A battery pack fault detection method, system, electronic device and storage medium

Abstract

The application discloses a battery pack fault detection method and system, electronic equipment and a storage medium, and relates to the technical field of energy storage detection. The method comprises the following steps: acquiring the voltages of all batteries in a battery pack; calculating the recursive correlation coefficient of the voltages of each adjacent two batteries by using a recursive correlation coefficient algorithm, and obtaining a plurality of recursive correlation coefficients; generating a pseudo two-dimensional image of a target battery according to a set time span and the recursive correlation coefficients; determining the detection result of the battery pack according to the pseudo two-dimensional image of the target battery and a battery fault detection model; the battery detection result comprises whether the target battery has a fault, and the fault level and the fault type when the fault exists; and the battery fault detection model is constructed based on an AlexNet model. The application can unify the detection standard of the batteries, and improve the accuracy and practicability of battery fault detection.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of energy storage detection, in particular to a battery pack fault detection method and system, an electronic device and a storage medium. BACKGROUND

[0002] Batteries play an irreplaceable role in new power grids and electric vehicle energy storage fields. However, the safety of these applications is still fragile in view of the frequently reported combustion accidents caused by thermal runaway of lithium-ion batteries. Serious faults can be easily detected from their obvious symptoms, while small faults are often hidden and ignored in their latent stage. Without proper maintenance, battery faults can easily deteriorate from subtle faults to uncontrolled faults. Therefore, it is imperative to develop effective detection methods for various faults of battery systems, take preventive measures and avoid disasters.

[0003] Existing technologies partly use developed battery models, while referring to historical data for optimization, but lack consideration of the time-varying and nonlinear characteristics of real battery systems. Therefore, the accuracy, reliability, generalization and subsequent detection results of the model have large deviations compared with the actual situation, and the practicability is not strong. Some technologies use insufficient data to explore the abstraction of fault patterns or rely on precise instruments and fine operations to propose subjective explanations of fault mechanisms, so it is difficult to obtain unified detection standards and cannot be applied in actual engineering with high safety requirements. SUMMARY

[0004] The purpose of the present application is to provide a battery pack fault detection method, system, electronic device and storage medium, which can unify the detection standards of batteries and improve the accuracy and practicability of battery fault detection.

[0005] To achieve the above purpose, the present application provides the following solutions:

[0006] A battery pack fault detection method, comprising:

[0007] obtaining the voltages of all target batteries in a battery pack;

[0008] calculating the recursive correlation coefficient of the voltages of each adjacent two target batteries by using a recursive correlation coefficient algorithm, to obtain a plurality of recursive correlation coefficients;

[0009] generating a pseudo two-dimensional image of a target battery according to a set time span and the recursive correlation coefficient;

[0010] determining a detection result of the battery pack according to the pseudo two-dimensional image of the target battery and a battery fault detection model; the battery detection result includes whether the target battery has a fault, and a fault level and a fault type when the target battery has a fault; the battery fault detection model is constructed based on an AlexNet model.

[0011] Optionally, the generating the pseudo two-dimensional image of the target battery according to the set time span and the recursive correlation coefficient comprises:

[0012] According to the spatial distribution sequence of all the target batteries in the battery pack and each recursive correlation coefficient, a recursive correlation coefficient sequence is constructed;

[0013] According to the recursive correlation coefficient sequence, a coefficient sequence curve image is generated;

[0014] According to the set time span, the coefficient sequence curve image is truncated to obtain a plurality of curve data segments;

[0015] According to the spatial distribution sequence of each curve data segment in the battery pack, an image gray value is obtained;

[0016] According to the image gray value, the pseudo two-dimensional image of the target battery is generated.

[0017] Optionally, the determination method of the battery fault detection model comprises:

[0018] Obtaining training data; the training data comprises a pseudo two-dimensional image of a training battery and a corresponding fault label; the fault label comprises whether the training battery has a fault, and a fault grade and a fault type when the training battery has a fault;

[0019] Constructing an AlexNet model;

[0020] Inputting the training data into the AlexNet model for training, and determining the trained AlexNet model as a battery fault detection model.

