CN-121978545-A - Lithium battery production process abnormality diagnosis method and system based on predicted capacity deviation

CN121978545ACN 121978545 ACN121978545 ACN 121978545ACN-121978545-A

Abstract

The invention discloses a lithium battery production process abnormality diagnosis method and system based on predicted capacity deviation, and belongs to the technical field of new energy batteries. The method comprises the steps of obtaining multi-procedure data in the battery production process, preprocessing the procedure data, training a deep learning model based on an FT-converter framework by using the preprocessed data, enabling the model to predict the discharge capacity of the battery according to the production process data, inputting the production process data of the battery to be diagnosed into the trained model to obtain the predicted capacity of the battery, comparing the predicted capacity with the actual capacity obtained by the battery capacity division test, and calculating capacity deviation between the predicted capacity and the actual capacity. The invention can efficiently and accurately identify the abnormal battery with the performance and the problem in the production process by utilizing the existing production data without adding additional tests, and provides a new technical approach for improving the overall quality and the safety of the battery system.

Inventors

Gong Xuanlin
CHEN BAOHUI
LI BO
GUO XIAOHAN
HU JINLI
HU HONGHUA
ZHU YINXIN
ZENG ZHIQIANG

Assignees

湖南防灾科技有限公司
国网湖南省电力有限公司防灾减灾中心
湖南省湘电试研技术有限公司

Dates

Publication Date: 20260505
Application Date: 20260408

Claims (10)

