Ultrasonic areal density measurement control method, system, and apparatus

Abstract

The application provides an ultrasonic surface density measurement control method, system and device, relates to the field of nondestructive testing control, and utilizes a supervision data pair of reference surface density data and reference waveform data constructed under various environmental conditions to realize end-to-end accurate mapping of surface density data and waveform data under various environments, so that the measurement accuracy is greatly improved. In addition, the method can obtain the surface density value corresponding to the spindle-shaped waveform data in real time by using the supervision data pair, and the measurement real-time performance of the ultrasonic surface density measurement device is improved.

Description

Technical Field

[0001] This invention relates to the field of nondestructive testing control, and in particular to an ultrasonic surface density measurement control method, system and equipment. Background Technology

[0002] In measurement scenarios such as lithium battery coated electrode sheets, ultrasonic areal density measuring equipment uses a transmitter to drive a transducer to generate an ultrasonic signal at a specific frequency. After the ultrasonic signal penetrates the medium under test, it is collected and received by the corresponding transducer at the receiver, ultimately obtaining specific spindle-shaped waveform data. This data fully carries the characteristic information of the interaction between the ultrasonic wave and the medium under test.

[0003] However, in real-world scenarios, such spindle-shaped waveform data is easily affected by various environmental factors, such as temperature and humidity fluctuations, air disturbances, air pressure changes, and differences in coupling states between the transducer and the measured medium. This can lead to unexpected differences in amplitude, frequency, pulse width, and other characteristics of the spindle-shaped waveform obtained when measuring the same measured medium under different environmental conditions.

[0004] Existing technologies primarily address measurement environment fluctuations by manually correcting environmental parameters and optimizing the measurement environment, but these methods cannot fundamentally solve the waveform differences caused by environmental fluctuations. Specifically, manual correction relies on operator experience, has limited accuracy, and cannot adapt to dynamically changing environments. Optimizing the measurement environment requires additional equipment costs and is difficult to implement in complex scenarios such as outdoor environments and production lines. Furthermore, the physical or mathematical models used in traditional areal density calculations can only establish a linear relationship between a single waveform parameter and areal density. When facing complex and changing environments, they cannot capture the complex nonlinear relationships between multiple parameters and areal density in a spindle-shaped waveform, nor can they handle the ambiguity caused by many-to-one datasets, resulting in low measurement accuracy and poor real-time performance. Summary of the Invention

[0005] In view of this, the purpose of the present invention is to provide an ultrasonic surface density measurement and control method, system and device. The method utilizes a monitoring data pair constructed from reference surface density data and reference waveform data under various environmental conditions to achieve end-to-end accurate mapping between surface density data and waveform data under various environments, which greatly improves the measurement accuracy. In addition, the method can obtain the surface density value corresponding to the spindle-shaped waveform data in real time using the monitoring data pair, which improves the measurement real-time performance of the ultrasonic surface density measuring device.

[0006] In a first aspect, embodiments of the present invention provide an ultrasonic surface density measurement control method, which is applied in the controller of an ultrasonic surface density measuring device, the ultrasonic surface density measuring device further comprising an ultrasonic transmitting module, an ultrasonic receiving module, a signal processing module, and a surface density calculation module; The method includes: After controlling the ultrasonic transmitting module to transmit ultrasonic signals to the medium under test, control the ultrasonic receiving module to receive the transmitted signal after the ultrasonic signal passes through the medium under test. The spindle-shaped waveform data of the transmission signal is obtained by controlling the signal processing module to extract features of the transmission signal through the spindle signal extraction strategy corresponding to the medium under test. Obtain a supervisory data pair constructed from the reference surface density data and reference waveform data of the medium under preset environmental conditions, and use the supervisory data to generate a convolutional feature map of the surface density data corresponding to the spindle-shaped waveform data in the control surface density calculation module. The channel dimension value control areal density calculation module based on convolutional feature map performs regression prediction calculation on spindle-shaped waveform data to obtain the areal density value of the medium under test.

[0007] Optionally, controlling the ultrasonic transmitting module to transmit ultrasonic signals to the medium under test includes: Acquire the first ultrasonic transducer in the ultrasonic transmitting module that meets the target transmission frequency and its corresponding transformer boost drive circuit. The resonant frequency of the transformer boost drive circuit is determined by the target transmission frequency; The resonant frequency control transformer boost drive circuit outputs excitation pulses to the first ultrasonic transducer, causing the first ultrasonic transducer to emit ultrasonic signals to the medium under test at the target emission frequency; wherein, the medium under test is located in the coaxial acoustic path between the ultrasonic transmitting module and the ultrasonic receiving module.

[0008] Optionally, the ultrasonic receiving module is controlled to receive the transmitted signal after the ultrasonic signal has passed through the medium under test, including: The specification matching strategy of the first ultrasonic transducer is determined based on the target transmission frequency and bandwidth value of the ultrasonic signal. The second ultrasonic transducer and its corresponding preamplifier that meet the specification matching strategy in the ultrasonic receiving module are obtained, and the timing synchronization frequency of the second ultrasonic transducer is obtained based on the target transmission frequency. The second ultrasonic transducer is controlled by timing synchronization frequency to receive the acoustic vibration electrical signal after the ultrasonic signal passes through the medium under test, and the spindle-shaped target signal corresponding to the acoustic vibration electrical signal is determined based on the specification matching strategy. After the acoustic vibration electrical signal is amplified in stages by controlling the preamplifier according to the spindle-shaped target signal, the transmission signal corresponding to the acoustic vibration electrical signal is obtained.

[0009] Optionally, the step of obtaining the spindle-shaped waveform data of the transmission signal by controlling the signal processing module to extract features from the transmission signal through the spindle-shaped signal extraction strategy corresponding to the medium under test includes: The analog electrical signal of the transmitted signal is converted from analog to digital using the spindle-shaped target signal control signal processing module to obtain the corresponding digital signal. The feature extraction strategy and noise suppression strategy of the medium under test are determined by the target waveform corresponding to the spindle-shaped target signal. The feature extraction strategy is used to control the signal processing module to extract the DC component features and high-frequency noise features contained in the digital signal; After the DC component characteristics and high-frequency noise characteristics are suppressed by the control signal processing module using a noise suppression strategy, the spindle-shaped waveform data of the transmitted signal is obtained.

[0010] Optionally, a supervisory data pair is constructed by acquiring the reference surface density data and reference waveform data of the medium under preset environmental conditions, including: By using the weighing values ​​of reference measurement media corresponding to different areal density grades, reference areal density data of the medium to be tested under each areal density grade are obtained; wherein, each areal density grade includes multiple reference measurement media; Under different environmental conditions, the ultrasonic surface density measuring device is controlled to repeatedly collect the transmission waveform signal of each reference measuring medium according to the preset collection quantity, so as to obtain the transmission waveform signal set corresponding to each reference measuring medium; After calculating the normalized waveform signal of each transmitted waveform signal by the signal extrema in the transmitted waveform signal set, the normalized waveform signal is subjected to sliding filtering calculation using a filter window of preset length to obtain the reference waveform data of the medium under test. The control surface density calculation module binds the reference waveform data and the reference surface density data according to the surface density level and environmental conditions to obtain the supervision data pair.

[0011] Optionally, the step of the control areal density calculation module binding the reference waveform data and the reference areal density data according to the areal density level and environmental conditions to obtain the supervision data pair includes: Obtain the initialized neural network model in the surface density calculation module; the neural network model includes a feature extraction layer and a fully connected layer; the feature extraction layer includes multiple sequentially connected deep convolutional blocks; each deep convolutional block includes a one-dimensional deep convolutional layer, a one-dimensional pointwise convolutional layer, a batch normalization layer, a linear rectified layer, and a max pooling layer; By using surface density levels and environmental conditions, the reference waveform data and reference surface density data are divided according to a preset ratio to obtain the training set, validation set and test set corresponding to the neural network model. After the control surface density calculation module inputs the training set into the neural network model, it sequentially obtains the feature maps output by the depth convolution blocks; The network weight parameters of the one-dimensional deep convolutional layer, one-dimensional pointwise convolutional layer, batch normalization layer, linear rectified layer, max pooling layer and fully connected layer in each deep convolutional block are updated according to the mean square error between the surface density prediction value and the benchmark surface density data in the validation set. When the mean square error meets the preset threshold condition, the update process of the depth convolution block is stopped. After binding the baseline waveform data and the surface density prediction value under the current network weight parameters, the supervised data pair is obtained.

