A nonlinear ultrasonic guided wave intelligent identification method for internal defects of a battery pole piece

Abstract

The application discloses a kind of nonlinear ultrasonic guided wave intelligent identification methods for internal defects of battery pole piece, comprising: building nonlinear ultrasonic guided wave measurement system, determine actual focal length and select the best receiving angle, excite and receive ultrasonic guided wave, carry out automated scanning;Preparation normal and contain defect battery pole piece, using preset phase excitation signal for measurement, after phase reversal and difference, obtain nonlinear ultrasonic dataset, second harmonic dataset and fundamental dataset;Using convolutional neural network to extract features from the dataset, and with the help of random forest-based permutation importance filtering feature set;Using screened feature set to train random forest model, obtain battery pole piece internal defect identification model;With the help of the measurement system, obtain the nonlinear ultrasonic time domain signal of target battery pole piece, use convolutional neural network to extract features, with the help of permutation importance filtering features, input the identification model to obtain defect identification result.

Description

Technical Field

[0001] This invention belongs to the field of quality inspection in the manufacturing stage of power batteries, and particularly relates to a nonlinear ultrasonic guided wave intelligent identification method for internal defects of battery electrodes. Background Technology

[0002] With the increasing popularity of new energy vehicles, the safety of vehicle batteries has gradually attracted attention, and quality inspection during the power battery manufacturing stage is crucial for safe operation. Battery electrodes are one of the core components of a battery, and their production is a complex process involving mixing, coating, drying, and calendering. Defects such as broken metal foil, agglomerates, and mixed particles can occur during these processes, leading to battery capacity decay, increased internal resistance, and shortened lifespan. In severe cases, they may even cause overheating or short circuits. Therefore, identifying and classifying electrode defects during battery manufacturing can ensure battery performance stability and safety, and extend battery lifespan.

[0003] Currently, defect detection in battery electrodes mainly employs methods such as visual inspection, X-ray inspection, and ultrasonic inspection. Visual inspection methods cannot penetrate the battery electrode material, making it difficult to detect internal defects. X-ray inspection suffers from low efficiency, high equipment cost, and significant radiation risk. Point-by-point inspection based on ultrasonic longitudinal waves has a long cycle time and high cost. Therefore, there is an urgent need to propose a nonlinear ultrasonic guided wave intelligent identification method for internal defects in battery electrodes. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention proposes a nonlinear ultrasonic guided wave intelligent identification method for internal defects in battery electrodes, which can monitor and effectively identify internal defects in battery electrodes in real time.

[0005] To achieve the above objectives, this invention provides a nonlinear ultrasonic guided wave intelligent identification method for internal defects in battery electrodes, comprising:

[0006] A nonlinear ultrasonic guided wave measurement system was built to determine the actual focal length and select the optimal receiving angle, excite and receive ultrasonic guided waves, and perform automated scanning.

[0007] Normal and defective battery electrodes are prepared, and measurements are performed using an excitation signal with a preset phase to obtain the raw dataset;

[0008] Based on the original dataset, time-domain signals of nonlinear ultrasound, second harmonic, and fundamental wave are extracted to obtain nonlinear ultrasound dataset, second harmonic dataset, and fundamental wave dataset.

[0009] Features are extracted from the dataset using a convolutional neural network, and the feature set is filtered using permutation importance based on random forest.

[0010] A random forest model was trained using the filtered feature set to obtain a model for identifying internal defects in battery electrodes.

[0011] The nonlinear ultrasonic time-domain signal of the target battery electrode is acquired using the measurement system, features are extracted using a convolutional neural network, features are filtered using permutation importance, and the results are input into the recognition model to obtain defect recognition results.

[0012] Optionally, the process of setting up a nonlinear ultrasonic guided wave measurement system includes:

[0013] The system consists of an arbitrary function generator, a pulse amplifier, an oscilloscope, a computer, a three-degree-of-freedom motion platform, an excitation transducer, and a receiving transducer.

[0014] The excitation signal is generated by an arbitrary function generator and is a sinusoidal tone burst with n periods and frequency f, with Hanning window applied, and the initial phase is 0° and 180°.

[0015] The center frequency of both the excitation transducer and the receiving transducer is f;

[0016] The excitation transducer is placed above the surface of the battery electrode, and its position is adjusted by a three-degree-of-freedom motion platform. The actual focal length is determined based on the maximum amplitude of the time-domain signal.

[0017] The angle between the axis of the receiving transducer and the battery electrode is determined by two-dimensional Fourier transform and ultrasonic guided wave dispersion curve to select the optimal receiving angle.