[0021] Optionally, the obtaining training data comprises:

[0022] Obtaining a pseudo two-dimensional image of a training battery;

[0023] Using a two-dimensional variational modal decomposition algorithm to decompose the pseudo two-dimensional image of the training battery to obtain each intrinsic modal sub-image;

[0024] Image texture features are extracted from each intrinsic modal sub-image to obtain image texture features;

[0025] According to the image texture, a fault label of the training battery is determined.

[0026] Optionally, the image texture features are extracted from each intrinsic modal sub-image to obtain image texture features, and the image texture features comprise:

[0027] Calculate a gray scale statistical feature quantity between two pixels with a set distance in each eigenmode sub-image by using a gray scale conjugate matrix algorithm; the gray scale statistical feature quantity includes energy, inverse variance, entropy, and contrast;

[0028] Determine an image texture feature according to the gray scale statistical feature quantity.

[0029] The application further provides a battery pack fault detection system, comprising:

[0030] A data acquisition module is configured to acquire voltages of all target batteries in a battery pack.

[0031] A data processing module is configured to calculate a recursive correlation coefficient of voltages of each adjacent two target batteries by using a recursive correlation coefficient algorithm, and obtain a plurality of recursive correlation coefficients.

[0032] An image generation module is configured to generate a pseudo two-dimensional image of a target battery according to a set time span and the recursive correlation coefficient.

[0033] A fault detection module is configured to determine a detection result of the battery pack according to the pseudo two-dimensional image of the target battery and a battery fault detection model; the battery detection result includes whether the target battery has a fault, and a fault level and a fault type when the target battery has a fault; the battery fault detection model is constructed based on an AlexNet model.

[0034] The application further provides an electronic device comprising a memory and a processor, wherein the memory is configured to store a computer program, and the processor is configured to run the computer program to enable the electronic device to execute the battery pack fault detection method according to the above.

[0035] The application further provides a computer readable storage medium storing a computer program, wherein the computer program is executed by a processor to implement the battery pack fault detection method as described above.

[0036] According to the embodiments of the application, the following technical effects are achieved:

[0037] The application discloses a battery pack fault detection method and system, electronic equipment and a storage medium, the method comprises the following steps: calculating the recursive correlation coefficient of the voltages of two adjacent target batteries in the battery pack by using a recursive correlation coefficient algorithm; generating a pseudo two-dimensional image of the target battery according to a set time span and a plurality of recursive correlation coefficients; and combining the time and space information in the pseudo two-dimensional image of the target battery can not only unify the detection standard of the model, but also improve the accuracy of battery fault detection; the battery fault detection model constructed based on the AlexNet model can also detect the batteries in the battery pack according to a unified standard, and obtain the detection results of whether the batteries have faults, the fault level and the fault type when the batteries have faults, thereby improving the practicability of battery fault detection. BRIEF DESCRIPTION OF DRAWINGS

[0038] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0039] Figure 1 The method flow chart of the battery pack fault detection method of the present application;

[0040] Figure 2 The logic flow diagram of the battery pack fault detection method in the present application;

[0041] Figure 3 The recursive correlation coefficient curve affected by different faults in the present application;

[0042] Figure 4 The structure block diagram of the battery pack fault detection system of the present application. DETAILED DESCRIPTION

[0043] The technical solutions in the embodiments of the present application will be described clearly and completely in the following with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0044] The purpose of the present application is to provide a battery pack fault detection method, system, electronic equipment and storage medium, which can unify the detection standard of the battery, improve the battery fault detection capability and practicability.

[0045] In order to make the above-mentioned purposes, characteristics and advantages of the present application more obvious and easy to understand, the present application will be further described in detail with reference to the drawings and specific embodiments.

[0046] Professional terms explanation:

[0047] P2I (pseudo 2D images): pseudo two-dimensional images.