[0012] Optionally, using supervised data, the control surface density calculation module generates a convolutional feature map of the surface density data corresponding to the spindle-shaped waveform data, including: The current network weight parameters of the deep convolutional blocks are obtained using supervised data, and the convolutional kernel and stride values ​​corresponding to each deep convolutional block are obtained. After the control surface density calculation module inputs the spindle-shaped waveform data into the neural network model under the current network weight parameters, it obtains the local temporal feature map output by the first depth convolution block through the convolution kernel and stride value. After inputting the local temporal feature map into the second deep convolutional block, the spindle wave envelope morphology feature map output by the second deep convolutional block is obtained through the convolution kernel and stride value. After inputting the spindle wave envelope morphology feature map into the third depth convolution block, the convolution feature map output by the third depth convolution block is obtained through the convolution kernel and stride value.

[0013] Optionally, the step of using the channel dimension value control of the convolutional feature map to perform regression prediction calculation on the spindle-shaped waveform data to obtain the surface density value of the medium under test includes: The dimension transformation strategy for the fully connected layer is generated based on the sequence length value and channel dimension value of the convolutional feature map. The one-dimensional feature vector corresponding to the spindle-shaped waveform data is obtained by controlling the surface density calculation module through a dimension transformation strategy. After performing regression prediction calculation on the one-dimensional feature vector using the fully connected layer control surface density calculation module under the current network weight parameters, the feature combination result of the spindle-shaped waveform data is obtained. The surface density value of the medium under test is then obtained by calculating the feature combination result through a preset linear activation function.

[0014] Secondly, the present invention provides an ultrasonic surface density measurement and control system, which is applied in the controller of an ultrasonic surface density measuring device. The ultrasonic surface density measuring device also includes an ultrasonic transmitting module, an ultrasonic receiving module, a signal processing module, and a surface density calculation module. The system includes: Ultrasonic signal control module: Used to control the ultrasonic transmitting module to transmit ultrasonic signals to the medium under test, and to control the ultrasonic receiving module to receive the transmitted signal after the ultrasonic signal passes through the medium under test. Waveform data acquisition module: Used to control the signal processing module to extract features from the transmission signal according to the spindle signal extraction strategy corresponding to the medium under test, and then obtain the spindle-shaped waveform data of the transmission signal. Convolutional feature map generation module: used to obtain the supervision data pair constructed by the reference surface density data and reference waveform data of the medium under the preset environmental conditions, and use the supervision data to generate the convolutional feature map of the surface density data corresponding to the spindle waveform data of the control surface density calculation module. Areal density value measurement and calculation module: This module controls the areal density calculation module based on the channel dimension value of the convolutional feature map to perform regression prediction calculation on the spindle-shaped waveform data and obtain the areal density value of the medium under test.

[0015] Thirdly, embodiments of the present invention also provide an ultrasonic surface density measuring device, which includes: an ultrasonic transmitting module, an ultrasonic receiving module, a signal processing module, a surface density calculation module, and a controller; the controller is connected to the ultrasonic transmitting module, the ultrasonic receiving module, the signal processing module, and the surface density calculation module respectively; The controller includes a processor and a memory. The memory stores computer-executable instructions that can be executed by the processor. The processor executes the computer-executable instructions to implement the ultrasonic surface density measurement control method provided in the first aspect.

[0016] This invention provides an ultrasonic areal density measurement control method, system, and device, applied in the controller of an ultrasonic areal density measuring device. The ultrasonic areal density measuring device further includes an ultrasonic transmitting module, an ultrasonic receiving module, a signal processing module, and an areal density calculation module. To address the problem of measurement result discrepancies caused by environmental fluctuations during the ultrasonic areal density measurement process, this method controls the ultrasonic transmitting module to transmit ultrasonic signals to the medium under test, and then controls the ultrasonic receiving module to receive the transmitted signal after the ultrasonic signal passes through the medium under test. Then, using a spindle signal extraction strategy corresponding to the medium under test, the signal processing module extracts features from the transmitted signal to obtain spindle-shaped waveform data. Subsequently, a supervisory data pair is constructed by acquiring the reference areal density data and the reference waveform data of the medium under test under preset environmental conditions. The supervisory data is used to control the areal density calculation module to generate a convolutional feature map of the areal density data corresponding to the spindle-shaped waveform data. Finally, based on the channel dimension value of the convolutional feature map, the areal density calculation module performs regression prediction calculations on the spindle-shaped waveform data to obtain the areal density value of the medium under test. This method utilizes supervised data pairs constructed from reference surface density data and reference waveform data under various environmental conditions to achieve end-to-end accurate mapping between surface density data and waveform data in various environments, significantly improving measurement accuracy. In addition, this method can obtain the surface density value corresponding to the spindle-shaped waveform data in real time using supervised data pairs, improving the real-time performance of ultrasonic surface density measuring equipment.

[0017] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention are realized and obtained through the structures particularly pointed out in the description and the drawings.

[0018] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0019] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0020] Figure 1 A flowchart of an ultrasonic surface density measurement and control method provided in an embodiment of the present invention; Figure 2A schematic diagram of the neural network model used in an ultrasonic surface density measurement and control method provided in an embodiment of the present invention; Figure 3 A flowchart of another ultrasonic surface density measurement and control method provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of an ultrasonic surface density measurement and control system provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of the structure of an ultrasonic surface density measuring device provided in an embodiment of the present invention; Figure 6 This is a schematic diagram of the controller in an ultrasonic surface density measuring device provided in an embodiment of the present invention.

[0021] icon: 100 - Ultrasonic signal control module; 200 - Waveform data acquisition module; 300 - Convolutional feature map generation module; 400 - Areal density value measurement and calculation module; 510 - Ultrasonic transmitting module; 520 - Ultrasonic receiving module; 530 - Signal processing module; 540 - Areal density calculation module; 550 - Controller; 101 - Processor; 102 - Memory; 103 - Bus; 104 - Communication interface. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] To facilitate understanding of this embodiment, a detailed description of the ultrasonic areal density measurement control method disclosed in this invention will be provided first. Specifically, this method is applied to the controller of an ultrasonic areal density measuring device, which further includes an ultrasonic transmitting module, an ultrasonic receiving module, a signal processing module, and an areal density calculation module. Based on this, as... Figure 1 As shown, the method includes: Step S101: After controlling the ultrasonic transmitting module to transmit ultrasonic signals to the medium under test, control the ultrasonic receiving module to receive the transmitted signal after the ultrasonic signal passes through the medium under test.

[0024] The controller first sends control commands to the ultrasonic transmitting module, driving the high-frequency ultrasonic transducer in the transmitting module (adapted to the measurement requirements of lithium battery coated electrodes). Through a transformer boost drive circuit with a preset frequency (e.g., 100kHz), a stable ultrasonic signal is generated and transmitted to the lithium battery coated electrode under test. Simultaneously, the controller synchronously controls the ultrasonic receiving module (set relative to the transmitting end with adjustable spacing, using an ultrasonic transducer matched to the transmitting end) to collect the transmission signal after the ultrasonic signal penetrates the lithium battery coated electrode under test in real time, ensuring that the receiving end and the transmitting end are synchronized in timing, reducing waveform misalignment caused by signal delay. During this process, the controller monitors the transmission power and timing in real time to ensure that the ultrasonic signal can effectively penetrate the electrode, while avoiding initial signal attenuation caused by environmental factors (such as air disturbance and differences in coupling state), providing a high-quality original transmission signal for subsequent waveform extraction.

[0025] Step S102: After the signal processing module extracts features from the transmission signal using the spindle signal extraction strategy corresponding to the medium under test, the spindle-shaped waveform data of the transmission signal is obtained.

[0026] The controller invokes the preset spindle signal extraction strategy, sends control commands to the signal processing module, performs a series of precise processing on the transmitted signal transmitted by the receiving module, and finally extracts the complete spindle-shaped waveform data.

