Battery quality online detection

Abstract

A battery quality online detection method and apparatus, and a device. The method comprises: in a non-contact manner, collecting an initial ultrasonic signal of the surface of a slurry, the slurry being obtained by mixing and stirring a positive electrode solid-state battery material and a negative electrode solid-state battery material (201); acquiring a target ultrasonic signal corresponding to the initial ultrasonic signal (202); acquiring an acoustic feature and a signal feature of the target ultrasonic signal, and fusing the acoustic feature and the signal feature to obtain a fused feature (203); on the basis of the fused feature, predicting material components of the slurry (204); and on the basis of the material components, detecting the battery quality corresponding to the slurry (205). Battery quality online detection is achieved during slurry stirring, and battery quality can be significantly improved.

Description

Online battery quality detection Technical Field

[0001] The present application relates to the field of battery management technology, and in particular to a method, device and equipment for online detection of battery quality. Background Art

[0002] Slurry stirring is a critical step in the lithium battery manufacturing process. It is a prerequisite for subsequent processes such as coating and roller pressing, and often determines battery quality. Slurry stirring involves mixing the positive and negative solid-state battery materials to form a slurry. This process involves stirring the slurry using methods such as the orbital rotation of paddles and the rotation of a dispersion disc. Solvents are added during the stirring process to form a slurry. After stirring is complete, the material can be discharged.

[0003] Different degrees of stirring affect the slurry's composition, and thus, battery quality. For example, uneven stirring can result in the slurry's composition not meeting expectations, leading to poor battery quality. However, there is no reliable method for measuring battery quality in the relevant technology.

[0004] Summary of the Invention

[0005] This application provides a method for online battery quality detection, comprising:

[0006] collecting an initial ultrasonic signal on a surface of a slurry in a non-contact manner, wherein the slurry is obtained by mixing and stirring a positive electrode solid-state battery material and a negative electrode solid-state battery material;

[0007] Acquiring a target ultrasonic signal corresponding to the initial ultrasonic signal;

[0008] Acquiring acoustic features and signal features of the target ultrasonic signal, and fusing the acoustic features and the signal features to obtain fused features;

[0009] predicting a material composition of the slurry based on the post-fusion features;

[0010] The battery quality corresponding to the slurry is detected based on the material composition.

[0011] The present application provides a battery quality online detection device, comprising:

[0012] An acquisition module is used to collect initial ultrasonic signals from the surface of a slurry in a non-contact manner; the slurry is obtained by mixing and stirring a positive solid-state battery material and a negative solid-state battery material;

[0013] an acquisition module, configured to acquire a target ultrasonic signal corresponding to the initial ultrasonic signal;

[0014] a processing module, configured to obtain acoustic features and signal features of the target ultrasonic signal, fuse the acoustic features and the signal features to obtain fused features, and predict the material composition of the slurry based on the fused features;

[0015] A detection module is used to detect the battery quality corresponding to the slurry based on the material composition.

[0016] The present application provides an electronic device comprising a processor and a machine-readable storage medium, wherein the machine-readable storage medium stores machine-executable instructions that can be executed by the processor; the processor is configured to execute the machine-executable instructions to implement the above-described example method for online battery quality detection.

[0017] It can be seen from the above technical solution that in the embodiment of the present application, by adopting laser ultrasonic non-contact to perform real-time signal acquisition of the slurry stirring process, and using machine learning algorithm to perform real-time analysis and judgment of the material composition of the slurry, the quality of the slurry stirring process is online detected, which is equivalent to realizing online detection of battery quality, which can significantly improve the battery quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments of the present application or the description of the prior art. The drawings described below are only some of the embodiments described in this application. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings of the embodiments of the present application.

[0019] FIG1 is a schematic diagram of a process for stirring a slurry in one embodiment of the present application;

[0020] FIG2 is a schematic flow chart of a method for online battery quality detection in one embodiment of the present application;

[0021] FIG3 is a schematic diagram of an online battery quality detection system in one embodiment of the present application;

[0022] FIG4 is a schematic diagram of the training process of a machine learning model in one embodiment of the present application;

[0023] FIG5 is a schematic diagram of an online battery quality detection process in one embodiment of the present application;

[0024] 6A and 6B are schematic diagrams of the structure of a feature fusion network in one embodiment of the present application;

[0025] FIG7 is a schematic structural diagram of an online battery quality detection device in one embodiment of the present application;

[0026] FIG8 is a hardware structure diagram of an electronic device in one embodiment of the present application. DETAILED DESCRIPTION

[0027] The terms used in the embodiments of this application are only for the purpose of describing specific embodiments and are not intended to limit this application. The singular forms "a," "the," and "the" used in this application and claims are also intended to include plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used herein refers to any or all possible combinations of one or more associated listed items.

[0028] It should be understood that although the terms first, second, third, etc. may be used to describe various information in the embodiments of the present application, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from each other. For example, without departing from the scope of the present application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the word "if" used may also be interpreted as "at the time of" or "when" or "in response to determining".

[0029] In an embodiment of the present application, a method for online battery quality detection is proposed, as shown in FIG2 , including steps 201 to 205 .

[0030] Step 201: collecting an initial ultrasonic signal from the surface of a slurry in a non-contact manner; wherein the slurry is obtained by mixing and stirring a positive electrode solid-state battery material and a negative electrode solid-state battery material.

[0031] Step 202: Acquire a target ultrasonic signal corresponding to the initial ultrasonic signal.

[0032] Step 203: Acquire the acoustic features and signal features of the target ultrasonic signal, and fuse the acoustic features and the signal features to obtain fused features.

[0033] Step 204: predict the material composition of the slurry based on the fused features.

[0034] Step 205: Detect the battery quality corresponding to the slurry based on the material composition.

[0035] Illustratively, non-contact collection of the initial ultrasonic signal from the surface of the slurry may include, but is not limited to: emitting a pulsed laser to the surface of the slurry via a pulsed laser, so that the slurry generates an initial ultrasonic signal through a thermoelastic effect after receiving the pulsed laser; and non-contact collection of the initial ultrasonic signal reflected from the surface of the slurry via a vibrometer.

[0036] Illustratively, the initial ultrasonic signal may include but is not limited to an ultrasonic A-scan signal.

[0037] Exemplarily, obtaining the target ultrasonic signal corresponding to the initial ultrasonic signal may include, but is not limited to: eliminating useless signals from the initial ultrasonic signal using a bandpass filter; and performing envelope processing on the ultrasonic signal after the useless signals have been eliminated to obtain the target ultrasonic signal. In one example, envelope processing involves extracting the envelope of the ultrasonic signal, i.e., a curve showing how the amplitude of the ultrasonic signal changes over time.

[0038] Exemplarily, obtaining the acoustic characteristics and signal characteristics of the target ultrasonic signal may include but is not limited to: determining the target acoustic attenuation corresponding to the target ultrasonic signal based on a priori functional relationship between the ultrasonic signal and acoustic attenuation; determining the target sound velocity corresponding to the target ultrasonic signal based on a priori functional relationship between the ultrasonic signal and sound velocity; determining the target acoustic impedance corresponding to the target ultrasonic signal based on a priori functional relationship between the ultrasonic signal and acoustic impedance; and determining the acoustic characteristics of the target ultrasonic signal based on the target acoustic attenuation, the target sound velocity, and the target acoustic impedance.