[0018] The spatial position of the receiving transducer is adjusted using a three-degree-of-freedom motion platform to perform automated scanning with a fixed step size and path.

[0019] Optionally, the process for preparing normal and defective battery electrodes includes:

[0020] Metal foil is selected as the substrate; aluminum foil is used for the positive electrode and copper foil is used for the negative electrode.

[0021] A slurry is prepared using active materials, conductive agents, binders, and solvents and then coated onto a substrate.

[0022] Pre-determined defects are created on an uncoated substrate, and microcracks, holes, or breaks are manufactured by laser cutting to prepare battery electrodes with broken foils; battery electrodes with foil foreign objects are prepared by attaching metal foils; battery electrodes with metal particles are prepared by placing metal particles.

[0023] The substrate is coated, dried, rolled, and cut to ensure that the battery electrodes meet manufacturing requirements.

[0024] Optionally, based on the original dataset, the time-domain signals of nonlinear ultrasound, second harmonic, and fundamental frequency are extracted, including:

[0025] In the original dataset, an excitation signal with a phase of 0° is used, and the received signal is a time-domain signal of nonlinear ultrasound, including acoustic linear parameters and acoustic nonlinear parameters.

[0026] Using phase reversal technology, the time-domain signal of the second harmonic is extracted from excitation signals with phases of 0° and 180°, mainly including acoustic nonlinear parameters;

[0027] The time-domain signal of the fundamental wave is extracted by subtracting the time-domain signal of the nonlinear ultrasound from the time-domain signal of the second harmonic wave, which mainly includes acoustic linear parameters.

[0028] Nonlinear ultrasound datasets, second harmonic datasets, and fundamental frequency datasets are generated.

[0029] Optionally, using a convolutional neural network to extract features from the dataset includes:

[0030] The nonlinear ultrasound dataset, second harmonic dataset, and fundamental frequency dataset are preprocessed. The preprocessing process includes padding the sequence to the same length, label encoding, and normalizing the signal data.

[0031] The preprocessed dataset is divided into a training set, a validation set, and a test set;

[0032] One-dimensional convolutional neural network models are used to extract deep features from preprocessed signal data. One-dimensional convolutional neural networks extract local features from data through convolution operations and reduce dimensionality and enhance feature invariance through pooling operations.

[0033] Optionally, the feature set can be filtered using permutation importance based on random forest, including:

[0034] For the features extracted by the convolutional neural network, feature selection is performed using permutation importance based on random forest;

[0035] This method combines embedding with model importance evaluation. After the random forest model is trained, each feature is randomly permuted and the decline in model performance is observed.

[0036] Based on the contribution of specific features in the quantification model prediction of the decline magnitude, features with high contribution are selected to form a filtered feature set.

[0037] Optionally, training a random forest model using the filtered feature set includes:

[0038] The filtered feature set is used as input to train a random forest classifier;

[0039] The hyperparameters of the model were optimized using 10-fold cross-validation and grid search.

[0040] The grid search system systematically traverses all possible combinations of the number of decision trees and the maximum depth of the decision trees;

[0041] Ten-fold cross-validation divides the training set into 10 subsets, performs 10 training and validation cycles, and calculates the average performance metric.

[0042] Select the hyperparameter combination with the best average performance to obtain a well-trained battery electrode internal defect identification model.

[0043] Optionally, acquiring the nonlinear ultrasonic time-domain signal of the target battery electrode using a measurement system includes:

[0044] Using the established nonlinear ultrasonic guided wave measurement system, an excitation signal with a phase of 0° was used to perform ultrasonic guided wave detection on battery electrodes with unknown mass.

[0045] By receiving signals through a transducer, a complete nonlinear ultrasonic time-domain signal is obtained.

[0046] The nonlinear ultrasound time-domain signal is preprocessed, including filling the sequence to the same length and normalizing the signal data to ensure data consistency.

[0047] Optionally, using convolutional neural networks to extract features includes:

[0048] For the preprocessed nonlinear ultrasonic time-domain signal, a one-dimensional convolutional neural network model is used for feature extraction. The one-dimensional convolutional neural network extracts local features in the signal through convolutional layers, and outputs a deep feature set by reducing dimensionality and enhancing feature invariance through pooling layers.

[0049] Optionally, feature selection using permutation importance and inputting it into the identification model to obtain defect identification results includes: using permutation importance based on random forest to select features extracted by the convolutional neural network;

[0050] Select a subset of features based on their importance scores;

[0051] The filtered feature subset is input into the trained battery electrode internal defect recognition model.

