An online screening method for lithium-ion battery cells with abnormal capacity

Abstract

The present invention relates to an online screening method for lithium-ion battery cells with abnormal capacity, comprising the following steps: constructing a historical data pool using historical cell process data to determine the normal range of process data; constructing an anomaly detection and classification model based on the historical data pool; performing an initial screening of online cells based on the normal range of process data, eliminating cells with process anomalies, and obtaining online cells for further testing; and re-screening the online cells for further testing based on the anomaly detection and classification model to obtain cells with abnormal capacity and cells with normal capacity. Compared with existing technologies, the present invention has the advantages of being able to accurately identify cells with abnormal capacity in real time and reducing screening costs and time.

Description

Technical Field

[0001] The present invention relates to the technical field of battery production and manufacturing, and in particular to an online screening method for lithium-ion battery cells with abnormal capacity. Background Art

[0002] Lithium-ion batteries are widely used in electronic devices and electric vehicles. During the lithium-ion battery manufacturing process, cells with abnormal capacity must be screened to ensure the consistency of subsequent battery packs and extend battery life.

[0003] Currently, lithium battery manufacturing typically involves calibrating battery capacity through a capacity grading process, and using the results to screen out cells with abnormal capacity. However, this process requires charging and discharging multiple cells in a grading cabinet, resulting in high equipment and electricity costs, and a time-consuming process. Summary of the Invention

[0004] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and provide an online screening method for lithium-ion battery cells with abnormal capacity, which can accurately identify cells with abnormal capacity in real time.

[0005] The purpose of the present invention can be achieved by the following technical solutions:

[0006] An online screening method for lithium-ion battery cells with abnormal capacity comprises the following steps:

[0007] Use historical cell process data to construct a historical data pool and determine the normal range of process data;

[0008] constructing an anomaly detection classification model based on the historical data pool;

[0009] Performing a preliminary screening of online battery cells according to the normal range of the process data, eliminating battery cells with process abnormalities, and obtaining online battery cells for further testing;

[0010] The online battery cells to be further detected are screened again based on the abnormality detection classification model to obtain battery cells with abnormal capacity and battery cells with normal capacity.

[0011] Furthermore, the historical data pool includes a set proportion of process data of battery cells with abnormal capacity.

[0012] Furthermore, the normal range of the process data is determined by upper and lower quantiles.

[0013] Furthermore, before constructing the anomaly detection classification model, the historical data pool is subjected to robust normalization processing, and the robust normalization process is as follows:

[0014]

[0015]

[0016] in, is the data pool after robust normalization, x j and are the historical data pool and the j-th dimension of the cell process data vector after robust normalization, x j,25 、x j,50 and x j,75 is the robust normalization parameter of the j-th dimension.

[0017] Furthermore, the construction process of the anomaly detection classification model includes:

[0018] Adopting the additive model and forward step-by-step algorithm, the boosting tree model is constructed with the binary classification tree as the basis function;

[0019] The boosted tree model is trained based on the historical data pool to obtain the optimal values ​​of the hyperparameters, thereby constructing an anomaly detection classification model.

[0020] Furthermore, a cross-validation hybrid grid search method is used to obtain the optimal value of the hyperparameter.

[0021] Furthermore, the hyperparameters include learning rate, maximum number of leaf nodes, tree depth and number of sub-classifiers.

[0022] Furthermore, it also includes:

[0023] The cells with normal capacity as described in the random inspection are put into the capacity separation process together with the cells with abnormal capacity to obtain the measured capacity of the cells;

[0024] The historical data pool, the normal range of process data, the anomaly detection classification model, and the optimal values ​​of hyperparameters are updated based on the measured capacity.

[0025] Furthermore, the step of updating the historical data pool includes:

[0026] Adding the battery cells that have undergone the capacity division process and their measured capacities to a historical data pool to update the historical data pool;

[0027] Determine whether the data volume of the historical data pool exceeds a certain threshold. If so, remove the oldest cell process data in the historical data pool to maintain the data volume of the historical data pool not exceeding the threshold. If not, do nothing.

[0028] Furthermore, it also includes determining whether the amount of online battery cell data newly added to the historical data pool exceeds a certain threshold. If so, the updated historical data pool is used to re-obtain the optimal value of the hyperparameter and construct an anomaly detection classification model.

