Method and system for predicting welding tensile force of cell tab

Abstract

The application discloses a kind of battery tab welding tension prediction method and system, method includes the acquisition of battery tab welding process multidimensional data, the multidimensional data includes ultrasonic welding data, incoming data and production environment data;The multidimensional data is pretreated, and is converted into time series data;Using long short-term memory network, the time series data is welded to tension prediction, and the next time welding tension prediction result is obtained;Based on the next time welding tension prediction result, early warning is carried out.The application provides an ultrasonic welding tension online detection scheme, improves the accuracy of welding tension prediction, improves the effectiveness of sampling inspection, reduces sampling frequency, reduces test cost, improves battery quality and yield.

Description

Technical Field

[0001] This invention relates to the field of lithium-ion battery technology, specifically to a method and system for predicting the welding pull force of battery cell tabs. Background Technology

[0002] Lithium-ion batteries are widely used in various electronic devices and vehicles due to their advantages such as high specific energy, long cycle life, low self-discharge, no memory effect, and no pollution. From the manufacturing of lithium battery cells to the assembly of battery packs, the welding process is a crucial step in battery manufacturing, directly affecting the cost, quality, safety, and consistency of the battery.

[0003] In battery production, the connection between foils and between foils and tabs of individual battery cells requires ultrasonic metal welding technology. Testing the welding pull force of the cell tabs to the nickel / copper / cover plate connection is a critical step. Currently, the commonly used testing method is periodic sampling, where cells from the production process are randomly selected, and then a tensile testing machine is used to perform welding pull force tests on them using appropriate fixtures. The test results are then used to evaluate the welding quality in actual production. For example, Chinese invention patent application CN113466033A discloses a method for testing the welding pull force of lithium battery tabs, in which a robotic arm of a tensile testing instrument holds the tab, and another robotic arm holds the overlap between the current collector and the adhesive tape, performing a pull force test on the tab.

[0004] However, the current technical solutions for ultrasonic welding tensile testing in the lithium battery industry are all to improve the accuracy of measurement by improving the testing methods or testing fixtures. They are all offline testing solutions, and the defects are: (1) The effectiveness of sampling inspection is poor. Welding quality can only be indirectly reflected by increasing the sampling inspection frequency. However, the tensile testing process is a destructive test and the testing cost is high. It cannot increase the sampling inspection frequency and cannot prevent batches of poor welding quality problems such as false welding and over-welding in the produced cells; (2) During the tensile testing process, since the test is performed by the staff manually, the direction of the stretching is prone to tilting, which can easily lead to inaccurate tensile values ​​displayed by the tensile testing machine. It cannot truly and effectively reflect the welding quality between the positive and negative tabs of the cell and the positive and negative soft connections during the production process, thus causing batches of poor welding quality problems such as false welding and over-welding in the produced cells, reducing the quality and yield of the battery.

[0005] The paper "Prediction of Mechanical Properties of Welded Joints Using Generalized Dynamic Fuzzy Neural Networks, Journal of Welding Engineering Technology I, Zhang Yongzhi" proposes to predict the mechanical properties of welded joints by establishing a generalized dynamic fuzzy neural network. This scheme uses predictions for the mechanical properties of laser-welded joints, focusing more on lifespan control, and does not predict the tensile force of ultrasonic welding.

[0006] Ultrasonic welding is a contact welding method, and real-time monitoring is not possible during the welding process. Current online solutions only monitor certain influencing factors. Ultrasonic welding differs from laser welding, where a laser beam shines from above and melts the strip metal through a certain motion, creating a "channel" that runs through the strip metal and the joint surface, forming a connection. During the welding process, the width and depth of the weld can be strictly controlled by the oscillation of the laser beam. Summary of the Invention

[0007] The technical problem to be solved by this invention is how to predict the tensile force trend in ultrasonic welding and improve the effectiveness of random inspection.

[0008] The present invention solves the above-mentioned technical problems through the following technical means:

[0009] On the one hand, a method for predicting the welding pull force of battery cell tabs is proposed, the method comprising:

[0010] Collect multidimensional data of the battery cell tab welding process, including ultrasonic welding data, incoming material data, and production environment data.