[0052] The model outputs the identification results of whether the battery electrodes are normal and the type of defect.

[0053] Technical advantages of this invention: This invention discloses a nonlinear ultrasonic guided wave intelligent identification method for internal defects in battery electrodes. By combining nonlinear ultrasonic guided waves with artificial intelligence algorithms, it effectively solves the problems of traditional visual inspection failing to detect internal defects, low efficiency and high radiation risk of X-ray inspection, and long C-scan period of ultrasonic longitudinal waves. Compared with traditional ultrasonic guided waves, nonlinear ultrasonic guided waves generate high-order harmonics during propagation, making them extremely sensitive to minute defects and significantly improving detection accuracy. Simultaneously, the artificial intelligence algorithm avoids interference from human factors such as lack of experience and misjudgment, achieving intelligent identification of internal defects. For battery electrodes with unknown production quality, this method enables real-time monitoring and effective identification of internal defects during manufacturing, allowing for timely detection and resolution of potential problems. It offers advantages such as low cost, high efficiency, and high intelligence, thereby significantly improving the quality and production efficiency of battery electrodes. Attached Figure Description

[0054] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0055] Figure 1 This is a flowchart illustrating a nonlinear ultrasonic guided wave intelligent identification method for internal defects of battery electrodes according to an embodiment of the present invention.

[0056] Figure 2 The ultrasonic transducer arrangement for ultrasonic guided wave detection of internal defects in battery electrodes according to an embodiment of the present invention;

[0057] Figure 3 These are typical received signals from normal battery electrodes and battery electrodes with different types of defects, as described in embodiments of the present invention.

[0058] Figure 4 The spectrum of nonlinear ultrasound, fundamental wave, and second harmonic wave is shown in the embodiments of the present invention.

[0059] Figure 5 This is a confusion matrix for intelligent identification of battery electrode defect types based on nonlinear ultrasonic datasets, fundamental frequency datasets, and second harmonic frequency datasets, as described in this embodiment of the invention. Detailed Implementation

[0060] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0061] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0062] like Figure 1 As shown, this embodiment provides a nonlinear ultrasonic guided wave intelligent identification method for internal defects of battery electrodes, including:

[0063] A nonlinear ultrasonic guided wave measurement system was constructed to excite and receive ultrasonic guided waves.

[0064] Different types of battery electrodes are prepared, and the same battery electrode is measured using an excitation signal with a preset phase to obtain a raw dataset. The different types of battery electrodes include normal battery electrodes and battery electrodes with different types of defects.

[0065] Based on the original dataset, time-domain signals of nonlinear ultrasound, second harmonic wave, and fundamental wave are extracted respectively to obtain nonlinear ultrasound dataset, second harmonic wave dataset, and fundamental wave dataset.

[0066] Based on the nonlinear ultrasound dataset, the second harmonic dataset, and the fundamental frequency dataset, the NLUGW_CNN-PI-RF model for identifying internal defects in battery electrodes is trained to obtain a well-trained model for identifying internal defects in battery electrodes.

[0067] Using the established nonlinear ultrasonic guided wave measurement system, the nonlinear ultrasonic time-domain signal of the target battery electrode is acquired. CNN is used for feature extraction, and permutation importance based on random forest is used for feature selection. The feature subset after feature selection is input into the trained battery electrode internal defect identification model NLUGW_CNN-PI-RF to obtain the identification result.

[0068] Specifically, the specific implementation process of the identification method described in this embodiment includes:

[0069] Step S1: Construct a nonlinear ultrasonic guided wave measurement system to excite and receive ultrasonic guided waves;

[0070] Step S2, Preparation x Ultrasonic guided wave measurements were conducted on various types of battery electrodes, including normal electrodes and electrodes with different types of defects. Excitation signals with phases of 0° and 180° were used to measure the same battery electrode separately. y This time, the amount of data generated is x ×2 y The original dataset;

[0071] Step S3: Extract the time-domain signals of nonlinear ultrasound, second harmonic, and fundamental wave from the original dataset to form a nonlinear ultrasound dataset, a second harmonic dataset, and a fundamental wave dataset.

[0072] Step S4: Based on the nonlinear ultrasonic dataset, second harmonic dataset, and fundamental frequency dataset, train the internal defect identification model NLUGW_CNN-PI-RF for battery electrode sheets, and carry out nonlinear ultrasonic guided wave intelligent identification of normal battery electrode sheets and battery electrode sheets with different types of defects.

