A method and system for detecting moisture content in an electric core baking process based on multi-sensor timing characteristics

Abstract

This invention discloses a method and system for detecting moisture content in the baking process of battery cells based on multi-sensor time-series features, relating to the field of lithium-ion battery cell manufacturing technology. The method includes: acquiring multiple target sensor data and corresponding moisture content labels for each baking step throughout the entire process cycle; preprocessing the target sensor data, followed by global feature extraction and local feature extraction; global feature extraction for all baking steps, and local feature extraction for each vacuum baking step within the baking process; optimizing and weighting the extracted features; constructing a moisture content detection model based on the LightGBM architecture, and training the moisture content detection model based on the optimization results, weight allocation results, and moisture content labels; and using the trained moisture content detection model to detect the moisture content during the battery cell baking process. This invention possesses technical advantages such as good real-time detection, comprehensive coverage, and high accuracy.

Description

Technical Field

[0001] This invention relates to the field of lithium-ion battery cell manufacturing technology, and in particular to a method and system for detecting moisture content in the battery cell baking process based on multi-sensor timing characteristics. Background Technology

[0002] In the lithium-ion battery cell manufacturing process, the baking process is a crucial step in removing residual moisture from the cell electrodes. The final moisture content of the cell directly affects the battery's cycle life and safety performance (for example, excessive moisture can easily lead to electrolyte decomposition, generating gas, causing cell swelling or even fire). Industry standards require that the final moisture content of power battery cells be controlled below 300 ppm, making precise control of the moisture content after baking essential. Currently, the industry commonly uses an "offline sampling inspection" model, where a portion of the cells are destructively sampled and inspected using a Karl Fischer moisture analyzer after the baking process to obtain the actual moisture content value. However, this model has significant limitations: firstly, the sampling ratio is low (usually 1%-5%), failing to cover the entire batch of cells and easily leading to missed detections of defective products; secondly, the detection has a strong lag (20-60 minutes from sampling to results), and if excessive moisture is found in a batch, the baked cells must be reworked, resulting in a significant waste of energy, time, and materials.

[0003] Existing moisture control technologies for battery cell baking processes mostly focus on static threshold control of parameters such as temperature and pressure in the baking equipment, such as setting fixed vacuum baking temperatures (80-120℃) and vacuum levels (≤150Pa). However, they do not consider the impact of batch-to-batch differences in battery cells (such as initial electrode humidity and cell thickness) and dynamic fluctuations in parameters during the process (such as temperature rise rate and vacuum level fluctuations) on the final moisture content. Some technologies attempt to predict moisture content by constructing simple regression models using single sensor data (such as temperature time series), but because they do not integrate multi-sensor information and do not extract features for key processes, the prediction accuracy is low (error rate ≥15%), which cannot meet the needs of industrial applications.

[0004] As battery cell manufacturing moves towards higher consistency and faster production cycles, traditional offline sampling inspection and single-parameter control methods are no longer suitable for large-scale production needs. There is an urgent need for a technology that can detect the final moisture content of battery cells in real time, across all batches, and with high precision, in order to achieve closed-loop control of the baking process, reduce rework costs, and improve product consistency. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method and system for detecting the moisture content of battery cell baking process based on multi-sensor time-series characteristics, thereby solving the technical problems of detection lag, limited coverage and low accuracy of the existing detection methods.

[0006] To achieve the above objectives, the present invention is implemented using the following technical solution:

[0007] In a first aspect, the present invention provides a method for detecting moisture content in a battery cell baking process based on multi-sensor time-series characteristics, comprising:

[0008] Acquire multiple target sensor data and corresponding moisture content labels for each baking step throughout the entire process cycle;

[0009] The target sensing data is preprocessed, and global feature extraction and local feature extraction are performed on the preprocessed target sensing data; the global feature extraction is performed on all baking steps, and the local feature extraction is performed on each vacuum baking step in the baking process.

[0010] The extracted global and local features are optimized and weighted.