[0011] The multidimensional data is preprocessed and converted into time-series data;

[0012] A long short-term memory network is used to predict the welding tensile force of the time series data to obtain the welding tensile force prediction result at the next time step.

[0013] An early warning is issued based on the predicted welding tensile force at the next moment.

[0014] Furthermore, the ultrasonic welding data includes welding process data, welding machine data, welding machine internal data, and CCD welding appearance inspection data; the incoming material data includes IQC incoming material surface roughness and warehousing time; and the production environment data includes temperature data and humidity data.

[0015] Further, the preprocessing of the multidimensional data into time-series data includes:

[0016] Based on the CCD welding appearance inspection data, the number of weld points and the morphological feature values ​​of the weld points are extracted.

[0017] Fourier transform is performed on the welding process data to extract frequency information during the welding process;

[0018] The welding machine data, the welding machine internal data, the number of weld points and the weld point morphology characteristics, the frequency information, the incoming material data and the production environment data are standardized to obtain data that is normally distributed.

[0019] The normally distributed data is converted into time series data.

[0020] Further, the frequency information includes the mean frequency spectrum, root mean square frequency spectrum, centroid frequency, root mean square frequency, and standard deviation frequency; the step of performing a Fourier transform based on the welding process data to extract frequency information during the welding process includes:

[0021] The Fourier transform of the welding process data is expressed as follows:

[0022]

[0023] In the formula: x(kΔtz) is the sampled value of the vibration signal; N is the number of sampling points; Δt is the sampling interval; k is the index of the discrete value in the time domain;

[0024] Based on the Fourier transform results, the spectral mean F during the welding process is extracted. mean , spectral root mean square value F mean square root The frequency centroid F1, root mean square frequency F2, and standard deviation frequency F3 are expressed by the following formula:

[0025]

[0026]

[0027]

[0028]

[0029]

[0030] In the formula: s(k) is the spectrum of the signal x(n), k = 1, 2, 3, ..., K-1, K; K is the number of spectra; f k This represents the frequency value of the k-th spectral line.

[0031] Furthermore, the Long Short-Term Memory network includes a convolutional layer, a max-pooling layer, an LSTM layer, and a fully connected layer connected in sequence;

[0032] The Long Short-Term Memory network is a pre-trained neural network used for predicting welding tensile strength, and the loss function used in the network training process is MAPE.

[0033] Furthermore, the early warning based on the predicted welding tensile force at the next moment includes:

[0034] Based on the predicted welding tensile force at the next moment from multiple predictions, the average value of the predicted welding tensile force is calculated.

[0035] If the average welding tensile force at i consecutive welding points shows a downward trend, an early warning will be issued.

[0036] Furthermore, the early warning based on the predicted welding tensile force at the next moment includes:

[0037] Based on the predicted welding tensile force at the next moment, the predicted value of the welding tensile force is calculated.

[0038] An early warning will be issued if the predicted welding tensile force is not within the set threshold range.

[0039] Furthermore, the early warning based on the predicted welding tensile force at the next moment includes:

[0040] The standard deviation is calculated based on the predicted welding tensile force at the next moment after multiple predictions.

[0041] Early warning is issued for regions where the standard deviation fluctuation exceeds the threshold.

[0042] Furthermore, before preprocessing the multidimensional data and converting it into time-series data, the method further includes:

[0043] Based on the ultrasonic welding data, determine whether the welding machine is malfunctioning;

[0044] If the welding machine malfunctions, it will be stopped and an alarm will be triggered.

[0045] Secondly, a battery cell tab welding pull force prediction system is proposed, the system comprising:

[0046] The data acquisition module is used to collect multi-dimensional data of the battery cell tab welding process, including ultrasonic welding data, incoming material data, and production environment data.

[0047] The data conversion module is used to preprocess the multidimensional data and convert it into time-series data;

[0048] The tensile force prediction module is used to predict the welding tensile force of the time series data using a long short-term memory network, and to obtain the welding tensile force prediction result for the next moment.

[0049] The early warning module is used to issue an early warning based on the predicted welding tensile force at the next moment.