[0073] Step S5: For battery electrodes with unknown production quality, nonlinear ultrasonic guided wave testing is carried out. The trained battery electrode internal defect identification model NLUGW_CNN-PI-RF is used to identify normal battery electrodes and battery electrodes with different types of defects.

[0074] Furthermore, in step S1, the nonlinear ultrasonic guided wave measurement system comprises an arbitrary function generator, a pulse amplifier, an oscilloscope, a computer, a three-degree-of-freedom motion platform, an excitation transducer, and a receiving transducer, and is used to excite and receive ultrasonic guided waves. Step S1 specifically includes:

[0075] Step S11: The excitation signal is generated by an arbitrary function generator, and a Hanning window is added. n Each cycle, frequency is f sinusoidal tone bursts, where the initial phases are 0° and 180°.

[0076] Step S12: To drive the excitation transducer, the energy of the signal is amplified by a pulse amplifier and enters the excitation transducer. The center frequencies of both the excitation transducer and the receiving transducer are... f .

[0077] Step S13: Place the excitation transducer above the surface of the battery electrode, fix the angle between the receiving transducer and the battery electrode, and automatically calibrate the focal length of the focused ultrasonic transducer based on the amplitude of the time domain signal using closed-loop feedback to determine the distance between the focused ultrasonic transducer and the surface of the battery electrode. Place the excitation transducer above the surface of the battery electrode (the distance should exceed the theoretical focal length of the focused ultrasonic transducer by 10mm). Using the Z-axis of a three-degree-of-freedom motion platform, move the focused ultrasonic transducer downwards in 0.5mm increments. Keep the angle between the receiving transducer and the battery electrode fixed, and collect the time-domain signal in real time, transmitting it to the host computer. Determine the corresponding distance based on the maximum amplitude of the time-domain signal. Using the Z-axis of the three-degree-of-freedom motion platform, adjust the receiving transducer to the distance corresponding to the maximum amplitude, then move it upwards by 5mm in 0.1mm increments. Continue moving the excitation transducer downwards, collecting the time-domain signal using the receiving transducer, and determining the corresponding distance based on the maximum amplitude of the time-domain signal to obtain the actual focal length of the excitation transducer.

[0078] Step S14: The excitation transducer is perpendicular to the surface of the battery electrode, and the distance between the transducer and the surface of the battery electrode is equal to one focal length. The angle between the axis of the receiving transducer and the battery electrode is adjustable, ranging from 35° to 90°, to receive ultrasonic guided waves of different modes. Using a 360° rotating displacement stage, the angle between the axis of the receiving transducer and the surface of the battery electrode is changed, successively selecting key angles such as 45°, 50°, 55°, 60°, 65°, 70°, 80°, and 90°. For each key angle, time-domain signals are received at fixed intervals along the length of the battery electrode, forming a dataset containing 200 time-domain signals. A two-dimensional Fourier transform is used to obtain the relationship between the ultrasonic guided wave number and frequency. The dispersion curve of the ultrasonic guided wave is introduced to determine the mode of the ultrasonic guided wave. Based on the amplitude and number of modes of the ultrasonic guided wave, the angle between the axis of the receiving transducer and the battery electrode is selected.

[0079] Step S15: Based on the step size and path, adjust the spatial position of the receiving transducer using a three-degree-of-freedom motion platform to achieve automated scanning of the battery electrode. The step size is 0.5mm, and the scanning area is 10mm × 10mm. The step size, path, and scanning area are set using software written in the host computer. The receiving transducer is moved using the three-degree-of-freedom motion platform to complete the scanning of the battery electrode. The host computer issues commands to collect the spatial position of the receiving transducer and acquires waveform data from the oscilloscope using a data cable. During the scanning process, the distance between the receiving transducer and the electrode surface is measured in real time. For offsets caused by uneven electrode surfaces, the Z-axis of the three-degree-of-freedom motion platform is automatically fine-tuned to ensure that the distance between the receiving transducer and the electrode surface equals the actual focal length.

[0080] Step S16: The excitation transducer converts the electrical signal into mechanical vibration. The ultrasonic guided wave is formed in the battery electrode and interacts with the defect. The ultrasonic guided wave carrying the defect information enters the receiving transducer and is converted into an electrical signal. This signal is displayed on the oscilloscope and stored in the computer.