[0039] The target ultrasonic signal is input to a trained feature extractor, which extracts signal features of the target ultrasonic signal and outputs the signal features.

[0040] Exemplarily, the fusion of the acoustic feature and the signal feature to obtain a fused feature may include, but is not limited to: if the acoustic feature includes multiple acoustic sub-features, for each acoustic sub-feature, matrix multiplying the acoustic sub-feature and the signal feature, and performing self-attention processing on the matrix-multiplied feature to obtain an attention weight matrix corresponding to the acoustic sub-feature; matrix multiplying the attention weight matrix and the target ultrasonic signal to obtain a fusion vector corresponding to the acoustic sub-feature. Feature fusion is performed on the multiple fusion vectors corresponding to the multiple acoustic sub-features to obtain a fused feature. The process of self-attention processing is described in detail below and will not be repeated here.

[0041] Exemplarily, performing feature fusion on multiple fusion vectors corresponding to the multiple acoustic sub-features to obtain fused features may include, but is not limited to: permuting and combining the multiple fusion vectors corresponding to the multiple acoustic sub-features to obtain multiple vector combinations, wherein each vector combination includes two fusion vectors; for each vector combination, performing matrix multiplication on the two fusion vectors in the vector combination, and performing self-attention processing on the features after the matrix multiplication to obtain an attention weight matrix corresponding to the vector combination; performing matrix multiplication on the attention weight matrix and the target ultrasonic signal to obtain a fusion vector corresponding to the vector combination. Performing weighted summation on the multiple fusion vectors corresponding to the multiple vector combinations to obtain the fused features.

[0042] Exemplarily, predicting the material composition of the slurry based on the fused features may include, but is not limited to: inputting the fused features into a trained material composition classifier, and having the material composition classifier predict and output the material composition of the slurry based on the fused features.

[0043] Exemplarily, detecting the battery quality corresponding to the slurry based on the material composition may include but is not limited to: if the material composition meets the expected material composition, determining that the battery quality meets the discharge conditions; if the material composition does not meet the expected material composition, determining that the battery quality does not meet the discharge conditions.

[0044] As can be seen from the above technical solutions, in the embodiments of the present application, ultrasonic signals on the surface of the slurry can be collected non-contactly, the material composition of the slurry can be predicted based on the ultrasonic signals, and the battery quality corresponding to the slurry can be detected based on the material composition, thereby providing a reliable detection method for battery quality, which can accurately detect the battery quality corresponding to the slurry. By using laser ultrasound to perform real-time signal acquisition of the slurry stirring process and using a machine learning algorithm to perform real-time analysis and judgment of the material composition, the quality of the slurry stirring process can be detected online. In the case where the slurry cannot be contacted during the slurry stirring process, ultrasonic signals can be well collected, and the characteristics of different material components can be extracted based on the ultrasonic signals. Online detection of battery quality can be achieved during the slurry stirring process, which can significantly improve battery quality.

[0045] Referring to Figure 1, which is a schematic diagram of the process of slurry stirring, the positive solid-state battery material and the negative solid-state battery material can be mixed to obtain a slurry, and the slurry is stirred by means of the revolution of the blade and the rotation of the dispersion disk. During the stirring process, a solvent is added to stir it into a slurry. After the stirring is completed, the material can be discharged. In an example, the revolution of the blade refers to the movement of the blade (also called the stirring blade) in the stirring equipment rotating around the central axis of the equipment, and the rotation of the dispersion disk refers to the self-rotation movement of the dispersion disk (or disperser) itself in the stirring equipment.

[0046] However, varying degrees of stirring can affect the slurry's composition, and thus battery quality. For example, uneven stirring can result in the slurry's composition not meeting expectations, leading to poor battery quality. However, there is no reliable method for measuring battery quality in the relevant art.

[0047] In response to the above findings, in the embodiments of the present application, laser ultrasound is used to collect real-time signals of the slurry stirring process, and a machine learning algorithm (such as a deep learning algorithm) is used to analyze and judge the material composition in real time, so that the quality of the slurry stirring process can be detected online. When the slurry cannot be contacted during the slurry stirring process, the ultrasonic signal can be well collected, and the characteristics of different material components can be extracted based on the ultrasonic signal. The online detection of battery quality is realized during the slurry stirring process, which can significantly improve the battery quality.

[0048] Figure 3 shows a schematic diagram of an online battery quality detection system based on laser ultrasound and machine learning. The system may include a pulsed laser, a vibrometer, and electronic equipment. The electronic equipment can be any type of device, such as a personal computer, terminal device, server, laptop, smartphone, or IoT device. There is no restriction on the type of electronic equipment.

[0049] Among them, the pulse laser is used to emit pulsed laser to the surface of the slurry, generating ultrasonic waves through the thermoelastic effect, and the vibrometer is used to receive the ultrasonic signals generated by the slurry. In this way, the slurry surface signal can be collected without contacting the material surface. It is a non-contact ultrasonic detection method. Electronic devices can support machine learning models (such as deep learning models) and can perform material composition analysis based on machine learning models. In one example, the thermoelastic effect refers to the mechanical strain phenomenon caused by thermal expansion or thermal contraction of an object when the object is subjected to temperature changes or temperature gradients.

[0050] In the embodiments of the present application, a training process of a machine learning model and an online battery quality detection process based on the machine learning model may be involved. The machine learning model may be a deep learning model or a neural network model. There is no restriction on this machine learning model as long as it can realize material composition analysis.

[0051] Exemplarily, the machine learning model may include a feature extractor, a feature fusion network and a material composition classifier, wherein the feature extractor is used to implement feature extraction, the feature fusion network is used to implement feature fusion, and the material composition classifier is used to classify ultrasonic signals to obtain material composition.

[0052] First, regarding the training process of the machine learning model, as shown in FIG4 , the process may include steps 401 to 408 .

[0053] Step 401: Acquire a sample data set, where the sample data set includes a plurality of initial ultrasonic signals and a label corresponding to each initial ultrasonic signal, where the label is used to represent the actual material composition corresponding to the sample slurry.

[0054] For example, the initial ultrasonic signal on the surface of a sample slurry (the slurry in the training process can be called a sample slurry, and the actual material composition corresponding to the sample slurry is known) can be collected non-contactly. The sample slurry is obtained by mixing and stirring the positive solid-state battery material and the negative solid-state battery material.

[0055] For example, a pulsed laser is emitted toward the surface of a sample slurry. Upon receiving the pulsed laser, the sample slurry generates an initial ultrasonic signal through the thermoelastic effect. This initial ultrasonic signal (i.e., laser ultrasound) reflected from the sample slurry surface can then be collected non-contact by a vibrometer.

