Multi-sensor data fusion method and system for monitoring copper busbar processing

By deploying multiple sensors to collect data during copper busbar processing and using a CNN-LSTM model for analysis, the problem of multi-source sensor information fusion in copper busbar processing was solved, enabling online prediction of material mechanical properties and improving the monitoring accuracy and production efficiency of copper busbar processing.

CN122858593BActive Publication Date: 2026-11-17ZHANGJIAGANG NANYUAN PHOTOELECTRIC TECH CO LTD
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202611364747.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2026-09-04
Publication Date
2026-11-17
Estimated Expiration
2046-09-04

AI Technical Summary

Technical Problem

Existing copper busbar processing monitoring methods are difficult to integrate multi-source sensor information, making it impossible to predict material mechanical properties online. Quality monitoring is lagging and inaccurate, resulting in low production control efficiency.

Method used

By deploying multiple sensors to collect multi-source physical field data during copper busbar processing, clock synchronization and signal noise reduction are performed. A data fusion prediction model based on CNN-LSTM architecture is used to capture the temporal dependence of force-thermal-acoustic dynamic evolution and output density, tensile strength and elongation indicators to achieve copper busbar processing quality analysis.

Benefits of technology

This technology enables the fusion of multi-sensor data and the mining of temporal features in copper busbar processing, improving the accuracy of online prediction of material mechanical properties and enhancing the accuracy and efficiency of real-time production monitoring.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122858593B_ABST
    Figure CN122858593B_ABST
Patent Text Reader

Abstract

This invention discloses a method and system for monitoring copper busbar processing based on multi-sensor data fusion, belonging to the field of copper busbar processing monitoring technology. The method includes: real-time acquisition of multi-source physical field data of the copper busbar; extraction of multi-dimensional feature vectors representing the temporal characteristics of drawing force, cross-sectional size change characteristics, high-frequency acoustic characteristics of the deformation zone, and temperature evolution characteristics, and calculation of the actual cross-sectional area reduction rate of the copper busbar; capture of temporal dependencies, outputting density, tensile strength, and elongation indices representing the current grain compactness of the copper busbar; performing copper busbar processing quality analysis, and using the quality analysis results as the processing monitoring results. This invention solves the technical problems of existing copper busbar processing monitoring methods, such as difficulty in integrating multi-source sensor information, inability to predict material mechanical properties online, and lagging and inaccurate quality monitoring. It achieves multi-sensor data fusion and temporal feature mining in copper busbar processing, enabling online prediction of material mechanical properties and improving the technical effect of real-time monitoring accuracy in copper busbar processing.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of copper busbar processing monitoring technology, specifically to a copper busbar processing monitoring method and system based on multi-sensor data fusion. Background Technology

[0002] In the continuous production process of copper busbar extrusion-drawing, most existing quality monitoring methods adopt offline sampling inspection mode, which can only obtain mechanical indicators such as density, tensile strength and elongation through post-processing sample testing. The detection has a strong lag and it is difficult to achieve online monitoring of the workpiece throughout the entire process. At the same time, traditional monitoring systems analyze data from various sensors independently, lacking the means to fuse and process multi-source physical field sensor data and mine time-series correlations. They cannot establish the mapping relationship between multiple parameters such as drawing force, temperature, acoustic emission, and cross-sectional dimensions and the internal structure and mechanical properties of the copper busbar. It is difficult to predict processing defects in real time, and it is impossible to achieve precise positioning and closed-loop process control for defective sections. This can easily lead to the generation of a large number of unqualified products and low production control efficiency.

[0003] Existing copper busbar processing monitoring methods suffer from technical problems such as difficulty in integrating multi-source sensor information, inability to predict material mechanical properties online, and lagging quality monitoring with insufficient accuracy. Summary of the Invention

[0004] This application provides a method and system for monitoring copper busbar processing by fusing multi-sensor data, which is used to address the technical problems of existing copper busbar processing monitoring methods, such as difficulty in fusing multi-source sensor information, inability to predict material mechanical properties online, and lagging and inaccurate quality monitoring.

[0005] In view of the above problems, this application provides a method and system for monitoring copper busbar processing by multi-sensor data fusion.

[0006] The first aspect of this application provides a method for monitoring copper busbar processing by fusing multi-sensor data, the method comprising: An online optical dimension measuring instrument is deployed at the discharge end of the extrusion equipment, and a drawing force sensor, an acoustic emission sensor attached to the side wall of the drawing die, and an infrared temperature measuring device pointing towards the drawing deformation zone are configured at the drawing equipment to collect multi-source physical field data of the copper busbar in real time during the extrusion and drawing process. The multi-source physical field data is clock-synchronized and denoised to extract a multi-dimensional feature vector containing the temporal characteristics of drawing force, cross-sectional size change characteristics, high-frequency acoustic characteristics of the deformation zone, and temperature evolution characteristics. The actual cross-sectional area reduction rate during the copper busbar drawing process is calculated. The multi-dimensional feature vector and the actual cross-sectional area reduction rate are input into a data fusion prediction model based on a CNN-LSTM architecture to capture the temporal dependency of the dynamic evolution of force-heat-acoustic during the drawing process. The model outputs density, tensile strength, and elongation indices representing the current grain compactness of the copper busbar. The processing quality of the copper busbar is analyzed based on the density, tensile strength, and elongation indices, and the quality analysis results are used as the processing monitoring results.

[0007] A second aspect of this application provides a copper busbar processing monitoring system based on multi-sensor data fusion, the system comprising: A multi-source physical field data acquisition module is used to deploy an online optical dimension measuring instrument at the discharge end of the extrusion equipment, and to configure a drawing force sensor, an acoustic emission sensor attached to the side wall of the drawing die, and an infrared temperature measuring device pointing towards the drawing deformation zone at the drawing equipment, to collect multi-source physical field data of the copper busbar in real time during the extrusion and drawing process; an actual cross-sectional area reduction calculation module is used to perform clock synchronization and signal noise reduction on the multi-source physical field data, extract a multi-dimensional feature vector containing drawing force time sequence characteristics, cross-sectional size change characteristics, high-frequency acoustic characteristics of the deformation zone, and temperature evolution characteristics, and calculate The actual cross-sectional area reduction rate during the copper busbar drawing process; the time-series dependency capture module, used to input the multi-dimensional feature vector and the actual cross-sectional area reduction rate into the data fusion prediction model based on the CNN-LSTM architecture, capture the time-series dependency of the dynamic evolution of force-heat-acoustic during the drawing process, and output the density index, tensile strength and elongation index representing the current grain compactness of the copper busbar; the processing monitoring result acquisition module, used to perform copper busbar processing quality analysis based on the density index, tensile strength and elongation index, and use the quality analysis result as the processing monitoring result.