1. The lithium battery production process abnormality diagnosis method based on the predicted capacity deviation is characterized by comprising the following steps of; step A, acquiring production process data, namely acquiring multi-dimensional production process data covering at least two working procedures in lithium battery production to form a process data vector corresponding to the lithium battery; training a predictive model, namely training a capacity predictive model based on an FT-transducer architecture by adopting historical process data vectors respectively corresponding to a plurality of lithium batteries and actual capacity values corresponding to the lithium batteries; step C, predicting the capacity of the battery to be diagnosed, and inputting a process data vector of the lithium battery to be diagnosed into a trained capacity prediction model to obtain a predicted capacity value corresponding to the lithium battery to be diagnosed; Step D, performing abnormality diagnosis, namely acquiring an actual capacity value after the capacity division process of the lithium battery to be diagnosed, calculating an absolute error between the predicted capacity value and the actual capacity value, and comparing the absolute error with a preset abnormality judgment threshold value; And B, a capacity prediction model of the FT-transducer architecture in the step B comprises the following structures: the characteristic tokenization module is used for uniformly converting the numerical type characteristics and the category type characteristics in the process data vector into characteristic embedded vectors with preset dimensions; A transducer encoder composed of at least one stacked transducer layer containing a multi-headed self-attention mechanism for processing information interaction of feature embedding vectors; And (3) a pre-measuring head, namely splicing CLSToken at the front end of the characteristic embedding sequence to form a transducer encoder input, and carrying out capacity prediction by utilizing the final output of the transducer encoder input.
2. The method according to claim 1, wherein the production process data of step a includes at least one or more of material lot information of an assembly process, weight information of a liquid injection process, process voltage and capacity information of a formation process, charge capacity and equipment information of a capacity division process, open circuit voltage and internal resistance information of each stage.
3. The method for diagnosing abnormal conditions in the production process of lithium batteries based on the predicted capacity deviation according to claim 1, wherein the production process data of the step a further comprises battery type, material lot number, weight before and after liquid injection, formation process voltage and capacity data, capacity-dividing process charging capacity, capacity-dividing equipment and channel number, environment data and time data.
4. The lithium battery production process abnormality diagnosis method based on the predicted capacity deviation according to claim 1, wherein the process data vector is core process data for representing the whole production process of a single lithium battery, and is formed by splicing two parts of original process parameters and characteristic engineering derivative parameters, and specifically comprises the following steps: Original process parameters: a. The assembly process is characterized by comprising a battery type and a material batch number of positive and negative electrode materials; b. the liquid injection procedure is characterized by comprising the weight of the battery before liquid injection, the primary liquid injection amount, the secondary liquid injection amount, the weight of the battery as a final finished battery and the weight of the battery reduced in the formation process of the battery; c. The formation process is characterized by comprising an initial voltage of the battery before formation, a final voltage of the battery after formation, charge and discharge capacity of the battery in each constant current/constant voltage stage of formation, temperature of the battery after formation and station number of the battery during formation; d. The capacity-dividing procedure is characterized by comprising the charge and discharge capacity of the battery in the capacity-dividing process of each constant current/constant voltage stage, the initial voltage and temperature of the battery before capacity division and the station number of the battery during capacity division; e. other electrochemical performance characteristics include open circuit voltage, alternating current internal resistance and aging K value of each stage; Characteristic engineering derivative parameters: a. The batch non-uniformity index Stds is that for the case that a single battery comprises a plurality of winding cores, the material batch number of each winding core is converted into a specific date and then mapped into a time frame value, and the standard deviation is calculated, wherein the specific calculation process is as follows: ; the method is used for quantifying the difference of material activity and coating uniformity caused by batch fluctuation of raw materials among different winding cores in the same battery; b. the formation resting voltage drop is used for calculating the difference value between the first-stage constant current charging ending voltage and the second-stage constant current charging starting voltage in the formation procedure, and capturing the polarization recovery capability of the battery cell and the interface stability of the battery cell in a short resting period; c. The equipment aging degree is that the day difference of the production date of the battery on the same day relative to the formal production date of the production line is d, and then: the method is used for capturing the influence of hidden environmental factors accumulated over time, such as equipment wear, process parameter fine adjustment and the like, on the battery capacity; d. And the number of the batteries processed in the same batch is calculated, and the number of the batteries processed in the same time by each device is calculated and bound with the corresponding batteries, so that the uniformity of thermal field distribution and current distribution in the cabinet machine can be influenced by the number of the batteries processed in the same batch.
5. The method for diagnosing a lithium battery production process abnormality based on a predicted capacity deviation according to claim 4, wherein the uniformly converting the numerical type feature and the class type feature in the process data vector into the feature embedding vector of the preset dimension comprises: numerical feature conversion of each numerical feature By association with a learnable weight vector Performing scalar multiplication expansion and adding a bias vector To generate an embedded vector The specific calculation formula is as follows: class feature conversion, each class feature First, a vector is obtained by single-hot encoding Then through a learning embedding matrix Find and add a bias vector To generate an embedded vector The specific calculation formula is as follows: finally, the converted product is Numerical characteristics of And Individual category features Stacked together to obtain an embedded matrix The method is characterized by comprising the following steps: 。
6. the method for diagnosing a lithium battery production process abnormality based on a predicted capacity deviation according to claim 4, wherein the input of the transducer encoder is a learnable [ CLS ] token spliced at the front end of the feature embedding vector sequence for aggregating global information.
7. The method for diagnosing abnormal conditions in the production process of lithium batteries based on the predicted capacity deviation according to claim 1, wherein the model training in the step B is performed with a minimum mean square error as an optimization target, and the threshold value of the minimum mean square error is ±1%.
8. The method for diagnosing a lithium battery production process abnormality based on a predicted capacity deviation according to claim 1, wherein the abnormality judgment threshold in step D is determined based on statistical characteristics of a predicted error distribution of the capacity prediction model on the verification data set, and the abnormality judgment threshold is ±1%.
9. The method for diagnosing abnormal conditions in the production process of the lithium battery based on the predicted capacity deviation according to claim 1, wherein after the lithium battery to be diagnosed is judged to be an abnormal battery in the step D, the method further comprises an abnormality attribution analysis step, namely, a SHAP analysis tool is adopted to calculate the characteristic importance of the process data vector of the lithium battery to be diagnosed, a key process or a key process parameter which causes the absolute error between the predicted capacity value and the actual capacity value is positioned, the key process is selected from at least one of an assembly process, a liquid injection process, a formation process and a capacity division process, and the key process parameter is selected from at least one of a liquid injection amount, a formation voltage, a formation capacity and a capacity division charging capacity.
10. A lithium battery production process abnormality diagnosis system, comprising: The data acquisition module is used for acquiring production process data, and acquiring multi-dimensional production process data covering at least two working procedures in lithium battery production to form a process data vector corresponding to the lithium battery; a model storage module for storing the capacity prediction model trained in the step B in claim 1; the calculation module is used for executing the calculation of the predicted capacity in the step C and the absolute error calculation of the predicted capacity and the actual capacity in the step D according to the claim 1; The diagnosis module is used for comparing the absolute error with an abnormality judgment threshold value and judging an abnormal battery in the step D according to claim 1.