[0056] Laser ultrasound uses laser pulses to stimulate stress pulses in the workpiece under test through thermoelastic or ablation effects. These stress pulses can generate ultrasonic signals of various waveforms. The propagated ultrasonic waves are then received by contact or non-contact methods to obtain workpiece information and defect characterization, such as workpiece thickness, internal and surface defects, and material parameters. To achieve this, a pulsed laser is emitted toward the surface of a sample slurry, and a vibrometer is used to non-contactly collect the initial ultrasonic signal reflected from the sample slurry surface.

[0057] For example, multiple sample slurries with different real material compositions can be obtained. For each sample slurry, an initial ultrasonic signal from the surface of the sample slurry can be collected non-contactly, and the real material composition corresponding to the initial ultrasonic signal can be determined. In summary, based on multiple sample slurries, a sample dataset can be obtained. The sample dataset includes multiple initial ultrasonic signals and a label corresponding to each initial ultrasonic signal, which is used to represent the real material composition corresponding to the sample slurry.

[0058] For example, the initial ultrasonic signal may include, but is not limited to, an ultrasonic A-scan signal. Of course, other types of ultrasonic signals may also be used, without limitation. The ultrasonic A-scan signal is a point-scanning signal and a waveform. The abscissa represents the propagation time or distance of the ultrasonic wave in the material being tested, and the ordinate represents the amplitude (magnitude) of the ultrasonic reflection wave.

[0059] Step 402: For each initial ultrasonic signal in the sample data set, a target ultrasonic signal corresponding to the initial ultrasonic signal is obtained. For example, the initial ultrasonic signal can be preprocessed to obtain the target ultrasonic signal corresponding to the initial ultrasonic signal. The purpose of preprocessing the initial ultrasonic signal is to remove useless signals from the ultrasonic signal and perform envelope processing on the ultrasonic signal.

[0060] Exemplarily, useless signals in the initial ultrasonic signal are eliminated using a bandpass filter to obtain a target ultrasonic signal corresponding to the initial ultrasonic signal. Alternatively, envelope processing is performed on the initial ultrasonic signal to obtain a target ultrasonic signal corresponding to the initial ultrasonic signal. Alternatively, useless signals in the initial ultrasonic signal are eliminated using a bandpass filter, and envelope processing is performed on the ultrasonic signal after useless signal elimination to obtain a target ultrasonic signal corresponding to the initial ultrasonic signal.

[0061] Among them, since the initial ultrasonic signal is a high-resolution wide-band signal, that is, the initial ultrasonic signal contains useless signals, the useless signals in the initial ultrasonic signal can be eliminated by a bandpass filter, and there is no restriction on the useless signal elimination process of the initial ultrasonic signal.

[0062] In order to extract the low-frequency modulation part in the initial ultrasonic signal, the initial ultrasonic signal may be subjected to envelope processing, such as performing Hilbert transform on the initial ultrasonic signal. There is no restriction on this transformation process.

[0063] Step 403: Acquire the acoustic characteristics (ie, acoustic performance characteristics) of the target ultrasonic signal.

[0064] For example, due to differences in Young's modulus, Poisson's ratio, and density, different materials have different acoustic characteristics. Therefore, the acoustic characteristics can be used as a basis for analyzing the material composition, that is, it is necessary to obtain the acoustic characteristics of the target ultrasonic signal. For example, the acoustic characteristics of the target ultrasonic signal can be obtained through prior knowledge, that is, the prior function relationship between the ultrasonic signal and the acoustic characteristics. In one example, Young's modulus is a physical quantity that describes the ability of a solid material to resist deformation, and Poisson's ratio is a physical quantity in material mechanics that describes the ratio of the transverse strain to the longitudinal strain of a material when it is subjected to force.

[0065] Exemplarily, the acoustic feature may include but is not limited to at least one of the following: acoustic attenuation, acoustic velocity, and acoustic impedance. Of course, the above are just a few examples, taking acoustic attenuation, sound velocity, and acoustic impedance as an example.

[0066] In one possible implementation, if the acoustic characteristic includes acoustic attenuation, which refers to the phenomenon in which the energy of a sound wave decreases with distance during propagation due to various reasons, then a target acoustic attenuation corresponding to the target ultrasonic signal can be determined based on a priori functional relationship between the ultrasonic signal and acoustic attenuation. For example, a functional relationship between the ultrasonic signal and acoustic attenuation can be preconfigured without any restrictions. The input of this functional relationship is the ultrasonic signal, and the output of this functional relationship is acoustic attenuation. Therefore, by querying this functional relationship using the target ultrasonic signal, the target acoustic attenuation can be obtained.

[0067] In one possible implementation, if the acoustic characteristic includes sound velocity, which refers to the speed at which sound waves propagate in a specific medium, then a target sound velocity corresponding to the target ultrasonic signal can be determined based on a priori functional relationship between the ultrasonic signal and the sound velocity. For example, a functional relationship between the ultrasonic signal and the sound velocity can be preconfigured without any restrictions. The input of this functional relationship is the ultrasonic signal, and the output of this functional relationship is the sound velocity. Therefore, by querying this functional relationship using the target ultrasonic signal, the target sound velocity can be obtained.

[0068] In one possible implementation, if the acoustic characteristic includes acoustic impedance, which refers to the resistance to sound wave propagation when a sound wave encounters a medium interface during propagation, then the target acoustic impedance corresponding to the target ultrasonic signal can be determined based on a prior functional relationship between the ultrasonic signal and the acoustic impedance. For example, a functional relationship between the ultrasonic signal and the acoustic impedance can be preconfigured without any restrictions. The input of this functional relationship is the ultrasonic signal, and the output of this functional relationship is the acoustic impedance. Therefore, by querying this functional relationship using the target ultrasonic signal, the target acoustic impedance can be obtained.

[0069] Exemplarily, after obtaining the target acoustic attenuation, the target sound velocity and the target acoustic impedance, the acoustic characteristics of the target ultrasonic signal can be determined based on the target acoustic attenuation, the target sound velocity and the target acoustic impedance, such as the acoustic characteristics including the target acoustic attenuation, the target sound velocity and the target acoustic impedance.

[0070] Step 404: Acquire the signal characteristics of the target ultrasonic signal.

[0071] Exemplarily, the machine learning model includes a feature extractor, which is used to implement feature extraction. For example, the feature extractor can be a convolutional neural network (CNN) or other networks. The structure of the feature extractor is not limited, as long as the feature extractor can extract features. Based on this, the target ultrasonic signal can be input into the feature extractor, which extracts the signal features of the target ultrasonic signal and outputs the signal features. The feature extractor performs operations such as convolution, thereby extracting the signal features of the target ultrasonic signal in the frequency domain.

[0072] Step 405: Fuse the acoustic feature and the signal feature to obtain a fused feature.

[0073] Exemplarily, the machine learning model includes a feature fusion network, and the feature fusion network is used to achieve feature fusion. There is no limitation on the feature fusion network, as long as it can achieve feature fusion. Based on this, the acoustic feature and the signal feature can be input into the feature fusion network, which fuses the acoustic feature and the signal feature to obtain a fused feature and outputs the fused feature.

[0074] Step 406: predict the material composition corresponding to the sample slurry based on the fused features.