[0029] Compared with the prior art, the present invention has the following beneficial effects:

[0030] (1) The present invention makes full use of the historical process data of battery cells, determines the normal range of battery cells and establishes an abnormality detection classification model, and screens the online battery cells twice. It can not only accurately identify battery cells with abnormal capacity in the manufacturing process, but also perform real-time detection, thus solving the problem of high economic and time costs for abnormal battery cell screening in the traditional capacity separation process.

[0031] (2) The present invention performs robust normalization on the data in the historical data pool, thereby weakening the impact of marginal data in the historical data pool on the model and ensuring that the detection results of the model are more accurate.

[0032] (3) The present invention inputs some of the screened normal cells and all the abnormal cells into the capacity division process for capacity measurement, and updates the previously determined historical data pool, the normal range of the cells, the abnormality detection classification model and the model hyperparameters online, so that they can be applied and updated in real time on the lithium battery production line, thereby further improving the accuracy of cell screening. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 Flow chart of the method of the present invention;

[0034] Figure 2 This is a diagram of the actual operation structure of the manufacturing production line according to an embodiment of the present invention;

[0035] Figure 3 It is the retention rate of different batches of online battery cells in the embodiment of the present invention. DETAILED DESCRIPTION

[0036] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0037] This embodiment provides an online screening method for lithium-ion battery cells with abnormal capacity, such as Figure 1 As shown, the following steps are included:

[0038] S1. Use historical cell process data to construct a historical data pool and determine the normal range of process data.

[0039] The cell manufacturing line provides a batch of historical cell process data X0 for the current line. This cell process data includes one or more of the following: electrolyte injection volume during the injection process, charge and discharge voltage, current, capacity, pressure, and temperature during the formation and recharging processes. This batch of data has been measured in the capacity grading process, and whether each capacity is abnormal is known. This data can be represented by y0. Specifically:

[0040]

[0041] Among them, the historical data pool X0 has a total of m0 battery cell data and n data dimensions. is the process data of the jth dimension of the i-th cell, and accordingly, y (i) represents whether the capacity of the i-th battery cell is abnormal, y (i) The value of the binary classification label is 0 or 1. If y (i) If it is 1, the capacity of the i-th battery cell is abnormal; if it is 0, the capacity is normal.

[0042] The historical data pool is constructed using the historical cell process data. The historical data pool should contain a certain proportion of abnormal data. Assuming the minimum proportion is r0, we have:

[0043]

[0044] Among them, m0 is the amount of battery data, r0 is the minimum ratio, I is the indicator function, y (i) Is the label indicating whether the i-th battery cell is abnormal.

[0045] According to the historical data pool X0, the normal range of process data is determined. Specifically, if the normal range of the jth process data dimension is set to the lower quantile l j and upper quantile u j Then the data of a cell to be tested x'=[x'1,x'2,...,x' n ] T The requirements for meeting the normal range of process data are:

[0046]

[0047] in, and are l of the jth dimension respectively j and u j Quantile, l j and u j Lower quantile and upper quantile respectively.

[0048] S2. Construct an anomaly detection classification model based on the historical data pool.

[0049] According to the historical data pool X0, the robust normalization parameter x must be determined according to each dimension j.j,25 、x j,50 and x j,75 , i.e. the 25%, 50% and 75% quantiles of each dimension, and calculate the robust normalized data For subsequent model training, the robust normalization process can be expressed as:

[0050]

[0051]

[0052] Among them, x j , are the historical data pool and the j-th dimension battery cell data vector after robust normalization.

[0053] The robustly normalized historical data is used to train the boosted tree model, which adopts the additive model and forward step-by-step algorithm, with the binary classification tree as the basis function. The additive model can be expressed as:

[0054]

[0055] Where, T(x; Θ a ) represents a binary classification tree, Θ a represents the hyperparameter of the decision tree, and N is the number of trees.