[0011] A moisture content detection model based on the LightGBM architecture was constructed, and the moisture content detection model was trained based on the optimization processing results, weight allocation results, and moisture content labels.

[0012] The moisture content detection model is trained to detect the moisture content during the cell baking process.

[0013] The present invention provides a method for detecting the moisture content of battery cells during the baking process. This method employs a dual-layer feature mechanism—global feature extraction and local feature extraction—to achieve accurate coverage of process information. By combining the extracted dual-layer features with a moisture content detection model based on the LightGBM architecture, the final moisture content of the battery cells can be accurately detected, avoiding missed detection of defective products due to detection bias. This invention enables real-time detection of the moisture content of battery cells in the baking chamber, replacing traditional Karl Fischer offline sampling detection, and allows for real-time, full-batch, and high-precision detection.

[0014] Optionally, the target sensing data includes sensing data output from the temperature sensor, moisture sensor, and air pressure sensor inside the battery cell baking oven.

[0015] This invention uses three types of sensor data related to the moisture content of the battery cell baking process—temperature data, moisture data, and air pressure data—inside the battery cell baking oven as target sensor data. For example, increased temperature accelerates moisture evaporation, and increased vacuum promotes moisture removal. By integrating these three types of time-series data, the accuracy of moisture content detection in the battery cell baking process is improved.

[0016] Optionally, the preprocessing of the target sensing data includes:

[0017] Outliers in the target sensing data are removed using the 3σ principle, and the moving average of multiple adjacent data points is used to replace the removed outliers.

[0018] For missing values ​​in the target sensing data, linear interpolation is used to repair them;

[0019] The start time of each baking step is determined by the switching signal of the baking step, and the target sensing data within the baking step is time-aligned based on the start time of each baking step.

[0020] This invention prevents the learning process from being distorted by a few extreme data points through outlier removal, allowing the model to focus more on the patterns in the main data and resulting in more robust predictions. Missing value repair ensures the integrity of the dataset, avoiding training errors or insufficient information utilization due to missing data, and ensuring the model can learn based on complete information. Time alignment ensures that all features are analyzed on the same time frame, making the causal relationships and time series patterns captured by the model truly effective, which is crucial for time series prediction.

[0021] Optionally, the global features extracted include:

[0022] For each item, the target sensor data includes the maximum, minimum, mean, median, range, variance, standard deviation, coefficient of variation, amplitude, frequency, area, mean slope, variance of slope, mean amplitude of abrupt change, number of abrupt changes, peak factor, impulse factor, and waveform factor in all baking steps.

[0023] The local features extracted include:

[0024] The target sensor data for each item includes the maximum, minimum, mean, median, range, variance, standard deviation, coefficient of variation, amplitude, area, mean slope, peak factor, impulse factor, and waveform factor in each vacuum baking step.

[0025] This invention employs a dual-layer feature mechanism of "global + local". The global layer extracts 18 types of dynamic and static features for all processes, while the local layer focuses on extracting 14 types of core features for each vacuum baking process. This achieves precise coverage of process information and provides data support for subsequent accurate detection of the final moisture content of the battery cell.

[0026] Optionally, the optimization processing and weight allocation of the extracted global and local features include:

[0027] The extracted global and local features are normalized.

[0028] Set the weight of each feature in the global feature set to 1;

[0029] Based on the local features of each vacuum baking step, the characteristic variance of each feature in the local features is calculated;

[0030] Feature terms with a variance less than a variance threshold are removed from the local features;

[0031] Calculate the mutual information entropy between each feature removed from the local features and the moisture content label, and select the feature with the largest mutual information entropy value as the preferred local features.

[0032] The weights of the preferred local features are calculated based on mutual information entropy:

[0033]

[0034] In the formula, For the first The weights and mutual information entropy of the preferred local features, To optimize the maximum and minimum values ​​of mutual information entropy of local features.