[0081] Furthermore, in step S2, the battery electrode is a metal current collector foil coated with active material on both sides. Normal battery electrodes and battery electrodes with different types of defects are prepared. The types of defective battery electrodes include, but are not limited to, battery electrodes with broken foil, battery electrodes with foreign matter in the foil, and battery electrodes with metal particles. Ultrasonic guided wave measurements are performed using a nonlinear ultrasonic guided wave measurement system to form a raw dataset. Step S2 specifically includes the following steps:

[0082] Step S21: Select metal foils that meet the requirements of battery electrodes as substrates, wherein aluminum foil is used for the positive electrode and copper foil is used for the negative electrode. Prepare a slurry using active materials, conductive agents, binders, and solvents, and adjust the rheological properties of the slurry to meet the uniformity and stability requirements of the coating process.

[0083] Step S22: Pre-determine defects on the uncoated aluminum / copper foil substrate. Microcracks, holes, or breaks are created on the foil using laser cutting, scribing, or other methods to prepare a battery electrode with a damaged foil. A battery electrode with foreign matter is prepared by attaching a metal foil to the surface of the foil. A battery electrode with metal particles is prepared by placing metal particles on the foil to simulate conductive foreign matter that may be introduced during the production process.

[0084] Step S24: Coating, drying, rolling, and cutting are performed on normal battery electrode substrates and substrates containing defects to ensure that the overall process of the battery electrode meets the battery manufacturing requirements, while defects are prepared in the finished battery electrode.

[0085] Step S25: Using a nonlinear ultrasonic guided wave measurement system, for x Ultrasonic guided wave measurements were performed on normal battery electrodes or battery electrodes with different types of defects. Excitation signals with phases of 0° and 180° were used to measure the same battery electrode separately. y This time, the amount of data generated is x ×2 y The original dataset.

[0086] Furthermore, in step S3, time-domain signals of nonlinear ultrasound, second harmonic wave, and fundamental wave are extracted from the original dataset to form a nonlinear ultrasound dataset, a second harmonic wave dataset, and a fundamental wave dataset. Step S3 specifically includes:

[0087] Step S31: In the original dataset, an excitation signal with a phase of 0° is used. The received signal is a time-domain signal of nonlinear ultrasound, which includes acoustic linear parameters and acoustic nonlinear parameters.

[0088] Step S32: Use phase reversal technology to extract the time domain signal of the second harmonic. The second harmonic mainly includes acoustic nonlinear parameters.

[0089] Step S33: Subtract the time-domain signal of the nonlinear ultrasound from the time-domain signal of the second harmonic to extract the time-domain signal of the fundamental wave. The fundamental wave mainly includes acoustic linear parameters.

[0090] Step S34: For the time-domain signals of nonlinear ultrasound, second harmonic, and fundamental wave, three independent datasets are formed: nonlinear ultrasound dataset, second harmonic dataset, and fundamental wave dataset, which are used for subsequent model training and evaluation.

[0091] Furthermore, in step S4, a battery electrode internal defect identification model, NLUGW_CNN-PI-RF, is trained based on the nonlinear ultrasound dataset, second harmonic dataset, and fundamental frequency dataset. Step S4 specifically includes:

[0092] Step S41: Preprocess the nonlinear ultrasound dataset, second harmonic dataset, and fundamental frequency dataset. The preprocessing process includes operations such as filling sequences to the same length, label encoding, and normalizing signal data to ensure data consistency and model adaptability.

[0093] Step S42: Divide the preprocessed dataset into a training set, a validation set, and a test set. The training set is used for learning and parameter tuning of the NLUGW_CNN-PI-RF model for identifying internal defects in battery electrodes. The validation set is used to evaluate the model's performance and tune hyperparameters. The test set is used for the final evaluation of the model's generalization ability. Determine an appropriate division ratio based on the size and requirements of the dataset.

[0094] Step S43: Use CNN to extract features from the preprocessed nonlinear ultrasound dataset, second harmonic dataset, and fundamental frequency dataset.

[0095] Step S44: Use permutation importance based on random forest to select features from the extracted feature data.

[0096] Step S45: Train the NLUGW_CNN-PI-RF model for identifying internal defects in battery electrodes. Based on the feature subset selected by feature selection, optimize the model's hyperparameters using methods such as cross-validation and grid search to ensure the accuracy of identifying internal defects in battery electrodes.