[0050] The advantages of this invention are:

[0051] (1) This invention provides an online detection scheme for ultrasonic welding tensile force. By acquiring multi-dimensional data of the cell tab welding process as input to a pre-trained long short-term memory network for welding tensile force prediction, the predicted value of welding tensile force is obtained. Based on the trend of the predicted value of welding tensile force, targeted sampling inspection is carried out to achieve dynamic sampling inspection, improve the effectiveness of sampling inspection and reduce the sampling inspection frequency, further reduce testing costs, and prevent batch of poor welding quality problems such as false welding and over-welding of produced cells. At the same time, it improves battery quality and yield, and achieves the dual benefits of human resources and quality.

[0052] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0053] Figure 1 This is a flowchart illustrating a method for predicting the welding pull force of battery cell tabs according to an embodiment of the present invention.

[0054] Figure 2 This is a block diagram illustrating the principle of majority data processing and welding tensile force prediction proposed in one embodiment of the present invention;

[0055] Figure 3 This is a structural diagram of a long short-term memory network in one embodiment of the present invention;

[0056] Figure 4 This is a schematic diagram of the overall process of the method for predicting the welding pull force of battery cell tabs according to an embodiment of the present invention;

[0057] Figure 5 This is a schematic diagram of the structure of a battery cell tab welding pull force prediction system proposed in an embodiment of the present invention. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0059] like Figure 1 As shown, the first embodiment of the present invention proposes a method for predicting the welding pull force of battery cell tabs, the method comprising the following steps:

[0060] S10. Collect multi-dimensional data of the battery cell tab welding process, including ultrasonic welding data, incoming material data and production environment data.

[0061] It should be noted that this embodiment collects multi-dimensional data during the battery cell tab welding process based on the principles of "people," "machine," "material," "method," and "environment."

[0062] S20. Preprocess the multidimensional data and convert it into time-series data;

[0063] S30. Use a long short-term memory network to predict the welding tensile force of the time series data to obtain the welding tensile force prediction result at the next moment.

[0064] S40. Based on the predicted welding tensile force at the next moment, issue an early warning.

[0065] This embodiment proposes an online detection scheme for ultrasonic welding tensile force. By acquiring multi-dimensional data of the battery cell tab welding process as input to a pre-trained long short-term memory network for welding tensile force prediction, a predicted welding tensile force value is obtained. Based on the trend of the predicted welding tensile force value, targeted sampling inspections are carried out to achieve dynamic sampling inspection, thereby improving the effectiveness of sampling inspections and reducing the sampling frequency. This further reduces testing costs and prevents batch defects such as poor welding and over-welding in the produced battery cells.

[0066] In one embodiment, the ultrasonic welding data includes welding process data, welding machine data, welding machine internal data, and CCD welding appearance inspection data; the incoming material data includes IQC incoming material surface roughness and warehousing time; and the production environment data includes temperature data and humidity data.

[0067] Among them, ultrasonic welding data and CCD welding appearance inspection data can be collected simply by activating the data acquisition function on the ultrasonic welding equipment. Incoming material data is obtained by sampling each batch of materials upon arrival at the factory, and production environment data is obtained through temperature and humidity sensors in the factory.

[0068] It should be noted that this embodiment uses a large number of actual cases of insufficient welding tensile strength and FTA analysis to identify the key factors affecting welding tensile strength. These factors directly affect welding tensile strength in actual production.

[0069] Specifically, the data during the ultrasonic welding process includes the vibration amplitude, vibration acceleration, and temperature of the welding head; the vibration amplitude, vibration acceleration, and temperature of the welding base; the ultrasonic welding machine data includes welding time, welding pressure, welding power, welding depth, and welding sound; the data inside the ultrasonic welding machine includes radio frequency current, radio frequency voltage, current and voltage frequency, and phase difference; and the CCD appearance inspection data after welding refers to the appearance data of the battery cell tabs after welding, captured by a CCD camera.

[0070] The incoming material data mainly refers to the data indicators monitored by IQC, such as the surface roughness of copper / aluminum foil, the surface roughness of copper / nickel tabs, and the time of warehousing.

