and trend analysis of different working states of the stack to ensure real-time monitoring effect of the system.

[0103] S6, using a deep learning model to identify abnormal patterns in the multi-dimensional data matrix to determine the possible leakage position

[0104] To accurately determine the potential leakage position of the vanadium flow battery stack, the method of the present application uses a deep learning model to identify abnormal patterns based on the generated multi-dimensional data matrix, thereby determining the possible leakage position. Through this step, the system can comprehensively analyze the fluid and chemical parameter data of each position in the stack, and accurately locate the leakage position based on the abnormal pattern recognition result. The design of the deep learning model is based on the combination of convolutional neural network (CNN) and long short-term memory network (LSTM) to extract spatial and time series features in the data matrix.

[0105] In this embodiment, the multi-dimensional data matrix is input into the convolution layer of the deep learning model, and the matrix data is convolved by the convolution kernel to extract the spatial features between different positions. Generally, the size and step of the convolution kernel are set according to different data features to ensure the accuracy of feature extraction. Therefore, in this embodiment, for chemical parameters with relatively dense spatial feature distribution (such as vanadium ion concentration and redox potential), a smaller convolution kernel size is used to accurately capture minor feature differences. The convolution operation generates an initial feature map containing the spatial feature distribution information of each position in the stack.

[0106] In this embodiment, the pooled feature map is subjected to a pooling operation to reduce the size of the feature map and reduce the amount of calculation. Specifically, the pooling operation retains the main features and reduces redundant information by extracting the extreme value or average value of the local area. As an option, the Max Pooling method is used to extract the maximum value from each convolution region to form a pooled feature map with significant features. Generally, the pooling operation helps to enhance the robustness of the features, thereby improving the extraction accuracy of the model for spatial features.

[0107] On the basis of the pooled feature map, a long short-term memory network (LSTM) is used to process the time series of the feature map to capture the time variation pattern of the fluid and chemical parameter data. Specifically, the pooled feature map is input into the LSTM layer, and the time series dependence of each position is analyzed by the LSTM unit. The structure of the LSTM unit can process the long and short time changes of the historical data, so that the model can adapt to the time dependence of the fluid and chemical parameters at different positions in the stack. As an option, the memory step of the LSTM layer is set to the periodic change time of the stack fluid to ensure accurate capture of the time characteristics of the periodic fluctuations.

[0108] As an option, the hyperparameters of the deep learning model are dynamically adjusted by the Bayesian optimization algorithm to further improve the accuracy of the abnormal pattern recognition. The Bayesian optimization iteratively updates the hyperparameters based on the detection effect of the model, so that the parameters such as the size of the convolution kernel, the step, the pooling window and the LSTM memory step are continuously adjusted in the optimization process to ensure that the model has the best adaptability under various data characteristics.

[0109] In this embodiment, the deep learning model is used to calculate the leakage probability of each position. When the leakage probability of a certain position exceeds a preset threshold, the system will determine that there is a leak at that position and issue an alarm signal. Generally, the threshold of the leakage probability is set according to experimental data to ensure sufficient sensitivity to the actual leakage position. In summary, the present embodiment realizes accurate positioning of the leakage position of the vanadium flow battery stack through multi-layer convolution, pooling and time series analysis, and can effectively determine the potential leakage area based on the distribution of the leakage probability.

[0110] The present application constructs a multi-level fluid mechanics model and a chemical reaction kinetics model, collects and monitors the fluid and chemical parameters inside the stack in real time, establishes a benchmark data through numerical simulation, and uses a deep learning model to perform abnormal pattern recognition on the multi-dimensional data matrix to accurately locate the potential leakage position. This method combines multi-dimensional data fusion and intelligent analysis technology of fluid and chemical characteristics, and can accurately detect the small leakage of the vanadium flow battery, providing effective technical support for the safe operation of the stack.

[0111] The vanadium flow battery stack assembly leak detection system described below can be mutually corresponding with the vanadium flow battery stack assembly leak detection method described above.