[0097] Furthermore, in step S5, nonlinear ultrasonic guided wave testing is performed on the battery electrode sheets whose quality is unknown during the manufacturing stage. A trained battery electrode internal defect identification model, NLUGW_CNN-PI-RF, is used to determine whether the battery electrode sheet is normal and to identify the type of defect. Step S5 specifically includes:

[0098] Step S51: Using the above-mentioned nonlinear ultrasonic guided wave measurement system, an excitation signal with a phase of 0° is used to perform ultrasonic guided wave detection on a battery electrode with unknown mass, and a complete nonlinear ultrasonic time domain signal is obtained.

[0099] Step S52: Preprocess the nonlinear ultrasonic time-domain signal, extract features from the preprocessed signal data using CNN, select features from the extracted feature data using permutation importance based on random forest, and input the selected feature subset into the trained battery electrode internal defect identification model NLUGW_CNN-PI-RF to determine whether the battery electrode is normal and identify the type of battery electrode defect.

[0100] An application example of this invention is as follows:

[0101] A nonlinear ultrasonic guided wave measurement system was constructed to excite and receive ultrasonic guided waves.

[0102] Furthermore, in this embodiment, a nonlinear ultrasonic measurement system was constructed. The excitation transducer was a focused ultrasonic transducer with a center frequency of 500kHz and a focal length of 38.1mm, and the receiving transducer was a focused ultrasonic transducer with a center frequency of 1MHz and a focal length of 50.8mm. The excitation signal was a tone burst with a Hanning window and 20 cycles, and the sampling frequency was 250M times / second.

[0103] Furthermore, this embodiment prepared four different types of positive electrode sheets for the battery, including: electrode sheets with damaged foil, electrode sheets containing foreign matter in the foil, electrode sheets containing metal particles, and normal electrode sheets. This embodiment used a nonlinear ultrasonic guided wave measurement system to perform ultrasonic guided wave testing on the above-mentioned electrode sheets, such as... Figure 2 As shown, the excitation transducer is perpendicular to the surface of the battery electrode, emitting ultrasonic waves. The axis of the receiving transducer is at 60° to the battery electrode, used to receive the ultrasonic guided wave signal. During the measurement, the scanning range of the receiving transducer is 10 mm × 10 mm, and the scanning step size is 0.5 mm. Excitation signals with phases of 0° and 180° were used, and each type of battery electrode was measured 441 times, resulting in a total of 3528 data points in the final raw dataset. Typical received signals for a normal battery electrode and battery electrodes with different types of defects are shown below for an excitation signal with a phase of 0°. Figure 3 As shown.

[0104] For the original dataset, time-domain signals of nonlinear ultrasound, second harmonic, and fundamental frequency are extracted respectively to form nonlinear ultrasound dataset, second harmonic dataset, and fundamental frequency dataset.

[0105] Furthermore, in this embodiment, when the phase of the excitation signal is 0° in the original dataset, the received signal is a nonlinear ultrasound time-domain signal. The phase of the excitation signal is set to 180°, and a phase inversion technique is used to extract the second harmonic time-domain signal. Finally, the second harmonic time-domain signal is subtracted from the nonlinear ultrasound time-domain signal to obtain the fundamental wave time-domain signal. Figure 4 The image shows typical spectra of nonlinear ultrasound, fundamental frequency, and second harmonic frequency.

[0106] The nonlinear ultrasound dataset, second harmonic dataset, and fundamental frequency dataset are preprocessed. The preprocessed signal data are then used for feature extraction using CNN. The extracted feature data are then used for feature selection based on permutation importance using random forest. The feature subset after feature selection is input into the trained battery electrode internal defect identification model NLUGW_CNN-PI-RF to carry out nonlinear ultrasonic guided wave intelligent identification of normal battery electrodes and battery electrodes with defects.

[0107] Furthermore, this embodiment utilizes a one-dimensional convolutional neural network (1D-CNN) model to extract deep-level features from the ultrasonic signal. 1D-CNN is a type of CNN used to process one-dimensional data (such as time series and signal data). Its core idea is to extract local features from the data through convolution operations and reduce dimensionality and enhance feature invariance through pooling operations. The extracted feature data is then used for feature selection using permutation importance based on random forest. The selected feature subset is then input into the trained battery electrode internal defect identification model NLUGW_CNN-PI-RF. This embodiment uses permutation importance based on random forest for feature selection. This method combines embedding with model importance evaluation. After the random forest model is trained, each feature is randomly permuted, and the decrease in model performance is observed, thereby quantifying the contribution of a specific feature in the model's prediction.