[0071] In one embodiment, step S20: preprocessing the multidimensional data and converting it into time-series data includes the following steps:

[0072] S21. Based on the CCD welding appearance inspection data, extract the number of weld points and the morphological feature values ​​of the weld points;

[0073] It should be noted that the extraction of the number and appearance features of weld points can be achieved using existing machine vision shape-based feature extraction and quantity statistics methods. In addition, the number of weld points after welding will vary depending on the process. Due to wear of the welding head or differences in welding position, the actual number of weld points is less than or equal to the number of weld points on the welding head. The size and shape of the weld points will directly affect the strength of the welding tensile force. Therefore, CCD appearance inspection data is extracted as input features.

[0074] S22. Perform Fourier transform on the welding process data to extract frequency information during the welding process;

[0075] It should be noted that the reason for extracting frequency information in this embodiment is that the ultrasonic system consists of an ultrasonic generator, an amplitude transformer, a transducer, and an ultrasonic welding head. The ultrasonic generator emits a set frequency wave, which is transmitted to the ultrasonic welding head through the amplitude transformer and the transducer. Extracting frequency information can directly reflect the actual frequency during the production process. If the actual frequency is lower than the set frequency, it will directly affect the welding effect.

[0076] S23. Standardize the welding machine data, the welding machine internal data, the number of weld points and the weld point morphology characteristic value, the frequency information, the incoming material data and the production environment data to obtain data that is normally distributed.

[0077] It should be noted that, in terms of features, the data domain can be divided into time domain data and frequency domain data. For time domain data, statistical features (mean, variance, median, etc.) are used. For frequency domain data, the original data is often subjected to Fourier transform and then processed in a similar statistical manner. Finally, the data is standardized, and the processed data conforms to a normal distribution with a mean of 0 and a standard deviation of 1.

[0078] An ultrasonic system includes a generator, transducer, amplitude transformer, and welding joint. Many factors affect welding tensile strength. The mechanical properties of the welding joint are one of the characteristics that affect welding tensile strength. Welding tensile strength is a combined process, characterizing the tensile strength after the fusion of two materials. Therefore, the characteristics of the welding process are mainly multimodal characteristics (time domain, frequency domain, and image).

[0079] S24. Convert the normally distributed data into time-series data.

[0080] It should be noted that, since the model is based on time series forecasting, the input data format is as follows:

[0081] Feature_(t-4,t-3,t-2,t-1,t)->(Label_t)

[0082] Where t represents the feature data corresponding to the current time t, t-1 represents the feature data corresponding to the previous time, and Label_t represents the welding tension at the current time.

[0083] In one embodiment, the frequency information includes the mean frequency, root mean square frequency, centroid frequency, root mean square frequency, and standard deviation frequency; step S22: performing a Fourier transform based on the welding process data to extract the frequency information during the welding process, specifically including the following steps:

[0084] The Fourier transform of the welding process data is expressed as follows:

[0085]

[0086] In the formula: x(kΔtz) is the sampled value of the vibration signal; N is the number of sampling points; Δt is the sampling interval; k is the index of the discrete value in the time domain;

[0087] Based on the Fourier transform results, the spectral mean F during the welding process is extracted. mean , spectral root mean square value F mean square root The frequency centroid F1, root mean square frequency F2, and standard deviation frequency F3 are expressed by the following formula:

[0088]

[0089]

[0090]

[0091]

[0092]

[0093] In the formula: the meaning of k is consistent, and it is the index of the discrete value in the time domain; f k Let be the frequency value of the k-th spectral line.

[0094] In one embodiment, such as Figures 2 to 3As shown, the Long Short-Term Memory network includes a convolutional layer, a max pooling layer, an LSTM layer, and a fully connected layer connected in sequence.

[0095] The Long Short-Term Memory (LSTM) network is a pre-trained neural network used for predicting welding tensile strength. The loss function used during network training is MAPE (Mean Absolute Percentage Error), as shown in the following formula:

[0096]

[0097] In the formula: n represents the amount of data, A_t represents the actual value of the t-th data point, and F_t represents the predicted value of the t-th data point.