[0075] Exemplarily, the machine learning model includes a material composition classifier, which is used to classify the ultrasonic signal to obtain the material composition. For example, the material composition classifier can be a classifier network. There are no restrictions on this material composition classifier, as long as it can classify the material composition. Based on this, the fused features can be input into the material composition classifier, which then predicts the predicted material composition corresponding to the initial ultrasonic signal based on the fused features. The predicted material composition corresponding to the initial ultrasonic signal is also the predicted material composition corresponding to the sample slurry, and outputs the predicted material composition.

[0076] Step 407: Determine the loss value corresponding to the initial ultrasonic signal based on the predicted material composition and the actual material composition corresponding to the sample slurry, and adjust the network parameters of the machine learning model (such as the feature extractor, the feature fusion network, and the material composition classifier) ​​based on the loss value to obtain an adjusted model.

[0077] Exemplarily, for each initial ultrasonic signal, the sample data set includes a label corresponding to the initial ultrasonic signal, and the label is used to represent the actual material composition corresponding to the sample slurry. Based on steps 402-406, the predicted material composition corresponding to the initial ultrasonic signal (i.e., the predicted material composition corresponding to the sample slurry) can be determined. In summary, it can be seen that the predicted material composition and the actual material composition corresponding to the initial ultrasonic signal can be obtained, and then the loss value corresponding to the initial ultrasonic signal is determined based on the predicted material composition and the actual material composition. If the predicted material composition is closer to the actual material composition, the smaller the loss value, and if the predicted material composition is farther away from the actual material composition, the larger the loss value.

[0078] For example, a multi-classification cross entropy loss function can be used to determine the loss value corresponding to the initial ultrasonic signal, or other loss functions can be used to determine the loss value corresponding to the initial ultrasonic signal, without limitation. See formula (1), which is an example of using a cross entropy loss function to determine the loss value:

[0079] In formula (1), p=[p0,…,p C-1] is a probability distribution, each element p i The probability that the initial ultrasonic signal belongs to the i-th category can be determined based on the probability distribution of the predicted material composition. For example, if there are C material compositions in total (i.e., the total number of categories is C), the predicted material composition can include the probability p0 corresponding to the 0th material composition (i.e., the probability that the initial ultrasonic signal belongs to the 0th category), the probability p1 corresponding to the 1st material composition (i.e., the probability that the initial ultrasonic signal belongs to the 1st category), and so on.

[0080] y=[y0,…,y C-1 ] is the one-hot encoding of the initial ultrasonic signal. When the initial ultrasonic signal belongs to the i-th category, y i =1, when the initial ultrasonic signal does not belong to the i-th category, y i = 0, the unique hot encoding can be determined based on the actual material composition. For example, there are C kinds of material components. If the actual material component is the 0th material component, then y0 is 1, [y1,…,y C-1 ] are all 0. If the real material composition is the first material composition, then y1 is 1, y0 is 0, [y2,…,y C-1 ] are all 0, and so on.

[0081] In summary, the probability distribution p = [p0,…,p C-1 ], based on the real material composition, determine the unique hot encoding y=[y0,…,y C-1 ], the cross entropy loss function shown in formula (1) determines the loss value corresponding to the initial ultrasonic signal, that is, the loss value corresponding to each initial ultrasonic signal can be obtained.

[0082] Based on the loss value corresponding to each initial ultrasonic signal, the target loss value can be determined. For example, the sum of the loss values ​​corresponding to all initial ultrasonic signals can be used as the target loss value, or the average of the loss values ​​corresponding to all initial ultrasonic signals can be used as the target loss value. There is no restriction on this.

[0083] For example, after obtaining the target loss value, the network parameters of the machine learning model can be adjusted based on the target loss value to obtain an adjusted model. For example, the network parameters of the feature extractor, the network parameters of the feature fusion network, and the network parameters of the material composition classifier can be adjusted using methods such as gradient descent. There are no restrictions on this adjustment process, and the adjustment goal is to reduce the target loss value.

[0084] Step 408: Determine whether the adjusted model has converged. If so, the adjusted model is used as the trained machine learning model and output, and the online battery quality detection process is implemented based on the machine learning model. If not, the adjusted model is used as the machine learning model to be trained, and the process returns to steps 402-407 to re-adjust the network parameters of the machine learning model to obtain the adjusted model.

[0085] For example, if the target loss value is less than a preset threshold (which can be configured based on experience), it is determined that the adjusted model has converged; if the target loss value is not less than the preset threshold, it is determined that the adjusted model has not converged.

[0086] For another example, if the number of iterations of the machine learning model reaches a threshold, it is determined that the adjusted model has converged; if the number of iterations of the machine learning model does not reach the threshold, it is determined that the adjusted model has not converged.

[0087] For another example, if the iteration time of the machine learning model reaches the time threshold, it is determined that the adjusted model has converged; if the iteration time of the machine learning model does not reach the time threshold, it is determined that the adjusted model has not converged.

[0088] At this point, the training process of the machine learning model is completed, and a trained machine learning model is obtained. The machine learning model may include a trained feature extractor, a feature fusion network, and a material composition classifier.

[0089] Second, regarding the online battery quality detection process, as shown in FIG. 5 , the process may include steps 501 to 507 .

[0090] Step 501: collecting an initial ultrasonic signal from the surface of a slurry in a non-contact manner; wherein the slurry is obtained by mixing and stirring a positive electrode solid-state battery material and a negative electrode solid-state battery material.

[0091] For example, the initial ultrasonic signal on the surface of the slurry (the slurry in the process of online battery quality detection can be called slurry, and the actual material composition corresponding to the unknown slurry needs to be predicted) can be collected non-contactly. The slurry can be obtained by mixing and stirring the positive solid-state battery material and the negative solid-state battery material.

[0092] For example, a pulsed laser is emitted toward the surface of the slurry. Upon receiving the pulsed laser, the slurry generates an initial ultrasonic signal through the thermoelastic effect. This initial ultrasonic signal, reflected from the slurry surface, can then be collected contactlessly by a vibrometer.

[0093] For example, the initial ultrasonic signal may include, but is not limited to, an ultrasonic A-scan signal. Of course, other types of ultrasonic signals may also be used, without limitation. The ultrasonic A-scan signal is a point-scanning signal and a waveform. The abscissa represents the propagation time or distance of the ultrasonic wave in the material being tested, and the ordinate represents the amplitude (magnitude) of the ultrasonic reflection wave.

[0094] Step 502: Acquire a target ultrasonic signal corresponding to the initial ultrasonic signal.

[0095] For example, the initial ultrasonic signal may be preprocessed to obtain a target ultrasonic signal corresponding to the initial ultrasonic signal. The purpose of preprocessing the initial ultrasonic signal is to remove useless signals from the initial ultrasonic signal and perform envelope processing on the initial ultrasonic signal.

[0096] Exemplarily, useless signals in the initial ultrasonic signal are eliminated using a bandpass filter to obtain a target ultrasonic signal corresponding to the initial ultrasonic signal. Alternatively, envelope processing is performed on the initial ultrasonic signal to obtain a target ultrasonic signal corresponding to the initial ultrasonic signal. Alternatively, useless signals in the initial ultrasonic signal are eliminated using a bandpass filter, and envelope processing is performed on the ultrasonic signal after useless signal elimination to obtain a target ultrasonic signal corresponding to the initial ultrasonic signal.