[0112] Please refer to the attached Figure 2 The present application also provides a vanadium flow battery stack assembly leak detection system, comprising:

[0113] A fluid mechanics modeling module 100 is configured to establish a multi-level fluid model of the vanadium flow battery stack to describe the flow characteristics of the electrolyte;

[0114] A numerical simulation module 200 is configured to simulate the normal state of the fluid inside the stack based on the fluid mechanics model and generate baseline fluid data;

[0115] A sensor module 300 includes a plurality of pressure, flow and chemical sensors for real-time acquisition of fluid, vanadium ion concentration, redox potential and pH data of the stack;

[0116] A data processing module 400 is configured to compare the real-time monitoring data with the baseline data and input the chemical data of the abnormal fluctuation position into a deep learning model;

[0117] A deep learning module 500 is configured to analyze the multi-dimensional feature data matrix and perform leak detection and positioning through a convolutional neural network and a long short-term memory network.

[0118] The system of the present embodiment can be used to perform the above-mentioned method embodiments, and has similar principles and technical effects, which will not be described here.

[0119] Although embodiments of the present application have been shown and described, it is to be understood that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A leak detection method for vanadium redox flow battery stack assembly, characterized in that, Includes the following steps: A multi-level fluid dynamics model is established based on the overall fluid flow state of the vanadium redox battery stack. The fluid dynamics model includes fluid models at the macroscopic, mesoscopic, and microscopic levels. Numerical simulation was used to obtain baseline pressure, velocity, and flow rate data of the electrolyte at different locations; Pressure and flow data at key locations on the fuel cell stack are collected in real time, and the real-time data is compared with reference data to detect any abnormal pressure or flow fluctuations. When abnormal fluctuations are detected, the chemical parameters in the electrolyte are monitored in real time, and the impact of leakage is analyzed through a chemical reaction kinetic model. The collected fluid and chemical parameters are fused using multi-dimensional features to form a multi-dimensional data matrix; Anomaly pattern recognition is performed on the multidimensional data matrix using a deep learning model to determine possible leak locations.

2. The method for leak detection in vanadium redox flow battery stack assembly according to claim 1, characterized in that, The macroscopic fluid model describes the relationship between pressure, velocity, and height of the fluid at different locations based on the principle of energy conservation in fluids, and determines the pressure and velocity distribution at different locations through Bernoulli's theorem.

3. The method for leak detection in vanadium redox flow battery stack assembly according to claim 1, characterized in that, The fluid model of the mesoscopic layer is based on the permeability of fluid in porous media. Darcy's law is used to model the permeation behavior at the seals and interfaces of the fuel cell stack in order to estimate the local permeation flow rate.

4. The method for leak detection in vanadium redox flow battery stack assembly according to claim 1, characterized in that, The fluid model of the micro-layer assumes that the leakage path is a thin capillary, and describes the flow rate and permeation flow of the electrolyte in the capillary or microcrack based on Poiseuille's flow formula.

5. A leak detection method for vanadium redox flow battery stack assembly according to claim 1, characterized in that, The steps for obtaining reference pressure, velocity, and flow rate data of the electrolyte at different locations through numerical simulation include: The fluid dynamics model was solved using numerical simulation software. First, a computational grid was divided within the vanadium redox flow battery stack, and the entire fluid region was discretized based on the fluid density, viscosity, and inlet and outlet boundary conditions. At the macroscopic level of the fuel cell stack, the energy conservation equation based on fluid density, velocity, and height is solved. The distribution relationship between pressure and velocity is established by using Bernoulli's theorem, which is satisfied by electrolyte density, velocity, and pressure at different locations of the fuel cell stack. The overall velocity and pressure distribution of the electrolyte at different locations are then calculated. At the mesoscopic level, Darcy's law model is used to calculate the local permeation flow rate at the seal or interface. The pressure and flow distribution in the seal or interface area are obtained based on the permeability coefficient, fluid viscosity and flow path length. At the microscopic level, assuming that the leakage path has capillary structure characteristics, the flow velocity and seepage flow rate in the microcrack are calculated using the Poiseuille flow formula to establish the baseline flow velocity and flow rate distribution in the microscopic flow region. The flow results at each level are integrated to obtain the baseline fluid pressure, velocity, and flow rate data distribution of the vanadium redox flow battery stack under normal operating conditions. The baseline data is stored in a database for subsequent comparison and analysis with real-time monitoring data.