[0108] This embodiment uses stratified sampling to divide the training and test sets. Stratified sampling ensures that the training and test sets have the same amount of data in both normal battery electrodes and battery electrodes with different types of defects. In this embodiment, 90% of the original dataset is used as the training set, and 10% is used as the test set. The test set is not used during the training phase of the NLUGW_CNN-PI-RF model and is used to verify the model's effectiveness and evaluate its generalization ability.

[0109] This embodiment employs a combination of 10-fold cross-validation and grid search to train the NLUGW_CNN-PI-RF model and select the optimal hyperparameter combination. Key hyperparameters of the NLUGW_CNN-PI-RF model include the number of decision trees and the maximum depth of the decision trees. Grid search systematically traverses all possible hyperparameter combinations to find the optimal configuration. 10-fold cross-validation is used to evaluate the performance of each hyperparameter set. First, the training set is evenly divided into 10 subsets, followed by 10 training and validation iterations: In the first fold, the first 9 subsets serve as the training set, and the 10th subset as the validation set; in the second fold, subsets 1 through 8 and the 10th subset serve as the training set, and the 9th subset as the validation set; and so on, until each subset has been used as the validation set. For each hyperparameter set, the average performance index of the 10 validation results is calculated, and the hyperparameter combination with the best average performance is finally selected. The combination of 10-fold cross-validation and grid search ensures the model's robustness and generalization ability in hyperparameter selection, providing a guarantee for model performance evaluation.

[0110] like Figure 5 As shown, this embodiment uses a confusion matrix to evaluate the performance of the NLUGW_CNN-PI-RF model. The percentages of different cells in the confusion matrix intuitively reflect the relationship between the model's predicted types and the true types. Experimental results show that the NLUGW_CNN-PI-RF model trained on different datasets exhibits significant differences in performance on the test set. Specifically, the model trained using the nonlinear ultrasound dataset demonstrates excellent predictive ability, achieving an accuracy of 94.92%. In contrast, the model trained using the fundamental wave dataset has a relatively low prediction accuracy of only 92.09%, while the model trained using the second harmonic dataset performs even worse, with an accuracy of only 75.71%. This indicates that the fundamental wave carries highly discriminative defect type information, but the second harmonic possesses defect type information that the former lacks.

[0111] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A nonlinear ultrasonic guided wave intelligent identification method for internal defects of battery electrodes, characterized in that, include: A nonlinear ultrasonic guided wave measurement system was built to determine the actual focal length and select the optimal receiving angle, excite and receive ultrasonic guided waves, and perform automated scanning. Normal and defective battery electrodes are prepared, and measurements are performed using an excitation signal with a preset phase to obtain the raw dataset; Based on the original dataset, time-domain signals of nonlinear ultrasound, second harmonic, and fundamental wave are extracted to obtain nonlinear ultrasound dataset, second harmonic dataset, and fundamental wave dataset. Features are extracted from the dataset using a convolutional neural network, and the feature set is filtered using permutation importance based on random forest. A random forest model was trained using the filtered feature set to obtain a model for identifying internal defects in battery electrodes. The nonlinear ultrasonic time-domain signal of the target battery electrode is acquired using the measurement system, features are extracted using a convolutional neural network, features are filtered using permutation importance, and the results are input into the recognition model to obtain defect recognition results. Extracting features from the dataset using a convolutional neural network includes: The nonlinear ultrasound dataset, second harmonic dataset, and fundamental frequency dataset are preprocessed. The preprocessing process includes padding the sequence to the same length, label encoding, and normalizing the signal data. The preprocessed dataset is divided into a training set, a validation set, and a test set; One-dimensional convolutional neural network models are used to extract deep features from preprocessed signal data. One-dimensional convolutional neural networks extract local features from data through convolution operations and reduce dimensionality and enhance feature invariance through pooling operations. Feature sets selected using permutation importance based on random forest include: For the features extracted by the convolutional neural network, feature selection is performed using permutation importance based on random forest; This method combines embedding with model importance evaluation. After the random forest model is trained, each feature is randomly permuted and the decline in model performance is observed. Based on the contribution of specific features in the quantification model prediction of the decline magnitude, features with high contribution are selected to form a filtered feature set.

