An ultrasonic welding quality detection method and system

Abstract

The present invention discloses a method and system for ultrasonic welding quality detection. By collecting vibration signal data on the welding seat of an ultrasonic welder and electrical box data of the ultrasonic welder, preprocessing the vibration signal data and the electrical box data, respectively extracting features from the preprocessed vibration signal data and electrical box data to obtain vibration feature data and electrical box feature data, inputting the vibration feature data and the electrical box feature data into multiple target learning models respectively, obtaining the detection results output by each target learning model, wherein the detection result output by the target learning model is a qualified probability value, and taking the average value of the qualified probability values output by all the target learning models to obtain the final ultrasonic welding quality detection result. It solves the technical problem that the existing method for evaluating the ultrasonic welding quality of the electrode tabs of the battery cell obtains the quality detection result through destructive testing, which is not conducive to popularization and use.

Description

Technical Field

[0001] The present invention relates to the technical field of ultrasonic welding, and in particular, to a method and system for detecting ultrasonic welding quality. Background Art

[0002] An ultrasonic power supply, also known as an ultrasonic generator, is used to convert electrical energy into a high-frequency alternating current signal that matches the ultrasonic transducer, and is a device for generating and providing ultrasonic energy to the ultrasonic transducer. The ultrasonic welding technology utilizes high-frequency vibration waves to be transmitted to the surfaces of two objects to be welded. Under pressure, the surfaces of the two objects rub against each other to form a fusion between molecular layers. With the development of the new energy industry, ultrasonic metal welding has gradually been applied to industries such as lithium batteries and semiconductors. The ultrasonic welding process in these industries is crucial. If there are defects or potential hazards in the welding quality, when the welded components are applied to products, there will inevitably be safety risks.

[0003] The electrode tab of a battery cell is a raw material for lithium-ion polymer battery products and is a metal conductor that leads out the positive and negative electrodes from the battery cell. The existing methods for evaluating the ultrasonic welding quality of electrode tabs of battery cells often obtain quality detection results through destructive tests such as shear force detection, which is not conducive to popularization and use. Summary of the Invention

[0004] The present invention provides a method and system for detecting ultrasonic welding quality, which is used to solve the technical problem that the existing methods for evaluating the ultrasonic welding quality of electrode tabs of battery cells obtain quality detection results through destructive tests, which is not conducive to popularization and use.

[0005] In view of this, the first aspect of the present invention provides a method for detecting ultrasonic welding quality, including:

[0006] Collecting vibration signal data on the welding seat of an ultrasonic welding machine and electrical box data of the ultrasonic welding machine;

[0007] Preprocessing the vibration signal data and the electrical box data, and respectively extracting features from the preprocessed vibration signal data and electrical box data to obtain vibration feature data and electrical box feature data;

[0008] Inputting the vibration feature data and the electrical box feature data into a plurality of target learning models respectively to obtain detection results output by each target learning model, wherein the detection result output by the target learning model is a qualified probability value;

[0009] Taking the average value of the qualified probability values output by all the target learning models to obtain the final ultrasonic welding quality detection result.

[0010] Optionally, it further includes:

[0011] Determine the welding quality grade of the final ultrasonic welding quality detection result according to the corresponding relationship between the preset welding quality grading levels and the qualified probability values, where the welding quality grades include qualified, marginally qualified, marginally unqualified, and unqualified.

[0012] Optionally, preprocess the vibration signal data and the electrical box data, and respectively extract features from the preprocessed vibration signal data and the electrical box data to obtain vibration feature data and electrical box feature data, including:

[0013] Segment the vibration signal data to obtain the vibration signal data corresponding to the pressing and preheating stage, the dry friction stage, and the plastic deformation welding stage in the ultrasonic welding process;

[0014] Perform noise filtering on the vibration signal data in the pressing and preheating stage, the dry friction stage, and the plastic deformation welding stage;

[0015] Respectively perform time-domain feature extraction, frequency-domain feature extraction, and time-frequency domain feature extraction on the vibration signal data in the pressing and preheating stage, the dry friction stage, and the plastic deformation welding stage after noise filtering;

[0016] Respectively splice the time-domain feature data, frequency-domain feature data, and time-frequency domain feature data in the pressing and preheating stage, the dry friction stage, and the plastic deformation welding stage to obtain vibration feature data;

[0017] Normalize the electrical box data to obtain the normalized electrical box data;

[0018] Extract features from the normalized electrical box data to obtain electrical box feature data.