[0035] This invention normalizes global and local features, and uses Min-Max normalization to map all features to the [0,1] interval to eliminate the influence of dimensions on model training, addressing the dimensional differences of different features. Redundant features are filtered out using feature variance and mutual information entropy, reducing feature dimensionality and improving model training efficiency. The correlation between local features (vacuum baking step) and moisture content is quantified using mutual information entropy, assigning high weights to highly correlated features to strengthen the model's contribution of core process information and improve detection accuracy.

[0036] Optionally, training the moisture content detection model based on the optimization processing results, weight allocation results, and moisture content labels includes:

[0037] A Bayesian optimization algorithm is used, with the root mean square error of cross-validation as the objective function, to automatically search for the optimal combination of hyperparameters.

[0038] The optimized global and local features are used as model inputs, and the moisture content label is used as the true value of the model output to construct a sample set;

[0039] The sample set is divided into a training set and a validation set according to a preset ratio. The weight allocation result is used as the initial parameter weights of the model, and the model is trained using a batch incremental training mode based on the training set.

[0040] After each training round, the root mean square error and mean absolute percentage error are used as evaluation metrics based on the validation set for validation until the validation is successful.

[0041] This invention employs a Bayesian optimization algorithm to automatically optimize hyperparameters such as learning rate, tree depth, and number of leaf nodes for different capacity cell models, thereby reducing generalization error. It also supports multi-production line model migration, requiring only minor adjustments to a small amount of target production line data for adaptation, thus shortening the deployment cycle.

[0042] Secondly, the present invention provides a cell baking process moisture content detection system based on multi-sensor time-series characteristics, comprising:

[0043] The data acquisition module is configured to acquire multiple target sensor data and corresponding moisture content labels for each baking step throughout the entire process cycle.

[0044] The feature extraction module is configured to preprocess the target sensing data, and to perform global feature extraction and local feature extraction on the preprocessed target sensing data; the global feature extraction is for all baking steps, and the local feature extraction is for each vacuum baking step in the baking process.

[0045] The feature processing module is configured to optimize and weight the extracted global and local features.

[0046] The model training module is configured to build a moisture content detection model based on the LightGBM architecture and train the moisture content detection model based on the optimization processing results, weight allocation results, and moisture content labels.

[0047] The model application module is configured to detect the moisture content of the battery cell baking process using the trained moisture content detection model.

[0048] Thirdly, the present invention provides an electronic device, including a processor and a storage medium;

[0049] The storage medium is used to store instructions;

[0050] The processor is configured to operate according to the instructions to perform the steps according to the method described above.

[0051] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.

[0052] Fifthly, the present invention provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the above-described method.

[0053] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:

[0054] The present invention provides a method and system for detecting the moisture content of battery cells during the baking process. This system employs a dual-layer feature mechanism—global and local feature extraction—to achieve accurate coverage of process information. By optimizing and weighting the extracted global and local features, redundant features are filtered out, and the model's contribution to core process information is strengthened. This not only improves model training efficiency but also enhances model prediction accuracy. Combining the extracted dual-layer features with a moisture content detection model based on the LightGBM architecture allows for accurate detection of the final moisture content of the battery cells, avoiding missed detection of defective products due to detection bias. This invention enables real-time detection of the moisture content of battery cells in the baking chamber, replacing traditional Karl Fischer offline sampling detection, and allows for real-time, full-batch, and high-precision detection. Attached Figure Description

[0055] Figure 1 This is a flowchart of a method for detecting moisture content in a battery cell baking process based on multi-sensor timing characteristics, provided in an embodiment of the present invention.

[0056] Figure 2 This is a diagram showing the gas pressure and process steps of the baking chamber provided in an embodiment of the present invention;

[0057] Figure 3 This is a step-by-step diagram showing the moisture content of the baking chamber in an embodiment of the present invention;

[0058] Figure 4 This is a temperature-process diagram of the baking cavity provided in an embodiment of the present invention. Detailed Implementation

[0059] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0060] Example 1

[0061] like Figure 1 As shown, this embodiment of the invention provides a method for detecting moisture content in a battery cell baking process based on multi-sensor time-series characteristics, including the following steps:

[0062] Step S1: Obtain multi-target sensor data and corresponding moisture content labels for each baking step throughout the entire process cycle.