[0097] Since the initial ultrasonic signal is a high-resolution, wide-band signal, i.e., it contains useless signals, a bandpass filter can be used to eliminate the useless signals. To extract the low-frequency modulation portion of the initial ultrasonic signal, envelope processing can be performed on the initial ultrasonic signal, such as performing a Hilbert transform on the initial ultrasonic signal.

[0098] Step 503: Acquire the acoustic characteristics (ie, acoustic performance characteristics) of the target ultrasonic signal.

[0099] For example, the acoustic characteristics of the target ultrasonic signal can be obtained through prior knowledge, that is, a priori functional relationship between the ultrasonic signal and the acoustic characteristics. The acoustic characteristics may include, but are not limited to, at least one of the following: acoustic attenuation, sound velocity, and acoustic impedance. Of course, the above are just a few examples.

[0100] Exemplarily, if the acoustic feature includes acoustic attenuation, then: based on the prior functional relationship between the ultrasonic signal and the acoustic attenuation, the target acoustic attenuation corresponding to the target ultrasonic signal can be determined. For example, the functional relationship between the ultrasonic signal and the acoustic attenuation can be pre-configured, and the target acoustic attenuation can be obtained by querying the functional relationship through the target ultrasonic signal. If the acoustic feature includes sound velocity, then: based on the prior functional relationship between the ultrasonic signal and the sound velocity, the target sound velocity corresponding to the target ultrasonic signal can be determined. For example, the functional relationship between the ultrasonic signal and the sound velocity can be pre-configured, and the target sound velocity can be obtained by querying the functional relationship through the target ultrasonic signal. If the acoustic feature includes acoustic impedance, then: based on the prior functional relationship between the ultrasonic signal and the acoustic impedance, the target acoustic impedance corresponding to the target ultrasonic signal can be determined. For example, the functional relationship between the ultrasonic signal and the acoustic impedance can be pre-configured, and the target acoustic impedance can be obtained by querying the functional relationship through the target ultrasonic signal.

[0101] Exemplarily, after obtaining the target acoustic attenuation, the target sound velocity and the target acoustic impedance, the acoustic characteristics of the target ultrasonic signal can be determined based on the target acoustic attenuation, the target sound velocity and the target acoustic impedance, such as the acoustic characteristics including the target acoustic attenuation, the target sound velocity and the target acoustic impedance.

[0102] Step 504: Acquire the signal characteristics of the target ultrasonic signal.

[0103] For example, a trained machine learning model may include a trained feature extractor for feature extraction. A target ultrasonic signal is input to the trained feature extractor, which extracts signal features of the target ultrasonic signal and outputs the features. The feature extractor performs operations such as convolution to extract frequency-domain signal features of the target ultrasonic signal.

[0104] Step 505: Fuse the acoustic feature and the signal feature to obtain a fused feature.

[0105] Exemplarily, the trained machine learning model includes a trained feature fusion network, and the feature fusion network is used to implement feature fusion. Based on this, the acoustic feature and the signal feature can be input into the feature fusion network, and after the feature fusion network obtains the acoustic feature and the signal feature, it fuses the acoustic feature and the signal feature to obtain a fused feature and outputs the fused feature.

[0106] Step 506: Predict the material composition of the slurry based on the fused features.

[0107] Exemplarily, the trained machine learning model includes a trained material composition classifier, which is used to classify ultrasonic signals to obtain material compositions. Based on this, the fused features can be input into the material composition classifier, which then predicts the material composition corresponding to the initial ultrasonic signal based on the fused features. The material composition corresponding to the initial ultrasonic signal is also the material composition of the slurry, and outputs the material composition of the slurry.

[0108] Step 507: Detect the battery quality corresponding to the slurry based on the material composition.

[0109] For example, an expected material composition can be pre-configured, and the expected material composition is the material composition of the best battery quality or the better battery quality. Based on this, if the material composition of the slurry meets the expected material composition (for example, the material composition is the same as or close to the expected material composition), it is determined that the battery quality meets the discharge conditions, and the slurry is allowed to be discharged, that is, the slurry stirring process is completed, and the battery is manufactured based on the stirred slurry. In this case, after the slurry is made into a battery, the battery quality is relatively high.

[0110] If the material composition of the slurry does not meet the expected material composition (such as the material composition is significantly different from the expected material composition), it is determined that the battery quality does not meet the discharge conditions. When the battery quality does not meet the discharge conditions, the slurry is not allowed to be discharged. Instead, the slurry continues to be stirred and the above process is repeated until the material composition of the slurry meets the expected material composition. Alternatively, when the battery quality does not meet the discharge conditions, the slurry is abandoned, that is, the battery is not made based on the slurry. Obviously, after the above treatment, the slurry that meets the expected material composition can be used to make batteries, and the slurry that does not meet the expected material composition will not be used to make batteries, thereby significantly improving the battery quality.

[0111] Both the machine learning model training process and the online battery quality detection process involve a feature fusion network, which is used to fuse acoustic and signal features to produce fused features. In one possible implementation, a schematic diagram of the feature fusion network structure can be seen in Figure 6A. This is merely an example of a feature fusion network and does not limit its structure.

[0112] For example, the acoustic feature may include multiple acoustic sub-features. FIG6A illustrates n acoustic sub-features as an example, where n can be a positive integer. For example, the n acoustic sub-features may be three acoustic sub-features, and the three acoustic sub-features may include target acoustic attenuation, target sound velocity, and target acoustic impedance.

[0113] For each acoustic sub-feature, matrix multiplication is performed on the acoustic sub-feature and the signal feature to obtain the matrix multiplied feature. After obtaining the matrix multiplied feature, the matrix multiplied feature can be input into the self-attention network, and the self-attention network performs self-attention processing on the matrix multiplied feature to obtain the attention weight matrix corresponding to the acoustic sub-feature. Among them, the self-attention network is a network implemented using the self-attention mechanism, such as a neural network. The self-attention network is used to extract the attention weight matrix of the input feature. Therefore, after the matrix multiplied feature is input into the self-attention network, the self-attention network can extract the attention weight matrix and output the attention weight matrix.

[0114] After obtaining the attention weight matrix corresponding to the acoustic sub-feature, the attention weight matrix and the target ultrasonic signal are matrix multiplied to obtain the fusion vector corresponding to the acoustic sub-feature.

[0115] As shown in Figure 6A, acoustic sub-feature 1 and the signal feature are matrix-multiplied, and the resulting feature is input into the self-attention network. The self-attention network performs self-attention processing on the matrix-multiplied feature to obtain the attention weight matrix 1 corresponding to acoustic sub-feature 1. The attention weight matrix 1 is matrix-multiplied with the target ultrasonic signal to obtain the fusion vector 1 corresponding to acoustic sub-feature 1.