[0098] In this embodiment, the Long Short-Term Memory (LSTM) network is used to process time-series data better, enabling the prediction of long-term tensile trends rather than the prediction of single tensile forces.

[0099] In one embodiment, step S40: issuing an early warning based on the predicted welding tensile force at the next moment includes the following steps:

[0100] Based on the predicted welding tensile force at the next moment from multiple predictions, the average value of the predicted welding tensile force is calculated.

[0101] If the average welding tensile force at i consecutive welding points shows a downward trend, an early warning will be issued.

[0102] Specifically, this embodiment can calculate the average value of the predicted welding tensile force based on the predicted values ​​of the welding tensile force at the next moment of the five predictions, and issue an early warning when the predicted welding tensile force of three consecutive welding points shows a downward trend.

[0103] In one embodiment, step S40: issuing an early warning based on the predicted welding tensile force at the next moment includes the following steps:

[0104] Based on the predicted welding tensile force at the next moment from multiple predictions, the average value of the predicted welding tensile force is calculated.

[0105] An early warning is issued when the average value of the predicted welding tensile force is not within the set threshold range.

[0106] It should be understood that the threshold range set in this embodiment is an empirical value obtained through a large number of experiments and used to compare with the average value of the predicted welding tensile force. The threshold is set between 90% and 95% of the value set according to the process requirements.

[0107] In one embodiment, step S40: issuing an early warning based on the predicted welding tensile force at the next moment includes the following steps:

[0108] The standard deviation is calculated based on the predicted welding tensile force at the next moment after multiple predictions.

[0109] Early warning is issued for regions where the standard deviation fluctuation exceeds the threshold.

[0110] It should be understood that the fluctuation threshold set in this embodiment is an empirical value obtained through a large number of experiments for comparison with the fluctuation value of the standard deviation. The standard deviation is set based on 90% to 95% of the actual production capacity.

[0111] In one embodiment, such as Figure 4 As shown, before step S20: preprocessing the multidimensional data and converting it into time-series data, the method further includes the following steps:

[0112] Based on the ultrasonic welding data, determine whether the welding machine is malfunctioning;

[0113] If the welding machine malfunctions, it will be stopped and an alarm will be triggered.

[0114] It should be understood that if the welding is normal, the process of step S20 can be carried out directly.

[0115] It should be noted that the anomaly detection algorithm for key welding parameters can adopt conventional SPC control, and the control range of SPC can be determined according to the actual process conditions. In this embodiment, when judging welding anomalies, if the input parameters of the welding machine exceed the control range, an alarm can be triggered directly without further prediction.

[0116] In addition, such as Figure 5 As shown, the second embodiment of the present invention also proposes a battery cell tab welding pull force prediction system, the system comprising:

[0117] Data acquisition module 10 is used to acquire multi-dimensional data of the battery cell tab welding process, including ultrasonic welding data, incoming material data and production environment data.

[0118] Data conversion module 20 is used to preprocess the multidimensional data and convert it into time series data;

[0119] The tensile force prediction module 30 is used to predict the welding tensile force of the time series data using a long short-term memory network to obtain the welding tensile force prediction result at the next moment.

[0120] The early warning module 40 is used to issue an early warning based on the predicted welding tensile force at the next moment.

[0121] This embodiment proposes an online detection scheme for ultrasonic welding tensile force. By acquiring multi-dimensional data of the battery cell tab welding process as input to a pre-trained long short-term memory network for welding tensile force prediction, the predicted welding tensile force value is obtained, thereby improving the accuracy of the welding tensile force prediction result, thereby improving the effectiveness of sampling inspection, reducing the sampling inspection frequency, further reducing testing costs, and preventing batches of poor welding quality problems such as incomplete welding and over-welding in the produced battery cells.

[0122] In one embodiment, the ultrasonic welding data includes welding process data, welding machine data, welding machine internal data, and CCD welding appearance inspection data; the incoming material data includes IQC incoming material surface roughness and warehousing time; and the production environment data includes temperature data and humidity data.

[0123] In one embodiment, the data conversion module 20 includes:

[0124] The feature extraction unit is used to extract the number of weld points and the morphological feature values ​​of weld points based on the CCD welding appearance inspection data.