[0063] Specifically in this embodiment, such as Figures 2 to 4 The temperature, moisture, and air pressure data inside the cell baking oven are three types of data that are closely related to the moisture content of the cell baking process. Therefore, the target sensing data includes the sensing data output by the temperature sensor, moisture sensor, and air pressure sensor inside the cell baking oven. The deployment parameters of the sensors are shown in Table 1.

[0064] Table 1: Sensor Deployment Parameters

[0065]

[0066] The PLC control system of the battery cell baking oven collects the switching signals (such as the trigger time of "preheating → preheating breathing") of 21 baking steps (some are repeated, 1 preheating, 1 preheating breathing, 9 vacuum baking, 9 vacuum breathing, and 1 baking completion), among which the "vacuum baking" process (9 in total) is the core stage of moisture removal. The system records the start timestamp of each step (accurate to the second) to ensure that the sensor timing data is accurately aligned with the steps.

[0067] The Doris database is used to store sensor timing data and process signals, and an index is created according to "oven number - cavity number - cell tray number - baking start time". At the same time, the moisture values ​​of sampled cells detected by the Karl Fischer moisture meter (1-5 cells per batch) are used as tags and associated with the sensor data of the corresponding batch to form a labeled training dataset.

[0068] This invention uses three types of sensor data related to the moisture content of the battery cell baking process—temperature data, moisture data, and air pressure data—inside the battery cell baking oven as target sensor data. For example, increased temperature accelerates moisture evaporation, and increased vacuum promotes moisture removal. By integrating these three types of time-series data, the accuracy of moisture content detection in the battery cell baking process is improved.

[0069] Step S2: Preprocess the target sensing data, and perform global feature extraction and local feature extraction on the preprocessed target sensing data; global feature extraction is performed on all baking steps, and local feature extraction is performed on each vacuum baking step in the baking process.

[0070] Step S2.1, Preprocessing the target sensing data includes:

[0071] (1) The outliers in the target sensing data are removed by the 3σ principle, and the moving average of multiple adjacent data points is used to replace the removed outliers.

[0072] If data points exceed A range (such as a sudden temperature jump to 200℃ or a sudden increase in air pressure to 200,000 Pa) is considered an abnormal value. The mean and standard deviation of the data points are used; the moving average of five adjacent data points is used to replace outliers to avoid data breaks and prevent a few extreme data points from distorting the model's learning process, making the model pay more attention to the main data patterns and the prediction results more robust.

[0073] (2) For missing values ​​in the target sensing data, linear interpolation is used for repair. The repair formula is as follows:

[0074]

[0075] In the formula, These represent the start and end times of the missing segment. For data points at the start time and end time Data values, For any time within the missing segment, This is the repair value.

[0076] Missing value repair ensures the integrity of the dataset, avoids model training errors or insufficient information utilization due to missing data, and ensures that the model can learn based on complete information.

[0077] (3) The start time of each baking step is determined by the switching signal of the baking step, and the target sensing data in the baking step is time-aligned based on the start time of each baking step.

[0078] Using the "preheating step start time" as a benchmark, time-series data can be segmented according to the step to form a "step-time series" associated dataset (such as "vacuum baking 1 - temperature time series" and "vacuum breathing 2 - air pressure time series"). Time alignment ensures that all features are analyzed under the same time benchmark, making the causal relationships and time series patterns captured by the model truly effective, which is crucial for time series prediction.

[0079] Step S2.2, Global features extracted include:

[0080] The maximum, minimum, mean, median, range, variance, standard deviation, coefficient of variation, amplitude, frequency, area, mean slope, variance slope, mean abrupt change amplitude, number of abrupt changes, peak factor, impulse factor, and waveform factor of each target sensor data in all baking steps are shown in Table 2.