[0056] The forward step-by-step algorithm for computing the boosted tree can be expressed as:

[0057] f a (x) = f a-1 (x)+T(x;Θ a )

[0058]

[0059] Where, f a-1 (x) is the current model, L is the empirical risk function, T(x; Θ a ) represents a binary classification tree, y (i) represents whether the capacity of the i-th cell is abnormal, x (i) is the vector of all dimensions of the i-th cell, where the hyperparameters of the next decision tree are determined by minimizing the empirical risk

[0060] The above process trains a boosted tree model f. The hyperparameters of the boosted tree are optimized through a combination of grid search and cross-validation. Critical hyperparameters for the model include the learning rate, maximum number of leaf nodes, tree depth, and number of subclassifiers. The optimization method is to traverse all hyperparameters and their possible values, train a boosted tree model for each value combination, and use this model to perform a 5-fold cross-validation on the historical data pool X0. The average cross-validation sensitivity is calculated as the score of this traversal, and the set of hyperparameter values ​​with the highest score is finally selected as the optimal value.

[0061] Specifically, assuming that the hyperparameter set Η={h1,h2,...,h R There are R hyperparameters to be optimized, among which the rth hyperparameter h r The value set of Total S r possible values, then for each hyperparameter and each value, a 5-fold cross validation of the boosted tree is performed, for a total of For each 5-fold cross validation, the robust normalized historical data pool Randomly divide the data into 5 equal parts, namely:

[0062]

[0063] in, The historical data pool is used for cross-validation and is randomly divided into a sub-dataset with a data volume of m0 / 5.

[0064] Then traverse each subset in turn as the verification set X val,k , the remaining 4 parts are used as training set X train,k , the corresponding classification label is y val,k and y train,k , train a boosted tree model f k ,Right now:

[0065] f k =train(X train,k ,y train,k ;Θ a )

[0066] Where, Θ a The hyperparameter values ​​corresponding to this cross validation are taken.

[0067] This improves the tree model f to make predictions on the validation set:

[0068] y pred,k =f k (X val,k )

[0069] The true positive rate TP of the boosted tree model on the validation set k and false negative rate FN k Statistics are performed, where TP k Is to determine the correct number of abnormal cells, FN k The number of normal cells mistakenly identified as abnormal cells:

[0070]

[0071]

[0072] Where I is the indicator function, i is the number of the cell in the verification set, and then the sensitivity score TPR is calculated k :

[0073] TPR k =TP k / (TP k +FN k )

[0074] Although using sensitivity as the judgment standard may result in some misjudgment of normal cells, it can minimize the omission of cells with abnormal capacity and prevent cells with abnormal capacity from entering the market. The final sensitivity score of this 5-fold cross-validation is the average of the scores obtained for each fold:

[0075]

[0076] Since all cases have been traversed, there must be a set of hyperparameter values The cross-validation sensitivity score is the highest, so this set of hyperparameter values ​​is the optimal value, and the anomaly detection classification model is trained under the optimal value. For online prediction:

[0077]

[0078]

[0079] S3. Performing a preliminary screening on the battery cells to be tested according to the normal range of the process data, eliminating battery cells with process abnormalities, and obtaining battery cells to be further tested.

[0080] As of now, the process screening, normalization, and anomaly detection classification models have been determined, and capacity anomaly detection and screening will be performed on online cells. Assume that the data of each batch of online cells is represented by m, which should include:

[0081] m≤bm0,b∈(0,1)

[0082] Where m is the amount of data in the historical data pool, and b is the ratio of the limited online batch data to the amount of data in the historical data pool. This formula aims to ensure that the amount of data in the historical data pool is sufficient relative to the number of online cells per batch, preventing underfitting of the boosted tree model and ensuring sufficient model reliability.

[0083] The purpose is to ensure that the amount of data in the historical data pool is large enough relative to the number of online battery cells in each batch, to prevent underfitting of the boosted tree model and to ensure that the model is sufficiently reliable.

[0084] First, process screening is performed on the online cells to remove the cell data that is out of the normal range of process data in any dimension. That is, the abnormal cell elimination set D is:

[0085]

[0086] Where, is the j-th dimension process data of the i-th cell on the line, The upper and lower quantiles of the normal data of the j-th dimension are determined based on the historical data pool. In this step, all battery cells that do not meet the normal range of process data in a certain dimension of the historical data pool are considered to have large process deviations and are directly judged as abnormal, and the online battery cells are retained for further inspection.