[0027] Specifically, the signal processing module first performs input adaptation and analog-to-digital conversion on the transmitted signal, converting the continuous analog electrical signal into a discrete digital signal. Then, through a preset preprocessing process (such as amplitude normalization and moving average filtering), it eliminates the effects of acquisition gain fluctuations, high-frequency sampling noise, and environmental interference (such as temperature and humidity fluctuations and electromagnetic interference). Finally, through a feature extraction algorithm, it accurately captures the spindle-shaped waveform data corresponding to the transmitted signal. This waveform data fully carries the characteristic information (such as amplitude, frequency, and pulse width) after the interaction between the ultrasonic wave and the coated electrode of the lithium battery under test, providing core data support for subsequent areal density calculation and ensuring that the waveform data can truly reflect the areal density characteristics of the electrode.

[0028] Step S103: Obtain the supervision data pair constructed from the reference surface density data and reference waveform data of the medium under preset environmental conditions, and use the supervision data to generate the convolution feature map of the surface density data corresponding to the spindle waveform data in the control surface density calculation module.

[0029] The controller first acquires a preset monitoring data pair, which is constructed from the reference areal density data and reference waveform data corresponding to the coated electrode of the lithium battery under test under various environmental conditions such as temperature, humidity, and air pressure. Then, the controller sends control commands to the areal density calculation module to perform feature extraction layer work, using the monitoring data pair as a constraint. Specifically, the spindle-shaped waveform data extracted in step S102 is subjected to layer-by-layer feature extraction, which sequentially completes depthwise convolution (capturing local temporal features), pointwise convolution (achieving channel feature fusion), batch normalization (eliminating amplitude drift caused by the environment), activation function mapping (adapting to nonlinear correlation), and max pooling downsampling (preserving key features and reducing the number of parameters). Finally, an areal density-correlated convolutional feature map corresponding to the spindle-shaped waveform data is generated. This feature map integrates multi-scale abstract features in the waveform that are strongly correlated with areal density, effectively avoiding the influence of waveform differences caused by environmental interference.

[0030] Step S104: The channel dimension value control module based on the convolutional feature map performs regression prediction calculation on the spindle-shaped waveform data to obtain the surface density value of the medium under test.

[0031] The controller uses the channel dimension values ​​of the convolutional feature map generated in step S103 (each channel corresponds to a specific abstract feature related to areal density) as the core control basis to drive the areal density calculation module to perform regression prediction calculation. Specifically, firstly, the high-dimensional convolutional feature map is converted into a low-dimensional feature vector, fully preserving the areal density-related features corresponding to each channel; then, the low-dimensional feature vector is input into multiple cascaded fully connected layers, deepening the feature combination layer by layer, strengthening the feature weights strongly correlated with areal density, weakening redundant feature interference, and solidifying the nonlinear mapping relationship between waveform features and areal density; finally, in the output layer, a linear activation function (adapting to the requirement of continuous areal density prediction) is used to map the fused global feature vector into continuous areal density prediction values, completing the end-to-end regression prediction of the spindle-shaped waveform data.

[0032] In addition, the controller verifies the validity of the predicted value (combined with the true value range of the supervised data pair), removes outliers, and outputs the final density value of the coated electrode sheet of the lithium battery to be tested. The whole process does not require manual intervention, realizing real-time and accurate measurement under environmental fluctuations, and effectively solving the problems of low accuracy and poor real-time performance of existing technologies.

[0033] Optionally, controlling the ultrasonic transmitting module to transmit ultrasonic signals to the medium under test includes the following steps: Step S201: Obtain the first ultrasonic transducer in the ultrasonic transmitting module that meets the target transmission frequency and its corresponding transformer boost drive circuit.

[0034] The controller first acquires the first ultrasonic transducer in the ultrasonic transmitting module that meets the target transmission frequency of 100KHz (adapting to the requirements of lithium battery coated electrode surface density measurement). This transducer is a high-frequency transducer and its specifications match those of the ultrasonic receiving module, ensuring acoustic path coordination. Simultaneously, it acquires the transformer boost drive circuit that is matched with this transducer. This circuit has preset hardware parameters adapted to the 100KHz frequency, which can realize efficient energy coupling of the transducer and lay the hardware foundation for subsequent stable transmission of ultrasonic signals.

[0035] Step S202: Determine the resonant frequency of the transformer boost drive circuit by using the target transmission frequency.

[0036] The controller uses the target transmission frequency of 100KHz as its core basis and combines it with the inherent operating frequency of the first ultrasonic transducer to determine the resonant frequency of the transformer boost drive circuit. By adjusting the circuit parameters, the resonant frequency is made to be completely aligned with the target transmission frequency and the transducer's inherent frequency, so as to avoid excitation signal loss and reduced transducer transmission efficiency caused by frequency mismatch, and ensure that the energy output by the drive circuit can be converted into ultrasonic signals to the greatest extent.

[0037] Step S203: The resonant frequency control transformer boost drive circuit outputs excitation pulses to the first ultrasonic transducer, so that the first ultrasonic transducer transmits ultrasonic signals to the medium under test according to the target transmission frequency; wherein, the medium under test is located in the coaxial acoustic path between the ultrasonic transmitting module and the ultrasonic receiving module.

[0038] The controller uses a determined resonant frequency to send control commands to the transformer boost drive circuit. The drive circuit outputs a 100kHz high-voltage excitation pulse, which acts on the first ultrasonic transducer, causing the transducer to generate mechanical resonance and stably transmit high-frequency ultrasonic signals to the medium under test (such as lithium battery coated electrodes) at the target transmission frequency of 100kHz. The medium under test is pre-placed in the coaxial acoustic path between the ultrasonic transmitting module and the ultrasonic receiving module to ensure that the ultrasonic waves can penetrate the medium under test perpendicularly, reducing signal attenuation or distortion caused by acoustic path offset, and providing a stable original signal source for the subsequent receiving module to collect the transmitted signal.

[0039] Optionally, controlling the ultrasonic receiving module to receive the transmitted signal after the ultrasonic signal has passed through the medium under test includes the following steps: Step S301: Determine the specification matching strategy of the first ultrasonic transducer based on the target transmission frequency and bandwidth value of the ultrasonic signal.

[0040] The controller determines the specification matching strategy for the second ultrasonic transducer (receiver) based on the target transmission frequency of the ultrasonic signal (100kHz) and the bandwidth parameters of the first ultrasonic transducer at the transmitting end. The core requirement of this strategy is that the operating frequency and bandwidth of the receiving transducer are fully compatible with those of the transmitting transducer, ensuring that the ultrasonic transmission signal with a frequency of 100kHz after passing through the coated electrode of the lithium battery under test can be accurately captured. At the same time, it takes into account the signal receiving sensitivity and adapts to the scenario of signal attenuation after medium penetration, laying the foundation for accurate signal reception in the future.

[0041] Step S302: Obtain the second ultrasonic transducer and its corresponding preamplifier in the ultrasonic receiving module that meet the specification matching strategy, and obtain the timing synchronization frequency of the second ultrasonic transducer based on the target transmission frequency.

[0042] The controller selects a second ultrasonic transducer from the ultrasonic receiving module that meets the above-mentioned specification matching strategy, and simultaneously obtains the preamplifier built into the transducer (used to amplify weak electrical signals and suppress noise); based on the target transmission frequency of 100KHz, it synchronously calculates and obtains the timing synchronization frequency of the second ultrasonic transducer, so that the timing of the receiving end is completely synchronized with that of the transmitting end, avoiding spindle waveform distortion or signal loss caused by timing misalignment, and ensuring the timing consistency between the received transmitted signal and the transmitted signal.

[0043] Step S303: Use timing synchronization frequency control to receive the acoustic vibration electrical signal of the ultrasonic signal after it passes through the medium under test, and determine the spindle-shaped target signal corresponding to the acoustic vibration electrical signal based on the specification matching strategy.

[0044] The controller uses a predetermined timing synchronization frequency to send control commands to the second ultrasonic transducer, driving the transducer to receive the acoustic vibration signal after the ultrasonic signal passes through the test medium (lithium battery coated electrode) in real time. Through the piezoelectric effect of the transducer, the acoustic vibration signal is converted into a corresponding weak acoustic vibration electrical signal. At the same time, based on a preset specification matching strategy, signals that meet the 100KHz frequency range and have a spindle-shaped profile are selected from the converted electrical signals and identified as spindle-shaped target signals. Clutter signals caused by environmental interference are eliminated to ensure the effectiveness of the received signal.

[0045] Step S304: After the preamplifier controls the spindle-shaped target signal to amplify the acoustic vibration electrical signal in stages, the transmission signal corresponding to the acoustic vibration electrical signal is obtained.