[0019] Optionally, the electrical box data includes power, three-phase current, three-phase voltage, phase difference, and impedance angle.

[0020] Optionally, the multiple target learning models include a random forest algorithm model.

[0021] Optionally, before inputting the vibration feature data and the electrical box feature data into the multiple target learning models respectively to obtain the detection results output by each target learning model, it further includes:

[0022] Obtain the vibration signal data samples, electrical box data samples, and corresponding welding quality labels in the ultrasonic welding process, where the welding quality labels are obtained by performing a tearing force test on the welded object;

[0023] Preprocess the vibration signal data samples and the electrical box data samples, and respectively extract features from the preprocessed vibration signal data samples and the electrical box data samples to obtain vibration feature sample data and electrical box feature sample data;

[0024] Construct a sample data set consisting of vibration characteristic sample data, electrical box characteristic sample data, and corresponding welding quality labels, and divide the sample data set into a training set, a validation set, and a test set;

[0025] Use the training set, the validation set, and the test set respectively to train the model, tune the model parameters, and test all learning models to obtain the corresponding target learning models.

[0026] The second aspect of the present invention provides an ultrasonic welding quality detection system, including:

[0027] A data acquisition module for acquiring vibration signal data on the welding seat of the ultrasonic welder and electrical box data of the ultrasonic welder;

[0028] A feature extraction module for preprocessing the vibration signal data and the electrical box data, and respectively extracting features from the preprocessed vibration signal data and electrical box data to obtain vibration feature data and electrical box feature data;

[0029] A welding quality detection module for respectively inputting the vibration feature data and the electrical box feature data into a plurality of target learning models to obtain the detection results output by each target learning model, wherein the detection result output by the target learning model is a qualified probability value;

[0030] A result output module for taking the average value of the qualified probability values output by all target learning models to obtain the final ultrasonic welding quality detection result.

[0031] Optionally, the result output module is further configured to:

[0032] Determine the welding quality grade of the final ultrasonic welding quality detection result according to the corresponding relationship between the preset welding quality division grades and the qualified probability values, wherein the welding quality grades include qualified, critically qualified, critically unqualified, and unqualified.

[0033] Optionally, the feature extraction module is specifically configured to:

[0034] Segment the vibration signal data to obtain the vibration signal data corresponding to the pressing and preheating stage, the dry friction stage, and the plastic deformation welding stage during the ultrasonic welding process;

[0035] Perform noise filtering processing on the vibration signal data in the pressing and preheating stage, the dry friction stage, and the plastic deformation welding stage;

[0036] Respectively perform time-domain feature extraction, frequency-domain feature extraction, and time-frequency domain feature extraction on the vibration signal data in the pressing and preheating stage, the dry friction stage, and the plastic deformation welding stage after noise filtering processing;

[0037] Splice the time-domain characteristic data, frequency-domain characteristic data, and time-frequency domain characteristic data of the pressing preheating stage, dry friction stage, and plastic deformation welding stage respectively to obtain vibration characteristic data;

[0038] Normalize the electrical box data to obtain the normalized electrical box data;

[0039] Extract features from the normalized electrical box data to obtain electrical box feature data.

[0040] Optionally, it further includes a model training module;

[0041] The model training module is used for:

[0042] Obtain vibration signal data samples, electrical box data samples, and corresponding welding quality labels of the ultrasonic welding process, where the welding quality labels are obtained by performing a tensile force test on the welded object;

[0043] Preprocess the vibration signal data samples and electrical box data samples, and extract features from the preprocessed vibration signal data samples and electrical box data samples respectively to obtain vibration feature sample data and electrical box feature sample data;

[0044] Construct a sample data set consisting of vibration feature sample data, electrical box feature sample data, and corresponding welding quality labels, and divide the sample data set into a training set, a validation set, and a test set;

[0045] Use the training set, validation set, and test set to train the model, tune the model parameters, and test all learning models respectively to obtain the corresponding target learning models.