[0048] 2D-VMD (2D variational mode decomposition): two-dimensional variational mode decomposition.

[0049] PCA (principal component analysis): principal component analysis.

[0050] IRCC: (Improved recursive correlation coefficient): improved recursive correlation coefficient.

[0051] M Δk : image gray matrix.

[0052] IMFs (intrinsic mode functions): intrinsic mode functions.

[0053] NCM: ternary lithium battery (the present application uses 3.7V, 3.5Ah).

[0054] ESC (External short circuit): external short circuit.

[0055] ISC (Internal short circuit): internal short circuit.

[0056] OHT (Over heat): overheat.

[0057] PCs: principal components.

[0058] No fault: no fault.

[0059] Minor: mild.

[0060] Medium: moderate.

[0061] Serious: serious.

[0062] CNN (Convolutional Neural Network): convolutional neural network.

[0063] AlexNet: a CNN model.

[0064] As Figure 1 shown, the present application provides a battery pack fault detection method, comprising:

[0065] Step 100: Obtain the voltages of all target batteries in the battery pack.

[0066] Step 200: Calculate the recursive correlation coefficient of the voltages of each adjacent two target batteries using the recursive correlation coefficient algorithm, to obtain a plurality of recursive correlation coefficients.

[0067] Step 300: Generate a pseudo two-dimensional image of the target battery according to the set time span and the recursive correlation coefficient.

[0068] Step 400: Determine the detection result of the battery pack according to the pseudo two-dimensional image of the target battery and the battery fault detection model; the battery detection result includes whether the target battery has a fault, and the fault level and fault type when the fault exists; the battery fault detection model is constructed based on the AlexNet model.

[0069] The determination method of the battery fault detection model is:

[0070] First, obtain the training data; the training data includes the pseudo two-dimensional image of the training battery and the corresponding fault label; the fault label includes whether the training battery has a fault, and the fault level and fault type when the fault exists.

[0071] Second, construct the AlexNet model.

[0072] Third, input the training data into the AlexNet model for training, and determine the trained AlexNet model as the battery fault detection model.

[0073] The acquisition process of the training data specifically includes:

[0074] First, obtain the pseudo two-dimensional image of the training battery.

[0075] Second, decompose the pseudo two-dimensional image of the training battery using a two-dimensional variational modal decomposition algorithm to obtain each intrinsic modal sub-image.

[0076] Third, perform image texture feature extraction on each intrinsic modal sub-image to obtain image texture features. The specific process includes:

[0077] Calculate the gray scale statistical feature quantity between two pixels at a set distance in each intrinsic modal sub-image using a gray conjugate matrix algorithm; the gray scale statistical feature quantity includes energy, inverse variance, entropy, and contrast; determine the image texture features according to the gray scale statistical feature quantity.

[0078] Fourth, determine the fault label of the training battery according to the image texture.

[0079] As a specific implementation of step 300, it comprises:

[0080] According to the spatial distribution order of all target batteries in the battery pack and each recursive correlation coefficient, a recursive correlation coefficient sequence is constructed; a coefficient sequence curve image is generated according to the recursive correlation coefficient sequence; a plurality of curve data segments are obtained by truncating the coefficient sequence curve image according to a set time span; an image gray value is obtained according to the spatial distribution order of each curve data segment in the battery pack; and a pseudo two-dimensional image of the target battery is generated according to the image gray value.

[0081] On the basis of the above method, the specific processing process is shown as Figure 2 .

[0082] First, design the fault signal and signal processing.