[0046] The controller, based on the amplitude range of the spindle-shaped target signal, directs the preamplifier to perform graded amplification of the weak acoustic vibration electrical signal. Graded amplification avoids signal overload or under-amplification caused by a single amplification factor. While efficiently amplifying the signal amplitude to meet the recognition requirements of subsequent signal processing modules, the amplifier's filtering characteristics further suppress high-frequency noise and environmental interference. After amplification, a stable and clear transmission signal is output and transmitted to the signal processing module, providing high-quality signal support for subsequent spindle-shaped waveform data extraction and areal density calculation.

[0047] Optionally, the step of obtaining the spindle-shaped waveform data of the transmission signal by controlling the signal processing module to extract features from the transmission signal through the spindle-shaped signal extraction strategy corresponding to the medium under test includes the following steps: Step S401: After the analog electrical signal of the transmission signal is converted into digital signal by the spindle-shaped target signal control signal processing module, the digital signal corresponding to the transmission signal is obtained.

[0048] The controller uses a pre-determined spindle-shaped target signal as a reference and sends a conversion command to the signal processing module. This drives the built-in analog-to-digital converter (ADC) unit to discretize and sample the analog electrical signal of the transmitted signal. During the sampling process, a sampling rate and sampling resolution matching the 100kHz ultrasonic frequency are set, while the number of sampling points is limited to 256 to ensure that the dimensions of the converted digital signal are fully adapted to the [C,T]=[1,256] required by the input layer of the deep learning model. After the conversion is completed, a digital signal corresponding one-to-one with the transmitted signal is obtained. This signal completely retains the timing characteristics of the spindle-shaped waveform, but still contains interference components such as DC components and high-frequency noise.

[0049] Step S402: Determine the feature extraction strategy and noise suppression strategy of the medium under test by the target waveform corresponding to the spindle-shaped target signal.

[0050] The controller determines the feature extraction strategy and noise suppression strategy corresponding to the medium under test based on the typical characteristics of the spindle-shaped target signal (such as the symmetrical envelope shape with large amplitude in the middle and small amplitude at both ends, a core frequency range of 100KHz, and a stable baseline level).

[0051] The feature extraction strategy clarifies the identification rules for DC component features (static offset of signal baseline) and the judgment criteria for high-frequency noise features (signal components with frequencies higher than 100KHz), ensuring accurate location of interference features; The noise suppression strategy employs digital mean subtraction to remove the DC component, uses a moving average filtering algorithm with a window length of 5 to filter out high-frequency noise, and combines amplitude normalization to eliminate the influence of acquisition gain fluctuations, ensuring the consistency of the processed waveform.

[0052] Specifically, the corresponding filtering algorithm is as follows: ;in, Let be the amplitude of the i-th data point after filtering. Let J be the magnitude of the j-th data point before filtering. The sliding window size is used to efficiently filter out random noise in the signal while preserving the core features of the spindle-shaped waveform to the greatest extent. This avoids distortion of the effective signal caused by preprocessing operations and provides stable and reliable basic data for subsequent waveform feature extraction and accurate surface density calculation.

[0053] Step S403: Use the feature extraction strategy to control the signal processing module to extract the DC component features and high-frequency noise features contained in the digital signal.

[0054] The controller uses a predetermined feature extraction strategy to drive the signal processing module to perform feature analysis on the converted digital signal. For the DC component characteristics, the static offset of the signal baseline is determined by calculating the mean of the digital signal over the entire time range. This offset mainly comes from the inherent offset of the transducer and the static noise of the circuit. To target high-frequency noise characteristics, frequency domain analysis (such as Fast Fourier Transform) is used to identify signal components with frequencies higher than 100kHz. These components are mainly caused by environmental electromagnetic interference, mechanical vibration, and thermal noise from the sampling circuit. After extraction, quantitative data of the interference characteristics are generated, providing a clear target for subsequent noise suppression processing.

[0055] Step S404: After the noise suppression strategy is used to control the signal processing module to perform noise suppression processing on the DC component characteristics and high-frequency noise characteristics, the spindle-shaped waveform data of the transmitted signal is obtained.

[0056] The controller utilizes a noise suppression strategy to drive the signal processing module to perform targeted processing on the extracted DC component features and high-frequency noise features, including the following three types of operations: Perform DC component stripping operation: The signal baseline is calibrated to zero level by digital mean subtraction, eliminating the influence of static offset on waveform amplitude measurement and preserving the effective alternating signal of the spindle waveform; Perform high-frequency noise filtering: Use a moving average filtering algorithm with a window length of 5 to smooth the digital signal after removing the DC component, filter out high-frequency interference components, correct waveform glitches, and restore the complete outline of the spindle-shaped waveform. Amplitude normalization is performed: the processed signal is standardized to eliminate the influence of acquisition gain fluctuations; finally, spindle-shaped waveform data with high signal-to-noise ratio that is strongly correlated with the surface density of the medium under test is obtained. This data can be directly input into the deep learning module for subsequent feature extraction and surface density prediction.

[0057] Optionally, acquiring a supervisory data pair constructed from the reference surface density data and reference waveform data of the medium under preset environmental conditions includes the following steps: Step S501: Using the weighing values ​​of the reference measurement media corresponding to different surface density grades, obtain the reference surface density data of the medium to be tested under each surface density grade; wherein, each surface density grade includes multiple reference measurement media.

[0058] The controller performs high-precision weighing calibration based on reference measurement media (i.e., preset lithium battery coated electrode standard sheets) at different areal density levels to obtain reference areal density data for the medium under test at each areal density level. Specifically, for each areal density level, a sample group containing multiple independent reference measurement media (i.e., K standard sheets corresponding to each areal density level) is configured. Each reference measurement medium is accurately weighed using a high-precision weighing device, and the weighing value of each medium is recorded as the true value of the reference areal density. (in Represents the surface density grade number. (Representing the medium index within the same level), ultimately integrating them to obtain a benchmark areal density dataset covering all levels. The core of this step is to provide accurate "ground truth labels" for subsequent supervised learning, ensuring the calibration accuracy of the areal density data and laying the data foundation for model training.

[0059] Step S502: Under different environmental conditions, the ultrasonic surface density measuring device is controlled to repeatedly collect the transmission waveform signal of each reference measuring medium according to the preset collection quantity, so as to obtain the transmission waveform signal set corresponding to each reference measuring medium.

[0060] The controller simulates the real-world measurement environment of an industrial site according to preset environmental conditions. It includes L preset environmental conditions covering temperature and humidity fluctuations, air pressure changes, and airflow interference. Under each condition, the controller controls the transmission, reception, and signal acquisition process of the ultrasonic surface density measuring device according to a preset number of repetitions (P≥10 times). For each reference measurement medium, a specified number of transmission signal acquisitions are performed under each environmental condition, ultimately obtaining the transmission waveform signal set corresponding to each reference measurement medium. (Where l represents the environmental condition number, and p represents the number of repeated data collections.) This step simulates acquisition scenarios under multiple operating conditions to fully reproduce the real interference factors in the field measurement, ensuring that the transmitted waveform data contains diverse characteristics under environmental interference and improving the anti-interference capability of subsequent models.

[0061] Step S503: After calculating the normalized waveform signal of each transmitted waveform signal through the signal extrema in the transmitted waveform signal set, the normalized waveform signal is subjected to sliding filtering calculation using a filter window of preset length to obtain the reference waveform data of the medium under test.

[0062] The controller performs a unified preprocessing operation on the transmission waveform signal set of each reference measurement medium to generate high-quality reference waveform data. The specific steps are as follows: Extreme value normalization: Based on the waveform data in the transmitted waveform signal set, calculate the maximum value of each waveform. and minimum value According to the normalization formula The original transmitted signal is normalized to eliminate waveform amplitude interference caused by gain fluctuations and differences in medium coupling states during the acquisition process, thus achieving standardization and unification of different waveform data; among these... The waveform data after normalization. The waveform data before normalization.

[0063] Moving average filtering: A moving average filter window of preset length (e.g., window length N=5) is used to filter the normalized waveform data. Moving average filtering is performed to efficiently filter out high-frequency sampling noise and glitches caused by environmental disturbances, while preserving the core amplitude and envelope of the spindle waveform to the greatest extent possible, thus avoiding distortion of the effective signal. The final processed data is a reference waveform of the medium under test with high signal-to-noise ratio and clear features, providing high-quality waveform support for subsequent binding with surface density data.