[0046] It can be seen from the above technical solutions that the ultrasonic welding quality detection method and system provided by the present invention have the following advantages:

[0047] For the ultrasonic welding quality detection method provided by the present invention, on the one hand, by obtaining the vibration signal data and electrical box data of ultrasonic welding, after preprocessing and feature extraction of the vibration signal data and electrical box data, multiple learning models are used for welding quality detection, and finally the average value of the detection results of multiple learning models is taken as the final ultrasonic welding quality detection result. The entire quality detection process is non-destructive to the battery cells and the battery cell tabs. On the other hand, the vibration signal data is process data and the electrical box data is result data. Using the process data and result data together as the input of the learning model improves the accuracy of the ultrasonic welding quality detection result, can avoid a large number of destructive tensile force verifications, and greatly reduces the cost, solving the technical problem that the existing ultrasonic welding quality evaluation method for battery cell tabs obtains the quality detection result through destructive testing and is not conducive to popularization and use.

[0048] The ultrasonic welding quality detection system provided by the present invention is used to execute the ultrasonic welding quality detection method provided by the present invention. Its principle and the achieved technical effects are the same as those of the ultrasonic welding quality detection method provided by the present invention, and will not be elaborated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other related drawings can be obtained based on these drawings.

[0050] Figure 1 It is a schematic flowchart of an ultrasonic welding quality detection method provided in the present invention;

[0051] Figure 2 It is a schematic diagram of vibration signal data segmentation provided in the present invention;

[0052] Figure 3 It is a schematic structural diagram of an ultrasonic welding quality detection system provided in the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0053] In order to enable those skilled in the art to better understand the solution of the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0054] For ease of understanding, please refer to Figure 1 , an embodiment of an ultrasonic welding quality detection method provided in the present invention includes:

[0055] Step 101: Collect vibration signal data on the welding seat of the ultrasonic welding machine and electrical box data of the ultrasonic welding machine.

[0056] It should be noted that in the embodiments of the present invention, the electrical box data of the ultrasonic welding machine is obtained through a PLC (Programmable Logic Controller). The electrical box data includes power, three-phase current, three-phase voltage, phase difference, and impedance angle. The vibration signal data is collected through an acceleration sensor and a data acquisition card installed on the welding seat of the ultrasonic welding machine.

[0057] Step 102: Preprocess the vibration signal data and the electrical box data, and extract features from the preprocessed vibration signal data and electrical box data respectively to obtain vibration feature data and electrical box feature data.

[0058] It should be noted that there is a lot of noise in the directly obtained vibration signal data and electrical box data. Therefore, it is necessary to preprocess the vibration signal data and the electrical box data. Perform normalization on the electrical box data to reduce the complexity of data processing and improve the detection efficiency. For the vibration signal data, perform operations such as noise filtering and digital filtering to filter out noise data and intercept effective frequency band data. For the preprocessed electrical box data, perform polynomial dimension elevation and PCA dimension reduction operations for feature extraction to obtain electrical box feature data. For the preprocessed vibration signal data, perform time-domain feature extraction, frequency-domain feature extraction, and time-frequency domain feature extraction respectively to obtain vibration feature data. Among them, the time-domain feature extraction of the vibration signal data specifically extracts the mean value, variance, and deviation features. The frequency-domain feature extraction specifically extracts the frequency, amplitude, and center frequency. The video domain feature extraction specifically extracts the real-time frequency and real-time amplitude.

[0059] In one embodiment, the process of preprocessing the vibration signal data includes:

[0060] S1: Segment the vibration signal to obtain the vibration signal data corresponding to the pressing and preheating stage, dry friction stage, and plastic deformation welding stage during the ultrasonic welding process. According to the vibration principle of ultrasonic welding, as Figure 2 shown, the ultrasonic welding process is divided into 3 stages: the pressing and preheating stage, the dry friction stage, and the plastic deformation welding stage. Segment the vibration signal data through a digital filter to segment the data of the pressing and preheating stage, dry friction stage, and plastic deformation welding stage in the vibration signal data.

[0061] S2: Perform noise filtering on the vibration signal data of the pressing and preheating stage, dry friction stage, and plastic deformation welding stage to remove noise interference.

[0062] S3: Perform time-domain feature extraction, frequency-domain feature extraction, and time-frequency domain feature extraction on the vibration signal data of the pressing and preheating stage, dry friction stage, and plastic deformation welding stage after noise filtering respectively. That is, perform time-domain feature extraction, frequency-domain feature extraction, and time-frequency domain feature extraction on the vibration signal data of the 3 stages during the ultrasonic welding process.

[0063] S4. Concatenate the time-domain characteristic data, frequency-domain characteristic data, and time-frequency domain characteristic data of the pressing and preheating stage, dry friction stage, and plastic deformation welding stage respectively to obtain vibration characteristic data. By concatenating the time-domain characteristic data, frequency-domain characteristic data, and time-frequency domain characteristic data of the pressing and preheating stage, dry friction stage, and plastic deformation welding stage, the time-domain characteristic data, frequency-domain characteristic data, and time-frequency domain characteristic data corresponding to the vibration signal data can be obtained, which are collectively referred to as vibration characteristic data.

