Control Processing Method and System Based on Gold Foil Processing

Abstract

The present invention provides a control processing method and system based on gold foil processing. The method includes: collecting real-time processing parameters of a target material in a gold foil processing device to generate an initial processing parameter set; determining a first control model adapted to the target material according to the hot pressing intensity gradient sequence in the initial processing parameter set; determining a second control model adapted to the target material according to the extension rate distribution curve in the initial processing parameter set; activating a target model in the first control model or the second control model based on the real-time thickness deviation data and surface uniformity fluctuation data during the gold foil processing, and generating a dynamic adjustment instruction through the target model; inputting the dynamic adjustment instruction into the actuator of the gold foil processing device to correct the pressure gradient of the hot pressing device or the rate distribution of the extension device. The present invention can improve the uniformity of the gold foil thickness distribution and the stability of the surface morphology in complex processing scenarios, and enhance the process quality and consistency of the gold foil finished product.

Description

Technical Field

[0001] The present invention relates to the technical fields of data processing and intelligent control, and particularly to a control processing method and system based on gold foil processing. Background Art

[0002] The control processing of gold foil processing aims to precisely control the gold foil thickness and surface uniformity by adjusting the parameters of the processing equipment. In the prior art, the dynamic correlation between the hot pressing strength gradient and the extension rate distribution has not been effectively modeled, resulting in a lag in thickness deviation compensation behind the real-time processing state change, and the surface uniformity fluctuation cannot be synergistically suppressed with the adjustment of the extension rate, causing over-pressing or extension tearing defects in local areas of the gold foil; at the same time, the static model cannot adapt to the hot pressing deformation differences of materials with different lattice structures, and parameter control mismatch is likely to occur in complex processing scenarios, leading to a decline in processing accuracy and insufficient process stability. Summary of the Invention

[0003] In view of this, the present invention provides a control processing method and system based on gold foil processing.

[0004] The technical solution of the embodiment of the present invention is implemented as follows:

[0005] On the one hand, the present invention provides a control processing method based on gold foil processing, the method includes: collecting real-time processing parameters of a target material in a gold foil processing device to generate an initial processing parameter set, the initial processing parameter set includes a hot pressing strength gradient sequence and an extension rate distribution curve; determining a first control model adapted to the target material according to the hot pressing strength gradient sequence in the initial processing parameter set, the first control model includes a dynamic mapping relationship between the hot pressing strength and the thickness change; determining a second control model adapted to the target material according to the extension rate distribution curve in the initial processing parameter set, the second control model includes a non-linear correlation rule between the extension rate and the surface uniformity; activating a target model in the first control model or the second control model based on real-time thickness deviation data and surface uniformity fluctuation data during the gold foil processing, and generating a dynamic adjustment instruction through the target model; inputting the dynamic adjustment instruction into an actuator of the gold foil processing device to correct the pressure gradient of the hot pressing device or the rate distribution of the extension device.

[0006] On the other hand, the present invention provides a control processing system, including a memory and a processor, the memory stores a computer program that can run on the processor, and when the processor executes the program, the steps in the above method are implemented.

[0007] The control processing method based on gold foil processing provided by the present invention generates an initial processing parameter set including a hot pressing strength gradient sequence and an extension rate distribution curve by collecting real-time processing parameters of the target material in the gold foil processing equipment, constructs the dynamic mapping relationship of the first control model based on the hot pressing strength gradient sequence and the non-linear correlation rule of the second control model based on the extension rate distribution curve, activates the target model in real time during the processing to generate dynamic adjustment instructions and correct the equipment parameters, so as to capture the correlation between the gold foil thickness and the surface uniformity by combining the coordinated change law of the hot pressing strength and the extension rate, accurately compensate the thickness deviation through the dynamic mapping relationship and suppress the surface uniformity fluctuation through the non-linear rule, and improve the control accuracy of the processing process. Moreover, through the multi-dimensional parameter fusion of the hot pressing strength gradient sequence and the extension rate distribution curve, the real-time perception ability of the deformation characteristics of the gold foil material is enhanced, the prediction ability of the thickness change trend is strengthened by using the first control model, and at the same time, the complex fluctuation mode of the surface uniformity is analyzed by the second control model to form a complementary and optimized control strategy. Further, based on the dynamic activation mechanism of the real-time thickness deviation and the surface uniformity fluctuation data, the core control requirements of different processing stages are adaptively matched, the key defect areas are preferentially corrected, and the parameter interference is eliminated through model cooperation, so as to maintain the uniformity of the gold foil thickness distribution and the stability of the surface morphology in complex processing scenarios, and finally improve the process quality and consistency of the gold foil finished product. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 FIG. is a schematic flowchart of the implementation of a control processing method based on gold foil processing provided by an embodiment of the present invention.

[0009] Figure 2 FIG. is a schematic diagram of the hardware entity of a control processing system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0010] An embodiment of the present invention provides a control processing method based on gold foil processing, and this method can be executed by a processor of a control processing system. For example, a computer system, a processing system of a gold foil processing device, etc. As Figure 1 shown, this method includes the following steps:

[0011] A control processing method based on gold foil processing, the method includes:

[0012] Step S100: Collect real-time processing parameters of the target material in the gold foil processing equipment, and generate an initial processing parameter set, where the initial processing parameter set includes a hot pressing strength gradient sequence and an extension rate distribution curve.

[0013] The hot pressing intensity gradient sequence refers to the sequence of pressure intensity changes applied by the hot pressing device at different time nodes during the gold foil processing. This sequence is collected by a pressure sensor array and contains the hot pressing intensity values and their time interval parameters for multiple consecutive processing stages. The stretching rate distribution curve characterizes the movement rate distribution law of the transmission system of the stretching device in different processing areas, which is jointly obtained by a displacement sensor and a speed measurement module, and includes the lateral stretching rate reference value and the longitudinal stretching rate change slope. The generation process of the initial processing parameter set is as follows: Dynamical physical property data of the target material during the hot pressing forming stage are captured in real time through a distributed sensing network. Among them, the multi-band pressure fluctuation detection technology is used to collect the hot pressing intensity gradient sequence, and the time series characteristics and spatial distribution characteristics of the pressure intensity are recorded synchronously; the stretching rate distribution curve is constructed based on the real-time rate sampling points obtained by a high-precision laser velocimeter, and a continuous and smooth rate distribution curve is generated through the cubic spline interpolation algorithm. During the implementation process, each data node of the hot pressing intensity gradient sequence includes a pressure intensity value, an action timestamp, and a spatial coordinate identifier, forming a multi-dimensional parameter matrix with spatio-temporal correlation; the stretching rate distribution curve maps the movement rates of different transmission shafts to a unified time reference coordinate system through normalization processing to ensure the coordination of the stretching rates in each area. Optionally, the acquisition frequency of the hot pressing intensity gradient sequence is not less than 1000 Hz, and the spatial resolution of the stretching rate distribution curve reaches the 0.1 mm level. The initial processing parameter set formed thereby can accurately reflect the real-time operating state of the processing equipment.

[0014] As an implementation manner, in step S100, real-time processing parameters of the target material in the gold foil processing equipment are collected to generate an initial processing parameter set, including:

[0015] Step S110: The lattice structure feature data of the target material in the pretreatment stage are synchronously collected through a multi-channel sensor. The lattice structure feature data include the grain size distribution map and the grain boundary density parameter.

[0016] A multi-channel sensor is a sensor that can collect data from multiple channels simultaneously and can be used to obtain various information of the target material in the pretreatment stage. The lattice structure feature data reflect the lattice structure characteristics inside the target material. Among them, the grain size distribution map shows the distribution of grain sizes in the target material, and grains of different sizes present different characteristics in the map; the grain boundary density parameter represents the density of grain boundaries in the target material. By synchronously collecting these data through a multi-channel sensor, the lattice structure characteristics of the target material in the pretreatment stage can be comprehensively understood. For example, for a certain gold foil material, the multi-channel sensor can simultaneously collect its grain size distribution map, showing the distribution range of grain sizes and the proportion of grains of different sizes, and can also obtain the grain boundary density parameter, reflecting the density of grain boundaries.

[0017] For example, the grain size distribution map refers to a visualization map of the microstructure of a metal crystal obtained by the combined scanning of an X-ray diffractometer and an electron backscatter diffraction system. This map records the diameter distribution data of different grains in the form of a two-dimensional matrix, and each matrix element corresponds to the equivalent diameter value of the grain at a specific coordinate point. The grain boundary density parameter characterizes the ratio of the total length of grain boundaries within a unit area to the surface area of the material, and is quantitatively calculated through the secondary electron imaging technology of a scanning electron microscope. The synchronous acquisition mechanism of the multi-channel sensor is specifically implemented as follows: at the initial moment when the target material enters the preprocessing stage, a data acquisition system based on the time-triggered protocol is started, so that the sampling clock signals of the X-ray diffraction channel, the electron backscatter channel, and the optical microscopy channel are kept strictly synchronized, ensuring that the spatial coordinate alignment accuracy of the grain size data and the grain boundary position data reaches the micron level. During the implementation process, an adaptive grid division algorithm is used to construct the grain size distribution map, dividing the material surface into several detection units, and local peak detection and neighborhood clustering analysis are performed within each unit to generate a grain size heat map with spatial continuity; the calculation of the grain boundary density parameter identifies the grain boundary contour line through the Sobel edge detection operator, and combines morphological dilation operations to eliminate measurement errors caused by microscopic pores. Optionally, the sampling resolution of the grain size distribution map is set to 0.5μm / pixel, and the update frequency of the grain boundary density parameter matches the material strain rate during the preprocessing stage.

[0018] Step S120: According to the peak intervals in the grain size distribution map, divide the target material into at least two sub-material clusters, and assign corresponding initial hot pressing gradient ranges and extension rate reference values to each sub-material cluster.

[0019] Exemplarily, the peak interval refers to the range of grain diameter values with the highest frequency of occurrence in the grain size distribution map, and the boundary of the diameter interval with a cumulative probability exceeding 70% in the distribution curve is identified through a probability density function fitting algorithm. The sub-material cluster division process uses an adaptive partitioning technique based on the DBSCAN spatial clustering algorithm to aggregate detection units with similar grain sizes and continuous spatial positions into independent control regions. The initial hot pressing gradient range is defined as the upper and lower limits of the pressure intensity required to be applied during the hot pressing forming stage for each sub-material cluster, and its value is dynamically calculated according to the grain size-pressure sensitivity curve: for sub-material clusters with smaller grain sizes, since the grain boundary slip resistance is higher, the initial hot pressing gradient range is set to a higher pressure interval (e.g., 200 - 300 MPa); conversely, sub-material clusters with larger grain sizes use a lower pressure interval (e.g., 150 - 250 MPa). The assignment of the extension rate reference value follows the grain boundary density-extension rate correlation rule and is determined by looking up a pre-stored empirical parameter table. For example, when the grain boundary density of a sub-material cluster exceeds the critical threshold, the extension rate reference value is proportionally reduced to prevent grain boundary cracking. For example, when the grain boundary density is 15μm / μm 2corresponding to the reference rate of 0.8 m / s and a density of 10 μm / μm 2 It is increased to 1.2 m / s during implementation. During the implementation process, the boundary contour of the sub-material clusters is optimized by the Voronoi diagram generation algorithm to ensure that the width of the transition region between adjacent clusters does not exceed two grain diameters, avoiding sudden changes in processing parameters.

[0020] Step S130: Extract the historical hot pressing strength compensation parameters and the extension rate correction coefficients that match each sub-material cluster from the preset parameter library.