[0064] Step S504: The control areal density calculation module binds the reference waveform data and the reference areal density data according to the areal density level and environmental conditions to obtain the supervision data pair.

[0065] The controller drives the areal density calculation module to process the preprocessed reference waveform data according to two core dimensions: areal density level and environmental conditions. With the corresponding reference surface density data Precise binding is performed. Specifically, each reference waveform data point under the same surface density level and environmental conditions is correlated one-to-one with the true value of the reference surface density of the reference measurement medium, constructing a supervisory data pair containing three-dimensional information of "reference surface density - reference waveform - environmental conditions". This data comprehensively covers waveform characteristics under different surface densities and environmental conditions, and can be directly used for training deep learning models. It supports the model in learning the end-to-end nonlinear mapping relationship between wave and surface density under environmental interference, ultimately achieving accurate measurement of ultrasonic surface density under complex conditions.

[0066] Optionally, the step of binding the reference waveform data and the reference surface density data according to the surface density level and environmental conditions in the control surface density calculation module to obtain the supervision data pair includes the following steps: Step S601: Obtain the initialized neural network model in the surface density calculation module; wherein, the neural network model includes a feature extraction layer and a fully connected layer; the feature extraction layer includes multiple sequentially connected deep convolutional blocks; each deep convolutional block includes a one-dimensional deep convolutional layer, a one-dimensional pointwise convolutional layer, a batch normalization layer, a linear rectified layer, and a max pooling layer.

[0067] The controller first acquires the neural network model with pre-initialized parameters from the areal density calculation module. This model is specifically designed for ultrasonic areal density regression prediction tasks, and its overall architecture is as follows: Figure 2 As shown, it includes two core parts: a feature extraction layer and a fully connected layer. The feature extraction layer consists of multiple one-dimensional depth convolutional blocks connected in series. The internal structure of each depth convolutional block is "one-dimensional depth convolutional layer - one-dimensional pointwise convolutional layer - batch normalization layer - linear rectified layer (ReLU) - max pooling layer". The model input dimension is adapted to the [1,256] specification of the preprocessed single-channel spindle waveform data. The output is a single node used to output the surface density prediction value, ensuring that the initial state of the model can directly support the subsequent training process.

[0068] Step S602: Using the surface density level and environmental conditions, the reference waveform data and reference surface density data are divided according to a preset ratio to obtain the training set, validation set and test set corresponding to the neural network model.

[0069] The controller uses both areal density level and environmental conditions as the basis for division. It performs hierarchical random division of the preprocessed reference waveform data and the corresponding reference areal density true values ​​according to a preset ratio (e.g., training set: validation set: test set = 7:2:1). The division process strictly follows the core principle: ensuring that multiple sets of reference waveform data corresponding to the same areal density level are evenly distributed in the three sets of training set, validation set and test set, while ensuring that the environmental conditions such as temperature, humidity and air pressure covered by the three sets are completely consistent, avoiding model overfitting due to uneven data distribution, and finally obtaining three types of datasets suitable for training, validation and testing of neural network models.

[0070] Step S603: After the control surface density calculation module inputs the training set into the neural network model, it sequentially obtains the feature maps output by the depth convolution blocks.

[0071] The controller controls the surface density calculation module to input the divided training set data into the initialized neural network model, driving multiple deep convolutional blocks in the feature extraction layer to perform feature extraction operations layer by layer: when the baseline waveform data in the training set is processed by the first deep convolutional block, the one-dimensional deep convolutional layer extracts the waveform temporal local features, the one-dimensional pointwise convolutional layer completes the channel dimension feature fusion, and then the batch normalization layer eliminates environmental amplitude drift, the ReLU layer introduces nonlinear mapping, and the max pooling layer downsamples and reduces dimensionality; subsequent deep convolutional blocks deepen feature extraction layer by layer based on this, and finally obtain the stage feature map output by each deep convolutional block in sequence, as well as the final high-dimensional feature map output by the feature extraction layer (size [128,32]).

[0072] Step S604: Obtain the surface density prediction value corresponding to the baseline waveform data in the feature map through the fully connected layer, and update the network weight parameters corresponding to the one-dimensional deep convolutional layer, one-dimensional pointwise convolutional layer, batch normalization layer, linear rectified layer, max pooling layer and fully connected layer in each deep convolutional block according to the mean square error between the surface density prediction value and the baseline surface density data in the validation set.

[0073] The fully connected layer of the controller-driven model processes the high-dimensional feature map output by the feature extraction layer. First, the feature map [128,32] is stretched into a one-dimensional feature vector through the Flatten layer, and then global feature fusion is completed through multiple fully connected layers to output the areal density prediction value corresponding to the baseline waveform data in the training set. Then, the mean square error (MSE) between the prediction value and the baseline areal density ground truth value in the validation set is calculated. With the goal of reducing the mean square error, all trainable parameters of the neural network model are iteratively updated through the backpropagation algorithm, including the weights and biases of the one-dimensional deep convolutional layer, one-dimensional pointwise convolutional layer, and batch normalization layer in each deep convolutional block, as well as the network parameters of the fully connected layer, to ensure that the model gradually learns the nonlinear mapping relationship between the baseline waveform data and the areal density ground truth value.

[0074] Step S605: When the mean square error meets the preset threshold condition, stop the update process of the depth convolution block, and bind the baseline waveform data and the surface density prediction value under the current network weight parameters to obtain the supervision data pair.

[0075] When the mean squared error of the validation set decreases to a preset threshold, or when the preset number of training epochs meets the convergence condition, the controller stops the iterative update process of the model parameters and locks the current network weight parameters. Then, the areal density prediction value output by the trained model is precisely associated and bound with the corresponding baseline waveform data and baseline areal density ground truth value, and finally a supervision data pair containing three-dimensional information of "baseline waveform data - baseline areal density ground truth value - areal density prediction value" is formed, which provides data support for subsequent model validation, testing and application in real-world scenarios.

[0076] Optionally, using supervised data, the control surface density calculation module generates a convolutional feature map of the surface density data corresponding to the spindle-shaped waveform data, including the following steps: Step S701: Use supervised data to obtain the current network weight parameters of the depth convolutional block, and obtain the convolutional kernel and stride value corresponding to each depth convolutional block.

[0077] Based on the constructed supervised data pairs, the controller retrieves the current network weight parameters of each deep convolutional block in the trained and converged neural network model. These parameters are the optimal parameters for feature extraction that are strongly correlated with surface density, learned by the model through supervised data. Simultaneously, it extracts the core computational configuration corresponding to each deep convolutional block, including the kernel size of the one-dimensional deep convolution. With step size value The number of output channels for one-dimensional pointwise convolution is also included to ensure that the operation rules of each convolutional block are completely consistent with the feature extraction logic of the supervised data during training, thus providing parameter support for the subsequent accurate generation of convolutional feature maps.

[0078] Step S702: After the control surface density calculation module inputs the spindle-shaped waveform data into the neural network model under the current network weight parameters, it obtains the local temporal feature map output by the first depth convolution block through the convolution kernel and stride value.

[0079] The controller controls the surface density calculation module, which inputs the preprocessed spindle-shaped waveform data (a single-channel one-dimensional tensor with dimensions [1,256]) into the neural network model loaded with the current network weight parameters; it drives the first deep convolution block to perform feature extraction operations on the input waveform data according to the obtained convolution kernel and stride values: it captures the local temporal details of the waveform through one-dimensional deep convolution (such as the amplitude changes in the neighborhood of the sampling point, the steepness of the rising / falling edge of the waveform), and then completes the channel dimension feature fusion through pointwise convolution, batch normalization to eliminate environmental amplitude drift, ReLU layer to introduce nonlinear mapping and max pooling layer downsampling dimensionality reduction, and finally outputs a local temporal feature map, which contains the basic temporal information of the spindle-shaped waveform, laying the foundation for subsequent high-order feature extraction.

[0080] Step S703: After inputting the local temporal feature map into the second deep convolutional block, obtain the spindle wave envelope morphological feature map output by the second deep convolutional block through the convolution kernel and stride value.