[0083] 1. According to the obtained voltage values of all batteries in the battery pack, through the recursive correlation coefficient of adjacent batteries:

[0084]

[0085]

[0086] Given discrete time series X = {x1, x2, … x k} and Y = {y1, y2, … y k}; Where x k = V x，k + ψ k , y k = v y，k + ψ k , V x，k and V y，k are the original signals (real-time measured voltages of two adjacent batteries). ψ is an alternating square wave used to operate the forgetting mechanism of the latest w samples, so as to balance between discarding redundant historical data and retaining sufficient fault evolution information. c(X,Y) k represents the recursive correlation coefficient of adjacent batteries w represents the length of the recursive correlation coefficient window; P k represents the sum of the products of the latest w x and y; Q k represents the sum of the latest w x; R k represents the sum of the latest w y; S k represents the sum of the squares of the latest w x; T k represents the sum of the squares of the latest w y; k represents the discrete time step sequence; x k-w represents the k-wth x; y k-w represents the k-wth y.

[0087] 2. The IRCC sequence of adjacent cells in the battery pack obtained in the previous step is truncated into data segments with a certain time span, and these parallel one-dimensional sequences are truncated into data segments with a certain time span. Collect them according to their spatial distribution order to form a matrix M Δk :

[0088]

[0089] Where c is a data segment with a certain time span, n is the number of cells in the battery pack, and Δk is the time size of the image in the sample. Then M Δk is normalized to [0, 255], corresponding to the gray scale (average) of the image pixel:

[0090]

[0091] M Δk is normalized to obtain the gray scale corresponding to the image. According to the gray scale of the image, the online updated P2Is are generated.

[0092] As Figure 3 shown, the IRCC sequence of all adjacent cells under different fault scenarios is plotted. Obviously, the IRCC of "no fault" provides a nearly smooth curve, which remains above 0.999, although there is a small fluctuation. For problematic cases, the IRCC curve shows obvious fluctuations. When ESC occurs, that is, a severe high-rate discharge, the IRCC curve first drops sharply, with an amplitude greater than half of the [0, 1] range, and then remains at a low level without obvious fluctuations until the ESC is released, because the external shunt branch dominates over the load current. For ISC faults, the IRCC also first experiences a drop (obviously more moderate than ESC), and then exhibits fluctuations similar to the basic outline of the load waveform. Here, the IRCC acts like a low-pass filter, because the internal shunt branch through the separator has lower impedance at high frequencies than the external load branch. For OHT faults, interestingly, the IRCC exhibits rich high-frequency components, where the IRCC acts like a differentiator to load changes. From the perspective of electrochemical processes, this phenomenon can be explained as high temperature increasing the conductivity of the electrolyte, thereby reducing the kinetic inertia and thus promoting the mass transfer of lithium ions. Based on the above analysis, the IRCC has obvious sensitivity and distinguishability to different battery faults.

[0093] Using the IRCC sequence obtained above, truncate it into data segments with a certain time span, and rearrange it according to its spatial distribution order to obtain an IRCC data matrix. The gray scale value data of the image is obtained through the data matrix, and the P2I texture map is made using the gray scale value.

[0094] 3、The P2I texture map in the previous step can be decomposed into a series of IMFs with different characteristic frequencies by 2D-VMD algorithm. Regarding decomposition, the number of IMFs, i.e. K, is a key parameter affecting the result: small K value means rough decomposition leading to loss of important information characteristics, while large K value makes the center frequencies of adjacent IMFs too close, resulting in mode aliasing and thus information redundancy. Comparing the IMF center frequencies under different K, when K exceeds 5, there will be IMFs less than 0.1, which may lead to aliasing between adjacent IMFs. Therefore, K is set to 5.

[0095] The texture features of the image are described according to the joint conditional probability between the image gray levels. Four statistical characteristic quantities of energy, inverse variance, entropy and contrast are obtained for principal component analysis. Four representative directions are set according to experience, θ = 0°, 45°, 90° and 135°, to achieve a balance between complexity and granularity. According to θ, the probability density can be obtained as:

[0096]

[0097] where i, j = 0, 1, 2…, L-1 are the pixel gray levels, L is the discrete level of image gray, θ and d are the angle and distance between two pixels to determine the granularity of texture extraction. Where x, y = 0, 1, 2…, N-1 and (x, y) are the coordinates of the pixels.