[0021] The preset parameter library is a database that pre-stores various parameters related to material processing. The historical hot pressing strength compensation parameters refer to the parameters recorded for compensating the hot pressing strength for different materials or material clusters during previous gold foil processing; the extension rate correction coefficient is a coefficient used to correct the extension rate. According to the characteristics of each sub-material cluster, the historical hot pressing strength compensation parameters and the extension rate correction coefficients that match it are extracted from the preset parameter library to adjust the hot pressing strength and the extension rate during subsequent processing. For example, for a specific sub-material cluster, the preset parameter library may record that the hot pressing strength compensation parameters required for this material cluster during previous processing are within a specific numerical range, and the extension rate correction coefficient is a specific coefficient value. For example, the historical hot pressing strength compensation parameters are stored in the process knowledge base module of a relational database, and its data structure includes fields such as sub-material cluster identifier, average grain size, pressure compensation amount, and effective timestamp. The matching process uses a multi-dimensional feature similarity retrieval algorithm: taking the skewness, kurtosis, and spatial coefficient of variation of the grain size distribution of the current sub-material cluster as the query vector, searching for the historical record with the smallest Euclidean distance in the preset parameter library, and extracting its corresponding hot pressing strength compensation parameters. The extraction of the extension rate correction coefficient is based on a fuzzy logic reasoning mechanism: establishing a fuzzy rule base between the grain boundary density, the reference value of the extension rate, and the correction coefficient. For example, when it is detected that the fluctuation amplitude of the grain boundary density exceeds 5%, the "high fluctuation - strong correction" rule is triggered, and the correction coefficient is increased from the reference value of 1.0 to 1.25. Optionally, the timeliness weight of the historical parameters is calculated by an exponential decay function to ensure that the influence weight of recent processing data on the correction coefficient accounts for more than 60%, thus adapting to the time-varying characteristics of material properties.

[0022] Step S140: Perform time-domain filtering on the historical hot pressing strength compensation parameters to generate an optimized hot pressing strength gradient sequence corresponding to each sub-material cluster.

[0023] Exemplarily, the time-domain filtering process can adopt a hybrid filtering scheme that combines a sliding window weighted average algorithm and a Kalman prediction model. The specific implementation process is as follows: Arrange the historical hot pressing strength compensation parameters in a time series. Perform median filtering within a sliding window of length N (N takes 20 - 30 sampling points) to remove impulse noise, and then use a first-order Kalman filter to predict the pressure compensation trend at the next moment. The generation of the optimized hot pressing strength gradient sequence follows the principle of pressure superposition: linearly superpose the filtered compensation parameters with the lower limit value of the initial hot pressing gradient range. For example, when the initial range is 200 - 300 MPa and the compensation parameter is +15 MPa, an optimized gradient sequence of 215 - 315 MPa is generated. During the dynamic adjustment process, the time interval of the gradient sequence is adaptively adjusted according to the thermal conductivity of the sub-material cluster: for the gold foil material with a higher thermal conductivity, the gradient sequence adopts dense time nodes (interval 0.5 s); otherwise, it is extended to an interval of 1.0 s to prevent the accumulation of thermal stress.

[0024] Step S150: Perform frequency-domain smoothing processing on the stretching rate correction coefficient to generate an optimized stretching rate distribution curve corresponding to each sub-material cluster.

[0025] Frequency-domain smoothing processing is an operation to smooth the signal in the frequency domain. It can reduce the high-frequency components in the signal and make the signal smoother. Performing frequency-domain smoothing processing on the stretching rate correction coefficient can eliminate the high-frequency fluctuations in the stretching rate correction coefficient and make the adjustment of the stretching rate more stable. Through this processing, an optimized stretching rate distribution curve corresponding to each sub-material cluster is generated, and this curve can more accurately reflect the change of the stretching rate of each sub-material cluster during the processing. For example, after frequency-domain smoothing processing, the stretching rate correction coefficient with original high-frequency fluctuations becomes more stable, and the generated optimized stretching rate distribution curve is smoother, which can better guide the stretching processing of gold foil. For example, frequency-domain smoothing processing converts the correction coefficient sequence to the frequency domain through the fast Fourier transform, and uses a Butterworth low-pass filter to suppress the high-frequency fluctuation components. The cut-off frequency is set to 1 / 3 of the mechanical resonance frequency of the stretching device to avoid exciting vibration modes. The reconstruction process of the optimized stretching rate distribution curve adopts a method that combines the inverse Fourier transform and cubic spline interpolation: on the basis of retaining the low-frequency trend components, perform the inverse transform on the filtered frequency-domain signal to obtain the time-domain reference curve, and then insert interpolation points to make the smoothness of the curve meet the C² continuity condition. For special material characteristics, the smoothing processing introduces an adaptive weight mechanism: when it is detected that the reference value of the stretching rate of the sub-material cluster exceeds 1.5 m / s, the cut-off frequency of the filter is automatically reduced to enhance the high-frequency suppression ability and prevent surface ripple defects caused by high-speed stretching.

[0026] Step S160: According to the optimized hot pressing strength gradient sequences and optimized stretching rate distribution curves of each sub-material cluster, fuse and generate an initial processing parameter set.

[0027] Fuse the optimized hot pressing strength gradient sequence and the optimized extension rate distribution curve obtained after processing each sub-material cluster, that is, integrate these data together to form an initial processing parameter set containing the hot pressing strength gradient sequence and the extension rate distribution curve. This set synthesizes the processing parameter information of each sub-material cluster and can provide a comprehensive and accurate basis for subsequent gold foil processing control. For example, the optimized hot pressing strength gradient sequences and the optimized extension rate distribution curves of different sub-material clusters are merged and sorted according to certain rules to finally form a complete initial processing parameter set. For example, the fusion generation process adopts an integrated method combining a spatio-temporal registration algorithm and a conflict resolution strategy. First, the optimized parameters of each sub-material cluster are mapped to a unified processing coordinate system through a coordinate transformation matrix to ensure the precise alignment of the time nodes of the hot pressing strength gradient sequence and the spatial positions of the extension rate distribution curve. The conflict resolution strategy mainly deals with two types of anomalies: when the hot pressing gradient difference between adjacent sub-material clusters exceeds the safety threshold, an intermediate gradient segment is inserted to make it smoothly connect; when there is a rate jump in the extension rate distribution curve at the cluster boundary, a Sigmoid function is used to generate a gradual transition region. The finally generated initial processing parameter set is stored in the form of a multi-dimensional tensor, including four index axes: the time dimension, the spatial dimension, the hot pressing strength value, and the extension rate value, supporting real-time retrieval and dynamic update. During the implementation process, the verification of the parameter set is executed through a digital twin simulation system: the initial parameters are input into the virtual processing model, and the thickness distribution and surface uniformity indexes are simulated and calculated. If the results exceed the expected range, the parameter rollback mechanism is triggered to re-execute the optimization process of steps S140 - S150.

[0028] As an implementation manner, when determining the first control model and the second control model, the following steps are included:

[0029] Step S10: Input the optimized hot pressing strength gradient sequence into a preset recurrent neural network model, and update the hidden layer weights through multi-period training iterations to generate the hot pressing strength compensation function in the first control model. The hot pressing strength compensation function is the target model in the first control model.

[0030] The preset recurrent neural network model is a neural network with feedback connections that can process sequential data and capture temporal information in the data. The optimized hot pressing intensity gradient sequence is obtained through time-domain filtering and more accurately reflects the variation law of the hot pressing intensity of each sub-material cluster during the processing. When the optimized hot pressing intensity gradient sequence is input into the recurrent neural network model, the model processes and learns the input data. During the multi-cycle training process, by continuously iteratively updating the weights of the hidden layer, the model can better fit the relationship between the hot pressing intensity and the thickness change. Finally, the hot pressing intensity compensation function in the first control model is generated. This function can predict the intensity value that needs to be compensated according to the current hot pressing intensity situation to ensure that the gold foil thickness meets the expectation. For example, in a gold foil processing scenario, when the optimized hot pressing intensity gradient sequence is input into the recurrent neural network model, after multiple training cycles, the model generates a hot pressing intensity compensation function that can accurately calculate the hot pressing intensity compensation value required to reach the target thickness according to the real-time change of the hot pressing intensity.

[0031] Step S20: Input the optimized extension rate distribution curve into the convolutional neural network model, and adjust the convolutional kernel parameters through the backpropagation algorithm to generate the extension rate correction mapping table in the second control model; the extension rate correction mapping table is the target model in the second control model.

[0032] The convolutional neural network model is good at processing data with spatial structures and can automatically extract features from the data. The optimized extension rate distribution curve is obtained through frequency-domain smoothing and more accurately reflects the variation of the extension rate of each sub-material cluster during the processing. When the optimized extension rate distribution curve is input into the convolutional neural network model, the model performs convolutional operations on the data using the convolutional kernel to extract the feature relationship between the extension rate and the surface uniformity. Through the backpropagation algorithm, the parameters of the convolutional kernel are adjusted according to the output error of the model, enabling the model to better learn the non-linear association rules between the extension rate and the surface uniformity. Finally, the extension rate correction mapping table in the second control model is generated. This table records the surface uniformity correction information corresponding to different extension rates and can be used to guide the adjustment of the extension rate. For example, in actual processing, when the optimized extension rate distribution curve is input into the convolutional neural network model and the convolutional kernel parameters are adjusted through the backpropagation algorithm, an extension rate correction mapping table is generated. According to this table, the extension rate can be adjusted according to different surface uniformity situations.

[0033] Step S30: After the recurrent neural network model and the convolutional neural network model converge, extract the key feature vectors from the hot pressing intensity compensation function and the extension rate correction mapping table respectively.

[0034] After the recurrent neural network model and the convolutional neural network model have undergone multiple rounds of training, their output errors gradually decrease and tend to stabilize, that is, they reach the convergence state. After convergence, key feature vectors need to be extracted from the hot pressing strength compensation function and the stretching rate correction mapping table. The key feature vectors are vectors that can represent the core features and performance of the model. By extracting these vectors, the adaptability of the model can be better evaluated and used for subsequent similarity matching and other operations. For example, after the model converges, through specific algorithms and methods, key feature vectors that can reflect the relationship between hot pressing strength and thickness change and the relationship between stretching rate and surface uniformity are extracted from the hot pressing strength compensation function and the stretching rate correction mapping table.

[0035] As an implementation manner, in step S30, the key feature vectors in the hot pressing strength compensation function and the stretching rate correction mapping table are extracted respectively, including:

[0036] Step S31: Traverse the weight distribution of each hidden layer node in the recurrent neural network model, extract the weight change trajectory that has a temporal correlation with the hot pressing strength gradient sequence, and generate the first weight matrix.

[0037] The weight distribution of the hidden layer nodes in the recurrent neural network model reflects the degree of importance the model attaches to different input data during the learning process. Traverse the weight distribution of each hidden layer node to find the weights that have a temporal correlation with the hot pressing strength gradient sequence. The change trajectory of these weights can reflect the influence of hot pressing strength on the model output at different time points. Organize these weight change trajectories to generate the first weight matrix. This matrix records the weight information related to the hot pressing strength gradient sequence, providing a basis for subsequent analysis of the characteristics of the hot pressing strength compensation function. For example, in the recurrent neural network model, by traversing the weights of the hidden layer nodes one by one, select those weights that are related to the temporal change of the hot pressing strength gradient sequence, and form the first weight matrix with their change trajectories.

[0038] Step S32: Perform a sliding correlation analysis on the first weight matrix with adjacent time windows to identify the target weight subset that maintains a stable fluctuation pattern within multiple training cycles.

[0039] The sliding correlation analysis with adjacent time windows is a method for analyzing the correlation of data in adjacent time intervals. Perform this analysis on the first weight matrix. By calculating the correlation between weights within different time windows, find those weights that maintain a stable fluctuation pattern within multiple training cycles. These weights constitute the target weight subset, which has an important stable influence on the output of the hot pressing strength compensation function. For example, perform a sliding correlation analysis on the first weight matrix, calculate the correlation coefficient of the weights within different time windows, and select the weights with a stable correlation coefficient and a consistent fluctuation pattern within multiple training cycles to form the target weight subset.

[0040] Step S33: Screen out the activated weight nodes directly connected to the output end of the hot pressing strength compensation function from the target weight subset, and extract their activation directions and amplitude data to generate a first feature vector set.