[0008] One or more technical solutions provided in this application have at least the following technical effects or advantages: An online optical dimension measuring instrument is deployed at the discharge end of the extrusion equipment to collect multi-source physical field data of copper busbars in real time during the extrusion and drawing processes. Clock synchronization and signal noise reduction are performed on the multi-source physical field data to extract multi-dimensional feature vectors containing temporal characteristics of drawing force, cross-sectional dimension changes, high-frequency acoustic characteristics of the deformation zone, and temperature evolution characteristics. The actual cross-sectional area reduction rate during the copper busbar drawing process is calculated. The temporal dependence of the force-thermal-acoustic dynamic evolution during the drawing process is captured, and density, tensile strength, and elongation indices representing the current grain density within the copper busbar are output. Copper busbar processing quality analysis is performed, and the results are used as processing monitoring results. This achieves the technical effect of realizing multi-sensor data fusion and temporal feature mining in copper busbar processing, online prediction of material mechanical properties, and improved real-time monitoring accuracy of copper busbar processing. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a schematic flowchart of a copper busbar processing monitoring method based on multi-sensor data fusion provided in an embodiment of this application. Figure 2 A schematic diagram of the structure of a copper busbar processing monitoring system based on multi-sensor data fusion provided in an embodiment of this application.

[0011] Figure labeling: 10 Multi-source physical field data acquisition module, 20 Actual cross-sectional area reduction ratio calculation module, 30 Time-series dependency capture module, 40 Processing monitoring result acquisition module. Detailed Implementation

[0012] This application provides a copper busbar processing monitoring method and system based on multi-sensor data fusion, which addresses the technical problems of existing copper busbar processing monitoring methods, such as difficulty in integrating multi-source sensor information, inability to predict material mechanical properties online, and lagging and inaccurate quality monitoring.

[0013] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0014] Example 1, as Figure 1As shown, this application provides a method for monitoring copper busbar processing based on multi-sensor data fusion, the method comprising: Step S100: Deploy an online optical dimension measuring instrument at the discharge end of the extrusion equipment, and configure a drawing force sensor, an acoustic emission sensor attached to the side wall of the drawing die, and an infrared temperature measuring device pointing to the drawing deformation zone at the drawing equipment to collect multi-source physical field data of the copper busbar in real time during the extrusion and drawing process.

[0015] Specifically, the system involves the zonal deployment and synchronous data acquisition of various types of sensors. An online optical dimension measuring instrument is fixedly installed at the extrusion equipment's discharge station to continuously collect information on the width and thickness of the extruded billet cross-section. A tension detection sensor is installed at the drawing station to capture real-time changes in the drawing force of the copper busbar. An acoustic emission sensor is tightly fitted and fixed to the side wall of the drawing die to continuously collect high-frequency acoustic signals generated by die deformation and metal friction. Simultaneously, an infrared temperature measuring device is deployed, with the temperature probe precisely aligned with the core area of ​​the copper busbar drawing deformation, to continuously collect transient temperature data in the deformation zone. All types of sensors are uniformly configured with the same microsecond-level acquisition clock, and synchronously collect original data on four types of multi-source physical fields—dimensional, mechanical, acoustic, and temperature—during the entire process of copper busbar extrusion forming and drawing deformation according to a preset fixed sampling period. This process completely preserves multi-dimensional original monitoring information of the entire copper busbar processing process, providing raw data input for subsequent data synchronization, noise reduction, and feature extraction.

[0016] Step S200: Perform clock synchronization and signal denoising on the multi-source physical field data, extract multi-dimensional feature vectors containing drawing force timing features, cross-sectional size change features, high-frequency acoustic features of the deformation zone and temperature evolution features, and calculate the actual cross-sectional area reduction rate during the copper busbar drawing process.

[0017] Specifically, the raw physical field data of the acquired dimensions, drawing force, acoustic emission, and infrared temperature are first synchronized using a unified clock based on microsecond-level timestamps. Then, a filtering algorithm is used to perform noise reduction preprocessing on the raw sensor signals. Within a preset sliding time window, the average drawing force, peak-to-peak drawing force, and variance of drawing force fluctuation are calculated for the noise-reduced drawing force data to obtain the time-series characteristics of the drawing force. Combined with the real-time width and thickness fluctuations of the extruded billet and the cross-sectional width-to-height ratio of the drawn product, the cross-sectional dimension change characteristics are obtained. Time-frequency domain analysis is performed on the reconstructed high-frequency acoustic signal to extract the root mean square (RMS), impact count rate, and energy proportion of the main frequency band. The high-frequency acoustic characteristics of the deformation zone are obtained. The temperature evolution characteristics are obtained by calculating the transient temperature rise rate and temperature gradient from the inlet to the outlet of the drawing deformation zone. The four types of features are directly spliced ​​to construct a multi-dimensional feature vector with fixed dimensions. At the same time, the initial cross-sectional area of ​​the billet is obtained by the optical measuring instrument at the extrusion end and the cross-sectional area of ​​the finished product is obtained by the optical measuring instrument at the drawing end. Based on the timestamp, the cross-sectional data of the billet and the finished product corresponding to the same copper busbar are matched. The actual cross-sectional area reduction rate of the copper busbar under the current time series is solved in real time according to the calculation rule of (initial cross-sectional area of ​​billet - cross-sectional area of ​​finished product) / initial cross-sectional area of ​​billet × 100%. The multi-dimensional feature vector and the actual cross-sectional area reduction rate are output synchronously for subsequent model input.

[0018] Step S300: Input the multidimensional feature vector and the actual cross-sectional area reduction ratio into the data fusion prediction model based on the CNN-LSTM architecture to capture the temporal dependence of the dynamic evolution of force-heat-sound during the drawing process, and output the density index, tensile strength and elongation index representing the current grain compactness of the copper busbar.

[0019] Specifically, a standardized input feature matrix is ​​constructed by first selecting average drawing force, drawing force fluctuation variance, actual cross-sectional area reduction rate, root mean square value of acoustic emission signal, and temperature rise rate in deformation zone from multi-dimensional feature vectors. This feature matrix, along with the actual cross-sectional area reduction rate, is then input into a CNN-LSTM data fusion prediction model trained to convergence using multi-source physical field features from historical processing batches, offline metallographic density detection data, and offline tensile test data. The CNN layer within the model extracts spatial hidden layer features of the cross-coupling of mechanical, temperature, acoustic, and dimensional multi-physical fields from the feature matrix. These hidden layer features are then received by the LSTM layer of the Long Short-Term Memory network to deeply mine the temporal dependency of the dynamic evolution of force-thermal-acoustic multi-physical fields during the copper busbar drawing process. Finally, regression calculations are performed through the fully connected layer at the end of the model to simultaneously output density indicators characterizing the grain compactness of the copper busbar, as well as tensile strength and elongation indicators reflecting the mechanical properties of the material, providing quantitative performance parameters for subsequent comprehensive evaluation of processing quality.

[0020] Step S400: Analyze the processing quality of copper busbars based on the density index, tensile strength index, and elongation index, and use the quality analysis results as the processing monitoring results.