[0064] Step 103. Input the vibration characteristic data and the electrical box characteristic data into multiple target learning models respectively to obtain the detection results output by each target learning model. The detection result output by the target learning model is a qualified probability value.

[0065] It should be noted that the target learning model is a pre-trained and tested learning model, such as a random forest algorithm model, a BP neural network model, and a CNN model, etc. Its input is vibration characteristic data and electrical box characteristics, and the output is a qualified probability value. The training and testing process of the learning model includes:

[0066] (1) Obtain the vibration signal data samples, electrical box data samples, and corresponding welding quality labels of the ultrasonic welding process. The welding quality labels can be qualified, marginally qualified, marginally unqualified, and unqualified. Among them, the welding quality labels are obtained by conducting a tensile force test on the welded object. For example, for a welded battery cell tab, if its tensile force is greater than 45N, it can be regarded as qualified and represented by OK; for a welded battery cell tab, if its tensile force is between 42N and 45N, it can be regarded as marginally qualified and represented by marginally OK; for a welded battery cell tab, if its tensile force is between 38N and 41N, it can be regarded as marginally unqualified and represented by marginally NG; for a welded battery cell tab, if its tensile force is less than 38N, it can be regarded as unqualified and represented by NG.

[0067] (2) Preprocess the vibration signal data samples and electrical box data samples, and extract features from the preprocessed vibration signal data samples and electrical box data samples respectively to obtain vibration characteristic sample data and electrical box characteristic sample data. Among them, the operations of preprocessing and feature extraction on the vibration signal data samples and electrical box data samples in this part are the same as the processing process in step 102, and will not be elaborated here.

[0068] (3) Construct a sample data set composed of vibration characteristic sample data, electrical box characteristic sample data, and corresponding welding quality labels, and divide the sample data set into a training set, a validation set, and a test set. Preferably, the ratio of the training set, the validation set, and the test set is 8:1:1.

[0069] (4) Use the training set, validation set, and test set to train the model, tune the model parameters, and test all learning models respectively to obtain the corresponding target learning models. The training set and validation set are used to train the model and tune the model parameters, and the test set is used to finally verify the detection effect of the model. Finally, a learning model with vibration feature data and electrical box features as inputs and a qualified probability value as the output can be obtained through verification.

[0070] Step 104: Take the average of the qualified probability values output by all target learning models to obtain the final ultrasonic welding quality detection result.

[0071] It should be noted that for each target learning model, there is a corresponding output result, that is, the qualified probability value. To improve the accuracy of the detection result, the average of the qualified probability values output by all target learning models is taken as the final ultrasonic welding quality detection result. According to the corresponding relationship between the preset welding quality grading and the qualified probability value, the welding quality grade of the final ultrasonic welding quality detection result can be determined. The welding quality grades include qualified, marginally qualified, marginally unqualified, and unqualified. For example, assume that the average result of the qualified probability values output by all target learning models is 80%. The corresponding relationship between the preset welding quality grading and the qualified probability value is: greater than 60% corresponds to qualified, between 56% and 60% corresponds to marginally qualified, between 50% and 55% corresponds to marginally unqualified, and less than 50% is regarded as unqualified.

[0072] The ultrasonic welding quality detection method provided by the present invention, on the one hand, by acquiring the vibration signal data and electrical box data of ultrasonic welding, after preprocessing and feature extraction of the vibration signal data and electrical box data, uses multiple learning models for welding quality detection, and finally takes the average of the detection results of multiple learning models as the final ultrasonic welding quality detection result. The entire quality detection process is non-destructive to the battery cell and the electrode tab of the battery cell. On the other hand, the vibration signal data is process data, and the electrical box data is result data. Using the process data and result data together as the input of the learning model improves the accuracy of the ultrasonic welding quality detection result, can avoid a large number of destructive tensile verifications, and greatly reduces the cost, solving the technical problem that the existing ultrasonic welding quality evaluation method for the electrode tab of the battery cell obtains the quality detection result through destructive testing, which is not conducive to popularization and use.