[0041] The target weight subset contains multiple weight nodes, but not all nodes directly affect the output of the hot pressing strength compensation function. Screen out the activated weight nodes directly connected to the output end of the hot pressing strength compensation function from the target weight subset. The activation of these nodes directly determines the output value of the hot pressing strength compensation function. Extract the activation directions (i.e., whether the weight is positive or negative, reflecting the enhancement or inhibition effect on the output) and amplitude data (i.e., the magnitude of the weight, reflecting the degree of influence on the output) of these activated weight nodes, and combine these data into a first feature vector set. This set can accurately describe the key features of the hot pressing strength compensation function. For example, in the target weight subset, find out those activated weight nodes directly connected to the output end of the hot pressing strength compensation function, record their activation directions and amplitudes, and generate a first feature vector set.

[0042] Step S34: Synchronously traverse the activation region distributions of the convolution kernels in the convolutional neural network model, extract the activation response patterns spatially associated with the extension rate distribution curve, and generate a second weight matrix.

[0043] When the convolution kernels in the convolutional neural network model process the input data, activation responses will be generated in different regions. Synchronously traverse the activation region distributions of the convolution kernels to find the activation response patterns spatially associated with the extension rate distribution curve. These patterns reflect the feature extraction situation of the convolution kernels when processing the extension rate distribution curve. Organize these activation response patterns to generate a second weight matrix. This matrix records the convolution kernel activation information related to the extension rate distribution curve, providing a basis for subsequent analysis of the features of the extension rate correction mapping table. For example, in the convolutional neural network model, simultaneously traverse the activation regions of each convolution kernel, screen out those activation response patterns related to the spatial features of the extension rate distribution curve, and form them into a second weight matrix.

[0044] Step S35: Perform cross-channel activation peak detection on the second weight matrix to locate the target convolution kernel groups repeatedly triggered in the key decision path of the extension rate correction mapping table.

[0045] Cross-channel activation peak detection is a method for detecting the activation peaks of data on different channels. This detection is performed on the second weight matrix. By analyzing the activation peaks on different channels, convolutional kernel groups that are repeatedly triggered in the key decision path of the extended rate correction mapping table are located. These target convolutional kernel groups play an important role in determining the output result of the extended rate correction mapping table, and their repeated triggering indicates their key influence on the relationship between the extended rate and surface uniformity. For example, cross-channel activation peak detection is performed on the second weight matrix to find convolutional kernel groups that have activation peaks repeatedly in the key decision steps of the extended rate correction mapping table, and they are located as target convolutional kernel groups.

[0046] Step S36: Separate the activation response sequences synchronized with the change of the surface uniformity index from the target convolutional kernel group, and extract the spatial frequency and intensity distribution data thereof to generate a second set of feature vectors.

[0047] The activation responses of multiple convolutional kernels are included in the target convolutional kernel group, but not all responses are synchronized with the change of the surface uniformity index. Separate the activation response sequences synchronized with the change of the surface uniformity index from the target convolutional kernel group. These sequences reflect the characteristics of the convolutional kernels when processing surface uniformity-related information. Extract the spatial frequency (i.e., the change frequency of the activation response in space) and intensity distribution data (i.e., the distribution of the intensity of the activation response in space) of these activation response sequences, and combine these data into a second set of feature vectors. This set can accurately describe the key characteristics of the relationship between the extended rate correction mapping table and surface uniformity. For example, in the target convolutional kernel group, screen out the activation response sequences synchronized with the change of the surface uniformity index, record their spatial frequency and intensity distribution, and generate a second set of feature vectors.

[0048] Step S37: Align the first set of feature vectors and the second set of feature vectors across the feature dimensions of the models respectively to eliminate the feature scale differences caused by different neural network structures.

[0049] Since the first set of feature vectors and the second set of feature vectors come from a recurrent neural network model and a convolutional neural network model respectively, these two models have different structures and feature extraction methods, resulting in possible differences in the feature dimensions and scales of the two sets of feature vectors. In order to be able to effectively compare and analyze them, it is necessary to align them across the feature dimensions of the models. Through specific algorithms and methods, adjust the dimensions and scales of the two sets of feature vectors to eliminate the feature scale differences caused by different neural network structures. For example, use methods such as normalization and standardization to process the first set of feature vectors and the second set of feature vectors to make their feature dimensions and scales consistent.

[0050] Step S38: According to the aligned feature vector set, extract the key feature vectors that are dynamically matched with the real-time processing parameters from the hot pressing strength compensation function and the stretching rate correction mapping table respectively.

[0051] After completing the cross-model feature dimension alignment of the first feature vector set and the second feature vector set, according to the aligned feature vector set, further extract the key feature vectors that are dynamically matched with the real-time processing parameters from the hot pressing strength compensation function and the stretching rate correction mapping table. The real-time processing parameters will change continuously during the gold foil processing process. These key feature vectors can dynamically adjust the output of the model according to the real-time parameters to better meet the requirements of the processing process. For example, combining the real-time hot pressing strength and stretching rate data, screen out the key feature vectors from the aligned feature vector set that can accurately reflect the current processing state, and use them for the dynamic adjustment of the hot pressing strength compensation function and the stretching rate correction mapping table respectively.

[0052] Step S40: Perform similarity matching between the key feature vectors and the lattice structure feature data collected in real time to verify the adaptability of the first control model and the second control model.

[0053] The key feature vectors represent the core features of the first control model and the second control model, while the lattice structure feature data collected in real time reflects the current lattice structure characteristics of the target material. Perform similarity matching between the key feature vectors and the lattice structure feature data collected in real time, and judge whether the first control model and the second control model are adapted to the current state of the target material by calculating the similarity index between the two, such as cosine similarity, etc. If the similarity is high, it means that the model can better reflect the processing characteristics of the target material and the adaptability is good; otherwise, it means that the model may need to be further adjusted. For example, calculate the cosine similarity between the key feature vectors and the lattice structure feature data collected in real time. If the similarity reaches a certain threshold, it is considered that the model is adapted; if the similarity is low, the model needs to be re-evaluated.

[0054] Step S50: When the similarity matching result is lower than the preset threshold, readjust the training cycles of the recurrent neural network model and the convolutional neural network model until the key feature vectors meet the adaptability conditions.

[0055] The preset threshold is a similarity critical value set in advance, which is used to determine whether the model is suitable for the target material. When the similarity matching result is lower than the preset threshold, it indicates that the first control model and the second control model have poor adaptability to the target material. At this time, it is necessary to readjust the training cycles of the recurrent neural network model and the convolutional neural network model. By increasing or decreasing the training cycles, the training intensity and the number of parameter updates of the model are changed, so that the model can better learn the characteristics of the target material. Continuously repeat the similarity matching process until the similarity between the key feature vector and the lattice structure feature data collected in real time reaches the preset threshold, that is, the adaptability condition is satisfied. For example, when the similarity matching result is lower than the preset threshold, increase the training cycles of the recurrent neural network model and the convolutional neural network model, and perform training and similarity matching again until the adaptability requirements are met.

[0056] Step S200: Determine a first control model suitable for the target material according to the hot pressing strength gradient sequence in the initial processing parameter set. The first control model includes the dynamic mapping relationship between the hot pressing strength and the thickness change.

[0057] The hot pressing strength gradient sequence reflects the change of the hot pressing strength during the processing. By analyzing and processing the hot pressing strength gradient sequence in the initial processing parameter set, a first control model suitable for the target material is determined. This model describes the dynamic mapping relationship between the hot pressing strength and the change of the gold foil thickness, that is, how the change of the hot pressing strength affects the change of the gold foil thickness. For example, when the hot pressing strength increases, the thickness of the gold foil may decrease accordingly. The first control model can accurately reflect this dynamic change relationship, so as to provide a basis for adjusting the hot pressing strength according to the change of the gold foil thickness in the subsequent process.

[0058] Step S300: Determine a second control model suitable for the target material according to the extension rate distribution curve in the initial processing parameter set. The second control model includes the non-linear correlation rule between the extension rate and the surface uniformity.

[0059] The extension rate distribution curve shows the change of the extension rate of the gold foil during the extension process. By studying and analyzing the extension rate distribution curve in the initial processing parameter set, a second control model suitable for the target material is determined. This model includes the non-linear correlation rule between the extension rate and the surface uniformity of the gold foil, that is, how the change of the extension rate affects the surface uniformity of the gold foil in a non-linear way. For example, when the extension rate is too fast, the surface of the gold foil may become uneven. The second control model can describe this complex non-linear relationship so as to adjust the extension rate according to the requirement of the surface uniformity during the processing.

[0060] Step S400: Based on the real-time thickness deviation data and surface uniformity fluctuation data during the gold foil processing, activate the target model in the first control model or the second control model, and generate a dynamic adjustment instruction through the target model.

[0061] The real-time thickness deviation data refers to the deviation data between the actual thickness and the expected thickness of the gold foil during the gold foil processing; the surface uniformity fluctuation data refers to the fluctuation of the surface uniformity of the gold foil during the processing. According to these real-time data, it is judged whether to activate the target model in the first control model or the second control model. If the real-time thickness deviation data exceeds a certain threshold, it indicates that there is a large deviation in the gold foil thickness, and at this time, the target model in the first control model is activated; if the surface uniformity fluctuation data exceeds a certain threshold, it indicates that there is a large fluctuation in the surface uniformity of the gold foil, and at this time, the target model in the second control model is activated. After activating the target model, a dynamic adjustment instruction is generated according to the rules and algorithms in the model to adjust the relevant parameters of the gold foil processing equipment. For example, when the real-time thickness deviation data exceeds the threshold, the hot pressing intensity compensation function in the first control model is activated, and a dynamic adjustment instruction for adjusting the hot pressing intensity is generated according to this function.

[0062] As an implementation manner, step S400, based on the real-time thickness deviation data and surface uniformity fluctuation data during the gold foil processing, activate the target model in the first control model or the second control model, and generate a dynamic adjustment instruction through the target model, includes:

[0063] Step S410: Monitor the thickness deviation trend line and surface uniformity oscillation amplitude during the gold foil processing in real time.

[0064] The thickness deviation trend line reflects the change trend of the gold foil thickness deviation over time. By monitoring it in real time, the development direction and degree of the gold foil thickness deviation can be understood. The oscillation amplitude of the surface uniformity represents the fluctuation range of the gold foil surface uniformity within a certain period of time. Monitoring this amplitude in real time can promptly detect abnormal changes in the gold foil surface uniformity. For example, the thickness and surface uniformity data of the gold foil are collected in real time through thickness sensors and surface detection devices, the thickness deviation trend line is plotted, and the oscillation amplitude of the surface uniformity is calculated. During specific implementation, for example, the following method can be referred to: continuously collect the longitudinal thickness data of the gold foil at a sampling frequency not lower than 100 Hz, and form a thickness deviation time series curve after moving average filtering processing. This curve takes the process standard thickness value as the reference line and displays the thickness deviation of each detection point in real time. The oscillation amplitude of the surface uniformity is quantitatively calculated through a multi-spectral imaging system. A high-resolution linear array CCD camera is installed at the outlet of the stretching device to capture the distribution of the reflected light intensity on the gold foil surface at a rate of 50 frames per second. The surface texture consistency parameters are extracted through the gray-level co-occurrence matrix algorithm, and combined with the oscillation frequency components obtained by wavelet transform decomposition to generate a uniformity oscillation waveform containing amplitude values and phase angles. During the monitoring process, the time window length of the thickness deviation trend line is set to the processing period of the most recent 30 seconds, and the calculation window of the oscillation amplitude of the surface uniformity is a 10-second moving average period to ensure that the detection data can reflect the immediate working conditions and avoid noise interference.

[0065] Step S420: When the thickness deviation trend line exceeds the first warning threshold and the oscillation amplitude of the surface uniformity is within the preset stable range, activate the hot pressing intensity compensation function in the first control model.