[0081] The controller inputs the local temporal feature map output from the first deep convolutional block into the second deep convolutional block. The second deep convolutional block, based on its corresponding convolutional kernel and stride value, further extracts the local temporal features, focusing on capturing the spindle wave envelope morphology features directly related to the surface density (such as the peak height, attenuation slope, pulse width, and other key indicators of the envelope). The recognition of the envelope features is enhanced through deep convolution, and multi-channel features are fused by point-by-point convolution. After batch normalization, activation, and pooling, the spindle wave envelope morphology feature map is output. This feature map has achieved the initial transformation from basic temporal features to surface density-related features.

[0082] Step S704: After inputting the spindle wave envelope morphology feature map into the third depth convolution block, obtain the convolution feature map output by the third depth convolution block through the convolution kernel and stride value.

[0083] The controller inputs the spindle wave envelope morphology feature map generated by the second deep convolution block into the third deep convolution block. The third deep convolution block, relying on its own convolution kernel and stride value, further abstracts and integrates the envelope morphology features, extracting high-order abstract features that are strongly correlated with surface density (such as the quantization features of the envelope decay trend and the nonlinear correlation features between waveform features and surface density). After a complete process of deep convolution, pointwise convolution, batch normalization, activation and pooling, the final output is a convolution feature map with a size of [128,32]. This feature map integrates multi-scale abstract features that are highly correlated with surface density in the spindle waveform and can be directly used for subsequent surface density regression prediction calculations.

[0084] Optionally, the step of using the channel dimension value control module of the convolutional feature map to perform regression prediction calculation on the spindle-shaped waveform data to obtain the surface density value of the medium under test includes the following steps: Step S801: Generate the dimension transformation strategy corresponding to the fully connected layer based on the sequence length value and channel dimension value of the convolutional feature map.

[0085] The controller first extracts the core parameters of the convolutional feature map output in step S704, namely the sequence length (32) and channel dimension (128). Combined with the input requirements of the fully connected layer, it generates the corresponding dimension transformation strategy. The core purpose of this strategy is to convert the high-dimensional convolutional feature map into a one-dimensional feature vector that is adapted to the operation of the fully connected layer. The transformation rule is as follows: the convolutional feature map of size [128,32] is linearly stretched through the Flatten layer, and the high-dimensional data of 128 channels and 32 temporal feature points in each channel are integrated into a single-dimensional feature vector, ensuring that the length of the vector after transformation is 128×32=4096. At the same time, the areal density correlation features corresponding to the dimension values ​​of each channel are completely preserved, laying the foundation for the feature combination and regression prediction of the subsequent fully connected layer.

[0086] Step S802: Use the dimension transformation strategy to control the surface density calculation module to obtain the one-dimensional feature vector corresponding to the spindle-shaped waveform data.

[0087] Following the generated dimension transformation strategy, the controller initiates the Flatten layer in the areal density calculation module to perform a dimension transformation operation on the convolutional feature map: the high-dimensional feature map [128,32] is linearly stretched channel by channel and time sequence to eliminate the dimensional differences between the channels and the sequence, generating a one-dimensional feature vector (dimensional [4096,1]) that corresponds one-to-one with the spindle-shaped waveform data. During the transformation process, the abstract features corresponding to the dimensional values ​​of each channel (such as amplitude attenuation features and envelope morphology features related to areal density) are strictly preserved to avoid feature loss and ensure that the one-dimensional feature vector can fully represent the relationship between the spindle-shaped waveform and the areal density, while also adapting to the input dimension requirements of the fully connected layer.

[0088] Step S803: After performing regression prediction calculation on the one-dimensional feature vector using the fully connected layer control surface density calculation module under the current network weight parameters, the feature combination result of the spindle-shaped waveform data is obtained. The surface density value of the medium to be tested is obtained by calculating the feature combination result through a preset linear activation function.

[0089] The controller calls the supervised data to train the current network weight parameters and controls the fully connected layer of the areal density calculation module to perform deep feature combination and regression prediction calculation on the one-dimensional feature vector: First, through the weight matrix operation of multiple fully connected layers, the features of each dimension in the one-dimensional feature vector are weighted and fused to strengthen the feature weights that are strongly correlated with areal density and weaken the interference of redundant features, so as to obtain the feature combination result corresponding to the spindle-shaped waveform data; then, the feature combination result is linearly mapped through a preset linear activation function (f(x)=x). Since areal density is a continuous variable, the linear activation function can directly output continuous real values, that is, the areal density prediction value of the medium to be measured; finally, the controller performs preliminary verification of the prediction value to ensure that it is within a reasonable measurement range, and finally outputs the accurate areal density value of the medium to be measured, completing the end-to-end regression prediction from convolutional features to areal density value.

[0090] The flowchart of the ultrasonic areal density measurement and control method based on relevant models in the field of deep learning is shown below. Figure 3 As shown, the specific steps include: Data Acquisition and Preprocessing: Ultrasonic waves are emitted from the transmitter of an ultrasonic surface density measuring device to multiple standard media samples with known surface densities. Simultaneously, the receiver acquires the raw spindle-shaped waveform data generated after the ultrasonic waves penetrate the medium. To recreate realistic interference scenarios in industrial settings, various common environmental conditions are actively introduced and simulated during the acquisition process, including temperature and humidity fluctuations at different gradients and airflow interference of varying intensities. For the same standard media sample with a fixed surface density, multiple sets of spindle-shaped data with significantly different waveforms generated under different environmental disturbances are collected. This constructs a complete raw dataset containing a "many-to-one" mapping relationship, perfectly matching the actual conditions of the field measurement. After the raw data acquisition is complete, a unified preprocessing operation is performed on all spindle-shaped waveform data. The core steps include noise reduction and numerical normalization. Through the preprocessing process, various noise interferences in the raw data are effectively eliminated, and the data is standardized, ultimately generating standardized input data.

[0091] End-to-end model construction: A deep learning neural network model for the regression task is constructed, consisting of an input layer, a CNN feature extraction layer, and an output layer. The size of the input layer matches the dimension of the sampling points in the preprocessed spindle-shaped waveform data, and the output layer is a single node corresponding to the predicted areal density value. The neural network model in the CNN feature extraction layer is designed to automatically learn the complex nonlinear mapping from the input waveform data to the output areal density value.

[0092] Model Training: The preprocessed dataset is divided into training, validation, and test sets. The training set is used to train the deep learning neural network. The core goal of the training is to minimize the loss function between the network's predicted areal density values ​​and the actual areal density values ​​of the samples. During training, the weight parameters in the network are continuously adjusted using the backpropagation algorithm. Since the training data includes multiple waveforms corresponding to the same areal density, the network is forced to learn the truly areal density-related, environmentally unaffected, invariant features of the waveforms. In this way, it acquires the ability to handle many-to-one mapping problems.

[0093] Model Deployment and Application: The trained deep learning neural network model is deployed into the measurement system. During actual measurement, spindle-shaped waveform data of the medium under test is acquired in real time under the current environment. After data preprocessing, the data is input into the trained model, which directly outputs the predicted areal density value of the medium under test.

[0094] As can be seen from the ultrasonic surface density measurement and control method mentioned in the above embodiments, this method uses the monitoring data pairs constructed by the reference surface density data and the reference waveform data under various environmental conditions to achieve end-to-end accurate mapping between surface density data and waveform data under various environments, which greatly improves the measurement accuracy. In addition, this method can obtain the surface density value corresponding to the spindle-shaped waveform data in real time using the monitoring data pairs, which improves the measurement real-time performance of the ultrasonic surface density measuring equipment.

[0095] Corresponding to the ultrasonic surface density measurement and control method provided in the foregoing embodiments, this invention provides an ultrasonic surface density measurement and control system. This system is applied in the controller of an ultrasonic surface density measuring device, which further includes an ultrasonic transmitting module, an ultrasonic receiving module, a signal processing module, and a surface density calculation module. For example... Figure 4 As shown, the system includes: Ultrasonic signal control module 100: used to control the ultrasonic transmitting module to transmit ultrasonic signals to the medium under test, and to control the ultrasonic receiving module to receive the transmitted signal after the ultrasonic signal passes through the medium under test. Waveform data acquisition module 200: Used to control the signal processing module to extract features from the transmission signal through the spindle signal extraction strategy corresponding to the medium under test, and obtain the spindle-shaped waveform data of the transmission signal. Convolutional feature map generation module 300: used to obtain a supervision data pair constructed from the reference surface density data and reference waveform data of the medium under preset environmental conditions, and to use the supervision data to generate a convolutional feature map of the surface density data corresponding to the spindle-shaped waveform data for the control surface density calculation module. Areal density value measurement and calculation module 400: Used to control the channel dimension value of the areal density calculation module based on the convolutional feature map to perform regression prediction calculation on the spindle-shaped waveform data to obtain the areal density value of the medium to be measured.