[0116] Similarly, we can perform matrix multiplication on the acoustic sub-feature n and the signal feature, and input the resulting feature into the self-attention network. The self-attention network then performs self-attention processing on the resulting feature to obtain the attention weight matrix n corresponding to the acoustic sub-feature n. We then perform matrix multiplication on the attention weight matrix n and the target ultrasonic signal to obtain the fusion vector n corresponding to the acoustic sub-feature n.

[0117] To sum up, the fusion vector corresponding to each acoustic sub-feature can be obtained, such as fusion vector 1, fusion vector 2, ..., fusion vector n. After obtaining the fusion vector corresponding to each acoustic sub-feature, the n fusion vectors corresponding to the n acoustic sub-features (including fusion vector 1, fusion vector 2, ..., fusion vector n) can be fused to obtain the fused feature.

[0118] In one possible implementation, a weighted summation may be performed on the n fusion vectors corresponding to the n acoustic sub-features to obtain a fused feature. For example, a weighted coefficient corresponding to each acoustic sub-feature is obtained, and the weighted coefficients corresponding to different acoustic sub-features may be the same or different. Based on the fusion vector and weighted coefficient corresponding to each acoustic sub-feature, a weighted summation may be performed on the n fusion vectors corresponding to the n acoustic sub-features to obtain a fused feature. For example, the fused feature is determined using the following formula:

[0119] W1*X1+W2*X2+...+W n *X n .

[0120] Where W1 represents the weighting coefficient corresponding to acoustic sub-feature 1, X1 represents the fusion vector 1 corresponding to acoustic sub-feature 1, W2 represents the weighting coefficient corresponding to acoustic sub-feature 2, X2 represents the fusion vector 2 corresponding to acoustic sub-feature 2, ..., W n represents the weighting coefficient corresponding to the acoustic sub-feature n, X n Represents the fusion vector n corresponding to acoustic sub-feature n.

[0121] In another possible implementation, a secondary feature fusion may be performed on the n fusion vectors corresponding to the n acoustic sub-features, thereby further fusing the acoustic features and signal features. A schematic diagram of the secondary feature fusion network structure can be seen in FIG6B . Of course, this is merely an example and is not intended to limit this structure.

[0122] For example, after obtaining n fusion vectors corresponding to n acoustic sub-features, the n fusion vectors are recorded as fusion vector 1, fusion vector 2, ..., fusion vector n. Combining fusion vector 1, fusion vector 2, ..., fusion vector n can generate multiple vector combinations, each of which includes two fusion vectors. For example, fusion vector 1 and fusion vector 2 form a vector combination, fusion vector 1 and fusion vector 3 form a vector combination, ..., fusion vector 1 and fusion vector n form a vector combination, fusion vector 2 and fusion vector 3 form a vector combination, ..., fusion vector n-1 and fusion vector n form a vector combination.

[0123] For each vector combination, matrix multiplication is performed on the two fused vectors within the vector combination to obtain the features after matrix multiplication. After obtaining the features after matrix multiplication, the features after matrix multiplication can be input into the self-attention network, which performs self-attention processing on the features after matrix multiplication to obtain the attention weight matrix corresponding to the vector combination. The self-attention network is used to extract the attention weight matrix of the input features. Therefore, after the features after matrix multiplication are input into the self-attention network, the self-attention network can extract the attention weight matrix and output the attention weight matrix.

[0124] After obtaining the attention weight matrix corresponding to the vector combination, the attention weight matrix and the target ultrasonic signal are matrix multiplied to obtain the fusion vector corresponding to the vector combination.

[0125] As shown in FIG6B , matrix multiplication can be performed on fusion vector 1 and fusion vector 2, and the resulting features are input into the self-attention network. The self-attention network then performs self-attention processing on the resulting features to obtain an attention weight matrix 1_2 (i.e., the attention weight matrix between fusion vector 1 and fusion vector 2). Matrix multiplication can be performed on the attention weight matrix 1_2 and the target ultrasonic signal to obtain a fusion vector 1_2 (i.e., the fusion vector between fusion vector 1 and fusion vector 2).

[0126] Similarly, we can perform matrix multiplication on the fusion vector n-1 and the fusion vector n, and input the resulting features into the self-attention network. The self-attention network then performs self-attention processing on the resulting features to obtain the attention weight matrix n-1_n. After obtaining the attention weight matrix n-1_n, we can perform matrix multiplication on the attention weight matrix n-1_n and the target ultrasonic signal to obtain the fusion vector n-1_n.

[0127] In summary, the fusion vector corresponding to each vector combination can be obtained, such as fusion vector 1_2, fusion vector 1_3, ..., fusion vector n-1_n. After obtaining the fusion vector corresponding to each vector combination, the multiple fusion vectors corresponding to the multiple vector combinations can be weighted summed to obtain the fused features. For example, the weighting coefficient corresponding to each vector combination is obtained. The weighting coefficients corresponding to different vector combinations can be the same or different. Based on the fusion vector and weighting coefficient corresponding to each vector combination, the multiple fusion vectors corresponding to the multiple vector combinations can be weighted summed to obtain the fused features. There is no restriction on the weighting process of the fusion vector.

[0128] Obviously, through the secondary feature fusion and the primary feature fusion, the acoustic features and signal features can be completely integrated. Therefore, only a weighted sum operation is needed to obtain the fused features.

[0129] As can be seen from the above technical solutions, in the embodiment of the present application, the ultrasonic signal on the surface of the slurry can be collected non-contactly, the material composition of the slurry can be predicted based on the ultrasonic signal, and the battery quality corresponding to the slurry can be detected based on the material composition, thereby providing a reliable detection method for battery quality, which can accurately detect the battery quality corresponding to the slurry. By using laser ultrasound to perform real-time signal acquisition of the slurry stirring process and using a machine learning algorithm to perform real-time analysis and judgment of the material composition, the quality of the slurry stirring process can be detected online. When the slurry stirring process cannot be contacted, the ultrasonic signal can be well collected, and the characteristics of different material components can be extracted based on the ultrasonic signal. The battery quality can be detected online during the slurry stirring process, which can significantly improve the battery quality. It is possible to perform real-time quality detection on the battery manufacturing process, which can be achieved by only using the ultrasonic signal of the laser ultrasound. It can detect the changes in the composition inside the battery and realize online real-time quality detection.

[0130] Based on the same application concept as the above method, an online battery quality detection device is proposed in an embodiment of the present application. FIG7 is a schematic structural diagram of the device, which may include the following modules.

[0131] The acquisition module 71 is used to collect the initial ultrasonic signal on the surface of the slurry in a non-contact manner; the slurry is obtained by mixing and stirring the positive solid-state battery material and the negative solid-state battery material.

[0132] The acquisition module 72 is configured to acquire a target ultrasonic signal corresponding to the initial ultrasonic signal.

[0133] The processing module 73 is used to obtain the acoustic characteristics and signal characteristics of the target ultrasonic signal, and fuse the acoustic characteristics and the signal characteristics to obtain fused characteristics; and predict the material composition of the slurry based on the fused characteristics.

[0134] The detection module 74 is configured to detect the battery quality corresponding to the slurry based on the material composition.