[0066] The first warning threshold is a preset critical value of the thickness deviation. When the thickness deviation trend line exceeds this threshold, it indicates that the gold foil thickness deviation has reached the level that needs to be adjusted. And the oscillation amplitude of the surface uniformity being within the preset stable range indicates that the gold foil surface uniformity is currently in a normal state. In this case, activate the hot pressing intensity compensation function in the first control model, which is used to compensate and adjust the hot pressing intensity according to the thickness deviation situation. For example, when the thickness deviation trend line exceeds the first warning threshold and the oscillation amplitude of the surface uniformity is within the preset stable range, start the hot pressing intensity compensation function in the first control model to prepare for adjusting the hot pressing intensity.

[0067] Step S430: Generate a dynamic pressure adjustment instruction for the hot pressing device according to the gradient compensation rule in the hot pressing intensity compensation function.

[0068] The gradient compensation rule in the hot pressing strength compensation function stipulates the specific method of how to adjust the hot pressing strength according to the thickness deviation. According to this rule, combined with the current thickness deviation data, a dynamic pressure adjustment instruction for the hot pressing device is generated. This instruction includes the pressure value that the hot pressing device needs to adjust and the adjustment method to ensure that the thickness of the gold foil can be restored to the expected range. For example, the hot pressing strength compensation function calculates the value of the hot pressing strength that needs to be increased or decreased based on the current thickness deviation, converts this information into a dynamic pressure adjustment instruction, and sends it to the hot pressing device. For example, the gradient compensation rule is encoded as a three-dimensional look-up table structure, whose input dimensions include the current pressure gradient value, the slope of the thickness deviation trend line, and the cumulative deviation amount, and the output is the compensation coefficient matrix of each pressure zone. The specific process of generating the dynamic pressure adjustment instruction is as follows: After normalizing the real-time thickness deviation data, it is input into the compensation function, and the compensation pressure value of each pressure zone is obtained through interpolation calculation, and it is converted into an opening adjustment instruction of the hydraulic servo valve in combination with the response characteristic curve of the hot pressing device. For example, when the thickness deviation in the central area is detected to reach +4μm, the compensation function generates an instantaneous pressure compensation amount of -8MPa in the corresponding zone, and the adjacent zones are weighted and allocated a gradient compensation value of -4MPa to -2MPa according to the distance. The instruction transmission adopts a timestamp synchronization protocol to ensure that the adjustment actions of each pressure zone are synchronously executed within ±5ms, avoiding stress unevenness caused by pressure wave propagation.

[0069] Step S440: When the oscillation amplitude of the surface uniformity exceeds the second warning threshold and the thickness deviation trend line is within the preset stable range, activate the extension rate correction mapping table in the second control model.

[0070] The second warning threshold is a preset critical value of the oscillation amplitude of the surface uniformity. When the oscillation amplitude of the surface uniformity exceeds this threshold, it indicates that there are large fluctuations in the surface uniformity of the gold foil. And the thickness deviation trend line being within the preset stable range indicates that the thickness of the gold foil is currently in a normal state. In this case, activate the extension rate correction mapping table in the second control model, which is used to correct and adjust the extension rate according to the surface uniformity fluctuations. For example, when the oscillation amplitude of the surface uniformity exceeds the second warning threshold and the thickness deviation trend line is within the preset stable range, start the extension rate correction mapping table in the second control model to prepare for adjusting the extension rate.

[0071] Step S450: Generate a dynamic rate adjustment instruction for the extension device according to the rate distribution rule in the extension rate correction mapping table.

[0072] The rate distribution rule in the extension rate correction mapping table stipulates the specific method of how to adjust the extension rate according to the surface uniformity fluctuation. According to this rule, combined with the current surface uniformity fluctuation data, a dynamic rate adjustment instruction for the extension device is generated. This instruction contains the rate value that the extension device needs to adjust and the adjustment method to ensure that the surface uniformity of the gold foil can be restored to the expected range. For example, the extension rate correction mapping table calculates the extension rate value that needs to be increased or decreased according to the current surface uniformity fluctuation, converts this information into a dynamic rate adjustment instruction, and sends it to the extension device. Exemplarily, the rate distribution rule is stored as two-dimensional polar coordinate grid data, and each grid cell is associated with a rate correction coefficient for a specific azimuth angle interval and the position of the transmission chain. The generation process of the dynamic rate adjustment instruction is to input the amplitude and phase information of the surface uniformity oscillation detected in real time into the correction mapping table, calculate the target rate value of each drive unit through bilinear interpolation, and then convert it into the speed-torque curve of the servo motor according to the dynamic model of the transmission system. For example, when it is detected that the uniformity amplitude exceeds the standard in the 90-degree azimuth angle area, the correction mapping table generates a rate increase instruction of +0.2 m / s in this area, and the adjacent 120-degree area is synchronously adjusted by -0.1 m / s to form a rate gradient balance. The instruction is sent using a hierarchical verification mechanism. First, the motion controller verifies the mechanical feasibility of the instruction to exclude dangerous instructions that exceed the upper limit of the motor torque or the transmission belt speed ratio limit.

[0073] Step S460: If both the thickness deviation trend line and the surface uniformity oscillation amplitude exceed the warning threshold, first execute the dynamic pressure adjustment instruction generated by the first control model, and delay the execution of the adjustment instruction of the second control model until the hot pressing device completes the correction operation.

[0074] When both the thickness deviation trend line and the surface uniformity oscillation amplitude exceed their respective warning thresholds, it indicates that there are relatively large problems with both the thickness and surface uniformity of the gold foil. Since the thickness problem may be more critical to the quality of the gold foil, the dynamic pressure adjustment instruction generated by the first control model is first executed to adjust the hot pressing intensity to quickly restore the thickness of the gold foil. After the hot pressing device completes the correction operation, the adjustment instruction of the second control model is executed to adjust the extension rate to improve the surface uniformity of the gold foil. For example, when both the thickness deviation trend line and the surface uniformity oscillation amplitude exceed the warning threshold, first adjust the pressure of the hot pressing device according to the dynamic pressure adjustment instruction generated by the first control model. After the hot pressing device completes the pressure adjustment and the thickness of the gold foil tends to be stable, then adjust the rate of the extension device according to the adjustment instruction generated by the second control model.

[0075] Step S500: Input the dynamic adjustment instruction into the actuator of the gold foil processing equipment to correct the pressure gradient of the hot pressing device or the rate distribution of the extension device.

[0076] The dynamic adjustment instructions are instructions for adjusting processing parameters generated based on real-time data and control models during the gold foil processing. These instructions are input into the actuator of the gold foil processing equipment, and the actuator corrects the pressure gradient of the hot pressing device or the rate distribution of the stretching device according to the instructions. For example, the dynamic pressure adjustment instructions of the hot pressing device are input into the actuator of the hot pressing device, and the actuator adjusts the pressure of each area in the hot pressing device according to the instructions to make the pressure gradient meet the adjustment requirements; the dynamic rate adjustment instructions of the stretching device are input into the actuator of the stretching device, and the actuator adjusts the rates at different positions of the stretching device according to the instructions to correct the rate distribution. Exemplarily, for the dual-overlimit condition handling, a priority arbitration strategy is adopted, and a quantitative influence coefficient matrix of thickness deviation and surface uniformity is established. When the influence coefficient of the thickness deviation exceeds 1.5 times that of the uniformity, it is determined that the hot pressing correction is the priority task. The specific implementation of the delayed execution mechanism includes: temporarily storing the rate adjustment instructions generated by the second control model in the circular buffer, and at the same time starting the monitoring thread for the completion of the hot pressing correction. When the pressure gradient feedback by the hot pressing device reaches 95% of the target value and remains stable for more than 3 seconds, the stretching adjustment instructions in the buffer are released. During the waiting period, the system continuously monitors the surface uniformity state. If its deterioration speed exceeds the preset safety threshold (such as an increase of more than 10% per minute), the hot pressing correction process is interrupted and the emergency collaborative adjustment mode is started. In the collaborative mode, the hot pressing compensation amount is reduced by 50% to free up system resources to synchronously execute part of the stretching rate adjustment to prevent the overall processing failure caused by the correction of a single parameter.

[0077] As an implementation manner, in step S500, inputting the dynamic adjustment instructions into the actuator of the gold foil processing equipment to correct the pressure gradient of the hot pressing device or the rate distribution of the stretching device includes:

[0078] Step S510: According to the gradient compensation value in the dynamic pressure adjustment instructions, adjust the instantaneous pressure values of each pressure partition in the hot pressing device in stages.

[0079] The gradient compensation value in the dynamic pressure adjustment instruction is used to compensate for the numerical value of the hot pressing strength deviation. According to this value, the instantaneous pressure values of each pressure zone in the hot pressing device are adjusted in stages. The staged adjustment can avoid adverse effects on the gold foil caused by excessive pressure adjustment, making the pressure adjustment smoother and more accurate. For example, when the dynamic pressure adjustment instruction requires increasing the pressure of a certain pressure zone, according to the gradient compensation value, the pressure increase process is divided into several stages, and a certain pressure value is increased in each stage to gradually achieve the required pressure adjustment effect. Exemplarily, the gradient compensation value is stored in the form of a three-dimensional pressure tensor, including an axial pressure component, a radial pressure component, and a time-phase parameter. The staged adjustment process adopts a pressure wavefront propagation control strategy, dividing the working surface of the hot pressing device into a 5×5 grid-like pressure zone, and the instantaneous pressure value of each zone is incrementally adjusted according to the corresponding element in the compensation value matrix. The adjustment stage is divided into a preheating period, a main adjustment period, and a stabilization period: in the preheating period, 10% of the compensation value is tentatively loaded, and the duration is set to 1.2 times the response delay time of the hydraulic system (usually 200 - 300 ms); in the main adjustment period, the remaining 90% of the compensation value is applied three times according to an exponential decay curve, and the loading interval for each time is 1 / 4 of the system mechanical resonance period (about 50 ms) to avoid exciting pressure oscillations; in the stabilization period, the final pressure value is maintained until the thickness feedback data meets the standard. For example, when the dynamic pressure adjustment instruction requires increasing the pressure value of zone A3 by 15 MPa, the system applies 1.5 MPa in the preheating period and loads 4.5 MPa, 6.0 MPa, and 3.0 MPa respectively three times in the main adjustment period to ensure the continuity of the second derivative of the pressure rise curve.

[0080] Step S520: After each pressure adjustment, collect the real-time feedback data of the gold foil thickness and calculate the deviation from the predicted value of the hot pressing strength compensation function.

[0081] After each adjustment of the pressure of the hot pressing device, the real-time thickness data of the gold foil is collected through a thickness sensor to obtain the real-time feedback data of the gold foil thickness. This data is compared with the predicted value of the hot pressing strength compensation function, and the deviation between the two is calculated. Through the deviation calculation, it can be understood whether the pressure adjustment has achieved the expected effect and whether further pressure adjustment is required. For example, after a pressure adjustment, the real-time thickness data of the gold foil is collected, compared with the thickness value predicted by the hot pressing strength compensation function, and the thickness deviation value is calculated.

[0082] Step S530: If the deviation value continues to decrease, maintain the current pressure gradient adjustment direction until the target thickness range is reached.

[0083] If the deviation value continues to decrease, it indicates that the current pressure gradient adjustment direction is correct and is developing in the direction of restoring the gold foil thickness to the target thickness range. In this case, maintain the current pressure gradient adjustment direction and continue the pressure adjustment until the thickness of the gold foil reaches the target thickness range. For example, after several consecutive pressure adjustments, if the calculated thickness deviation value gradually decreases, it indicates that the pressure adjustment is effective. Continue the pressure adjustment in the current adjustment direction until the gold foil thickness meets the requirements. Exemplarily, the determination criterion for continuous deviation reduction is that the absolute value decrement rate of the deviation in three consecutive sampling periods exceeds 5%, and the direction consistency coefficient is greater than 0.85. The maintenance of the pressure gradient adjustment direction is achieved by locking the servo valve control current. The update period of the control signal is extended to three times the normal period (about 300 ms). At the same time, start the pressure micro-oscillation suppression algorithm, and superimpose a high-frequency disturbance signal of ±0.5 MPa (frequency 50 Hz) near the target pressure value to eliminate the hysteresis effect caused by static friction. The boundary conditions of the target thickness range are set to ±1.5 times the standard deviation of the process standard value. When the actual thickness value enters this interval, the system automatically switches to the pressure maintenance mode and adjusts the working state of the hydraulic system to constant pressure output.