[0081] Table 2: Global features extracted from global features

[0082]

[0083] For the time-series data of the entire process cycle of three types of sensors—temperature, moisture, and air pressure—18 types of global statistical features were calculated, resulting in 54 global features.

[0084] Step S2.3, Local features extracted include:

[0085] The target sensor data for each item includes the maximum, minimum, mean, median, range, variance, standard deviation, coefficient of variation, amplitude, area, mean slope, peak factor, impulse factor, and waveform factor in each vacuum baking step.

[0086] For the nine core processes—4 (vacuum baking 1), 6 (vacuum baking 2), 8 (vacuum baking 3), 10 (vacuum baking 4), 12 (vacuum baking 5), 14 (vacuum baking 6), 16 (vacuum baking 7), 18 (vacuum baking 8), and 20 (vacuum baking 9)—local features were extracted for each process. For the three types of sensor data in each process, 14 types of local features were calculated (excluding indicators with low correlation to vacuum baking, such as "frequency" and "number of mutations"). Each vacuum baking process generated 3 types of sensors × 14 indicators = 42 features, resulting in a total of 378 local features across the nine processes. The 54 global features and 378 local features were integrated to form a 432-dimensional initial feature vector. Each feature vector corresponds to the baking process data and the final moisture content label within a chamber.

[0087] This invention employs a dual-layer feature mechanism of "global + local". The global layer extracts 18 types of dynamic and static features for all processes, while the local layer focuses on extracting 14 types of core features for each vacuum baking process. This achieves precise coverage of process information and provides data support for subsequent accurate detection of the final moisture content of the battery cell.

[0088] Step S3: Optimize and weight the extracted global and local features, specifically including:

[0089] Step S3.1: Normalize the extracted global and local features.

[0090] This invention normalizes global and local features and uses Min-Max normalization to map all features to the [0,1] interval to eliminate the influence of dimensions on model training, taking into account the differences in the dimensions of different features.

[0091] Step S3.2: Set the weight of each feature in the global features to 1.

[0092] Step S3.3: Based on the local features of each vacuum baking step, calculate the characteristic variance of each feature in the local features.

[0093] Step S3.4: Remove feature terms whose feature variance is less than the variance threshold from the local features.

[0094] In this embodiment, the variance threshold is set to 0.001, and features with variance ≤ 0.001 are removed (such as the average air pressure feature; if the variance of all average air pressures is calculated and the fluctuation is very small, it indicates that the feature has no change and has little impact on the result, so it can be removed), retaining 300-320 feature items.

[0095] Step S3.5: Calculate the mutual information entropy between each feature removed from the local features and the moisture content label, and select the feature with the largest number of mutual information entropy values ​​as the preferred local features.

[0096] The formula for calculating mutual information entropy is:

[0097]

[0098] In the formula, For variables and variables Mutual information entropy, It is a variable Values and variables Values Joint probability distribution at time, Variables Values and variables Values Marginal probability distribution at time, For variables and variables All possible values ​​for .

[0099] In this embodiment, the features are sorted from high to low according to mutual information entropy, and the first 280-300 features are retained (redundant features with low correlation are removed).

[0100] Step S3.6: Calculate the weights of the preferred local features based on mutual information entropy.

[0101]

[0102] In the formula, For the first The weights and mutual information entropy of the preferred local features, To optimize the maximum and minimum values ​​of mutual information entropy of local features.

[0103] Redundant features are filtered out using feature variance and mutual information entropy to reduce feature dimensionality and improve model training efficiency. Mutual information entropy is used to quantify the correlation between local features (vacuum baking step) and moisture content, assigning high weights to highly correlated features to strengthen the model's contribution of core process information and improve detection accuracy.

[0104] Step S4: Construct a moisture content detection model based on the LightGBM architecture, and train the moisture content detection model based on the optimization processing results, weight allocation results, and moisture content labels.

[0105] Specifically, in this embodiment, a moisture content detection model based on the LightGBM architecture is adopted, which is optimized for multi-feature nonlinear regression scenarios. The initial hyperparameter settings are shown in Table 3.