[0135] Exemplarily, when the acquisition module 71 acquires the initial ultrasonic signal on the surface of the slurry in a non-contact manner, it is specifically used to: emit a pulsed laser to the surface of the slurry through a pulsed laser, so that the slurry generates an initial ultrasonic signal through a thermoelastic effect after receiving the pulsed laser; and acquire the initial ultrasonic signal reflected by the surface of the slurry in a non-contact manner through a vibrometer; wherein the initial ultrasonic signal includes an ultrasonic A-scan signal.

[0136] Exemplarily, when the acquisition module 72 acquires the target ultrasonic signal corresponding to the initial ultrasonic signal, it is specifically used to: eliminate useless signals in the initial ultrasonic signal through a bandpass filter; and perform envelope processing on the ultrasonic signal after the useless signal is eliminated to obtain the target ultrasonic signal.

[0137] Exemplarily, when the processing module 73 obtains the acoustic characteristics and signal characteristics of the target ultrasonic signal, it is specifically used to: determine the target acoustic attenuation corresponding to the target ultrasonic signal based on the prior function relationship between the ultrasonic signal and the acoustic attenuation; determine the target sound velocity corresponding to the target ultrasonic signal based on the prior function relationship between the ultrasonic signal and the sound velocity; determine the target acoustic impedance corresponding to the target ultrasonic signal based on the prior function relationship between the ultrasonic signal and the acoustic impedance; determine the acoustic characteristics of the target ultrasonic signal based on the target acoustic attenuation, the target sound velocity and the target acoustic impedance; input the target ultrasonic signal into a trained feature extractor, and the feature extractor extracts the signal characteristics of the target ultrasonic signal and outputs the signal characteristics.

[0138] Exemplarily, when the processing module 73 fuses the acoustic feature and the signal feature to obtain a fused feature, it is specifically used for: if the acoustic feature includes multiple acoustic sub-features, for each acoustic sub-feature, performing matrix multiplication on the acoustic sub-feature and the signal feature, and performing self-attention processing on the feature after the matrix multiplication to obtain an attention weight matrix corresponding to the acoustic sub-feature; performing matrix multiplication on the attention weight matrix and the target ultrasonic signal to obtain a fusion vector corresponding to the acoustic sub-feature; performing feature fusion on multiple fusion vectors corresponding to multiple acoustic sub-features to obtain a fused feature.

[0139] Exemplarily, the processing module 73 performs feature fusion on multiple fusion vectors corresponding to multiple acoustic sub-features to obtain the fused features, which is specifically used for: combining multiple fusion vectors corresponding to multiple acoustic sub-features to obtain multiple vector combinations, wherein each vector combination includes two fusion vectors; for each vector combination, performing matrix multiplication on the two fusion vectors in the vector combination, and performing self-attention processing on the features after the matrix multiplication to obtain the attention weight matrix corresponding to the vector combination; performing matrix multiplication on the attention weight matrix and the target ultrasonic signal to obtain the fusion vector corresponding to the vector combination; performing weighted summation on multiple fusion vectors corresponding to multiple vector combinations to obtain the fused features.

[0140] Exemplarily, when the processing module 73 predicts the material composition of the slurry based on the fused features, it is specifically used to: input the fused features to a trained material composition classifier, and the material composition classifier predicts and outputs the material composition of the slurry based on the fused features; when the detection module 74 detects the battery quality corresponding to the slurry based on the material composition, it is specifically used to: if the material composition meets the expected material composition, determine that the battery quality meets the discharge conditions; if the material composition does not meet the expected material composition, determine that the battery quality does not meet the discharge conditions.

[0141] Based on the same application concept as the above method, an electronic device is proposed in an embodiment of the present application, as shown in Figure 8, the electronic device includes: a processor 81 and a machine-readable storage medium 82, the machine-readable storage medium 82 stores machine-executable instructions that can be executed by the processor 81; the processor 81 is used to execute the machine-executable instructions to implement the battery quality online detection method disclosed in the above example of this application.

[0142] Based on the same application concept as the above method, an embodiment of the present application also provides a machine-readable storage medium, on which a number of computer instructions are stored. When the computer instructions are executed by a processor, the online battery quality detection method disclosed in the above example of the present application can be implemented.

[0143] The machine-readable storage medium may be any electronic, magnetic, optical, or other physical storage device that may contain or store information, such as executable instructions, data, and the like. For example, the machine-readable storage medium may be: RAM (Random Access Memory), volatile memory, non-volatile memory, flash memory, a storage drive (such as a hard disk drive), a solid-state drive, any type of storage disk (such as a CD, DVD, etc.), or similar storage media, or a combination thereof.

[0144] The systems, devices, modules, or units described in the above embodiments may be implemented by a computer or entity, or by a product having certain functions. A typical implementation device is a computer, which may be in the form of a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email transceiver, game console, tablet computer, wearable device, or any combination of these devices.

[0145] For the convenience of description, the above devices are described as being divided into various units according to their functions. Of course, when implementing this application, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0146] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the embodiments of the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0147] The present application is described with reference to the flow chart and / or block diagram of the method, device (system) and computer program product according to the embodiment of the present application. It should be understood that each flow process and / or box in the flow chart and / or block diagram and the combination of the flow process and / or box in the flow chart and / or block diagram can be realized by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processing machine or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a device for realizing the function specified in one flow chart flow or multiple flows and / or one box or multiple boxes of the block diagram.

[0148] Moreover, these computer program instructions can also be stored in a computer-readable memory that can guide a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device, which implements the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0149] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0150] The foregoing is merely an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application should be included within the scope of the claims of the present application.

Claims

1. A battery quality online detection method, comprising: collecting an initial ultrasonic signal on the surface of a slurry in a non-contact manner, wherein the slurry is obtained by mixing and stirring a positive electrode solid-state battery material and a negative electrode solid-state battery material; Acquire a target ultrasonic signal corresponding to the initial ultrasonic signal; Acquiring an acoustic feature and a signal feature of the target ultrasonic signal, and fusing the acoustic feature and the signal feature to obtain a fused feature; predicting a material composition of the slurry based on the post-fusion features; The battery quality corresponding to the slurry is detected based on the material composition.

2. The method according to claim 1, characterized in that: The non-contact collection of the initial ultrasonic signal on the surface of the slurry comprises: emitting a pulsed laser to the surface of the slurry by means of a pulsed laser, so that the slurry generates the initial ultrasonic signal by means of a thermoelastic effect after receiving the pulsed laser; Collecting the initial ultrasonic signal reflected from the surface of the slurry in a non-contact manner by a vibrometer; Wherein, the initial ultrasonic signal includes an ultrasonic A-scan signal.

3. The method according to claim 1 or 2, characterized in that: The step of acquiring a target ultrasonic signal corresponding to the initial ultrasonic signal includes: Eliminating useless signals in the initial ultrasonic signal through a bandpass filter to obtain an ultrasonic signal after the useless signals are eliminated; The target ultrasonic signal is obtained by performing envelope processing on the ultrasonic signal after the useless signal is eliminated.