[0084] Step S540: If the deviation value shows an increasing trend, reverse the pressure gradient and re-trigger the parameter iteration of the hot pressing strength compensation function.

[0085] If the deviation value shows an increasing trend, it indicates that the current pressure gradient adjustment direction may be incorrect, and the pressure gradient needs to be reversed. At the same time, re-trigger the parameter iteration of the hot pressing strength compensation function, that is, recalculate the parameters of the hot pressing strength compensation function to find a more appropriate pressure adjustment method. For example, during the pressure adjustment process, if it is found that the thickness deviation value gradually increases, at this time, change the pressure adjustment direction, recalculate the parameters of the hot pressing strength compensation function, and perform the pressure adjustment again. Exemplarily, the warning trigger condition for deviation increase is that the deviation growth rate in two adjacent sampling periods exceeds 10%, or the direction consistency coefficient is less than -0.7. The reverse adjustment strategy adopts a hybrid algorithm combining Bang-Bang control and fuzzy logic: First, perform an emergency reverse compensation, and reverse the current pressure gradient value by 200% within 50 ms; then start the fuzzy inference engine to generate an accurate reverse adjustment coefficient based on the deviation increase rate, die temperature change rate, and material yield strength parameters. The parameter iteration process resets the training period counter of the hot pressing strength compensation function, adopts an incremental learning mechanism, inserts the current working condition data as a new sample at the head of the training set, and recalculates the hidden layer weight matrix. During the iteration process, the learning rate is increased to five times the initial value to accelerate the model convergence.

[0086] Step S550: Dynamically adjust the transmission speed of the stretching device in different sections according to the speed distribution parameters in the dynamic rate adjustment instruction.

[0087] The rate distribution parameters in the dynamic rate adjustment instruction specify the transmission rates that the stretching device needs to adjust in different sections. According to these parameters, the transmission rates of the stretching device in different sections are dynamically adjusted to make the rate distribution of the stretching device meet the adjustment requirements. For example, if the dynamic rate adjustment instruction requires increasing the stretching rate in a certain section and decreasing the stretching rate in another section, the transmission rates of the stretching device in the corresponding sections are adjusted according to these requirements. Exemplarily, the rate distribution parameters are encoded as an angular velocity - linear velocity mapping table in the polar coordinate system, and each section corresponds to the target linear velocity value within a specific azimuth angle range. The dynamic adjustment process uses a multi-axis synchronous interpolation algorithm. The drive system is decomposed into a main drive shaft and six driven shafts. The main shaft operates at the reference rate, and the driven shafts calculate the phase difference compensation amount according to the mapping table parameters. During the adjustment process, the acceleration of each drive shaft is limited within 5 m / s² to prevent the gold foil from wrinkling due to inertial overshoot. For example, when the dynamic rate adjustment instruction requires the linear velocity in the 120 - 150 degree section to increase by 0.3 m / s, the system first calculates the angular acceleration requirement of the corresponding driven shaft in this section, and then realizes a ramp-up of the speed in real time through the electronic gear ratio to ensure that the speed difference between adjacent sections does not exceed 0.1 m / s.

[0088] Step S560: During the rate adjustment process, the uniformity change of the gold foil surface is detected in real time by an optical sensor, and the correction coefficient in the stretching rate correction mapping table is dynamically updated according to the detection result.

[0089] During the process of adjusting the rate of the stretching device, an optical sensor is used to detect the uniformity change of the gold foil surface in real time. The optical sensor can obtain the uniformity data of the gold foil surface by detecting information such as the reflected light or transmitted light on the gold foil surface. According to the detection result, the correction coefficient in the stretching rate correction mapping table is dynamically updated, so that the correction coefficient can more accurately reflect the relationship between the current gold foil surface uniformity and the stretching rate, thereby better guiding the adjustment of the stretching rate. For example, when the optical sensor detects that the uniformity of a certain area on the gold foil surface becomes worse, according to this detection result, the correction coefficient corresponding to this area in the stretching rate correction mapping table is adjusted to adjust the stretching rate of this area.

[0090] As an implementation manner, in step S560, the uniformity change of the gold foil surface is detected in real time by an optical sensor, and the correction coefficient in the stretching rate correction mapping table is dynamically updated according to the detection result. Specifically, it may include:

[0091] Step S561: The multi-spectral imaging unit continuously collects the reflected light intensity distribution of the gold foil surface in different sections of the stretching device to generate a real-time surface image data set.

[0092] The multi - spectral imaging unit can collect the reflection light intensity distribution information of the gold foil surface in multiple spectral bands. By continuously collecting the reflection light intensity distribution of the gold foil surface in different sections of the stretching device through this unit, the collected data is sorted and stored to generate a real - time surface image dataset. This dataset contains the reflection light intensity information of the gold foil surface at different positions and in different spectral bands, and can reflect the microscopic characteristics and uniformity of the gold foil surface. For example, the multi - spectral imaging unit continuously collects the reflection light intensity of the gold foil surface in each section of the stretching device over a period of time, and combines these data into a real - time surface image dataset.

[0093] Step S562: Conduct gray - scale gradient analysis on the real - time surface image dataset, extract the surface texture direction consistency parameters and specular reflection intensity dispersion indexes corresponding to each stretching section, and generate a uniformity distribution index set.

[0094] Gray - scale gradient analysis (such as the Histogram of Oriented Gradients algorithm) is used to analyze the change of gray - scale values in an image. Conducting gray - scale gradient analysis on the real - time surface image dataset can extract the surface texture direction consistency parameters and specular reflection intensity dispersion indexes corresponding to each stretching section. The surface texture direction consistency parameter reflects the degree of consistency of the surface texture direction of the gold foil. The higher the consistency, the more uniform the surface; the specular reflection intensity dispersion index represents the degree of dispersion of the specular reflection intensity on the gold foil surface. The smaller the dispersion, the more uniform the surface. These parameters and indexes are sorted to generate a uniformity distribution index set, which can comprehensively reflect the uniformity of the gold foil surface in each stretching section. For example, through gray - scale gradient analysis, the surface texture direction consistency parameters and specular reflection intensity dispersion indexes of the gold foil surface in each stretching section are calculated, and these data are combined into a uniformity distribution index set.

[0095] Step S563: Compare the uniformity distribution index set with the reference uniformity template stored in the stretching rate correction mapping table section by section, and identify the abnormal stretching sections where the surface uniformity deviation exceeds the threshold under the current rate distribution.

[0096] The reference standard of the pre-set gold foil surface uniformity is the reference standard of the benchmark uniformity template stored in the extension rate correction mapping table. Compare the set of uniformity distribution indicators with the benchmark uniformity template section by section, that is, compare the uniformity indicators of each extension section on the gold foil surface with the benchmark template, and calculate the surface uniformity deviation of each section. When the surface uniformity deviation of a certain section exceeds the pre-set threshold, this section is identified as an abnormal extension section. For example, compare the surface texture direction consistency parameter and the specular intensity dispersion index of each extension section in the set of uniformity distribution indicators with the benchmark uniformity template, calculate the deviation, and if the deviation of a certain section exceeds the threshold, mark it as an abnormal extension section. As an exemplary implementation, the benchmark uniformity template can be derived from a statistical model under historical optimal process parameters and stored as a two-dimensional look-up table containing azimuth angle intervals, direction consistency reference values, and specular intensity dispersion reference values. The comparison process uses a dual difference analysis method: first calculate the absolute deviation between the real-time direction consistency parameter and the reference value, and mark it as direction abnormal when the deviation exceeds 15%; synchronously calculate the relative change rate of the specular intensity dispersion, and mark it as reflection abnormal when it exceeds 20%. The comprehensive determination logic for abnormal extension sections is: when a certain section meets both the direction abnormal and reflection abnormal conditions, or a single abnormal indicator exceeds twice the threshold, it is determined as a severely deviated section. For example, in the 60-120° section, the detected direction consistency deviation reaches 32% and the specular dispersion change rate is 18%. Although it does not meet the dual abnormal standard, since the direction deviation exceeds twice the threshold (15% × 2 = 30%), it is still identified as an abnormal section.

[0097] Step S564: According to the deviation direction and amplitude of the abnormal extension section, match the corresponding rate compensation mode from the historical correction records, and adjust the correction coefficient weight associated with the abnormal extension section in the extension rate correction mapping table based on the rate compensation mode.

[0098] The deviation direction and amplitude of the abnormal extension section reflect the degree and direction of the deviation of the surface uniformity of this section from the reference standard. Based on this information, search for a matching rate compensation pattern from the historical correction records, which store the rate adjustment patterns used in the past when dealing with similar surface uniformity problems. Based on the matched rate compensation pattern, adjust the correction coefficient weight associated with the abnormal extension section in the extension rate correction mapping table to change the extension rate adjustment method of this extension section. For example, if the deviation of the surface uniformity of the abnormal extension section is caused by too fast an extension rate, find the corresponding rate reduction compensation pattern from the historical correction records, and adjust the correction coefficient weight associated with this abnormal extension section in the extension rate correction mapping table to reduce the extension rate of this extension section. Exemplarily, the historical correction records are stored in the process knowledge graph of a relational database, and each record contains an abnormal type code (direction deviation / reflection deviation / composite deviation), a deviation amplitude range (0-15% / 15-30% / 30%+), and a hash value of the effective compensation pattern. The matching process uses a similarity retrieval algorithm based on a random forest: input the direction consistency deviation rate, the change rate of the reflection dispersion, and the azimuth position of the current abnormal section into a pre-trained matching model, and output the three closest historical cases. The specific operation of the rate compensation pattern adjustment is as follows: select the historical case with the highest matching degree, scale its compensation coefficient weight according to the current deviation amplitude ratio, and perform weighted fusion with the original weight in the extension rate correction mapping table. For example, when the historical case shows a 25% direction deviation and a +0.12 correction coefficient is used, and a 30% deviation is detected currently, the correction coefficient is increased to +0.15 (0.12×30 / 25 = 0.144≈0.15), and the weight value of this section in the mapping table is adjusted from 1.0 to 1.15.

[0099] Step S565: After the correction coefficient weight is adjusted, trigger a local rate fine-tuning operation of the extension device, and synchronously update the acquisition period of the real-time surface image dataset to verify whether the corrected surface uniformity distribution reaches the target convergence range.

[0100] After adjusting the correction coefficient weights associated with the abnormal extension section in the extension rate correction mapping table, the extension device is triggered to perform a local rate fine-tuning operation on this abnormal extension section, that is, to make a small adjustment to the extension rate of this extension section. At the same time, the acquisition period of the real-time surface image dataset is updated synchronously to obtain the corrected gold foil surface uniformity data more timely. By analyzing the updated real-time surface image dataset, it is verified whether the corrected surface uniformity distribution reaches the target convergence range, that is, whether it is close to or meets the standard of the reference uniformity template. For example, after adjusting the correction coefficient weights, the extension device fine-tunes the extension rate of the abnormal extension section, and at the same time shortens the acquisition period of the real-time surface image dataset, acquires the corrected surface image data, and analyzes whether the surface uniformity is improved. Exemplarily, the local rate fine-tuning operation is implemented by a distributed motion control system, converting the corrected extension rate correction mapping table into independent rate setting values for six process sections, and generating a trapezoidal velocity curve instruction including the target rate, acceleration limit, and transition time. The triggering mechanism adopts the hardware-in-the-loop verification method. Before the instruction is issued, the execution effect is first simulated through the digital twin system, and the actual device adjustment is triggered only when the predicted uniformity improvement rate exceeds 50%. The acquisition period update strategy is as follows: the sampling frequency is increased to 240 fps within three feedback cycles (25 ms) after the fine-tuning is executed, and then dynamically adjusted according to the slope of the uniformity change. If the improvement rate meets the expectation, it is gradually restored to the reference sampling rate. The target convergence range is defined as the composite condition that the deviation of the direction consistency parameter ≤ 5% and the change rate of the specular intensity dispersion ≤ 8%. When verifying, the double-index requirements need to be met simultaneously in three consecutive sampling cycles.