[0087] S4. Re-screen the online cells to be further inspected based on the anomaly detection classification model to obtain cells with abnormal capacity and cells with normal capacity.

[0088] Perform robust normalization on online cell data, for all cell data vectors x' in each dimension j , the normalized formula is:

[0089]

[0090] Among them, x j,25 、x j,50 and x j,75 The 25%, 50%, and 75% quantiles of the j-th dimension determined for the historical data pool.

[0091] The normalized data is then fed into the anomaly detection classification model Make a prediction and get the predicted value

[0092]

[0093] y' pred There are data of cells predicted to have normal capacity, totaling m n There are also data on cells with abnormal capacity, totaling m p Then obviously we have:

[0094] mn +m p =m

[0095] For cells with abnormal predicted capacity, they should be added to the elimination set:

[0096]

[0097] S5. Extract some of the cells with normal capacity and enter the capacity separation process together with all the cells with abnormal capacity to obtain the measured capacity of the cells.

[0098] For cells with abnormal predicted capacity, the capacity division process should be further verified. At the same time, in order to balance the updated data set, a certain proportion of normal predicted capacity data should be randomly sampled for capacity division. If the sampling ratio is r G , then the amount of data required for the capacity division process is m G for:

[0099] m G =m p +r G m n

[0100] Among them, m n The predicted result is the number of cells with normal capacity, m p The predicted number of cells with abnormal capacity.

[0101] S6. Update the historical data pool, normal range, abnormal detection classification model, and optimal values ​​of hyperparameters based on the measured capacity.

[0102] After the volume fractionation process, m can be determined. G Whether the actual data is abnormal, that is, the true value vector After this, the historical data pool can be updated, namely:

[0103]

[0104] After updating the historical data pool, the following actions should be monitored and implemented simultaneously:

[0105] (1) Check whether the historical data pool is overflowing. The historical data pool should set an upper limit for the amount of data according to the actual process conditions, which is set as m max If the amount of data in the historical data pool exceeds the upper limit after the update, that is, m0+m G >m max , then the earliest m should be removed according to the data time max -(m0+m G ) pieces of cell data to prevent premature data from interfering with the screening. In this embodiment, m max Preferably 50,000.

[0106] (2) Whether the normal process data range and robust normalization parameters should be updated. After the cumulative update data volume reaches a certain amount, in order to ensure the timeliness of process screening, the normal process data range and robust normalization parameters should be updated in the manner described above. When the amount of data for each batch of battery cells on the line is not too small, each update of the historical data pool can correspond to an update of the normal process data range and robust normalization parameters.

[0107] (3) Whether the anomaly screening model and its optimal hyperparameters should be updated. After the cumulative updated data volume reaches a certain amount, in order to ensure the timeliness of the anomaly detection classification model, the updated historical data pool should be used to perform grid search and 5-fold cross-validation to optimize the hyperparameters as described above, and the boosted tree model should be retrained using the optimal hyperparameters.

[0108] The above operation is repeated for each batch of cells produced online, resulting in a set D of cells with abnormal capacity, thus achieving the goal of online real-time screening. This step enables real-time application and updates on the lithium battery production line, further improving the accuracy of cell screening.

[0109] To verify the effectiveness of the above method, this embodiment uses a lithium-ion battery manufacturing production line as an application scenario and describes in detail the operation process of the actual application scenario. Figure 2 In this application scenario, online real-time cell data is first screened against the normal range of process data to identify cells outside the normal range, i.e., abnormal cells, and cells awaiting further inspection. Robust normalization is then performed on the historical data pool to obtain a robustly normalized data pool. An anomaly detection and classification model is then constructed based on this data pool. Cells awaiting further inspection are then screened again using this model, which classifies them as abnormal or normal. Normal cells are then inspected and shipped as qualified. Furthermore, a certain percentage of the normal cells are sampled and sent to the capacity separation process along with all abnormal cells for capacity measurement. The measured capacity from the capacity separation process is then used to regularly update the historical data pool, the normal process data range, the anomaly detection and classification model, and its parameters. If the historical data pool is updated due to excessive data overflow, the oldest data is discarded. In this application scenario, 176,594 online cells were verified, and all cells with abnormal capacity were screened out with a screening accuracy of 100%. The retention rates of different online batches are shown in the attached figure. Figure 3As shown. The retention rate is defined as the ratio of the number of cells in each batch that are deemed qualified by the system and not screened out as abnormal to the total number of cells in that batch. These cells, with the exception of a few for balancing spot checks, do not need to undergo the capacity separation process. As can be seen from the figure, the system involved in the present invention can screen out all cells with abnormal capacities while ensuring a high retention rate, thereby significantly reducing the number of cells requiring physical capacity separation. This is expected to reduce the cost of capacity separation testing and the number of capacity separation equipment, thereby improving the overall economic efficiency of lithium-ion battery production lines.