[0125] The frequency extraction unit is used to perform Fourier transform based on the welding process data to extract frequency information during the welding process;

[0126] The standardization unit is used to standardize the welding machine data, the welding machine internal data, the number of weld points and the weld point morphology characteristics, the frequency information, the incoming material data and the production environment data to obtain data that is normally distributed.

[0127] A conversion unit is used to convert the normally distributed data into time-series data.

[0128] In one embodiment, the frequency information includes the mean frequency spectrum, the root mean square frequency spectrum, the centroid of the frequency, the root mean square frequency, and the standard deviation frequency; the frequency extraction unit is specifically used to perform the following steps:

[0129] The Fourier transform of the welding process data is expressed as follows:

[0130]

[0131] In the formula: x(kΔtz) is the sampled value of the vibration signal; N is the number of sampling points; Δt is the sampling interval; k is the index of the discrete value in the time domain;

[0132] Based on the Fourier transform results, the spectral mean F during the welding process is extracted. mean , spectral root mean square value F mean square rootThe frequency centroid F1, root mean square frequency F2, and standard deviation frequency F3 are expressed by the following formula:

[0133]

[0134]

[0135]

[0136]

[0137]

[0138] In the formula: k = 1, 2, 3, ..., K-1, K; K is the spectrum number; f k This represents the frequency value of the k-th spectral line.

[0139] In one embodiment, the long short-term memory network includes a convolutional layer, a max pooling layer, an LSTM layer, and a fully connected layer connected in sequence.

[0140] The Long Short-Term Memory (LSTM) network is a pre-trained neural network used for predicting welding tensile force. The loss function used during network training is MAPE.

[0141] In one embodiment, the early warning module 40 includes a first early warning unit, specifically used to perform the following steps:

[0142] Based on the predicted welding tensile force at the next moment from multiple predictions, the average value of the predicted welding tensile force is calculated.

[0143] If the predicted welding tensile strength at i consecutive welding points shows a downward trend, an early warning will be issued.

[0144] In one embodiment, the early warning module 40 includes a second early warning unit, specifically used to perform the following steps:

[0145] Based on the predicted welding tensile force at the next moment from multiple predictions, the average value of the predicted welding tensile force is calculated.

[0146] An early warning is issued when the average value of the predicted welding tensile force is not within the set threshold range.

[0147] In one embodiment, the early warning module 40 includes a third early warning unit, specifically used to perform the following steps:

[0148] The standard deviation is calculated based on the predicted welding tensile force at the next moment after multiple predictions.

[0149] Early warning is issued for regions where the standard deviation fluctuation exceeds the threshold.

[0150] In one embodiment, the system further includes a welding machine anomaly detection unit, used for:

[0151] Based on the ultrasonic welding data, determine whether the welding machine is malfunctioning;

[0152] If the welding machine malfunctions, it will be stopped and an alarm will be triggered.

[0153] It should be noted that other embodiments or implementation methods of the battery cell tab welding pull prediction system of the present invention can refer to the above-described method embodiments, and will not be repeated here.

[0154] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0155] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0156] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for predicting the tensile force during battery cell tab welding, characterized in that, The method includes: Collect multidimensional data of the battery cell tab welding process. The multidimensional data includes ultrasonic welding data, incoming material data and production environment data. The incoming material data includes the surface roughness of the incoming material and the warehousing time. The ultrasonic welding data includes welding process data, welding machine data, welding machine internal data and CCD welding appearance inspection data. The multidimensional data is preprocessed and converted into time-series data, including extracting the number of weld points and weld point morphology feature values ​​based on the CCD welding appearance inspection data; performing Fourier transform on the welding process data to extract frequency information during the welding process; standardizing the welding machine data, the welding machine internal data, the number of weld points and weld point morphology feature values, the frequency information, the incoming material data, and the production environment data to obtain normally distributed data; and converting the normally distributed data into time-series data. The welding tensile force is predicted by using a long short-term memory network to obtain the welding tensile force prediction result at the next time step. The long short-term memory network is a pre-trained neural network for welding tensile force prediction, which includes a convolutional layer, a max pooling layer, an LSTM layer and a fully connected layer connected in sequence. An early warning is issued based on the predicted welding tensile force at the next moment.