[0101] Step S566: If the target convergence range is not reached, re-extract the updated uniformity distribution index set and iteratively execute the deviation identification and correction coefficient weight adjustment until the surface uniformity indexes of all extension sections meet the preset process standards.

[0102] If the verification result shows that the corrected surface uniformity distribution does not reach the target convergence range, it means that the current adjustment is not sufficient and further adjustment is needed. Re-extract the updated uniformity distribution index set, perform deviation identification again, find out the abnormal extension sections that still have problems, and adjust the correction coefficient weights according to the new deviation situation. Repeat this process, continuously iteratively execute the deviation identification and correction coefficient weight adjustment until the surface uniformity indexes of all extension sections meet the preset process standards. For example, after one adjustment, it is found that the surface uniformity of some extension sections still does not reach the target convergence range. Re-extract the uniformity distribution index set, perform deviation analysis again, adjust the correction coefficient weights, and continue to adjust until the surface uniformity of all extension sections meets the requirements.

[0103] As an implementation manner, the method provided by the present invention further includes a feedback calibration stage after dynamically adjusting the instruction execution, and specifically performs the following operations:

[0104] Step S600: Collect the final thickness distribution data and the surface uniformity detection map after the gold foil processing is completed.

[0105] After the gold foil processing is completed, the final thickness distribution data of the gold foil is collected by a thickness detection device, and this data reflects the thickness change of the gold foil after the entire processing process. At the same time, a surface detection device is used to obtain the surface uniformity detection map of the gold foil, and this map shows the distribution of the surface uniformity of the gold foil. For example, a high-precision thickness sensor is used to measure the thickness of each position of the gold foil to obtain the final thickness distribution data; an optical detection device is used to scan the surface of the gold foil to generate the surface uniformity detection map. Exemplarily, the final thickness distribution data is obtained by a high-precision laser scanning array, which is composed of 48 groups of confocal displacement sensors, distributed at intervals of 5 mm along the gold foil transmission direction, and each sensor captures the axial thickness value at a sampling frequency of 1 kHz, and a three-dimensional thickness distribution cloud map is generated after Kalman filtering for noise reduction. The surface uniformity detection map is generated by a multi-spectral confocal microscopy system, scanning the surface of the gold foil at two wavelengths of visible light (532 nm) and near-infrared (785 nm), and combining the Moiré fringe interference technology to quantify the surface roughness parameters to form a three-dimensional topology map including azimuth coordinates, roughness values, and texture direction angles.

[0106] Step S700: Compare the final thickness distribution data with the predicted thickness range of the first control model to generate a hot pressing model error signal.

[0107] The predicted thickness range of the first control model is the gold foil thickness range predicted according to the dynamic mapping relationship between the hot pressing intensity and the thickness change. The collected final thickness distribution data is compared with this predicted thickness range, and the difference between the two is calculated to generate a hot pressing model error signal. This signal reflects the accuracy of the first control model in predicting the gold foil thickness, as well as the difference between the influence of the hot pressing intensity on the gold foil thickness in the actual processing process and the model prediction. For example, the thickness value of each position in the final thickness distribution data is compared with the predicted thickness range of the first control model, the deviation value is calculated, and these deviation values are combined into a hot pressing model error signal. Exemplarily, the comparison process uses a three-dimensional space registration algorithm to align the coordinates of the measured thickness distribution cloud map with the predicted thickness grid output by the first control model, and calculate the absolute deviation amount at each registration point.

[0108] Step S800: Perform an overlap analysis on the surface uniformity detection map and the expected uniformity template of the second control model to generate a stretching model error signal.

[0109] The expected uniformity template of the second control model is a reference standard for the surface uniformity of gold foil set according to the non-linear correlation rule between the stretching rate and the surface uniformity. The overlapping degree analysis is carried out on the surface uniformity detection map and this expected uniformity template, that is, the similarity between the two is compared, and the overlapping degree is calculated. According to the difference in the overlapping degree, a stretching model error signal is generated. This signal reflects the accuracy of the second control model in predicting the surface uniformity of gold foil, as well as the difference between the influence of the stretching rate on the surface uniformity of gold foil in the actual processing process and the model expectation. For example, the surface uniformity detection map and the expected uniformity template are compared through image analysis technology, the overlapping degree is calculated, and a stretching model error signal is generated according to the deviation of the overlapping degree. Exemplarily, the overlapping degree analysis adopts a hybrid algorithm combining phase correlation method and feature matching: first, Gabor filtering processing is performed on the surface uniformity detection map to extract the texture direction feature map; then, multi-scale pyramid matching is carried out with the expected uniformity template to calculate the similarity integral value at each scale. The generation of the stretching model error signal is, for example, that the main error component extracts the low-frequency difference component through discrete cosine transform, and the secondary error component adopts local binary pattern analysis to capture the high-frequency texture deviation. The error signal is stored in polar coordinate form, and each azimuth angle interval is associated with the radial error value, the tangential error value, and the comprehensive error intensity.

[0110] Step S900: Adjust the gradient compensation rule in the first control model according to the systematic error component in the hot pressing model error signal.

[0111] The hot pressing model error signal contains a systematic error component and a random error component. The systematic error component is the error caused by the defects of the model itself or some fixed factors in the processing process. According to the systematic error component in the hot pressing model error signal, the gradient compensation rule in the first control model is adjusted to improve the accuracy and adaptability of the model.

[0112] Step S1000: Optimize the stretching rate correction coefficient in the second control model according to the random error component in the stretching model error signal.

[0113] The random error component in the stretching model error signal is the error caused by some random factors in the processing process, such as small differences in materials, small changes in the environment, etc. According to the random error component in the stretching model error signal, the stretching rate correction coefficient in the second control model is optimized to improve the adaptability of the model to random factors and the prediction accuracy of the surface uniformity of gold foil. For example, analyze the random error component in the stretching model error signal, and adjust the stretching rate correction coefficient in the second control model according to this error situation, so that the model can better cope with the random changes in the processing process and more accurately adjust the stretching rate to ensure the surface uniformity of gold foil.

[0114] Step S1100: Synchronously update the adjusted gradient compensation rule and the extension rate correction coefficient to the historical parameter library of the initial processing parameter set.

[0115] Synchronously update the gradient compensation rule in the first control model and the extension rate correction coefficient in the second control model, which have been adjusted and optimized, to the historical parameter library of the initial processing parameter set. The historical parameter library is used to store various parameters and model information in previous processing. The updated parameters and coefficients can provide more accurate references for subsequent gold foil processing, improving the quality and efficiency of processing. For example, write the adjusted gradient compensation rule and the extension rate correction coefficient into the historical parameter library. In subsequent processing, when encountering similar materials or processing situations, these updated parameters and coefficients can be directly obtained from the historical parameter library and applied to the new processing process.

[0116] As an implementation, the process of adjusting the gradient compensation rule and the extension optimization rate correction coefficient includes the following steps:

[0117] Step S910: Identify the steady-state deviation component in the hot pressing model error signal and calculate the steady-state compensation amount by the moving window averaging method.

[0118] The steady-state deviation component in the hot pressing model error signal refers to the relatively stable deviation part in the error signal that does not change rapidly with time. Identify this steady-state deviation component through a specific signal processing method, and then use the moving window averaging method to process it. The moving window averaging method is a method for smoothing signals. By averaging the signals within a certain time window, a relatively stable value, that is, the steady-state compensation amount, is obtained. This steady-state compensation amount is used to compensate for the steady-state deviation in the hot pressing model and improve the accuracy of the model. For example, analyze the hot pressing model error signal to find the steady-state deviation component, and use the moving window averaging method to average this component within a set time window to obtain the steady-state compensation amount. Exemplarily, the identification of the steady-state deviation component is based on the time series stationarity test, and the ADF unit root test method is used to screen out the error components with significant non-periodicity. The window width of the moving window averaging method is set to an integer multiple of the processing cycle (such as 30 - 50 sampling points), and the window step size is equal to the mechanical response cycle of the hot pressing device (such as 200 ms). The calculation of the steady-state compensation amount is implemented as the weighted average of the error values within the window, and the weight coefficients are set according to the importance grading of the spatial position: the weight of the central region is 1.2, and the weight of the edge region is 0.8. For example, when the steady-state deviation of +2.8 μm is detected in the central region within five consecutive windows, the steady-state compensation amount for this region is calculated to be +3.1 MPa (such as 2.8 μm × 1.1 μm / MPa conversion coefficient).

[0119] Step S920: Adjust the reference pressure gradient in the hot pressing intensity compensation function according to the steady-state compensation amount.

[0120] According to the calculated steady-state compensation amount, adjust the reference pressure gradient in the hot pressing intensity compensation function. The reference pressure gradient is an important parameter in the hot pressing intensity compensation function, which determines the basic change trend of the hot pressing intensity during the processing. By adjusting the reference pressure gradient, the steady-state deviation in the hot pressing model can be compensated, making the adjustment of the hot pressing intensity more accurately meet the actual change requirements of the gold foil thickness. For example, increase or decrease the reference pressure gradient in the hot pressing intensity compensation function according to the steady-state compensation amount, so that the adjustment of the hot pressing intensity can better adapt to the processing conditions of the gold foil.

[0121] Step S1010: Identify the high-frequency oscillation components in the stretching model error signal and extract the effective correction components through a low-pass filter.

[0122] The high-frequency oscillation components in the stretching model error signal refer to the parts with relatively high change frequencies in the error signal. These high-frequency oscillations may be caused by some random interferences or rapidly changing factors during the processing. Identify these high-frequency oscillation components through signal processing methods, and then use a low-pass filter to process them. A low-pass filter is a filter that allows low-frequency signals to pass through while suppressing high-frequency signals. Through the low-pass filter, the effective correction components in the stretching model error signal can be extracted, that is, those low-frequency signal components that are of practical significance for the stretching rate correction. For example, perform spectral analysis on the stretching model error signal to find the high-frequency oscillation components, and use a low-pass filter to filter the signal to extract the effective correction components.

[0123] Step S1020: Adjust the rate distribution weights in the stretching rate correction mapping table according to the effective correction components.

[0124] According to the extracted effective correction components, adjust the rate distribution weights in the stretching rate correction mapping table. The rate distribution weights determine the adjustment amplitude and method of the stretching rate in different sections of the stretching device. By adjusting the rate distribution weights, the stretching rate can be more accurately adjusted according to the effective correction components in the stretching model error signal to improve the surface uniformity of the gold foil. For example, increase or decrease the rate distribution weight of a certain section in the stretching rate correction mapping table according to the effective correction components, so that the stretching rate of this section is adjusted accordingly.

[0125] Step S1110: Replace the original model parameters with the updated hot pressing intensity compensation function and stretching rate correction mapping table.

[0126] Replace the original model parameters with the adjusted hot pressing strength compensation function and the stretching rate correction mapping table, so as to update the parameters in the first control model and the second control model. The updated model parameters can more accurately reflect the relationship between the hot pressing strength and the thickness change, and between the stretching rate and the surface uniformity during the gold foil processing, thereby improving the control accuracy of the model and the quality of gold foil processing. For example, store the updated hot pressing strength compensation function and the stretching rate correction mapping table into the model to replace the original parameters, and use the updated model for control and adjustment during the subsequent processing.