[0096] As can be seen from the ultrasonic surface density measurement and control system mentioned in the above embodiments, the system utilizes the monitoring data pairs constructed by the reference surface density data and reference waveform data under various environmental conditions to achieve end-to-end accurate mapping between surface density data and waveform data under various environments, which greatly improves the measurement accuracy. In addition, the system can obtain the surface density value corresponding to the spindle-shaped waveform data in real time using the monitoring data pairs, which improves the measurement real-time performance of the ultrasonic surface density measuring equipment.

[0097] The ultrasonic surface density measurement and control device provided in this embodiment of the invention has the same implementation principle and technical effect as the aforementioned ultrasonic surface density measurement and control method embodiment. For the sake of brevity, any parts not mentioned in the device embodiment can be referred to the corresponding content in the aforementioned ultrasonic surface density measurement and control method embodiment.

[0098] This embodiment also provides an ultrasonic surface density measuring device, such as... Figure 5 As shown, the ultrasonic surface density measuring device includes: an ultrasonic transmitting module 510, an ultrasonic receiving module 520, a signal processing module 530, a surface density calculation module 540, and a controller 550; the controller 550 is connected to the ultrasonic transmitting module 510, the ultrasonic receiving module 520, the signal processing module 530, and the surface density calculation module 540 respectively.

[0099] The ultrasonic transmitting module 510 includes a high-frequency ultrasonic transducer and a transformer boost drive circuit. Under the command of the controller, the boost drive circuit provides a high-voltage excitation pulse with resonant matching to the transducer, driving the transducer to generate an ultrasonic signal of a specific frequency (e.g., 100 kHz), which is then directed perpendicularly towards the medium under test (e.g., a lithium battery coated electrode). The core function of the ultrasonic transmitting module 510 is to generate a stable, energy-matched ultrasonic transmission signal, providing a foundation for subsequent penetration and measurement.

[0100] Test medium: When an ultrasonic signal penetrates the test medium, physical effects such as attenuation, dispersion, and phase shift will occur. The signal shape (such as a spindle-shaped waveform) will carry characteristic information that is strongly related to the surface density of the medium, and will also be superimposed with the influence of environmental interference factors (temperature, humidity, coupling state, etc.).

[0101] The ultrasonic receiving module 520 includes an ultrasonic transducer and a preamplifier matched with the transmitting end; under the timing synchronization control of the controller, it receives the ultrasonic transmission signal after penetrating the medium under test; then, it converts the sound wave vibration into a weak electrical signal through the piezoelectric effect; then, the weak electrical signal is amplified in stages by the preamplifier, while suppressing noise interference, and outputting a stable electrical signal to the signal processing module 530.

[0102] The signal processing module 530 includes an analog-to-digital converter (ADC) unit and a filtering preprocessing unit (including operations such as DC removal, high-frequency noise removal, normalization, and moving average filtering). This signal processing module 530 can perform analog-to-digital conversion on the analog electrical signal output from the receiving module to generate a discrete digital signal. During subsequent preprocessing, by removing the DC component, filtering high-frequency sampling noise, normalizing the amplitude, and performing moving average filtering, it ultimately generates standardized spindle-shaped digital waveform data (dimensions [1, 256]), providing standardized input for the deep learning model.

[0103] The surface density calculation module 540 includes a deep learning module (containing a data storage unit and a model operation unit), corresponding to the aforementioned one-dimensional depthwise separable convolutional and fully connected layer model.

[0104] The data storage unit is used to store trained deep learning models, multi-condition supervision data pairs, waveform data, and surface density ground truth. The model operation unit is used to load the preprocessed digital waveform, run the model to complete feature extraction (3-layer depth separable convolution to generate [128,32] convolutional feature map), dimensionality reduction and fully connected regression prediction, and finally output the predicted surface density value of the medium under test in the form of waveform display and surface density display.

[0105] The controller 550 serves as the control center of the device, enabling the coordinated operation of various modules through control signals: controlling the driving frequency and resonance matching of the transmitting module; controlling the timing synchronization of the receiving module to avoid waveform misalignment; scheduling the preprocessing process of the signal processing module; and controlling the model loading and operation scheduling of the surface density calculation module to ensure timing synchronization and parameter matching throughout the entire process.

[0106] Specifically, the controller 550, such as Figure 6 As shown, it includes a processor 101 and a memory 102; wherein, the memory 102 is used to store one or more computer instructions, which are executed by the processor to implement the steps of the ultrasonic surface density measurement and control method described above.

[0107] Figure 6 The controller 550 shown also includes a bus 103 and a communication interface 104. The processor 101, the communication interface 104 and the memory 102 are connected via the bus 103.

[0108] The memory 102 may include high-speed random access memory (RAM) and may also include non-volatile memory, such as at least one disk storage device. The bus 103 may be an ISA bus, PCI bus, or EISA bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 6 The symbol is represented by a single double-headed arrow, but this does not mean that there is only one bus or one type of bus.

[0109] The communication interface 104 is used to connect to at least one user terminal and other network units through a network interface, and to send encapsulated IPv4 packets or IPv4 packets to the user terminal through the network interface.

[0110] Processor 101 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of processor 101 or by instructions in software form. The processor 101 can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it can also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this disclosure. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this disclosure can be directly manifested as execution by a hardware decoding processor, or execution by a combination of hardware and software modules in the decoding processor. The software module can reside in a mature storage medium in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory 102. The processor 101 reads the information in memory 102 and, in conjunction with its hardware, completes the steps of the method described in the foregoing embodiments.

[0111] This invention also provides a storage medium storing a computer program, which, when executed by a processor, performs the steps of the ultrasonic surface density measurement and control method described in the foregoing embodiments.

[0112] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, devices, and methods can be implemented in other ways. The system embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the coupling or direct coupling or communication connection shown or discussed may be through some communication interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0113] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0114] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0115] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0116] Finally, it should be noted that the above-described embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for controlling ultrasonic surface density measurement, characterized in that, The method is applied to the controller of an ultrasonic surface density measuring device, which further includes an ultrasonic transmitting module, an ultrasonic receiving module, a signal processing module, and a surface density calculation module. The method includes: After controlling the ultrasonic transmitting module to transmit ultrasonic signals to the medium under test, the ultrasonic receiving module is controlled to receive the transmitted signal of the ultrasonic signal after passing through the medium under test. By controlling the signal processing module to extract the DC component features and suppress the high-frequency noise features in the transmitted signal through the spindle signal extraction strategy corresponding to the medium under test, the spindle waveform data of the transmitted signal is obtained. A supervisory data pair is constructed by acquiring the reference surface density data and reference waveform data of the medium under preset environmental conditions. The supervisory data pair is used to control the surface density calculation module to generate a spindle wave envelope morphology feature map of the spindle waveform data, and the spindle wave envelope morphology feature map is used to generate a convolution feature map of the surface density data corresponding to the spindle waveform data. Based on the channel dimension value of the convolutional feature map, the areal density calculation module is controlled to perform regression prediction calculation on the spindle-shaped waveform data to obtain the areal density value of the medium under test. Acquire a supervisory data pair constructed from the reference surface density data and reference waveform data of the medium under preset environmental conditions, including: By using the weighing values ​​of reference measurement media corresponding to different areal density grades, reference areal density data of the medium to be tested under each areal density grade are obtained; wherein, each areal density grade includes multiple reference measurement media; Under different environmental conditions, the ultrasonic surface density measuring device repeatedly collects the transmission waveform signal of each reference measuring medium according to the preset collection quantity, thereby obtaining the transmission waveform signal set corresponding to each reference measuring medium; After calculating the normalized waveform signal of each transmitted waveform signal by the signal extrema in the transmitted waveform signal set, the normalized waveform signal is subjected to sliding filtering calculation using a filter window of preset length to obtain the reference waveform data of the medium under test. The control module binds the reference waveform data and the reference surface density data according to the surface density level and the environmental conditions to obtain the supervision data pair.