[0021] Specifically, the system retrieves the pre-stored acceptable thresholds for three indicators: density, tensile strength, and elongation. The density, tensile strength, and elongation indicators output by the CNN-LSTM model are compared one by one with their respective preset acceptable thresholds, and the deviation of each indicator is calculated. A weighted summation algorithm is used to perform a comprehensive calculation on the deviations of all indicators to obtain a comprehensive quality score. Based on the comprehensive quality score, the current copper busbar is classified into four quality levels: high-quality, qualified, requiring correction, and scrap. Simultaneously, the defect type is determined based on the combination of exceeding the limits of the three indicators: when the density is below the acceptable threshold and the tensile strength and elongation are simultaneously below their corresponding thresholds, it is determined to be a loose defect caused by insufficient area reduction leading to excessive grain spacing; when the density is above the acceptable threshold... When the pass threshold is exceeded or the tensile strength exceeds the threshold but the elongation is below the threshold, it is judged as an overwork hardening defect caused by excessive deformation or abnormal temperature rise. The quality grade and defect type are integrated to generate a complete quality analysis result. Then, the quality analysis result is bound to the corresponding copper busbar timing stroke position and pushed to the industrial dashboard for visualization in real time. If it is determined to be a product that needs correction, a mold parameter correction signal is sent to the control system and the corrected die size is output to the extrusion die processing terminal. If it is determined to be a scrap product, a control command is sent to the flying shear and inkjet coding mechanism at the drawing end to mark the unqualified copper busbar section online or automatically cut it out. Finally, the complete quality analysis result containing the grade, defect, and linkage control command is output as the copper busbar processing monitoring result.

[0022] In one possible implementation, step S200 further includes: Step S210: Within a preset sliding time window, calculate the average pulling force, peak-to-peak value of the pulling force, and variance of the pulling force fluctuation after noise reduction to obtain the temporal characteristics of the pulling force.

[0023] Step S220: Calculate the real-time width fluctuation, thickness fluctuation, and cross-sectional width-to-height ratio change rate of the extruded billet and the drawn finished product to obtain the cross-sectional dimension change characteristics.

[0024] Step S230: Perform time-frequency domain analysis on the reconstructed high-frequency acoustic signal, extract the root mean square (RMS) value, impact count rate, and energy proportion of the main frequency band, and obtain the high-frequency acoustic characteristics of the deformation zone.

[0025] Step S240: Calculate the transient temperature rise rate and the maximum temperature gradient from the inlet to the outlet of the drawing deformation zone to obtain the temperature evolution characteristics.

[0026] Step S250: The time sequence features of the pulling force, the cross-sectional size change features, the high-frequency acoustic features, and the temperature evolution features are spliced ​​together to construct a multi-dimensional feature vector with fixed dimensions.

[0027] Specifically, a fixed-duration sliding time window is defined to capture the continuous pulling force time series data after noise reduction. All pulling force sampling values ​​within the window are traversed. First, the arithmetic mean of all sampling points within the window is calculated to obtain the average pulling force. Then, the maximum and minimum pulling forces within the window are extracted and the difference between them is calculated to obtain the peak pulling force. Next, the mean value of the squared difference between each sampling value within the window and the average pulling force is calculated to obtain the pulling force fluctuation variance. The calculated average pulling force, peak-to-peak pulling force, and pulling force fluctuation variance are combined to form a complete pulling force time series feature, which is used to characterize the magnitude of the force, the amplitude of the force fluctuation, and the stability of the force during the copper busbar pulling process.

[0028] The real-time width and thickness values ​​of the billet collected by the optical dimension measuring instrument at the extrusion end and the real-time width and thickness values ​​of the finished product collected by the optical dimension measuring instrument at the drawing end are read. The width fluctuation is obtained by subtracting the real-time width of the billet from the standard width of the billet, and the thickness fluctuation is obtained by subtracting the real-time thickness of the billet from the standard thickness of the billet. The width-to-height ratio of the billet section and the width-to-height ratio of the finished product section are calculated separately. The width-to-height ratio of the finished product section is subtracted from the width-to-height ratio of the billet section and then divided by the width-to-height ratio of the billet section to obtain the change rate of the width-to-height ratio of the cross-section. The width fluctuation, thickness fluctuation, and change rate of the width-to-height ratio of the cross-section obtained are integrated into the cross-sectional dimension change characteristics, which accurately reflects the deformation fluctuation law of the cross-sectional dimensions of the copper busbar from the extruded billet to the drawn finished product.

[0029] First, a fast Fourier transform is performed on the continuous high-frequency acoustic time-domain signal acquired by the acoustic emission sensor on the mold sidewall and after noise reduction and reconstruction to achieve joint analysis of the time and frequency domains. Within a complete sliding time window, the root mean square (RMS) value of the time-domain signal is calculated to characterize the overall acoustic energy level. A fixed amplitude threshold is set to count the number of pulses whose signal amplitude exceeds the threshold within the window and divide it by the window duration to obtain the impact count rate. Multiple preset frequency intervals are divided and the signal energy of each frequency band is calculated separately. The main frequency band with the highest energy is selected, and the energy ratio of the main frequency band is obtained by dividing the total energy of the main frequency band by the total energy of the entire frequency band. The three types of parameters, namely the root mean square (RMS), the impact count rate, and the energy ratio of the main frequency band, are combined to construct the high-frequency acoustic characteristics of the deformation zone, which quantitatively reflects the acoustic signal intensity, pulse frequency, and energy distribution characteristics generated by metal friction and mold deformation in the drawing deformation zone.

[0030] The real-time inlet and outlet temperatures of the drawing deformation zone are simultaneously read from the infrared thermometer. Two sets of temperature values ​​corresponding to the first and last moments of the same sliding time window are recorded. The transient temperature rise rate is obtained by dividing the difference between the final outlet temperature and the initial inlet temperature by the window sampling duration. The temperature data of all sampling points within the window are traversed, and the ratio of the temperature difference between adjacent temperature measurement points to the distance between the points is calculated. The maximum value is extracted as the highest temperature gradient. The calculated transient temperature rise rate and the highest temperature gradient are integrated to form a temperature evolution characteristic, which fully characterizes the heating rate and spatial temperature difference of the copper busbar in the drawing deformation zone.

[0031] The average pull-out force, peak-to-peak pull-out force, and pull-out force fluctuation variance output from step S210, the width fluctuation, thickness fluctuation, and cross-sectional width-to-height ratio change rate output from step S220, the root mean square value (RMS), impact count rate, and main frequency band energy ratio output from step S230, and the transient temperature rise rate and highest temperature gradient output from step S240 are extracted sequentially. All scalar parameters under the four types of features are then concatenated sequentially, and the arrangement order and total number of features of each batch of output features are uniformly constrained to construct a multidimensional feature vector with constant dimension and uniform input format, ensuring the standardization of subsequent surface reduction rate calculation and CNN-LSTM model input data structure.

[0032] In one possible implementation, step S200 further includes: Step S260: The cross-sectional width and thickness of the extruded billet are obtained in real time by an online optical dimension measuring instrument deployed at the discharge end of the extrusion equipment, and the initial cross-sectional area of ​​the billet is calculated.

[0033] Step S270: The cross-sectional width and thickness of the drawn product are obtained in real time by an online optical dimension measuring instrument deployed at the discharge end of the drawing equipment, and the cross-sectional area of ​​the product is calculated.

[0034] Step S280: Based on microsecond-level timestamp alignment, obtain the corresponding cross-sectional area of ​​the same copper busbar micro-segment under extrusion and drawing conditions, and calculate the actual cross-sectional area reduction rate under the current time sequence node in real time according to the area reduction rate calculation rules.