[0127] As an implementation manner, the method provided by the present invention further includes performing the following preprocessing operations before the device is started, which may specifically include:

[0128] Step S1: Analyze the lattice symmetry identifier and the atomic spacing coding segment in the lattice type code corresponding to the target material to generate a lattice feature description vector.

[0129] The lattice type code corresponding to the target material is a code used to represent the lattice structure characteristics of the target material. Among them, the lattice symmetry identifier represents the symmetry property of the lattice, and the atomic spacing coding segment represents the spacing information between atoms. By analyzing the lattice type code, the lattice symmetry identifier and the atomic spacing coding segment are extracted, and these information are combined into a vector, that is, the lattice feature description vector. This vector can comprehensively describe the lattice structure characteristics of the target material and provide a basis for subsequent material matching and parameter selection. For example, analyze the lattice type code of the target material, extract the lattice symmetry identifier and the atomic spacing coding segment therein, and combine these information according to certain rules into a vector to obtain the lattice feature description vector.

[0130] Step S2: Input the lattice feature description vector into the pre-loaded material property database for multi-dimensional similarity retrieval, and screen out a candidate reference template set whose similarity to the current lattice type code exceeds the matching threshold.

[0131] The pre-loaded material property database stores the property information of various materials and related reference templates. Input the generated lattice feature description vector into this database for multi-dimensional similarity retrieval, that is, compare the similarity degree of the lattice feature description vector with the feature vectors of each material in the database from multiple angles. Screen out the material reference templates whose similarity to the current lattice type code exceeds the matching threshold to form a candidate reference template set. The matching threshold is a pre-set similarity critical value, and only the reference templates whose similarity exceeds this threshold will be selected. For example, compare the lattice feature description vector with the feature vectors of each material in the material property database, calculate the similarity, and screen out the reference templates whose similarity exceeds the matching threshold to form a candidate reference template set.

[0132] Step S3: Perform the following operations on each candidate reference template in the candidate reference template set: Extract the average thickness deviation value and the surface uniformity compliance rate from its historical application records, and calculate the comprehensive process efficiency score.

[0133] Each candidate reference template in the candidate reference template set has its historical application records, which contain some process indicators of the template in the previous gold foil processing, such as the average thickness deviation value and the surface uniformity compliance rate. The average thickness deviation value reflects the average deviation degree between the actual thickness and the expected thickness of the gold foil when using this template to process the gold foil; the surface uniformity compliance rate indicates the proportion of the surface uniformity of the gold foil reaching the process standard when using this template to process the gold foil. According to these indicators, calculate the comprehensive process efficiency score of each candidate reference template, which comprehensively considers the performance of the template in controlling the thickness and surface uniformity of the gold foil. For example, for each template in the candidate reference template set, extract the average thickness deviation value and the surface uniformity compliance rate from its historical application records, and calculate the comprehensive process efficiency score according to a certain calculation method.

[0134] Step S4: Prioritize the candidate reference template set according to the comprehensive process efficiency score, and select the candidate reference template with the highest score as the optimal reference hot pressing gradient template.

[0135] According to the calculated comprehensive process efficiency score, prioritize the candidate reference template set, and the template with a higher score is ranked more in the front. Select the candidate reference template with the highest score as the optimal reference hot pressing gradient template. This template has the best performance in controlling the thickness and surface uniformity of the gold foil and can provide the most suitable hot pressing gradient reference for the subsequent gold foil processing. For example, sort the candidate reference template set in descending order according to the comprehensive process efficiency score, and select the template ranked first as the optimal reference hot pressing gradient template.

[0136] Step S5: Extract the set of extension rate reference curves that have a process correlation with the optimal reference hot pressing gradient template from the material property database.

[0137] The material property database stores the process information of various materials, including the correlation between the hot pressing gradient and the extension rate. Extract the set of extension rate reference curves that have a process correlation with the optimal reference hot pressing gradient template from this database. These curves reflect the influence of different extension rates on the processing effect when using the optimal reference hot pressing gradient template to process the gold foil. For example, search for the extension rate reference curves related to the optimal reference hot pressing gradient template in the material property database, and form these curves into a set of extension rate reference curves.

[0138] Step S6: Verify the coordination between the rate distribution and the hot pressing gradient for each curve in the extended rate reference curve set, eliminate the abnormal curves with rate mutations or hot pressing conflicts, and generate an optimized extended rate benchmark curve set.

[0139] Verify the coordination between the rate distribution and the hot pressing gradient for each curve in the extended rate reference curve set, that is, check whether the rate distribution in the curve can work in coordination with the hot pressing gradient in the optimal reference hot pressing gradient template, and whether there are rate mutations or hot pressing conflicts. A rate mutation means that the extended rate changes significantly in a short period of time, and a hot pressing conflict means that the interaction between the hot pressing gradient and the extended rate causes problems in the processing. Eliminate the curves with these abnormal conditions, and form the remaining curves into an optimized extended rate benchmark curve set. For example, analyze each curve in the extended rate reference curve set, check for rate mutations or hot pressing conflicts, and eliminate the abnormal curves to obtain the optimized extended rate benchmark curve set.

[0140] As an implementation method, in step S6, verifying the coordination between the rate distribution and the hot pressing gradient for each curve in the extended rate reference curve set may specifically include:

[0141] Step S61: Extract the rate distribution data of a single curve to be verified from the extended rate reference curve set, and simultaneously obtain the hot pressing gradient sequence aligned with the curve to be verified in the optimal reference hot pressing gradient template.

[0142] Select a curve to be verified from the extended rate reference curve set, and extract the rate distribution data of this curve, which reflects the change of the extended rate over time or position. At the same time, obtain the hot pressing gradient sequence aligned with the curve to be verified in the optimal reference hot pressing gradient template to ensure that the rate distribution data and the hot pressing gradient sequence are corresponding in time for coordination verification. For example, select a curve from the extended rate reference curve set, extract its rate distribution data, and at the same time find the hot pressing gradient sequence corresponding to this curve in the optimal reference hot pressing gradient template.

[0143] Step S62: Mark the corresponding coordination timestamps on the curve to be verified according to the phase nodes of the hot pressing gradient sequence, and generate a rate distribution curve with time series marks.

[0144] The phase nodes of the hot pressing gradient sequence refer to the time points at which the hot pressing gradient changes significantly. According to these phase nodes, corresponding collaborative timestamp identifiers are marked on the curve to be verified, so that the curve to be verified has chronological information, and a rate distribution curve with time series marks is generated. This curve can clearly show the corresponding relationship between the extension rate and the hot pressing gradient in time, facilitating subsequent collaborative analysis. For example, according to the phase nodes of the hot pressing gradient sequence, the corresponding timestamps are marked on the curve to be verified to form a rate distribution curve with time series marks.

[0145] Step S63: Traverse each collaborative timestamp identifier in the rate distribution curve with time series marks, detect whether the rate change slope in its adjacent time interval exceeds a preset mutation threshold, and identify the abnormal time interval with rate mutation.

[0146] Traverse each collaborative timestamp identifier in the rate distribution curve with time series marks, and calculate the rate change slope in its adjacent time interval. The rate change slope represents the change rate of the extension rate in adjacent time. Compare the calculated rate change slope with the preset mutation threshold. When the slope exceeds the threshold, it indicates that there is a rate mutation in this adjacent time interval, and this time interval is identified as an abnormal time interval. For example, for each collaborative timestamp identifier in the rate distribution curve with time series marks, calculate the rate change slope between the adjacent time points before and after it. If the slope exceeds the preset mutation threshold, then mark this adjacent time interval as an abnormal time interval.

[0147] Step S64: When an abnormal time interval is identified, trace back the hot pressing gradient value at the corresponding time node in the optimal reference hot pressing gradient template, and judge whether the pressure fluctuation direction of the hot pressing gradient value has a reverse interference with the rate mutation direction.

[0148] When an abnormal time interval with rate mutation is identified, trace back the hot pressing gradient value at the corresponding time node in the optimal reference hot pressing gradient template. Analyze the pressure fluctuation direction of the hot pressing gradient value, that is, whether the hot pressing intensity increases or decreases, and whether it has a reverse interference with the rate mutation direction, that is, whether the extension rate increases or decreases. Reverse interference means that the change direction of the hot pressing gradient is contradictory to the change direction of the extension rate, which may cause problems in the processing process. For example, when an abnormal time interval is identified, check the hot pressing gradient value at the corresponding time in the optimal reference hot pressing gradient template, and judge whether the hot pressing intensity increases while the extension rate suddenly decreases, or the hot pressing intensity decreases while the extension rate suddenly increases. If there is such a contradictory situation, it is considered that reverse interference has occurred.

[0149] Step S65: If the reverse interference intensity exceeds the conflict threshold, mark the abnormal time interval as a pressure conflict area, and calculate the starting position and duration length of the pressure conflict area in the rate distribution curve.

[0150] When it is determined that the pressure fluctuation direction of the thermal pressure gradient value and the rate mutation direction produce reverse interference, the reverse interference intensity is calculated. The reverse interference intensity is compared with the conflict threshold. If it exceeds the threshold, the abnormal time interval is marked as a pressure conflict area. At the same time, the starting position and duration of the pressure conflict area in the rate distribution curve are calculated for subsequent processing. For example, the reverse interference intensity is calculated by a certain calculation method. If it exceeds the conflict threshold, the abnormal time interval is marked as a pressure conflict area, and its start time and end time in the rate distribution curve are recorded, and the duration is calculated.

[0151] Step S66: According to the starting position and duration of the pressure conflict area, a rate correction interval is defined on the curve to be verified, and an alternative rate smoothing segment is generated based on the correction records of similar conflicts in the historical parameter library.

[0152] According to the starting position and duration of the pressure conflict area, a rate correction interval is defined on the curve to be verified, which includes the pressure conflict area and a certain range around it. Based on the correction records of similar conflicts in the historical parameter library, an alternative rate smooth segment is generated. This smooth segment is used to replace the part of the rate mutation and pressure conflict in the curve to be verified, making the rate distribution smoother and more reasonable. For example, according to the start and end time of the pressure conflict area, a rate correction interval is determined on the curve to be verified, and correction records of similar conflicts are found from the historical parameter library, and an alternative rate smooth segment is generated based on these records.

[0153] Step S67: inserting the alternative rate smoothing segment into the rate correction interval to generate an optimized extended rate curve, and deleting abnormal data points in the original rate distribution curve that overlap with the pressure conflict area.

[0154] The generated alternative rate smooth segment is inserted into the rate correction interval of the curve to be verified, replacing the problematic part of the original curve to generate an optimized extended rate curve. At the same time, the abnormal data points in the original rate distribution curve that overlap with the pressure conflict area are deleted to make the optimized extended rate curve more accurate and reliable. For example, the alternative rate smooth segment is inserted into the specified position of the curve to be verified, and the abnormal data points corresponding to the pressure conflict area in the original curve are deleted to form an optimized extended rate curve.

[0155] Step S68: performing integrity check on all optimized extension rate curves that have completed the replacement insertion operation in the extension rate reference curve set, and removing residual curves that still contain uncorrected conflict areas.

[0156] Perform integrity verification on all the optimized extension rate curves obtained through substitution insertion operations in the extension rate reference curve set, and check whether there are still uncorrected conflict regions in the curves. If there are uncorrected conflict regions, it means that there are still problems with the curve, and it will be excluded. Through integrity verification, ensure that the optimized extension rate curve set only contains curves without conflicts and abnormalities. For example, conduct a detailed inspection on each optimized extension rate curve to find out whether there are still regions with rate mutations or pressure conflicts. If there are, the curve will be excluded from the set.

[0157] Step S69: Merge the optimized extension rate curves that pass the verification into an optimized extension rate reference curve set.

[0158] Merge the optimized extension rate curves that pass the integrity verification together to form an optimized extension rate reference curve set. The curves in this set have good synergy in terms of rate distribution and hot pressing gradient, and can provide a more reasonable extension rate reference for gold foil processing. For example, integrate all the optimized extension rate curves that pass the verification into a set to obtain the optimized extension rate reference curve set.