[0098] Thus a plurality of feature vectors are obtained. Then PCA is used to further refine the feature vectors, so that more than 90% of the variance information of the original feature vectors can be contained in several PCs matrices.

[0099] Finally, the model is trained and the fault detection result is obtained.

[0100] The PCs matrix obtained in the previous step can be divided into 10 categories, with 9 different fault states plus one healthy state, and multiple P2Is are generated for each state. All samples are used to train the AlexNet model in the CNN algorithm, and the rest are used for method testing.

[0101] The AlexNet model is trained by machine learning. The model can detect the fault level of the battery in the battery pack: No fault, Minor, Medium, Serious, and classify its fault type as: ISC, ESC, OHT. Then, the invention has good results for isolation between different fault types, with an accuracy of 97.67%. The overall fault classification accuracy is 84.67%.

[0102] Based on the above embodiment, the following beneficial effects are achieved:

[0103] The current method for estimating the internal parameters of the battery by using the behavior mechanism modeling combined with prior knowledge accumulation to evaluate the fault of the battery pack has an inaccurate battery model, insufficient data, and does not consider the electrochemical reaction mechanism inside the battery, but the actual process is more complex than the estimation. The real-time measurement data of the battery pack is used in the application, which avoids the estimation error of the battery model, and the data is obtained by real-time measurement, which is sufficient. The obtained data completely contains the electrochemical reaction mechanism inside the battery, and the accuracy is higher. Another method depends on accurate instruments and fine operation to propose a subjective explanation of the fault mechanism, and it is difficult to obtain a unified diagnostic standard, which may lead to different diagnostic evaluation results between different battery packs. The large-scale battery pack fault diagnosis method based on the space-time escape image proposed in the application can accurately judge the type and severity of the fault, avoid false alarms and omissions, and the overall fault isolation accuracy can reach 97.67%. In addition, the scheme also performs better in fault grading, and the overall accuracy is 84.67%.

[0104] As Figure 4 shown, the application also provides a battery pack fault detection system, comprising:

[0105] A data acquisition module is configured to acquire the voltage of all target batteries in the battery pack.

[0106] A data processing module is configured to calculate the recursive correlation coefficient of the voltage of each adjacent two target batteries by using a recursive correlation coefficient algorithm, and obtain a plurality of recursive correlation coefficients.

[0107] An image generation module is configured to generate a pseudo two-dimensional image of the target battery according to a set time span and the recursive correlation coefficient.

[0108] A fault detection module is configured to determine the detection result of the battery pack according to the pseudo two-dimensional image of the target battery and a battery fault detection model; the battery detection result includes whether the target battery has a fault, and the fault level and the fault type when the fault exists; and the battery fault detection model is constructed based on an AlexNet model.

[0109] The application also provides an electronic device comprising a memory and a processor, wherein the memory is configured to store a computer program, and the processor is configured to run the computer program to enable the electronic device to perform the battery pack fault detection method according to the above.

[0110] The application also provides a computer readable storage medium storing a computer program, wherein the computer program is executed by a processor to implement the battery pack fault detection method as described above.

[0111] The various embodiments described in this specification are intended to be exemplary only. The various embodiments were chosen and described in order to best explain the principles of the application and the practical application thereof. The best mode of the application has been described in detail, however, it will be apparent to those skilled in the art that various modifications and changes can be made to the specific embodiments described without departing from the spirit and scope of the application. Accordingly, the scope of the application should be limited only by the appended claims.

[0112] The principles and implementations of the present application have been described in the specification with specific examples. The above description of the embodiments is only for the purpose of understanding the core idea of the present application; at the same time, for those skilled in the art, according to the idea of the present application, the specific embodiments and the scope of application will be changed. In view of the above, the content of the specification should not be understood as a limitation of the present application.