Description

Lithium battery production process abnormality diagnosis method and system based on predicted capacity deviation Technical Field The invention relates to the technical field of new energy batteries, in particular to a lithium battery production process abnormality diagnosis method and system based on prediction capacity deviation. Background In mass production of lithium ion batteries, ensuring high consistency of each single battery is a key to ensure safe and long-life operation of the final battery system. Currently, the industry commonly adopts static indexes such as capacity, voltage, self-discharge (K value) and the like of a measured battery to carry out screening and grading. However, these methods only ensure that the selected cells are similar in static parameters, but cannot effectively discriminate "hidden" abnormal cells with potential performance degradation or safety risk, which are caused by small and unrecorded fluctuations in the production process (such as uneven electrolyte infiltration, micro-wrinkles on the separator, instantaneous abnormality in the state of the device, etc.). These cells, while acceptable in initial static parameters, may fail prematurely during long-term cycling, severely affecting the overall reliability of the battery system. The addition of additional physical detection procedures (such as multiplying power discharge, direct current internal resistance and the like) can improve the screening effect, but can significantly increase the production time and the equipment cost. Therefore, a new method for performing deep diagnosis and identification of hidden abnormalities of a battery in a low-cost and efficient manner using existing production data is needed. Disclosure of Invention The embodiment of the invention provides a lithium battery production process abnormality diagnosis method and system based on prediction capacity deviation, which are used for solving the problems in the background technology. The lithium battery production process abnormality diagnosis method based on the predicted capacity deviation comprises the following steps of; step A, acquiring production process data, namely acquiring multi-dimensional production process data covering at least two working procedures in lithium battery production to form a process data vector corresponding to the lithium battery; Training a predictive model, namely training a capacity predictive model based on an FT-transducer (deep learning architecture based on a multi-head self-attention mechanism) architecture by adopting historical process data vectors respectively corresponding to a plurality of lithium batteries and actual capacity values corresponding to the lithium batteries; step C, predicting the capacity of the battery to be diagnosed, and inputting a process data vector of the lithium battery to be diagnosed into a trained capacity prediction model to obtain a predicted capacity value corresponding to the lithium battery to be diagnosed; Step D, performing abnormality diagnosis, namely acquiring an actual capacity value after the capacity division process of the lithium battery to be diagnosed, calculating an absolute error between the predicted capacity value and the actual capacity value, and comparing the absolute error with a preset abnormality judgment threshold value; And B, a capacity prediction model of the FT-transducer architecture in the step B comprises the following structures: Feature Tokenizer (feature tokenization module) which is used for uniformly converting the numerical type features and the category type features in the process data vector into feature embedded vectors with preset dimensions; a transducer encoder composed of at least one stacked transducer layer containing a multi-headed self-attention mechanism for processing information interaction of feature embedding vectors; And (3) a pre-measuring head, namely splicing CLSToken (classified word elements) at the front end of the characteristic embedding sequence to form input of a transducer encoder, and carrying out capacity prediction by utilizing the final output of the transducer encoder. The production process data of the step A comprises at least one or more of material batch information of an assembling process, weight information of a liquid injection process, process voltage and capacity information of a formation process, charging capacity and equipment information of a capacity division process, open circuit voltage and internal resistance information of each stage. The production process data of the step A further comprises battery type, material batch number, weight before and after liquid injection, formation process voltage and capacity data, capacity-dividing process charging capacity, capacity-dividing equipment and channel number, environment data and time data. The process data vector is used for representing core process data of a single lithium battery in the whole production process, is formed by splic