[0110] If the above functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0111] It will be understood by those skilled in the art that the embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of the present invention may be implemented in various computer languages, for example, the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0112] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1A device that provides the functions specified in a block or multiple blocks.

[0113] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0114] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0115] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0116] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

[0117] The above describes in detail the preferred embodiments of the present invention. It should be understood that those skilled in the art can make numerous modifications and variations based on the concepts of the present invention without inventive effort. Therefore, any technical solutions that can be derived by those skilled in the art through logical analysis, reasoning, or limited experimentation based on the concepts of the present invention and the prior art should be within the scope of protection defined by the claims.

Claims

1. An online screening method for lithium-ion battery cells with abnormal capacity, characterized in that: The following steps are involved: Use historical cell process data to construct a historical data pool and determine the normal range of process data; constructing an anomaly detection classification model based on the historical data pool; Performing a preliminary screening of online battery cells according to the normal range of the process data, eliminating battery cells with process abnormalities, and obtaining online battery cells for further testing; Based on the anomaly detection classification model, the online battery cells to be further detected are screened again to obtain battery cells with abnormal capacity and battery cells with normal capacity; Before constructing the anomaly detection classification model, the historical data pool is subjected to robust normalization processing. The robust normalization process is as follows: in, is the data pool after robust normalization, x j and are the historical data pool and the j-th dimension of the cell process data vector after robust normalization, x j,25 、x j,50 and x j,75 is the robust normalization parameter of the j-th dimension; The construction process of the anomaly detection classification model includes: Adopting the additive model and forward step-by-step algorithm, the boosting tree model is constructed with the binary classification tree as the basis function; Training the boosted tree model based on the historical data pool to obtain optimal values ​​of hyperparameters, thereby constructing an anomaly detection classification model; Also includes: The cells with normal capacity as described in the random inspection are put into the capacity separation process together with the cells with abnormal capacity to obtain the measured capacity of the cells; The historical data pool, the normal range of process data, the anomaly detection classification model, and the optimal values ​​of hyperparameters are updated based on the measured capacity.

2. The method for online screening of lithium-ion battery cells with abnormal capacity according to claim 1, characterized in that: The historical data pool includes a set proportion of process data of battery cells with abnormal capacity.

3. The online screening method for lithium-ion battery cells with abnormal capacity according to claim 1, characterized in that: The normal range of the process data is determined by the upper and lower quantiles.

4. The online screening method for lithium-ion battery cells with abnormal capacity according to claim 1, characterized in that: A cross-validation hybrid grid search method is used to obtain the optimal values ​​of the hyperparameters.

5. The online screening method for lithium-ion battery cells with abnormal capacity according to claim 1, characterized in that: The hyperparameters include learning rate, maximum number of leaf nodes, tree depth, and number of sub-classifiers.

6. The method for online screening of lithium-ion battery cells with abnormal capacity according to claim 1, characterized in that: The steps of updating the historical data pool include: Add the cells that have gone through the capacity division process and their measured capacities to the historical data pool to update the historical data pool; Determine whether the data volume of the historical data pool exceeds a certain threshold. If so, remove the oldest cell process data in the historical data pool to maintain the data volume of the historical data pool not exceeding the threshold. If not, do nothing.

7. The method for online screening of lithium-ion battery cells with abnormal capacity according to claim 6, characterized in that: It also includes determining whether the amount of online battery cell data newly added to the historical data pool exceeds a certain threshold. If so, the updated historical data pool is used to re-obtain the optimal value of the hyperparameters and construct an anomaly detection classification model.

Images