[0035] Specifically, the instantaneous sampling data of the cross-sectional width and thickness of the extruded billet are retrieved in real time from the online optical dimension measuring instrument at the discharge end of the extrusion equipment. The width and thickness values ​​of the billet under the same timestamp are multiplied to calculate the initial cross-sectional area corresponding to the extruded copper busbar. This provides basic cross-sectional parameters for subsequent comparison of the cross-sectional area of ​​the drawn finished product and calculation of the actual cross-sectional area reduction ratio.

[0036] The system reads the instantaneous cross-sectional width and thickness sampling data of the drawn finished product collected by the online optical dimension measuring instrument installed at the discharge end of the drawing equipment in real time. The width and thickness values ​​of the finished product corresponding to the same timestamp are multiplied to calculate the cross-sectional area of ​​the current copper busbar drawn finished product in real time. This is used to complete the quantitative calculation of the subsequent cross-sectional area reduction rate in conjunction with the initial cross-sectional area of ​​the extruded billet.

[0037] Based on the unified timekeeping benchmark of the production line, the cross-sectional data collected by the optical measuring instruments at both ends of the extrusion and drawing are all stamped with microsecond-level hardware timestamps. The nearest neighbor matching algorithm is used to complete the high-precision timestamp alignment of the two sets of time series data, the initial cross-sectional area of ​​the billet and the cross-sectional area of ​​the finished product. The set of data with the smallest time difference is selected to match the same copper busbar micro-segment. Then, according to the calculation rule of (initial cross-sectional area of ​​billet - cross-sectional area of ​​finished product) ÷ initial cross-sectional area of ​​billet × 100%, the two sets of cross-sectional area values ​​are substituted into the calculation to solve the actual cross-sectional area reduction rate of the copper busbar at the current time node in real time. This eliminates the mismatch problem of the front and rear cross-sectional data caused by the fluctuation of the copper busbar conveying speed and ensures that the calculation result of the area reduction rate is completely matched with the corresponding multi-dimensional feature vector time series.

[0038] In one possible implementation, step S280 further includes: The rule for calculating the reduction rate is to subtract the finished product's cross-sectional area from the initial cross-sectional area of ​​the billet, divide the result by the initial cross-sectional area of ​​the billet, and multiply the result by 100% to obtain the actual cross-sectional reduction rate.

[0039] Specifically, the rule for calculating the reduction rate first uses the initial cross-sectional area of ​​the extruded billet, which is precisely matched with the timestamp, to subtract the cross-sectional area of ​​the drawn copper busbar micro-segment to obtain the difference in cross-sectional area reduction. Then, this difference is divided by the initial cross-sectional area of ​​the billet to obtain the cross-sectional shrinkage ratio. Finally, the obtained ratio is multiplied by 100% to convert it into a percentage form, and the actual cross-sectional reduction rate corresponding to the current copper busbar micro-segment drawing process is output, which quantitatively reflects the overall degree of cross-sectional reduction during the copper busbar drawing process.

[0040] In one possible implementation, step S300 further includes: Step S310: Construct the average pull-out force, pull-out force fluctuation variance, actual cross-sectional area reduction ratio, root mean square value of acoustic emission signal, and temperature rise rate of deformation zone from the multidimensional feature vector into an input feature matrix.

[0041] Step S320: Use the convolutional neural network (CNN) layer to extract the hidden layer features in the feature matrix that are spatially coupled with force, heat, sound, and size.

[0042] Step S330: Utilize the Long Short-Term Memory (LSTM) network layer to receive the hidden layer features, extract the temporal dynamic features that evolve with the drawing stroke, and output the density index, tensile strength, and elongation index through regression prediction by a fully connected layer.

[0043] Specifically, four feature scalars—average pull-out force, pull-out force fluctuation variance, root mean square value of acoustic emission signal, and temperature rise rate in the deformation zone—are first extracted from the generated multidimensional feature vector. Then, the actual cross-sectional area reduction rate calculated synchronously is retrieved. These five features are arranged in a fixed order as a single-row feature sequence. In this preferred embodiment, the sliding time window duration is set to 200ms, and the sliding step size is set to 50ms. Multiple sets of time-series sampling data are acquired using a sliding truncation method. Each sampling moment corresponds to one row of feature sequences, and the arrangement of the five features in each row is strictly uniform. This forms a two-dimensional input feature matrix with the number of rows equal to the number of sampling points within the sliding time window and a fixed number of columns of 5. All feature values ​​within the matrix are normalized and scaled using Z-score standardization. The Z-score standardization calculation formula is: In the formula, x is the original feature value. This is the mean of the feature across the entire historical training sample set. The standard deviation of this feature across the entire historical training sample set is used. The mean and standard deviation parameters obtained from the training set are directly applied to the feature matrix during the online inference stage to avoid leaking real-time collected data. This transforms each physical quantity feature to a value range with a mean of 0 and a variance of 1, eliminating the influence of dimensional differences between features such as pull-out force, temperature, acoustics, and surface reduction rate. This completes the construction of the standardized input feature matrix, adapting to the extraction and computation requirements of subsequent CNN convolutional layers for two-dimensional matrix format data.

[0044] The normalized two-dimensional input feature matrix is ​​fed into the convolutional neural network (CNN) layer. In this embodiment, the CNN layer uses two sets of one-dimensional convolution-batch normalization-ReLU-max pooling modules stacked in series, i.e., a total of two one-dimensional convolutional layers. Each convolutional layer is followed by a one-dimensional batch normalization layer, a ReLU activation layer, and a one-dimensional max pooling layer. The first one-dimensional convolutional layer has 16 convolutional kernels with kernel sizes of 3 and 5 (multi-scale convolutional kernels), a stride of 1, and a "same" padding mode to ensure that the temporal sequence length does not shrink drastically. The second one-dimensional convolutional layer has 32 convolutional kernels with kernel sizes of 3 and 5, a stride of 1, and a "same" padding mode. The pooling layers uniformly employ one-dimensional max pooling (MaxPool1D) with a kernel size of 2 and a stride equal to the kernel size of 2, without overlapping pooling. The activation function is uniformly the ReLU function. Batch normalization uses one-dimensional BN, and mean-variance standardization is applied to each channel of the convolutional output. Multiple sets of one-dimensional convolutional kernels of different sizes are used to perform sliding convolution operations row-by-row on the feature matrix: the size 3 convolutional kernel focuses on capturing short-term local abrupt changes such as pull-out force mutations, acoustic pulses, and temperature rise spikes; the size 5 convolutional kernel focuses on capturing medium-period correlation features such as surface area reduction drift and slow process fluctuations. These methods capture the local spatial correlations between multiple features including pull-out force, temperature, acoustics, and surface area reduction, uncovering the coupling correlation patterns of mutual constraints and influences between pull-out force, heat, sound, and cross-sectional dimensions, and outputting multi-channel convolutional feature maps. The convolutional operation output feature map undergoes one-dimensional batch normalization to correct numerical distribution, eliminating covariate bias within the network and accelerating model convergence. Then, a ReLU activation function is used to introduce nonlinear mapping to eliminate gradient saturation, strengthen the nonlinear coupling expression between multiple physics fields, and avoid the vanishing gradient phenomenon in deep networks. A one-dimensional max-pooling layer then compresses the temporal dimension while retaining key coupling information, filtering out invalid feature components caused by sensor noise disturbances, reducing the number of parameters, and suppressing model overfitting. Through layer-by-layer operations using two sets of convolution-BN-ReLU-max-pooling modules, deep hidden features of the cross-coupling of force, heat, sound, and cross-sectional dimensions are extracted layer by layer. Linearly irrelevant redundant information between original features is removed, and interference features with weak correlation to the evolution of the copper busbar's internal grain structure are eliminated. The CNN layer output still maintains a compressed hidden feature sequence with a temporal dimension, and the output format is adapted to the input requirements of subsequent LSTM layers, providing high-value feature input for mining the temporal dependencies of the pulling process evolution.