2. The method for predicting the welding pull force of battery cell tabs as described in claim 1, characterized in that, The production environment data includes temperature data and humidity data.

3. The method for predicting the welding pull force of battery cell tabs as described in claim 1, characterized in that, The frequency information includes the mean frequency spectrum, root mean square frequency spectrum, centroid frequency, root mean square frequency, and standard deviation frequency; the extraction of frequency information during the welding process by performing a Fourier transform based on the welding process data includes: The Fourier transform of the welding process data is expressed as follows: In the formula: These are the sampled values ​​of the vibration signal; N This represents the number of sampling points; The sampling interval; k The index of the discrete value in the time domain; Based on the Fourier transform results, the spectral mean value of the welding process is extracted. F mean , Root mean square value of spectrum F mean square root The frequency centroid F1, root mean square frequency F2, and standard deviation frequency F3 are expressed by the following formula: In the formula: s ( k ) is a signal x ( n ) spectrum, K is the spectrum number; Indicates the first k Frequency values ​​of each spectral line.

4. The method for predicting the welding pull force of the battery cell tabs as described in claim 1, characterized in that, The loss function used in the training process of the Long Short-Term Memory network is MAPE, which is expressed by the following formula: In the formula: This represents the actual value of the t-th data point. This represents the predicted value of the t-th data point.

5. The method for predicting the welding pull force of the battery cell tabs as described in claim 1, characterized in that, The early warning based on the predicted welding tensile force at the next moment includes: Based on the predicted welding tensile force at the next moment from multiple predictions, the average value of the predicted welding tensile force is calculated. If continuous i If the average welding tensile force at a welding point shows a downward trend, an early warning will be issued.

6. The method for predicting the welding pull force of battery cell tabs as described in claim 1, characterized in that, The early warning based on the predicted welding tensile force at the next moment includes: Based on the predicted welding tensile force at the next moment, the predicted value of the welding tensile force is calculated. An early warning is issued when the average value of the predicted welding tensile force is not within the set threshold range.

7. The method for predicting the welding pull force of the battery cell tabs as described in claim 1, characterized in that, The early warning based on the predicted welding tensile force at the next moment includes: The standard deviation is calculated based on the predicted welding tensile force at the next moment after multiple predictions. Early warning is issued for regions where the standard deviation fluctuation exceeds the threshold.

8. The method for predicting the welding pull force of battery cell tabs as described in any one of claims 1-7, characterized in that, Before preprocessing the multidimensional data and converting it into time-series data, the method further includes: Based on the ultrasonic welding data, determine whether the welding machine is malfunctioning; If the welding machine malfunctions, it will be stopped and an alarm will be triggered.

9. A battery cell tab welding pull force prediction system, characterized in that, The system includes: The data acquisition module is used to collect multi-dimensional data of the battery cell tab welding process. The multi-dimensional data includes ultrasonic welding data, incoming material data and production environment data. The incoming material data includes the surface roughness of the incoming material and the warehousing time. The ultrasonic welding data includes welding process data, welding machine data, welding machine internal data and CCD welding appearance inspection data. The data conversion module is used to preprocess the multidimensional data and convert it into time-series data. This includes extracting the number of weld points and weld point morphology features based on the CCD welding appearance inspection data; performing Fourier transform on the welding process data to extract frequency information during the welding process; standardizing the welding machine data, the welding machine internal data, the number of weld points and weld point morphology features, the frequency information, the incoming material data, and the production environment data to obtain normally distributed data; and converting the normally distributed data into time-series data. The tensile strength prediction module is used to predict the welding tensile strength of the time series data using a long short-term memory network to obtain the welding tensile strength prediction result at the next moment. The long short-term memory network is a pre-trained neural network for welding tensile strength prediction, which includes a convolutional layer, a max pooling layer, an LSTM layer and a fully connected layer connected in sequence. The early warning module is used to issue an early warning based on the predicted welding tensile force at the next moment.

Images