4. The method according to any one of claims 1 to 3, characterized in that: The step of acquiring the acoustic characteristics and signal characteristics of the target ultrasonic signal comprises: Determining a target sound attenuation corresponding to the target ultrasonic signal based on a priori functional relationship between the ultrasonic signal and the sound attenuation; Determining a target sound speed corresponding to the target ultrasonic signal based on a priori functional relationship between the ultrasonic signal and the sound speed; Determining a target acoustic impedance corresponding to the target ultrasonic signal based on a priori functional relationship between the ultrasonic signal and the acoustic impedance; determining an acoustic characteristic of the target ultrasonic wave signal based on the target acoustic attenuation, the target sound velocity, and the target acoustic impedance; The target ultrasonic signal is input to a trained feature extractor, and the feature extractor extracts signal features of the target ultrasonic signal and outputs the signal features.

5. The method according to any one of claims 1 to 4, characterized in that: The fusing the acoustic feature and the signal feature to obtain a fused feature includes: If the acoustic feature includes multiple acoustic sub-features, for each acoustic sub-feature, Performing matrix multiplication on the acoustic sub-feature and the signal feature, and performing self-attention processing on the feature after the matrix multiplication to obtain an attention weight matrix corresponding to the acoustic sub-feature; Performing matrix multiplication on the attention weight matrix and the target ultrasonic signal to obtain a fusion vector corresponding to the acoustic sub-feature; Feature fusion is performed on multiple fusion vectors corresponding to the multiple acoustic sub-features to obtain the fused feature.

6. The method according to claim 5, characterized in that The step of performing feature fusion on the multiple fusion vectors corresponding to the multiple acoustic sub-features to obtain the fused feature includes: Combining a plurality of fusion vectors corresponding to the plurality of acoustic sub-features to obtain a plurality of vector combinations, wherein each of the vector combinations includes two fusion vectors; For each of the vector combinations, Performing matrix multiplication on two fused vectors in the vector combination, and performing self-attention processing on the features after the matrix multiplication, to obtain an attention weight matrix corresponding to the vector combination; Performing matrix multiplication on the attention weight matrix and the target ultrasonic signal to obtain a fusion vector corresponding to the vector combination; A weighted sum is performed on a plurality of fused vectors corresponding to the plurality of vector combinations to obtain the fused feature.

7. The method according to any one of claims 1 to 6, characterized in that: The predicting of the material composition of the slurry based on the post-fusion characteristics comprises: The fused features are input to a trained material component classifier, and the material component classifier predicts and outputs the material composition of the slurry based on the fused features.

8. The method according to any one of claims 1 to 7, characterized in that: The detecting the battery quality corresponding to the slurry based on the material composition includes: If the material composition is as expected, it is determined that the battery quality corresponding to the slurry meets the discharge conditions; If the material composition does not meet expectations, it is determined that the battery quality corresponding to the slurry does not meet the discharge conditions.

9. A battery quality online detection device, comprising: A collection module, used for collecting initial ultrasonic signals on the surface of the slurry in a non-contact manner; The slurry is obtained by mixing and stirring the positive electrode solid-state battery material and the negative electrode solid-state battery material; An acquisition module, used for acquiring a target ultrasonic signal corresponding to the initial ultrasonic signal; A processing module, used for acquiring the acoustic features and signal features of the target ultrasonic signal, and fusing the acoustic features and the signal features to obtain fused features; predicting a material composition of the slurry based on the post-fusion features; A detection module is used to detect the battery quality corresponding to the slurry based on the material composition.

10. The device according to claim 9, It is characterized in that in, The acquisition module is specifically used to collect the initial ultrasonic signal on the surface of the slurry in a non-contact manner: emit a pulse laser to the surface of the slurry through a pulse laser, so that the slurry generates the initial ultrasonic signal through a thermoelastic effect after receiving the pulse laser; and collect the initial ultrasonic signal reflected by the surface of the slurry in a non-contact manner through a vibrometer; Wherein, the initial ultrasonic signal includes an ultrasonic A-scan signal; Wherein, when the acquisition module acquires the target ultrasonic signal corresponding to the initial ultrasonic signal, it is specifically used to: Eliminating useless signals in the initial ultrasonic signal through a bandpass filter to obtain an ultrasonic signal after the useless signals are eliminated; performing envelope processing on the ultrasonic signal after the useless signals are eliminated to obtain the target ultrasonic signal; Wherein, when the processing module obtains the acoustic characteristics and signal characteristics of the target ultrasonic signal, it is specifically used to: determine the target acoustic attenuation corresponding to the target ultrasonic signal based on the prior function relationship between the ultrasonic signal and the acoustic attenuation; determine the target sound velocity corresponding to the target ultrasonic signal based on the prior function relationship between the ultrasonic signal and the sound velocity; determine the target acoustic impedance corresponding to the target ultrasonic signal based on the prior function relationship between the ultrasonic signal and the acoustic impedance; determine the acoustic characteristics of the target ultrasonic signal based on the target acoustic attenuation, the target sound velocity and the target acoustic impedance; input the target ultrasonic signal to a trained feature extractor, and the feature extractor extracts the signal characteristics of the target ultrasonic signal and outputs the signal characteristics; Wherein, when the processing module fuses the acoustic feature and the signal feature to obtain a fused feature, it is specifically used to: if the acoustic feature includes multiple acoustic sub-features, for each acoustic sub-feature, perform matrix multiplication on the acoustic sub-feature and the signal feature, and perform self-attention processing on the feature after the matrix multiplication to obtain an attention weight matrix corresponding to the acoustic sub-feature; perform matrix multiplication on the attention weight matrix and the target ultrasonic signal to obtain a fusion vector corresponding to the acoustic sub-feature; perform feature fusion on multiple fusion vectors corresponding to the multiple acoustic sub-features to obtain the fused feature; Among them, the processing module performs feature fusion on multiple fusion vectors corresponding to the multiple acoustic sub-features to obtain the fused features, which is specifically used for: combining the multiple fusion vectors corresponding to the multiple acoustic sub-features to obtain multiple vector combinations, wherein each of the vector combinations includes two fusion vectors; for each of the vector combinations, performing matrix multiplication on the two fusion vectors in the vector combination, and performing self-attention processing on the features after the matrix multiplication to obtain the attention weight matrix corresponding to the vector combination; performing matrix multiplication on the attention weight matrix and the target ultrasonic signal to obtain the fusion vector corresponding to the vector combination; performing weighted summation on the multiple fusion vectors corresponding to the multiple vector combinations to obtain the fused features; Among them, when the processing module predicts the material composition of the slurry based on the fused features, it is specifically used to: input the fused features to a trained material composition classifier, and the material composition classifier predicts and outputs the material composition of the slurry based on the fused features; when the detection module detects the battery quality corresponding to the slurry based on the material composition, it is specifically used to: if the material composition meets expectations, determine that the battery quality corresponding to the slurry meets the discharge conditions; if the material composition does not meet expectations, determine that the battery quality corresponding to the slurry does not meet the discharge conditions.

11. An electronic device comprising a processor and a machine-readable storage medium, characterized in that: The machine-readable storage medium stores machine-executable instructions that can be executed by the processor; the processor is used to execute the machine-executable instructions to implement any method described in claims 1-8.

Images