[0045] The compressed latent feature sequence output from the CNN layer, which preserves the temporal dimension, is input into a stacked LSTM network. In this embodiment, the LSTM network is configured as a two-layer stacked structure, with 64 hidden units in each LSTM layer. The activation functions for the forget gate, input gate, and output gate within the LSTM are sigmoid functions, while the activation function for cell state updates is tanh. The network input format is configured with batch_first=True to adapt to the two-dimensional temporal input feature format. A dropout layer is set between the two LSTM layers, with a dropout rate of 0.2, to reduce the risk of overfitting in the temporal network. The LSTM unit selectively discards historical latent feature information with low correlation to the copper busbar drawing quality through a forget gate, updates the cell state through an input gate, and retains the temporal changes in force, heat, sound, and size coupling features over time during the drawing process. The output gate controls the output of the effective temporal features at the current moment. The double-layer stacked LSTM structure can progressively mine the dynamically evolving temporal correlation features as the copper busbar drawing process progresses, fully capturing the long-term dependencies between temporal data before and after the drawing process. The temporal feature vector output from the LSTM layer is fed into the subsequent fully connected layer. The fully connected layer has two layers: the first fully connected layer has 128 neurons with the ReLU activation function, and the second fully connected layer has 3 output neurons without an activation function, performing linear regression fitting. Finally, it synchronously outputs density indicators characterizing the grain compactness of the current copper busbar, as well as tensile strength and elongation indicators reflecting the material's mechanical properties, providing quantitative performance parameters for subsequent copper busbar processing quality grading and defect identification.

[0046] In one possible implementation, step S300 further includes: The data fusion prediction model is obtained through supervised training based on the multi-source physical field characteristics of historical processing batches, offline metallographic density detection results, and offline tensile test data until convergence.

[0047] This CNN-LSTM data fusion prediction model employs supervised learning for offline training. First, it collects historical processing data from multiple batches of copper busbars, gathering original multi-source physical field sensor data for each segment of the copper busbar and performing a complete feature extraction process to obtain standardized multi-dimensional feature vectors and input feature matrices as training input samples. In this preferred embodiment, a total of 1200 labeled samples are collected, covering various actual production conditions including standard processes, insufficient surface reduction, excessive deformation, and abnormal temperature rise in the deformation zone. Simultaneously, offline metallographic testing is performed on samples taken from the same batch of copper busbars to obtain true density labels, and tensile mechanical tests are conducted to measure true tensile strength and elongation labels. The input feature matrix is ​​paired with the three corresponding true performance labels to construct a complete labeled training dataset. The dataset is randomly divided in a 7:2:1 ratio: 70% of the samples are used as the training set for network weight iteration updates, 20% as the validation set for training effect evaluation and overfit monitoring, and the remaining 10% as an independent test set for final evaluation of the model's generalization ability after training. During the training phase, the mean squared error (MSE) is used as the loss function, with the specific form of the loss function as follows: In the formula, N represents the number of training samples in each batch; , , These are the true density, tensile strength, and elongation labels for the i-th sample, respectively. , , These represent the predicted values ​​of the model's output. The Adam optimizer is used to iteratively update the network parameters of the CNN convolutional layers, LSTM gate weights, and fully connected layers; the initial learning rate of the optimizer is set to... The training process employs a learning rate decay strategy. When the validation set loss shows no decrease for five consecutive iterations, the learning rate decays to 0.5 times its original value. The batch size is set to 32, and the maximum number of iterations (epochs) is set to 200. After each iteration, the loss value is calculated using the validation set to monitor overfitting. This is combined with an early stopping mechanism. The specific triggering conditions for the early stopping mechanism are: monitoring the validation set loss value; when the validation set loss no longer decreases for eight consecutive iterations, or the decrease in loss is less than a preset threshold. If the early stopping condition is not triggered, the model is considered to have converged, and training is terminated directly to avoid overfitting. If the early stopping condition is not triggered, training continues until the maximum number of iterations, 200, and then stops. After training terminates, all network weight parameters at this point are saved to obtain the data fusion prediction model that has completed training convergence. Subsequently, the model's prediction accuracy is verified using an independent test set. After passing the test, it can be directly put into production line for real-time prediction of copper busbar density, tensile strength, and elongation.

[0048] In one possible implementation, step S400 further includes: Step S410: Obtain the preset density qualification threshold, tensile strength qualification threshold and elongation qualification threshold.

[0049] Step S420: Compare the predicted density, tensile strength and elongation indices with the corresponding preset density qualification threshold, tensile strength qualification threshold and elongation qualification threshold, and calculate the deviation of each index.

[0050] Step S430: Use a weighted summation algorithm to comprehensively evaluate the deviation of each indicator, generate a comprehensive quality score, and classify the copper busbar into four quality levels based on the comprehensive quality score: high-quality, qualified, requiring correction, and scrap, as the quality analysis result.

[0051] Specifically, based on national standards for copper busbar products, customer forming performance technical requirements, and offline testing statistics of qualified workpieces produced in long-term mass production, the standard qualified ranges for the three performance indicators of density, tensile strength, and elongation are determined in advance. The lower limit of each indicator range is extracted as a preset qualified threshold and stored in the system parameter library. When the online quality analysis process of copper busbar is started, the system automatically retrieves the qualified thresholds for density, tensile strength, and elongation from the parameter library and allocates them to the quality judgment calculation unit. This provides a judgment benchmark for comparing the model-predicted indicators with the standard limits and calculating the deviation of the indicators.

[0052] The density, tensile strength, and elongation indices predicted in real time by the CNN-LSTM model are compared one by one with the corresponding preset qualified thresholds retrieved by the system. For each performance index, a uniform deviation calculation formula is used: Deviation = |Real-time predicted index value − Corresponding preset qualified threshold| ÷ Corresponding preset qualified threshold. The deviation values ​​of the three indices and thresholds are substituted into the values ​​and calculated independently to obtain the deviation of density, tensile strength, and elongation. The deviation values ​​after normalization are used to quantify the degree of deviation of each performance index from the qualified standard, eliminating the judgment interference caused by the inconsistency of the dimensions and numerical ranges of the three indices, and providing a uniform deviation parameter for the subsequent weighted comprehensive quality score calculation.