[0073] For the sake of easy understanding, an embodiment of an ultrasonic welding quality detection system is provided in the present invention, including:

[0074] A data acquisition module, configured to acquire the vibration signal data on the welding seat of the ultrasonic welding machine and the electrical box data of the ultrasonic welding machine;

[0075] A feature extraction module, which is used to preprocess vibration signal data and electrical box data, respectively extract features from the preprocessed vibration signal data and electrical box data to obtain vibration feature data and electrical box feature data;

[0076] A welding quality detection module, which is used to input the vibration feature data and electrical box feature data into multiple target learning models respectively, and obtain the detection results output by each target learning model. The detection result output by the target learning model is a qualified probability value;

[0077] A result output module, which is used to take the average value of the qualified probability values output by all target learning models to obtain the final ultrasonic welding quality detection result.

[0078] The result output module is further used for:

[0079] According to the corresponding relationship between the preset welding quality grading levels and the qualified probability values, determine the welding quality grade of the final ultrasonic welding quality detection result, where the welding quality grades include qualified, critically qualified, critically unqualified, and unqualified.

[0080] The feature extraction module is specifically used for:

[0081] Segment the vibration signal data to obtain the vibration signal data corresponding to the pressing and preheating stage, dry friction stage, and plastic deformation welding stage in the ultrasonic welding process;

[0082] Perform noise filtering processing on the vibration signal data in the pressing and preheating stage, dry friction stage, and plastic deformation welding stage;

[0083] Perform time-domain feature extraction, frequency-domain feature extraction, and time-frequency domain feature extraction on the vibration signal data in the pressing and preheating stage, dry friction stage, and plastic deformation welding stage after noise filtering processing respectively;

[0084] Concatenate the time-domain feature data, frequency-domain feature data, and time-frequency domain feature data in the pressing and preheating stage, dry friction stage, and plastic deformation welding stage respectively to obtain vibration feature data;

[0085] Normalize the electrical box data to obtain the normalized electrical box data;

[0086] Extract features from the normalized electrical box data to obtain electrical box feature data.

[0087] It also includes a model training module;

[0088] The model training module is used for:

[0089] Obtain the vibration signal data samples, electrical box data samples, and corresponding welding quality labels during the ultrasonic welding process, where the welding quality labels are obtained by performing a peel strength test on the welded object;

[0090] Preprocess the vibration signal data samples and electrical box data samples, and respectively extract features from the preprocessed vibration signal data samples and electrical box data samples to obtain vibration feature sample data and electrical box feature sample data;

[0091] Construct a sample data set composed of vibration feature sample data, electrical box feature sample data, and corresponding welding quality labels, and divide the sample data set into a training set, a validation set, and a test set;

[0092] Use the training set, the validation set, and the test set to train the model, tune the model parameters, and test all learning models respectively to obtain the corresponding target learning models.

[0093] The electrical box data includes power, three-phase current, three-phase voltage, phase difference, and impedance angle.

[0094] The multiple target learning models include a random forest algorithm model.

[0095] The ultrasonic welding quality detection system provided by the present invention is used to execute the ultrasonic welding quality detection method provided by the present invention. Its principle and the achieved technical effects are the same as those of the ultrasonic welding quality detection method provided by the present invention, and will not be elaborated here.

[0096] As mentioned above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An ultrasonic welding quality detection method, characterized in that, Including: Collecting vibration signal data on the welding seat of the ultrasonic welding machine and the electrical box data of the ultrasonic welding machine; Preprocessing the vibration signal data and the electrical box data, respectively extracting features from the preprocessed vibration signal data and the electrical box data to obtain vibration feature data and electrical box feature data; Inputting the vibration feature data and the electrical box feature data into multiple target learning models respectively to obtain the detection results output by each target learning model, where the detection result output by the target learning model is a qualified probability value; Taking the average value of the qualified probability values output by all target learning models to obtain the final ultrasonic welding quality detection result; Also including: Determining the welding quality grade of the final ultrasonic welding quality detection result according to the corresponding relationship between the preset welding quality division grade and the qualified probability value, where the welding quality grade includes qualified, critically qualified, critically unqualified, and unqualified; Preprocessing the vibration signal data and the electrical box data, respectively extracting features from the preprocessed vibration signal data and the electrical box data to obtain vibration feature data and electrical box feature data, including: Segmenting the vibration signal data to obtain the vibration signal data in the pressing and preheating stage, dry friction stage, and plastic deformation welding stage corresponding to the ultrasonic welding process; Performing noise filtering processing on the vibration signal data in the pressing and preheating stage, dry friction stage, and plastic deformation welding stage; Respectively performing time-domain feature extraction, frequency-domain feature extraction, and time-frequency domain feature extraction on the vibration signal data in the pressing and preheating stage, dry friction stage, and plastic deformation welding stage after noise filtering processing; Respectively splicing the time-domain feature data, frequency-domain feature data, and time-frequency domain feature data in the pressing and preheating stage, dry friction stage, and plastic deformation welding stage to obtain vibration feature data; Normalizing the electrical box data to obtain the normalized electrical box data; Extracting features from the normalized electrical box data to obtain electrical box feature data.