Claims

1. A battery pack fault detection method, characterized in that, include: Obtain the voltage of all target cells in the battery pack; Using the recursive correlation coefficient algorithm, the recursive correlation coefficient of the voltage of each two adjacent target batteries is calculated to obtain multiple recursive correlation coefficients; A pseudo-two-dimensional image of the target battery is generated based on the set time span and the recursive correlation coefficient. The detection results of the battery pack are determined based on the pseudo-two-dimensional image of the target battery and the battery fault detection model; the detection results of the battery pack include whether the target battery has a fault, and if so, the fault level and fault type; the battery fault detection model is built based on the AlexNet model; The step of generating a pseudo-two-dimensional image of the target battery based on a set time span and the recursive correlation coefficient specifically includes: Based on the spatial distribution order of all target batteries in the battery pack and each of the recursive correlation coefficients, a recursive correlation coefficient sequence is constructed. Generate a coefficient sequence curve image based on the recursive correlation coefficient sequence; The coefficient sequence curve image is truncated according to a set time span to obtain multiple curve data segments; The image grayscale value is obtained according to the spatial distribution order of each curve data segment in the battery pack; A pseudo-two-dimensional image of the target battery is generated based on the image grayscale values.

2. The battery pack fault detection method according to claim 1, characterized in that, The method for determining the battery fault detection model is as follows: Acquire training data; the training data includes pseudo-two-dimensional images of the training battery and corresponding fault labels; the fault labels include whether the training battery has a fault, and the fault level and fault type when a fault exists; Build the AlexNet model; The training data is input into the AlexNet model for training, and the trained AlexNet model is determined as the battery fault detection model.

3. The battery pack fault detection method according to claim 2, characterized in that, The acquisition of training data specifically includes: Obtain a pseudo-two-dimensional image of the training battery; The pseudo-two-dimensional image of the training battery is decomposed using a two-dimensional variational mode decomposition algorithm to obtain the intrinsic mode sub-images. Image texture features are extracted from each intrinsic modality sub-image to obtain image texture features; The fault labels of the training batteries are determined based on the image texture.

4. The battery pack fault detection method according to claim 3, characterized in that, Image texture features are extracted from each intrinsic modality sub-image to obtain image texture features, specifically including: The gray-level statistical features between two pixels at a predetermined distance in each intrinsic modality sub-image are calculated using the gray-level conjugate matrix algorithm; the gray-level statistical features include energy, inverse variance, entropy, and contrast. Image texture features are determined based on the grayscale statistical features.

5. A battery pack fault detection system, characterized in that, include: The data acquisition module is used to acquire the voltage of all target batteries in the battery pack; The data processing module is used to calculate the recursive correlation coefficient of the voltage of each two adjacent target batteries using a recursive correlation coefficient algorithm, thereby obtaining multiple recursive correlation coefficients. An image generation module is used to generate a pseudo-two-dimensional image of the target battery based on a set time span and the recursive correlation coefficient. The fault detection module is used to determine the detection result of the battery pack based on the pseudo-two-dimensional image of the target battery and the battery fault detection model; the detection result of the battery pack includes whether the target battery has a fault, and the fault level and fault type when a fault exists; the battery fault detection model is built based on the AlexNet model; The step of generating a pseudo-two-dimensional image of the target battery based on a set time span and the recursive correlation coefficient specifically includes: Based on the spatial distribution order of all target batteries in the battery pack and each of the recursive correlation coefficients, a recursive correlation coefficient sequence is constructed. Generate a coefficient sequence curve image based on the recursive correlation coefficient sequence; The coefficient sequence curve image is truncated according to a set time span to obtain multiple curve data segments; The image grayscale value is obtained according to the spatial distribution order of each curve data segment in the battery pack; A pseudo-two-dimensional image of the target battery is generated based on the image grayscale values.

6. An electronic device, characterized in that, The device includes a memory and a processor, the memory being used to store a computer program, and the processor running the computer program to cause the electronic device to perform the battery pack fault detection method according to any one of claims 1-4.

7. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the battery pack fault detection method as described in any one of claims 1-4.

Images