[0053] Based on the density deviation weight, tensile strength deviation weight, and elongation deviation weight set according to the product process requirements, the deviations of the three indicators obtained in step S420 are multiplied by their respective corresponding weights and then summed to obtain the total weighted deviation value. Then, a fixed conversion formula is used to perform a reverse mapping operation on the total weighted deviation value to generate a comprehensive quality score in the range of 0 to 100. The smaller the total weighted deviation value, the higher the comprehensive quality score. The system internally presets four threshold intervals. When the comprehensive quality score is in the highest interval, it is judged as a high-quality product; the next level interval is judged as a qualified product; the third interval is judged as a product that needs correction; and the lowest interval is judged as a defective product. The calculated comprehensive quality score is bound and integrated with the corresponding quality level to form a basic quality analysis result that includes quantitative scores and grading results.

[0054] In one possible implementation, step S430 further includes: Step S431: When the density index is lower than the preset density qualification threshold, and the tensile strength and elongation index are both lower than the tensile strength qualification threshold and the elongation qualification threshold, it is determined that there is an excessively large grain gap or loose defect in the current copper busbar due to insufficient reduction rate.

[0055] Step S432: When the density index is higher than the preset density qualification threshold, but the elongation index is lower than the preset elongation threshold and the tensile strength index is higher than the tensile strength qualification threshold and the elongation qualification threshold, it is determined that the current copper busbar has a work hardening over-defect caused by excessive deformation or abnormal temperature rise in the deformation zone.

[0056] Step S433: Add the identified defect types to the quality analysis results.

[0057] Specifically, the system reads the predicted density, tensile strength, and elongation indices in real time and performs logical judgments against the corresponding qualified thresholds stored in the system. When the density index is less than the preset qualified density threshold, and the tensile strength index is less than the preset qualified tensile strength threshold and the elongation index is also less than the preset qualified elongation threshold, a defect judgment conclusion is triggered. This index characteristic corresponds to insufficient actual cross-sectional area reduction in the drawing process, insufficient plastic deformation of the metal, and the inability of the internal grains of the copper busbar to be fully squeezed together, making it difficult to effectively reduce the grain gap, thus resulting in excessive grain gap and loose micro-defects. The system completes the defect type identification and incorporates the defect type information into the quality analysis results to support process traceability and online control.

[0058] The model outputs predicted indices for density, tensile strength, and elongation. These indices are then logically verified against pre-stored threshold values. If the density index exceeds the preset threshold, the tensile strength index exceeds the preset threshold, and the elongation index falls below the preset threshold, the current copper busbar segment is deemed to have an overwork hardening defect. This defect is caused by the deformation of the drawing section exceeding the process design range, or by abnormal temperature rise in the drawing deformation area, leading to continuous dislocation proliferation in the metal grains and severe work hardening. This results in increased material strength and a significant decrease in plasticity. The identified overwork hardening defect type is incorporated into the quality analysis results to guide online adjustments of process parameters such as drawing speed, die reduction ratio, and cooling parameters.

[0059] The system acquires the defect type information obtained from the judgment. If the corresponding defect judgment condition is met, the defects such as excessively large grain gaps and looseness or excessive work hardening are written into the quality analysis results. If no defect judgment condition is triggered, the defect is marked as having no microstructure. The defect identification information is then packaged and integrated with the previously obtained comprehensive quality score and copper busbar quality grade to form a complete quality analysis result. This result is then sent to the production line control system to provide a complete basis for process parameter adjustment, marking of non-conforming sections, and production data traceability.

[0060] In one possible implementation, step S430 further includes: The quality analysis results are transmitted to the industrial dashboard in real time for visual presentation and are associated with the current timing travel position of the copper busbar.

[0061] When the quality analysis result indicates that the product needs correction, the control system is triggered to generate a mold parameter correction signal and send the corrected die size to the extrusion die processing terminal.

[0062] When the quality analysis result is scrap, a control command is sent in real time to the flying shear or marking actuator at the back end of the drawing equipment to mark or automatically cut and remove the unqualified copper busbar sections online.

[0063] Specifically, after the complete quality analysis results for the copper busbar micro-segment are packaged, the comprehensive quality score, quality level, and defect type information are pushed to the on-site industrial dashboard in real time via industrial Ethernet. The index values, grade labels, and defect information are dynamically displayed according to the preset visualization template. At the same time, the real-time travel position time sequence data of the copper busbar collected by the production line encoder is retrieved, and the travel position information is linked one-to-one with the quality analysis results of the corresponding segment. The specific spatial location of the defective copper busbar is marked on the production line simulation view of the industrial dashboard, which facilitates on-site maintenance personnel to quickly locate abnormal segments and realize synchronous visual monitoring of processing quality and workpiece position.

[0064] When the quality analysis results determine that the current copper busbar segment needs correction, the built-in process closed-loop control logic is immediately triggered and a die parameter correction signal is generated. Combining the actual cross-sectional area reduction rate and mechanical property deviation data of the copper busbar segment, the appropriate extrusion die correction die size is obtained by reverse calculation based on the preset process mapping model. Then, the correction die size parameters are sent to the extrusion die processing terminal through the industrial communication bus, so that the terminal can complete the die die parameter adjustment or machining correction. This optimizes the initial cross-sectional size of the subsequent copper busbar billet, so that the actual area reduction rate of the drawing process returns to the reasonable range of the process, thereby improving the microstructure and mechanical properties of the copper busbar from the source.

[0065] When the current copper busbar segment quality analysis result is determined to be scrap, the system immediately links and binds the timing and travel position information to generate precise control commands. These commands are then sent in real time to the coding mechanism and flying shear actuator at the back end of the drawing equipment via the industrial PLC bus. Based on the preset processing logic, the scrap segment is precisely marked online or automatically rejected: For continuous segments of non-conforming copper busbars, the coding equipment is controlled to spray a unique defect traceability mark and scrap code onto the corresponding workpiece surface, completing the visual marking of the abnormal segment; at the same time, the flying shear mechanism is triggered to precisely move according to the preset length threshold, automatically cutting, separating, and rejecting the marked scrap copper busbar segment online. This achieves real-time sorting and isolation of non-conforming copper busbar segments, preventing defective workpieces from flowing into the next process and ensuring the overall quality stability of the finished product.

[0066] Using the multi-sensor data fusion copper busbar processing monitoring method described above, on-site tests were conducted under four typical drawing conditions on the production line. Multi-source sensor signals were simultaneously collected and input into a CNN-LSTM data fusion prediction model for online prediction. The model output results were then compared with measured data obtained from offline metallographic inspection and tensile testing. The test data are shown in Table 1. Table 1 Comparison of online prediction results and offline measured data of the method of the present invention under different process conditions.

[0067] As shown in Table 1, under four typical working conditions—standard process, insufficient reduction rate, excessive deformation, and abnormal temperature rise—the predicted values ​​of density, tensile strength, and elongation obtained by the method of this invention have small deviations from the offline measured values, and the prediction accuracy meets the requirements of online monitoring. The results obtained by quality judgment based on the predicted indicators are consistent with the conclusions of offline sample testing, and can accurately identify defects such as loose grains and excessive work hardening, thus realizing online prediction and real-time quality grading of the mechanical properties of copper busbars.