2. The ultrasonic welding quality detection method according to claim 1, wherein, The electrical box data includes power, three-phase current, three-phase voltage, phase difference, and impedance angle.

3. The ultrasonic welding quality detection method according to claim 1, characterized in that The multiple target learning models include a random forest algorithm model.

4. The ultrasonic welding quality detection method according to claim 1, characterized in that, Before inputting the vibration feature data and the electrical box feature data into multiple target learning models respectively to obtain the detection results output by each target learning model, it also includes: Obtaining the vibration signal data sample, electrical box data sample, and corresponding welding quality label of the ultrasonic welding process, where the welding quality label is obtained by performing a tear strength test on the welded object; Preprocessing the vibration signal data sample and the electrical box data sample, respectively extracting features from the preprocessed vibration signal data sample and the electrical box data sample to obtain vibration feature sample data and electrical box feature sample data; Constructing a sample data set composed of vibration feature sample data, electrical box feature sample data, and corresponding welding quality labels, and dividing the sample data set into a training set, a validation set, and a test set; Respectively using the training set, the validation set, and the test set to train the model, adjust the model parameters, and test all learning models to obtain the corresponding target learning models.

5. An ultrasonic welding quality detection system, characterized in that, Including: A data acquisition module for collecting vibration signal data on the welding seat of the ultrasonic welding machine and the electrical box data of the ultrasonic welding machine; A feature extraction module, which is used to preprocess vibration signal data and electrical box data, and respectively extract features from the preprocessed vibration signal data and electrical box data to obtain vibration feature data and electrical box feature data; A welding quality detection module, which is used to respectively input the vibration feature data and the electrical box feature data into multiple target learning models to obtain the detection results output by each target learning model, wherein the detection result output by the target learning model is a qualified probability value; A result output module, which is used to take the average value of the qualified probability values output by all target learning models to obtain the final ultrasonic welding quality detection result; The result output module is also used for: According to the corresponding relationship between the preset welding quality grading levels and the qualified probability values, determine the welding quality grade of the final ultrasonic welding quality detection result, wherein the welding quality grades include qualified, critically qualified, critically unqualified, and unqualified; Specifically, the feature extraction module is used for: Segment the vibration signal data to obtain the vibration signal data corresponding to the pressing and preheating stage, dry friction stage, and plastic deformation welding stage during the ultrasonic welding process; Perform noise filtering processing on the vibration signal data in the pressing and preheating stage, dry friction stage, and plastic deformation welding stage; Respectively perform time-domain feature extraction, frequency-domain feature extraction, and time-frequency domain feature extraction on the vibration signal data in the pressing and preheating stage, dry friction stage, and plastic deformation welding stage after noise filtering processing; Respectively splice the time-domain feature data, frequency-domain feature data, and time-frequency domain feature data in the pressing and preheating stage, dry friction stage, and plastic deformation welding stage to obtain vibration feature data; Normalize the electrical box data to obtain the normalized electrical box data; Extract features from the normalized electrical box data to obtain electrical box feature data.

6. The ultrasonic welding quality detection system according to claim 5, characterized in that It also includes a model training module; The model training module is used for: Obtain the vibration signal data samples, electrical box data samples, and corresponding welding quality labels during the ultrasonic welding process, wherein the welding quality labels are obtained by performing a tearing force test on the welded object; Preprocess the vibration signal data samples and electrical box data samples, and respectively extract features from the preprocessed vibration signal data samples and electrical box data samples to obtain vibration feature sample data and electrical box feature sample data; Construct a sample data set composed of vibration feature sample data, electrical box feature sample data, and corresponding welding quality labels, and divide the sample data set into a training set, a validation set, and a test set; Respectively use the training set, the validation set, and the test set to perform model training, model parameter tuning, and testing on all learning models to obtain the corresponding target learning models.

Images