2. The nonlinear ultrasonic guided wave intelligent identification method for internal defects of battery electrodes as described in claim 1, characterized in that, The process of setting up a nonlinear ultrasonic guided wave measurement system includes: The system consists of an arbitrary function generator, a pulse amplifier, an oscilloscope, a computer, a three-degree-of-freedom motion platform, an excitation transducer, and a receiving transducer. The excitation signal is generated by an arbitrary function generator and is a sinusoidal toneburst with n periods and frequency f, with Hanning window applied and initial phase of 0° and 180°. The center frequency of both the excitation transducer and the receiving transducer is f; The excitation transducer is placed above the surface of the battery electrode, and its position is adjusted by a three-degree-of-freedom motion platform. The actual focal length is determined based on the maximum amplitude of the time-domain signal. The angle between the axis of the receiving transducer and the battery electrode is determined by two-dimensional Fourier transform and ultrasonic guided wave dispersion curve to select the optimal receiving angle. The spatial position of the receiving transducer is adjusted using a three-degree-of-freedom motion platform to perform automated scanning with a fixed step size and path.

3. The nonlinear ultrasonic guided wave intelligent identification method for internal defects of battery electrodes as described in claim 1, characterized in that, The process of preparing normal and defective battery electrodes includes: Metal foil is selected as the substrate; aluminum foil is used for the positive electrode and copper foil is used for the negative electrode. A slurry is prepared using active materials, conductive agents, binders, and solvents and then coated onto a substrate. Pre-determined defects are created on an uncoated substrate, and microcracks, holes, or breaks are manufactured by laser cutting to prepare battery electrodes with broken foils; battery electrodes with foil foreign objects are prepared by attaching metal foils; battery electrodes with metal particles are prepared by placing metal particles. The substrate is coated, dried, rolled, and cut to ensure that the battery electrodes meet manufacturing requirements.

4. The nonlinear ultrasonic guided wave intelligent identification method for internal defects of battery electrodes as described in claim 1, characterized in that, Based on the original dataset, the time-domain signals of nonlinear ultrasound, second harmonic, and fundamental frequency were extracted, including: In the original dataset, an excitation signal with a phase of 0° is used, and the received signal is a time-domain signal of nonlinear ultrasound, including acoustic linear parameters and acoustic nonlinear parameters. Using phase reversal technology, the time-domain signal of the second harmonic is extracted from excitation signals with phases of 0° and 180°, mainly including acoustic nonlinear parameters; The time-domain signal of the fundamental wave is extracted by subtracting the time-domain signal of the nonlinear ultrasound from the time-domain signal of the second harmonic wave, which mainly includes acoustic linear parameters. Nonlinear ultrasound datasets, second harmonic datasets, and fundamental frequency datasets are generated.

5. The nonlinear ultrasonic guided wave intelligent identification method for internal defects of battery electrodes as described in claim 1, characterized in that, Training a random forest model using the selected feature set includes: The filtered feature set is used as input to train a random forest classifier; The hyperparameters of the model were optimized using 10-fold cross-validation and grid search. The grid search system systematically traverses all possible combinations of the number of decision trees and the maximum depth of the decision trees; Ten-fold cross-validation divides the training set into 10 subsets, performs 10 training and validation cycles, and calculates the average performance metric. Select the hyperparameter combination with the best average performance to obtain a well-trained battery electrode internal defect identification model.

6. The nonlinear ultrasonic guided wave intelligent identification method for internal defects of battery electrodes as described in claim 1, characterized in that, The acquisition of nonlinear ultrasonic time-domain signals of the target battery electrode using a measurement system includes: Using the established nonlinear ultrasonic guided wave measurement system, an excitation signal with a phase of 0° was used to perform ultrasonic guided wave detection on battery electrodes with unknown mass. By receiving signals through a transducer, a complete nonlinear ultrasonic time-domain signal is obtained. The nonlinear ultrasound time-domain signal is preprocessed, including filling the sequence to the same length and normalizing the signal data to ensure data consistency.

7. The nonlinear ultrasonic guided wave intelligent identification method for internal defects of battery electrodes as described in claim 1, characterized in that, Feature extraction using convolutional neural networks includes: For the preprocessed nonlinear ultrasonic time-domain signal, a one-dimensional convolutional neural network model is used for feature extraction. The one-dimensional convolutional neural network extracts local features in the signal through convolutional layers, and outputs a deep feature set by reducing dimensionality and enhancing feature invariance through pooling layers.

8. The nonlinear ultrasonic guided wave intelligent identification method for internal defects of battery electrodes as described in claim 1, characterized in that, By using permutation importance to select features, the input identification model obtains defect identification results, including: using permutation importance based on random forest to select features extracted by the convolutional neural network; Select a subset of features based on their importance scores; The filtered feature subset is input into the trained battery electrode internal defect recognition model. The model outputs the identification results of whether the battery electrodes are normal and the type of defect.

Images