[0068] Example 2, based on the same inventive concept as the copper busbar processing monitoring method using multi-sensor data fusion in the foregoing examples, such as... Figure 2As shown, this application provides a copper busbar processing monitoring system based on multi-sensor data fusion. The system and method embodiments in this application are based on the same inventive concept. The system includes: The multi-source physical field data acquisition module 10 is used to deploy an online optical dimension measuring instrument at the discharge end of the extrusion equipment, and to configure a drawing force sensor, an acoustic emission sensor attached to the side wall of the drawing die, and an infrared temperature measuring device pointing to the drawing deformation zone at the drawing equipment, so as to collect multi-source physical field data of copper busbars in real time during the extrusion and drawing process.

[0069] The actual cross-sectional area reduction ratio calculation module 20 is used to perform clock synchronization and signal noise reduction on the multi-source physical field data, extract multi-dimensional feature vectors including drawing force timing features, cross-sectional size change features, high-frequency acoustic features of the deformation zone and temperature evolution features, and calculate the actual cross-sectional area reduction ratio during the copper busbar drawing process.

[0070] The temporal dependency capture module 30 is used to input the multidimensional feature vector and the actual cross-sectional area reduction rate into the data fusion prediction model based on the CNN-LSTM architecture, capture the temporal dependency of the dynamic evolution of force-heat-sound during the drawing process, and output the density index, tensile strength and elongation index representing the current grain compactness inside the copper busbar.

[0071] The processing monitoring result acquisition module 40 is used to perform copper busbar processing quality analysis based on the density index, tensile strength and elongation index, and use the quality analysis result as the processing monitoring result.

[0072] Furthermore, the system is also used to implement the following functions: Within a preset sliding time window, the average drawing force, peak-to-peak value of the drawing force, and variance of the drawing force fluctuation are calculated to obtain the temporal characteristics of the drawing force. The real-time width fluctuation, thickness fluctuation, and aspect ratio change rate of the extruded billet and the drawn finished product are calculated to obtain the cross-sectional dimension change characteristics. The reconstructed high-frequency acoustic signal is analyzed in the time-frequency domain to extract the root mean square (RMS), impact count rate, and energy proportion of the main frequency band to obtain the high-frequency acoustic characteristics of the deformation zone. The transient temperature rise rate and maximum temperature gradient from the inlet to the outlet of the drawing deformation zone are calculated to obtain the temperature evolution characteristics. The temporal characteristics of the drawing force, the cross-sectional dimension change characteristics, the high-frequency acoustic characteristics, and the temperature evolution characteristics are spliced ​​together to construct a multidimensional feature vector with fixed dimensions.

[0073] Furthermore, the system is also used to implement the following functions: An online optical dimension measuring instrument deployed at the discharge end of the extrusion equipment is used to acquire the cross-sectional width and thickness of the extruded billet in real time and calculate the initial cross-sectional area of ​​the billet. An online optical dimension measuring instrument deployed at the discharge end of the drawing equipment is used to acquire the cross-sectional width and thickness of the drawn finished product in real time and calculate the cross-sectional area of ​​the finished product. Based on microsecond-level timestamp alignment, the corresponding cross-sectional area of ​​the same copper busbar micro-segment under extrusion and drawing states is acquired, and the actual cross-sectional area reduction rate under the current time node is calculated in real time according to the area reduction rate calculation rules.

[0074] Furthermore, the system is also used to implement the following functions: The rule for calculating the reduction rate is to subtract the finished product's cross-sectional area from the initial cross-sectional area of ​​the billet, divide the result by the initial cross-sectional area of ​​the billet, and multiply the result by 100% to obtain the actual cross-sectional reduction rate.

[0075] Furthermore, the system is also used to implement the following functions: The average drawing force, drawing force fluctuation variance, actual cross-sectional area reduction ratio, root mean square value of acoustic emission signal, and temperature rise rate in deformation zone from the multidimensional feature vector are used to construct the input feature matrix. The hidden layer features of force, heat, sound, and size spatial cross-coupling in the feature matrix are extracted using the convolutional neural network (CNN) layer. The hidden layer features are received using the long short-term memory (LSTM) layer, and the temporal dynamic features that evolve with the drawing stroke are extracted. The density index, tensile strength, and elongation index are then output through regression prediction by the fully connected layer.

[0076] Furthermore, the system is also used to implement the following functions: The data fusion prediction model is obtained through supervised training based on the multi-source physical field characteristics of historical processing batches, offline metallographic density detection results, and offline tensile test data until convergence.

[0077] Furthermore, the system is also used to implement the following functions: Obtain preset density, tensile strength, and elongation qualification thresholds; compare the predicted density, tensile strength, and elongation indices with their corresponding preset density, tensile strength, and elongation qualification thresholds to calculate the deviation of each index; use a weighted summation algorithm to comprehensively evaluate the deviation of each index, generate a comprehensive quality score, and classify the copper busbars into four quality levels—high-quality, qualified, requiring correction, and scrap—based on the comprehensive quality score, as the quality analysis result.

[0078] Furthermore, the system is also used to implement the following functions: When the density index is lower than the preset density qualification threshold, and the tensile strength and elongation indices are both lower than the tensile strength qualification threshold and the elongation qualification threshold, it is determined that there is an excessively large grain gap or looseness defect in the current copper busbar due to insufficient reduction area; when the density index is higher than the preset density qualification threshold, but the elongation index is lower than the preset elongation threshold and the tensile strength index is higher than the tensile strength qualification threshold and the elongation qualification threshold, it is determined that there is an excessive work hardening defect in the current copper busbar due to excessive deformation or abnormal temperature rise in the deformation zone; the determined defect type is added to the quality analysis results.

[0079] Furthermore, the system is also used to implement the following functions: The quality analysis results are transmitted to the industrial dashboard in real time for visual display and are associated with the current timing stroke position of the copper busbar. When the quality analysis result indicates that the product needs correction, the control system is triggered to generate a mold parameter correction signal and send the corrected die size to the extrusion die processing terminal. When the quality analysis result indicates that the product is defective, control commands are sent to the flying shear or marking actuator at the back end of the drawing equipment in real time to mark or automatically cut and remove the unqualified copper busbar sections online.

[0080] It should be noted that the order of the embodiments described above is for descriptive purposes only and does not represent the superiority or inferiority of the embodiments. Specific embodiments of this specification have been described above. Furthermore, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0081] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0082] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application intends to include such modifications and variations.