[0106] Table 3: Initial hyperparameters of the moisture content detection model

[0107]

[0108] The Bayesian optimization algorithm is used, with the root mean square error of cross-validation as the objective function, to automatically search for the optimal combination of hyperparameters, including:

[0109] a. Define the search space: Set the value range of each hyperparameter (e.g., learning rate 0.001-0.01, tree depth 5-10) to construct the hyperparameter search space;

[0110] b. Initialize samples: Randomly select 5 sets of initial hyperparameters, train the model and calculate the corresponding RMSE;

[0111] c. Iterative optimization: Based on the mapping relationship between historical hyperparameters and RMSE, the optimal hyperparameter candidates are predicted through a Gaussian process model. In each iteration, 3 sets of candidate hyperparameters are trained, RMSE is calculated, and historical samples are updated.

[0112] d. Termination condition: After 30 iterations, select the hyperparameter combination with the smallest RMSE as the optimal model hyperparameters (e.g., the optimal hyperparameters for a certain square battery cell are: learning rate 0.005, tree depth 7, number of leaf nodes 30, number of trees 50, reg_alpha 0.3).

[0113] Model training and validation:

[0114] The optimized global and local features are used as model inputs, and the moisture content label is used as the true value of the model output to construct a sample set;

[0115] The sample set is divided into a training set and a validation set according to a preset ratio (7:3). The training set is used for model parameter learning, and the validation set is used for generalization ability evaluation.

[0116] The weight allocation results are used as the initial parameter weights of the model, and the model is trained using a batch incremental training mode based on the training set. For every 1,000 new labeled data, incremental fine-tuning is performed based on the historical best parameters (iteration 50 epochs) to avoid the efficiency loss caused by full retraining.

[0117] After each training round, the root mean square error and mean absolute percentage error are used as evaluation metrics based on the validation set for validation until the validation is successful.

[0118]

[0119]

[0120] In the formula, These are the root mean square error and the mean absolute percentage error, respectively. For the first The true and predicted values ​​output by the model for each sample; This represents the number of samples.

[0121] In this embodiment, the verification is considered successful if the verification set RMSE ≤ 4ppm and MAPE ≤ 8%.

[0122] This invention employs a Bayesian optimization algorithm to automatically optimize hyperparameters such as learning rate, tree depth, and number of leaf nodes for different capacity cell models, thereby reducing generalization error. It also supports multi-production line model migration, requiring only minor adjustments to a small amount of target production line data for adaptation, thus shortening the deployment cycle.

[0123] Step S5: Detect the moisture content of the battery cell during the baking process using the trained moisture content detection model.

[0124] The trained moisture content detection model is exported in ONNX format and deployed to edge computing nodes on the production line (such as industrial servers), supporting real-time data input (the input time for feature vectors of each cell is ≤1s) and prediction output.

[0125] In summary, the cell baking process moisture content detection method provided by this invention achieves accurate coverage of process information through a dual-layer feature mechanism of global and local feature extraction. By combining the extracted dual-layer features with a moisture content detection model based on the LightGBM architecture, the final moisture content of the cell can be accurately detected, avoiding missed detection of defective products due to detection bias. This invention enables real-time detection of cell moisture content in the baking chamber, replacing traditional Karl Fischer offline sampling detection, and allows for real-time, full-batch, and high-precision detection.

[0126] Based on the moisture content detection results of the battery cell baking process, the following follow-up steps can be achieved:

[0127] 1) Real-time result output:

[0128] Visualization: The predicted moisture content of each batch of battery cells is displayed in real time on the production line monitoring screen, with different colors used to indicate the status (green: predicted moisture, red: predicted moisture exceeds the standard).

[0129] Data push: The prediction results (including oven, chamber, cell number, baking start time, predicted moisture value, and feature vector) are pushed to the MES system in real time and stored in a relational database (such as Doris), supporting historical queries by chamber and time range.