6. The method for leak detection in vanadium redox flow battery stack assembly according to claim 1, characterized in that, The steps of monitoring the chemical parameters in the electrolyte in real time when abnormal fluctuations are detected, and analyzing the impact of leakage through a chemical reaction kinetic model, include: The concentrations of divalent and trivalent vanadium ions in the electrolyte were measured using a vanadium ion selective electrode, and their initial reference concentrations were recorded. A differential equation for concentration change is established based on a kinetic model of vanadium ion redox reaction, wherein the kinetic model includes the rate constants of both the oxidation and reduction reactions; The rate constant in the redox kinetic model was used to calculate the rate of change of vanadium ion concentration at different temperatures, where the temperature dependence of the rate constant was described by the Arrhenius equation. By monitoring the changing trend of vanadium ion concentration in real time, abnormal values ​​of concentration fluctuations can be identified and the impact of leakage can be determined. The detected vanadium ion concentration fluctuation data is input into the calculation system to determine the significance of the concentration change, so as to identify the concentration change trend caused by the leak and determine the type of leaked component.

7. A method for leak detection in vanadium redox flow battery stack assembly according to claim 6, characterized in that, The step of monitoring the chemical parameters in the electrolyte in real time when abnormal fluctuations are detected, and analyzing the impact of leakage through a chemical reaction kinetic model, further includes: The redox potential data of the electrolyte is acquired in real time, and the data is collected by a redox potential sensor; Based on the concentration ratio of divalent and trivalent vanadium ions, the redox potential of the electrolyte is calculated using the Nernst equation, which includes the standard electrode potential at room temperature and the logarithmic relationship between the vanadium ion concentration. When the redox potential fluctuation exceeds the preset threshold, the fluctuation data is recorded as an abnormal fluctuation, and it is determined whether the fluctuation indicates a leak in the vanadium electrolyte. Multiple data collections and trend analyses were performed on the abnormally fluctuating redox potential data to confirm whether the fluctuations in redox potential remained stable within the abnormal range, in order to further determine the changes in chemical composition at the leak location.

8. A method for leak detection in vanadium redox flow battery stack assembly according to claim 1, characterized in that, The step of fusing multi-dimensional features of the collected fluid and chemical parameters to form a multi-dimensional data matrix includes: The collected real-time pressure, flow rate, and electrolyte chemical parameter data are segmented according to time series to form a basic dataset with time series characteristics, and the dataset is then normalized. Convolution operations are applied to the normalized data to extract the spatial feature matrix in order to obtain the feature distribution at different locations and chemical states. The temporal changes of the feature matrix are processed using a long short-term memory network to form a multidimensional data matrix that incorporates temporal features.

9. A method for leak detection in vanadium redox flow battery stack assembly according to claim 1, characterized in that, Its features are, The step of using a deep learning model to perform abnormal pattern recognition on the multidimensional data matrix to determine possible leakage locations includes: A multidimensional data matrix is ​​input into the convolutional layer of a deep learning model. Multiple convolutional kernels are used to perform convolution operations on the input data to generate an initial feature map, followed by pooling operations. The pooled feature map is processed using a long short-term memory network to form a temporal feature vector. The temporal feature vector is processed using a classification layer to generate the leakage probability distribution at each location; The most likely leak location is determined based on the leak probability distribution of each location, and a leak warning signal is issued when the leak probability exceeds a preset threshold.

10. A vanadium redox flow battery stack assembly leak detection system, used to perform the method as described in any one of claims 1-9, characterized in that, include: The fluid dynamics modeling module is used to build a multi-level fluid model of the vanadium redox flow battery stack and describe the flow characteristics of the electrolyte. The numerical simulation module is used to simulate the normal state of the fluid inside the fuel cell stack based on the fluid dynamics model and generate reference fluid data. The sensor module includes multiple pressure, flow, and chemical sensors for real-time acquisition of fluid, vanadium ion concentration, redox potential, and pH data of the fuel cell stack. The data processing module is used to compare real-time monitoring data and benchmark data, and input the chemical data of abnormal fluctuation locations into the deep learning model; The deep learning module is used to analyze multi-dimensional feature data matrices and perform leak detection and localization through convolutional neural networks and long short-term memory networks.

Images