Claims

1. A method for copper bar processing monitoring by multi-sensor data fusion, characterized by, The method includes: An online optical dimension measuring instrument is deployed at the discharge end of the extrusion equipment, and a drawing force sensor, an acoustic emission sensor attached to the side wall of the drawing die, and an infrared temperature measuring device pointing to the drawing deformation zone are configured at the drawing equipment to collect multi-source physical field data of the copper busbar in real time during the extrusion and drawing process. Clock synchronization and signal denoising are performed on the multi-source physical field data. Multidimensional feature vectors containing drawing force timing features, cross-sectional size change features, high-frequency acoustic features of the deformation zone and temperature evolution features are extracted, and the actual cross-sectional area reduction rate during the copper busbar drawing process is calculated. The multidimensional feature vector and the actual cross-sectional area reduction ratio are input into the data fusion prediction model based on the CNN-LSTM architecture to capture the temporal dependence of the dynamic evolution of force-heat-sound during the drawing process, and output the density index, tensile strength and elongation index representing the current grain compactness inside the copper busbar. The processing quality of copper busbars was analyzed based on the density, tensile strength and elongation indices, and the results of the quality analysis were used as the processing monitoring results. Clock synchronization and signal denoising are performed on the multi-source physical field data to extract a multi-dimensional feature vector containing tensile force timing features, cross-sectional size change features, high-frequency acoustic features of the deformation zone, and temperature evolution features, including: Within a preset sliding time window, the average pulling force, peak-to-peak pulling force, and variance of pulling force fluctuation of the noise-reduced pulling force data are calculated to obtain the temporal characteristics of the pulling force. Calculate the real-time width fluctuation, thickness fluctuation, and cross-sectional width-to-height ratio change rate of the extruded billet and the drawn finished product to obtain the cross-sectional dimension change characteristics; Time-frequency domain analysis was performed on the reconstructed high-frequency acoustic signal to extract the root mean square (RMS), impact count rate, and energy proportion of the main frequency band, thereby obtaining the high-frequency acoustic characteristics of the deformation region. The transient temperature rise rate and maximum temperature gradient from the inlet to the outlet of the drawing deformation zone are calculated to obtain the temperature evolution characteristics. The time-series characteristics of the pulling force, the cross-sectional size change characteristics, the high-frequency acoustic characteristics, and the temperature evolution characteristics are spliced ​​together to construct a multi-dimensional feature vector with fixed dimensions; And calculate the actual cross-sectional area reduction rate during the copper busbar drawing process, including: By deploying an online optical dimension measuring instrument at the discharge end of the extrusion equipment, the cross-sectional width and thickness of the extruded billet are obtained in real time, and the initial cross-sectional area of ​​the billet is calculated. By deploying an online optical dimension measuring instrument at the discharge end of the drawing equipment, the cross-sectional width and thickness of the drawn finished product are obtained in real time, and the cross-sectional area of ​​the finished product is calculated. Based on microsecond-level timestamp alignment, the corresponding cross-sectional area of ​​the same copper busbar micro-segment under extrusion and drawing conditions is obtained, and the actual cross-sectional area reduction rate under the current time node is calculated in real time according to the area reduction rate calculation rules. The average pull-out force, pull-out force fluctuation variance, actual cross-sectional area reduction ratio, root mean square value of acoustic emission signal, and temperature rise rate of deformation zone in the multidimensional feature vector are used to construct the input feature matrix. The hidden layer features related to the spatial cross-coupling of force, heat, sound, and size in the feature matrix are extracted using the convolutional neural network (CNN) layer. The hidden layer features are received by the Long Short-Term Memory (LSTM) network, the temporal dynamic features that evolve with the drawing stroke are extracted, and the density index, tensile strength and elongation index are output by regression prediction through a fully connected layer.

2. The copper busbar processing monitoring method based on multi-sensor data fusion as described in claim 1, characterized in that, The rule for calculating the reduction rate is to subtract the finished product's cross-sectional area from the initial cross-sectional area of ​​the billet, divide the result by the initial cross-sectional area of ​​the billet, and multiply the result by 100% to obtain the actual cross-sectional reduction rate.

3. The copper busbar processing monitoring method based on multi-sensor data fusion as described in claim 1, characterized in that, The data fusion prediction model is obtained through supervised training based on the multi-source physical field characteristics of historical processing batches, offline metallographic density detection results, and offline tensile test data until convergence.

4. The copper busbar processing monitoring method based on multi-sensor data fusion as described in claim 1, characterized in that, The processing quality of copper busbars was analyzed based on the aforementioned density, tensile strength, and elongation indices, and the results of the quality analysis were used as processing monitoring results, including: Obtain the preset density qualification threshold, tensile strength qualification threshold, and elongation qualification threshold; The predicted density, tensile strength and elongation indices are compared with the corresponding preset density qualification threshold, tensile strength qualification threshold and elongation qualification threshold, respectively, and the deviation of each index is calculated. A weighted summation algorithm is used to comprehensively evaluate the deviation of each indicator, generate a comprehensive quality score, and classify the copper busbars into four quality levels based on the comprehensive quality score: high-quality, qualified, requiring correction, and scrap, which are used as the quality analysis results.

5. The copper busbar processing monitoring method based on multi-sensor data fusion as described in claim 4, characterized in that, Also includes: When the density index is lower than the preset density qualification threshold, and the tensile strength and elongation index are both lower than the tensile strength qualification threshold and the elongation qualification threshold, it is determined that there is a defect in the current copper busbar caused by insufficient reduction rate, resulting in excessive grain gap or looseness. When the density index is higher than the preset density qualification threshold, but the elongation index is lower than the preset elongation threshold and the tensile strength index is higher than the tensile strength qualification threshold and the elongation qualification threshold, it is determined that the current copper busbar has a work hardening over-defect caused by excessive deformation or abnormal temperature rise in the deformation zone. The identified defect types are added to the quality analysis results.

6. The copper busbar processing monitoring method based on multi-sensor data fusion as described in claim 5, characterized in that, include: The quality analysis results are transmitted to the industrial dashboard in real time for visual presentation and are associated with the current timing position of the copper busbar. When the quality analysis result indicates that the product needs to be corrected, the control system is triggered to generate a mold parameter correction signal and send the corrected die size output in reverse to the extrusion die processing terminal. When the quality analysis result is scrap, a control command is sent in real time to the flying shear or marking actuator at the back end of the drawing equipment to mark or automatically cut and remove the unqualified copper busbar sections online.

7. A copper busbar processing monitoring system based on multi-sensor data fusion, characterized in that, The system is used to implement the copper busbar processing monitoring method based on multi-sensor data fusion as described in any one of claims 1-6, and the system comprises: The multi-source physical field data acquisition module is used to deploy an online optical dimension measuring instrument at the discharge end of the extrusion equipment, and to configure a drawing force sensor, an acoustic emission sensor attached to the side wall of the drawing die, and an infrared temperature measuring device pointing to the drawing deformation zone at the drawing equipment, so as to collect multi-source physical field data of copper busbars in real time during the extrusion and drawing process. The actual cross-sectional area reduction ratio calculation module is used to perform clock synchronization and signal noise reduction on the multi-source physical field data, extract multi-dimensional feature vectors including drawing force timing features, cross-sectional size change features, high-frequency acoustic features of the deformation zone and temperature evolution features, and calculate the actual cross-sectional area reduction ratio during the copper busbar drawing process. The temporal dependency capture module is used to input the multidimensional feature vector and the actual cross-sectional area reduction rate into the data fusion prediction model based on the CNN-LSTM architecture, capture the temporal dependency of the dynamic evolution of force-heat-sound during the drawing process, and output the density index, tensile strength and elongation index representing the current grain compactness of the copper busbar. The processing monitoring result acquisition module is used to analyze the processing quality of copper busbars based on the density index, tensile strength and elongation index, and use the quality analysis results as the processing monitoring results.

Citation Information

Patent Citations

  • Copper wire drawing process quality control method based on acoustic emission signals

    CN120394591A

  • Method and system for detecting residual stress of drawn copper bar

    CN120724843A

  • Intelligent prediction and process optimization method for forming performance of hot-rolled titanium strip

    CN122352687A