[0130] 2) Anomaly detection and alarm:

[0131] Judgment rule: Set a qualified moisture content threshold (300ppm for power batteries), or if the predicted moisture content of the cells in the chamber is ≥300ppm (power batteries), the batch is judged as abnormal.

[0132] Alarm Trigger: When a batch is abnormal, a text message is sent to the process engineer (including the abnormal oven, chamber, baking start time, batch number, current predicted average moisture content, and possible causes of the abnormality, such as "temperature variance of vacuum baking step 3 in oven 3, chamber 2 exceeds the standard").

[0133] 3) Process closed-loop adjustment:

[0134] Parameter adjustment logic: If a batch is determined to be abnormal, the system automatically sends a parameter adjustment command to the oven PLC to optimize parameters for subsequent vacuum baking steps (e.g., increase temperature by 5-10℃, extend time by 5-10 minutes). The adjustment rules are as follows:

[0135] If the moisture content is predicted to be 300-400ppm (power battery): increase the vacuum baking temperature by 5℃ and extend the baking time by 5min;

[0136] If the predicted moisture content is >400ppm (power battery): increase the vacuum baking temperature by 10℃ and extend the baking time by 10min;

[0137] Effect verification: After parameter adjustment, the predicted moisture content of subsequent cells is monitored in real time. If the predicted moisture content of 10 consecutive cells is ≤280ppm, the parameter adjustment is automatically stopped.

[0138] 4) Anomaly tracing and analysis:

[0139] Feature backtracking: For abnormal cells, automatically extract their global and local features (such as the average temperature and pressure variance of the two vacuum baking steps), compare them with the feature distribution of historical normal cells, and generate a "feature deviation report";

[0140] Root cause identification: Based on the characteristic deviation report and combined with process knowledge (such as low average temperature may lead to insufficient moisture removal), possible root causes of abnormalities are recommended (such as "aging of oven heating tubes leading to insufficient vacuum baking temperature") to assist engineers in quickly troubleshooting.

[0141] Example 2

[0142] This invention provides a cell baking process moisture content detection system based on multi-sensor time-series characteristics, comprising:

[0143] The data acquisition module is configured to acquire multiple target sensor data and corresponding moisture content labels for each baking step throughout the entire process cycle.

[0144] The feature extraction module is configured to preprocess the target sensing data and perform global and local feature extraction on the preprocessed target sensing data. Global feature extraction is performed on all baking steps, while local feature extraction is performed on each vacuum baking step within the baking process.

[0145] The feature processing module is configured to optimize and weight the extracted global and local features.

[0146] The model training module is configured to build a moisture content detection model based on the LightGBM architecture, and train the moisture content detection model based on the optimization results, weight allocation results, and moisture content labels.

[0147] The model application module is configured to detect the moisture content of the battery cell baking process using a trained moisture content detection model.

[0148] Example 3

[0149] Based on the cell baking process moisture content detection method provided in Embodiment 1, this embodiment of the invention provides an electronic device, including a processor and a storage medium;

[0150] Storage media are used to store instructions;

[0151] The processor is used to perform operations according to instructions to execute the steps according to the method described above.

[0152] Example 4

[0153] Based on the cell baking process moisture content detection method provided in Embodiment 1, this embodiment of the invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above method.

[0154] Example 5

[0155] Based on the cell baking process moisture content detection method provided in Embodiment 1, this embodiment of the invention provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the above method.

[0156] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied 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.

[0157] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations 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, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0158] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0159] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0160] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for detecting moisture content in battery cell baking process based on multi-sensor time-series characteristics, characterized in that, include: Acquire multiple target sensor data and corresponding moisture content labels for each baking step throughout the entire process cycle; The target sensing data is preprocessed, and global feature extraction and local feature extraction are performed on the preprocessed target sensing data; the global feature extraction is performed on all baking steps, and the local feature extraction is performed on each vacuum baking step in the baking process. The extracted global and local features are optimized and weighted. A moisture content detection model based on the LightGBM architecture was constructed, and the moisture content detection model was trained based on the optimization processing results, weight allocation results, and moisture content labels. The moisture content detection model is trained to detect the moisture content during the battery cell baking process.