2. The ultrasonic surface density measurement and control method according to claim 1, characterized in that, Controlling the ultrasonic transmitting module to transmit ultrasonic signals to the medium under test includes: Obtain the first ultrasonic transducer in the ultrasonic transmitting module that meets the target transmission frequency and its corresponding transformer boost drive circuit. The resonant frequency of the transformer boost drive circuit is determined by the target transmission frequency; The transformer boost drive circuit is controlled by the resonant frequency to output an excitation pulse to the first ultrasonic transducer, so that the first ultrasonic transducer emits an ultrasonic signal to the medium under test according to the target emission frequency; wherein the medium under test is located in the coaxial acoustic path between the ultrasonic transmitting module and the ultrasonic receiving module.

3. The ultrasonic surface density measurement and control method according to claim 2, characterized in that, Controlling the ultrasonic receiving module to receive the transmitted signal of the ultrasonic signal after it has passed through the medium under test includes: The specification matching strategy of the first ultrasonic transducer is determined based on the target transmission frequency and bandwidth value of the ultrasonic signal; Obtain the second ultrasonic transducer and its corresponding preamplifier in the ultrasonic receiving module that meet the specification matching strategy, and obtain the timing synchronization frequency of the second ultrasonic transducer based on the target transmission frequency; The timing synchronization frequency is used to control the second ultrasonic transducer to receive the acoustic vibration electrical signal after the ultrasonic signal passes through the medium under test, and the spindle-shaped target signal corresponding to the acoustic vibration electrical signal is determined based on the specification matching strategy. After the preamplifier controls the spindle-shaped target signal to amplify the acoustic vibration electrical signal in stages, the transmission signal corresponding to the acoustic vibration electrical signal is obtained.

4. The ultrasonic surface density measurement and control method according to claim 3, characterized in that, The step of controlling the signal processing module to extract features from the transmission signal using the spindle-shaped signal extraction strategy corresponding to the medium under test, and obtaining the spindle-shaped waveform data of the transmission signal, includes: The signal processing module is controlled by the spindle-shaped target signal to perform analog-to-digital conversion on the analog electrical signal of the transmitted signal, thereby obtaining the digital signal corresponding to the transmitted signal. The feature extraction strategy and noise suppression strategy of the medium under test are determined by the target waveform corresponding to the spindle-shaped target signal. The feature extraction strategy is used to control the signal processing module to extract the DC component features and high-frequency noise features contained in the digital signal; After the signal processing module performs noise suppression processing on the DC component characteristics and the high-frequency noise characteristics using the noise suppression strategy, the spindle-shaped waveform data of the transmitted signal is obtained.

5. The ultrasonic surface density measurement and control method according to claim 1, characterized in that, The step of controlling the areal density calculation module to bind the reference waveform data and the reference areal density data according to the areal density level and the environmental conditions to obtain the supervision data pair includes: Obtain the initialized neural network model in the surface density calculation module; wherein, the neural network model includes a feature extraction layer and a fully connected layer; the feature extraction layer includes multiple sequentially connected deep convolutional blocks; each deep convolutional block includes a one-dimensional deep convolutional layer, a one-dimensional pointwise convolutional layer, a batch normalization layer, a linear rectified layer, and a max pooling layer; Using the areal density level and the environmental conditions, the reference waveform data and the reference areal density data are divided according to a preset ratio to obtain the training set, validation set and test set corresponding to the neural network model; After the areal density calculation module inputs the training set into the neural network model, it sequentially obtains the feature maps output by the depth convolution block. The fully connected layer obtains the surface density prediction value corresponding to the baseline waveform data in the feature map, and updates the network weight parameters corresponding to the one-dimensional deep convolutional layer, the one-dimensional pointwise convolutional layer, the batch normalization layer, the linear rectified layer, the max pooling layer and the fully connected layer in each deep convolutional block according to the mean square error between the surface density prediction value and the baseline surface density data in the validation set. When the mean square error meets the preset threshold condition, the update process of the depth convolution block is stopped. After binding the baseline waveform data and the surface density prediction value under the current network weight parameters, the supervision data pair is obtained.

6. The ultrasonic surface density measurement and control method according to claim 5, characterized in that, Using the supervised data to control the areal density calculation module to generate a convolutional feature map of the areal density data corresponding to the spindle-shaped waveform data, including: The current network weight parameters of the depth convolutional block are obtained using the supervision data, and the convolutional kernel and stride values ​​corresponding to each depth convolutional block are obtained. After the areal density calculation module inputs the spindle-shaped waveform data into the neural network model under the current network weight parameters, it obtains the local temporal feature map output by the first depth convolution block through the convolution kernel and the stride value. After the local temporal feature map is input into the second deep convolutional block, the spindle wave envelope morphology feature map output by the second deep convolutional block is obtained through the convolutional kernel and the stride value. After the spindle wave envelope morphology feature map is input into the third depth convolution block, the convolution feature map output by the third depth convolution block is obtained through the convolution kernel and the stride value.

7. The ultrasonic surface density measurement and control method according to claim 6, characterized in that, The step of controlling the areal density calculation module to perform regression prediction calculation on the spindle-shaped waveform data based on the channel dimension value of the convolutional feature map to obtain the areal density value of the medium under test includes: The dimension transformation strategy corresponding to the fully connected layer is generated based on the sequence length value and channel dimension value of the convolutional feature map; The dimension transformation strategy is used to control the surface density calculation module to obtain the one-dimensional feature vector corresponding to the spindle-shaped waveform data. Using the fully connected layer under the current network weight parameters, the areal density calculation module performs regression prediction calculation on the one-dimensional feature vector to obtain the feature combination result of the spindle-shaped waveform data. The areal density value of the medium under test is obtained by calculating the feature combination result through a preset linear activation function.

8. An ultrasonic surface density measurement and control system, characterized in that, The system is applied in the controller of an ultrasonic surface density measuring device, which also includes an ultrasonic transmitting module, an ultrasonic receiving module, a signal processing module, and a surface density calculation module. The system includes: Ultrasonic signal control module: used to control the ultrasonic transmitting module to transmit ultrasonic signals to the medium under test, and to control the ultrasonic receiving module to receive the transmitted signal of the ultrasonic signal after it passes through the medium under test. Waveform data acquisition module: used to control the signal processing module to perform feature extraction and noise suppression processing on the DC component features and high-frequency noise features in the transmission signal according to the spindle signal extraction strategy corresponding to the medium under test, so as to obtain the spindle-shaped waveform data of the transmission signal; Convolutional feature map generation module: used to obtain a supervision data pair constructed from the reference surface density data and reference waveform data of the medium under preset environmental conditions, use the supervision data pair to control the surface density calculation module to generate the spindle wave envelope morphology feature map of the spindle waveform data, and generate the convolutional feature map of the surface density data corresponding to the spindle waveform data through the spindle wave envelope morphology feature map. Areal density value measurement and calculation module: used to control the areal density calculation module to perform regression prediction calculation on the spindle-shaped waveform data based on the channel dimension value of the convolution feature map, so as to obtain the areal density value of the medium under test; In the process of acquiring the supervisory data pair constructed from the reference areal density data and reference waveform data of the medium under test under preset environmental conditions, the convolutional feature map generation module is further configured to: acquire the reference areal density data of the medium under test for each areal density level using the weighing values ​​of the reference measurement media corresponding to different areal density levels; wherein, each areal density level includes multiple reference measurement media; under different environmental conditions, after controlling the ultrasonic areal density measuring device to repeatedly acquire the transmission waveform signal of each reference measurement media according to a preset acquisition quantity, a transmission waveform signal set corresponding to each reference measurement media is obtained; after calculating the normalized waveform signal of each transmission waveform signal through the signal extreme values ​​in the transmission waveform signal set, a sliding filter calculation is performed on the normalized waveform signal using a filter window of preset length to obtain the reference waveform data of the medium under test; and after controlling the areal density calculation module to bind the reference waveform data and the reference areal density data according to the areal density level and the environmental conditions, the supervisory data pair is obtained.

9. An ultrasonic surface density measuring device, characterized in that, The ultrasonic areal density measuring device includes: an ultrasonic transmitting module, an ultrasonic receiving module, a signal processing module, an areal density calculation module, and a controller; the controller is connected to the ultrasonic transmitting module, the ultrasonic receiving module, the signal processing module, and the areal density calculation module respectively. The controller includes a processor and a memory, the memory storing computer-executable instructions that can be executed by the processor, and the processor executing the computer-executable instructions to implement the ultrasonic areal density measurement control method mentioned in any one of claims 1 to 7.

Images