2. The method for detecting moisture content in battery cell baking process based on multi-sensor time-series characteristics according to claim 1, characterized in that, The target sensing data includes sensing data output from temperature sensors, moisture sensors, and air pressure sensors inside the battery cell baking oven.

3. The method for detecting moisture content in battery cell baking process based on multi-sensor time-series characteristics according to claim 1, characterized in that, The preprocessing of the target sensing data includes: Outliers in the target sensing data are removed using the 3σ principle, and the moving average of multiple adjacent data points is used to replace the removed outliers. For missing values ​​in the target sensing data, linear interpolation is used to repair them; The start time of each baking step is determined by the switching signal of the baking step, and the target sensing data within the baking step is time-aligned based on the start time of each baking step.

4. The method for detecting moisture content in battery cell baking process based on multi-sensor time-series characteristics according to claim 1, characterized in that, The global features extracted include: For each item, the target sensor data includes the maximum, minimum, mean, median, range, variance, standard deviation, coefficient of variation, amplitude, frequency, area, mean slope, variance of slope, mean amplitude of abrupt change, number of abrupt changes, peak factor, impulse factor, and waveform factor in all baking steps. The local features extracted include: The target sensor data for each item includes the maximum, minimum, mean, median, range, variance, standard deviation, coefficient of variation, amplitude, area, mean slope, peak factor, impulse factor, and waveform factor in each vacuum baking step.

5. The method for detecting moisture content in battery cell baking process based on multi-sensor time-series characteristics according to claim 1, characterized in that, The optimization and weight allocation of the extracted global and local features includes: The extracted global and local features are normalized. Set the weight of each feature in the global feature set to 1; Based on the local features of each vacuum baking step, the characteristic variance of each feature in the local features is calculated; Feature terms with a variance less than the variance threshold are removed from the local features; Calculate the mutual information entropy between each feature removed from the local features and the moisture content label, and select the feature with the largest number of mutual information entropy values ​​as the preferred local features. The weights of the preferred local features are calculated based on mutual information entropy: In the formula, For the first The weights and mutual information entropy of the preferred local features, To optimize the maximum and minimum values ​​of mutual information entropy of local features.

6. The method for detecting moisture content in battery cell baking process based on multi-sensor time-series characteristics according to claim 1, characterized in that, The training of the moisture content detection model based on the optimization results, weight allocation results, and moisture content labels includes: A Bayesian optimization algorithm is used, with the root mean square error of cross-validation as the objective function, to automatically search for the optimal combination of hyperparameters. The optimized global and local features are used as model inputs, and the moisture content label is used as the true value of the model output to construct a sample set; The sample set is divided into a training set and a validation set according to a preset ratio. The weight allocation result is used as the initial parameter weights of the model, and the model is trained using a batch incremental training mode based on the training set. After each training round, the root mean square error and mean absolute percentage error are used as evaluation metrics based on the validation set for validation until the validation is successful.

7. A cell baking process moisture content detection system based on multi-sensor time-series characteristics, characterized in that, include: The data acquisition module is configured to acquire multiple target sensor data and corresponding moisture content labels for each baking step throughout the entire process cycle. The feature extraction module is configured to preprocess the target sensing data, and to perform global feature extraction and local feature extraction on the preprocessed target sensing data; the global feature extraction is for all baking steps, and the local feature extraction is for each vacuum baking step in the baking process. The feature processing module is configured to optimize and weight the extracted global and local features. The model training module is configured to build a moisture content detection model based on the LightGBM architecture and train the moisture content detection model based on the optimization processing results, weight allocation results, and moisture content labels. The model application module is configured to detect the moisture content of the battery cell baking process using the trained moisture content detection model.

8. An electronic device, characterized in that, Including processor and storage media; The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the steps of the method according to any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the steps of the method according to any one of claims 1-6.

10. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method described in any one of claims 1-6.

Images