[0159] Step S7: According to the historical processing stability indicators of each optimized extension rate reference curve in the optimized extension rate reference curve set, select the optimized extension rate reference curve with the smallest fluctuation coefficient as the optimal extension rate reference curve.

[0160] Each curve in the optimized extension rate reference curve set has its historical processing stability indicator. The fluctuation coefficient is an important indicator to measure the stability of the curve. The smaller the fluctuation coefficient, the smaller the fluctuation of the curve during historical processing and the better the stability. According to the historical processing stability indicators of each optimized extension rate reference curve, select the curve with the smallest fluctuation coefficient as the optimal extension rate reference curve, which can provide the most stable extension rate reference for gold foil processing. For example, calculate the fluctuation coefficient for each curve in the optimized extension rate reference curve set and select the curve with the smallest fluctuation coefficient as the optimal extension rate reference curve.

[0161] Step S8: Align the optimal reference hot pressing gradient template and the optimal extension rate reference curve in process time sequence to generate the reference configuration of the initial processing parameter set.

[0162] Align the process timing of the optimal reference hot-pressing gradient template and the optimal extension rate reference curve, that is, ensure that the hot-pressing gradient and the extension rate are matched in time, so that the two can work together to complete the processing of the gold foil. Through the process timing alignment, a reference configuration of the initial processing parameter set is generated, which provides an initial and reasonable reference for the hot-pressing intensity and extension rate parameters for the subsequent gold foil processing. For example, align the optimal reference hot-pressing gradient template and the optimal extension rate reference curve in chronological order, and combine the aligned parameters into a reference configuration of the initial processing parameter set.

[0163] As an implementation manner, the method provided by the present invention may further include:

[0164] Step S1200: Synchronously collect the reflected light intensity distribution and the transmitted light attenuation map of the gold foil surface in the driving direction of the stretching device through a multi-angle optical scanning unit, and generate a multi-modal surface topography data set.

[0165] The multi-angle optical scanning unit can scan the gold foil surface from multiple angles, and synchronously collect the reflected light intensity distribution and the transmitted light attenuation map of the gold foil surface in the driving direction of the stretching device. The reflected light intensity distribution reflects the reflection characteristics of the gold foil surface at different angles, and the transmitted light attenuation map shows the attenuation of light when passing through the gold foil. Integrate the collected data to generate a multi-modal surface topography data set, which contains rich microstructural and topographical information of the gold foil surface. For example, the multi-angle optical scanning unit scans the gold foil surface from different angles, simultaneously obtains the reflected light intensity distribution and the transmitted light attenuation map, and combines them into a multi-modal surface topography data set.

[0166] Step S1300: Perform spatial frequency decomposition on the multi-modal surface topography data set, separate the low-frequency component representing the thickness change and the high-frequency component representing the micro-cracks, and generate a thickness distribution map and a crack feature map.

[0167] Spatial frequency decomposition is a method for processing images or data, which is used to decompose the data into components of different frequencies. Perform spatial frequency decomposition on the multi-modal surface topography data set to separate the low-frequency component representing the thickness change and the high-frequency component representing the micro-cracks. The low-frequency component reflects the overall change of the gold foil surface thickness, and the high-frequency component reflects the characteristics of the micro-cracks on the gold foil surface. According to the separated low-frequency component and high-frequency component, generate a thickness distribution map and a crack feature map respectively, which can visually display the thickness change and micro-crack situation of the gold foil surface. For example, use a spatial frequency decomposition algorithm to process the multi-modal surface topography data set, separate the low-frequency component and the high-frequency component, draw a thickness distribution map according to the low-frequency component, and draw a crack feature map according to the high-frequency component.

[0168] Step S1400: Input the thickness distribution map into the pre-trained surface defect recognition model, and detect the gradient direction consistency of the thickness mutation region by sliding the convolution kernel, so as to identify the local thickness anomaly band perpendicular to the transmission direction of the stretching device.

[0169] The pre-trained surface defect recognition model is obtained by training with a large amount of data and can identify various defects on the gold foil surface. Input the thickness distribution map into this model, and the model slides the convolution kernel on the map to detect the gradient direction consistency of the thickness mutation region. The gradient direction consistency reflects the law and continuity of the thickness change. When the gradient directions are inconsistent, there may be thickness anomalies. Through detection, the local thickness anomaly bands perpendicular to the transmission direction of the stretching device are identified, and these anomaly bands may affect the quality and performance of the gold foil. For example, input the thickness distribution map into the surface defect recognition model, the model slides the convolution kernel on the map, calculates the gradient direction of each region, finds the regions where the gradient directions are inconsistent and perpendicular to the transmission direction of the stretching device, and identifies them as local thickness anomaly bands.

[0170] Step S1500: Input the crack feature map into the surface defect recognition model, extract the included angle data between the propagation direction of the crack branch and the principal stress direction, and screen out the accelerating expansion cracks with an included angle less than the preset angle.

[0171] Input the crack feature map into the surface defect recognition model, and the model analyzes the crack feature map to extract the included angle data between the propagation direction of the crack branch and the principal stress direction. The principal stress direction refers to the direction of the main stress suffered by the gold foil during the processing. The included angle between the propagation direction of the crack branch and the principal stress direction will affect the crack propagation speed. Screen out the cracks with an included angle less than the preset angle. These cracks will accelerate the expansion under the action of the principal stress and are called accelerating expansion cracks. For example, input the crack feature map into the surface defect recognition model, the model calculates the included angle between the propagation direction of each crack branch and the principal stress direction, and screens out the cracks with an included angle less than the preset angle as accelerating expansion cracks.

[0172] Step S1600: Generate a coordinate set of the hot pressing compensation region associated with the first control model according to the width and gradient direction of the local thickness anomaly band, and calculate the pressure compensation priority weight of the corresponding region.

[0173] Determine the area that needs hot pressing compensation according to the width and gradient direction of the identified local thickness anomaly zone. Generate a coordinate set of the hot pressing compensation area associated with the first control model, which specifies the specific locations that need hot pressing compensation. At the same time, calculate the pressure compensation priority weight for the corresponding area according to the severity of the local thickness anomaly zone and its impact on the quality of the gold foil. The higher the priority weight, the more urgently this area needs hot pressing compensation. For example, determine its specific position on the gold foil surface according to the width and gradient direction of the local thickness anomaly zone, generate a coordinate set of the hot pressing compensation area, and calculate the pressure compensation priority weight for each area according to the severity of the anomaly zone.

[0174] Step S1700: Generate a coordinate set of the extension suppression area associated with the second control model according to the included angle data and branch length of the accelerating expanding crack, and calculate the rate suppression urgency level for the corresponding area.

[0175] Determine the area that needs extension suppression according to the included angle data and branch length of the selected accelerating expanding crack. Generate a coordinate set of the extension suppression area associated with the second control model, which specifies the specific locations that need extension suppression. At the same time, calculate the rate suppression urgency level for the corresponding area according to the expansion speed of the accelerating expanding crack and its harm to the quality of the gold foil. The higher the urgency level, the more urgently this area needs to suppress the extension rate. For example, determine its specific position on the gold foil surface according to the included angle data and branch length of the accelerating expanding crack, generate a coordinate set of the extension suppression area, and calculate the rate suppression urgency level for each area according to the expansion of the crack.

[0176] Step S1800: Input the coordinate set of the hot pressing compensation area and the pressure compensation priority weight into the first control model to generate a dynamic pressure compensation instruction set covering the local thickness anomaly zone.

[0177] Input the generated coordinate set of the hot pressing compensation area and the pressure compensation priority weight into the first control model. The first control model generates a dynamic pressure compensation instruction set covering the local thickness anomaly zone based on this information and the dynamic mapping relationship between the hot pressing intensity and the thickness change. This instruction set includes the pressure values and adjustment methods that need to be adjusted for each hot pressing compensation area to eliminate the local thickness anomaly zone and make the gold foil thickness more uniform. For example, input the coordinate set of the hot pressing compensation area and the pressure compensation priority weight into the first control model, and the model calculates the hot pressing intensity that needs to be increased or decreased for each area based on these data to generate a dynamic pressure compensation instruction set.

[0178] Step S1900: Input the coordinate set of the extension suppression area and the rate suppression urgency level into the second control model to generate a gradient rate decay instruction set for suppressing the accelerating expanding crack.

[0179] Input the generated extended suppression region coordinate set and rate suppression urgency level into the second control model. Based on this information and combined with the non-linear correlation rule between the extension rate and surface uniformity, the second control model generates a gradient rate decay instruction set for suppressing the accelerating expanding crack. This instruction set contains the extension rate values and adjustment methods that need to be adjusted for each extended suppression region to suppress the expansion of the accelerating expanding crack and ensure the quality of the gold foil surface. For example, input the extended suppression region coordinate set and rate suppression urgency level into the second control model, and the model calculates the extension rate that needs to be reduced for each region based on this data and generates a gradient rate decay instruction set.

[0180] Step S2000: Adjust the pressure distribution of the hot pressing device in the compensation region according to the dynamic pressure compensation instruction set, and execute the gradient rate decay instruction set to reduce the driving speed of the extending device in the suppression region.

[0181] According to the generated dynamic pressure compensation instruction set, adjust the pressure distribution of the hot pressing device in the hot pressing compensation region to make the hot pressing intensity meet the requirements of the instruction set to eliminate the local thickness abnormal band. At the same time, execute the gradient rate decay instruction set to reduce the driving speed of the extending device in the extended suppression region to suppress the expansion of the accelerating expanding crack. For example, send the dynamic pressure compensation instruction set to the actuator of the hot pressing device, and the actuator adjusts the pressure of the hot pressing device in the compensation region according to the instruction; send the gradient rate decay instruction set to the actuator of the extending device, and the actuator reduces the driving speed of the extending device in the suppression region according to the instruction.

[0182] Step S2100: After the adjustment operation is executed, re-collect the multi-modal surface topography data sets of the compensation region and the suppression region, and verify whether the gradient direction consistency of the thickness abnormal band is restored within the allowable threshold and whether the included angle of the accelerating expanding crack increases to the safe range.

[0183] After the adjustment operations of the pressure of the hot pressing device and the speed of the extending device are executed, re-use the multi-angle optical scanning unit to collect the multi-modal surface topography data sets of the compensation region and the suppression region. By analyzing the newly collected data sets, verify whether the gradient direction consistency of the local thickness abnormal band is restored within the allowable threshold, that is, whether the thickness change becomes more uniform; at the same time, verify whether the included angle of the accelerating expanding crack increases to the safe range, that is, whether the expansion of the crack is effectively suppressed. For example, re-collect the multi-modal surface topography data sets, analyze the thickness distribution map and the crack feature map, and check whether the gradient direction consistency of the local thickness abnormal band and the included angle of the accelerating expanding crack meet the requirements.

[0184] Step S2200: If the verification condition is not met, iteratively update the pressure compensation priority weight and the rate suppression urgency level, and regenerate the dynamic pressure compensation instruction set and the gradient rate decay instruction set until the anomaly is eliminated.

[0185] If the verification result shows that the gradient direction consistency of the local thickness anomaly zone has not been restored within the allowable threshold, or the included angle of the accelerating expanding crack has not increased to the safe range, it indicates that the adjustment operation is not sufficient and further adjustment is needed. Iteratively update the pressure compensation priority weight and the rate suppression urgency level, and regenerate the dynamic pressure compensation instruction set and the gradient rate decay instruction set according to the updated weight and level. Repeat this process, continuously making adjustments and validations until the anomalies of the local thickness anomaly zone and the accelerating expanding crack are eliminated. For example, adjust the pressure compensation priority weight and the rate suppression urgency level according to the verification result, regenerate the instruction set, perform the adjustment operation again, and then verify again until the anomaly is eliminated.

[0186] Step S2300: After the anomaly is eliminated, record the final adjustment parameters and the verification result into the historical parameter library, and update the local feature detection rule of the surface defect recognition model.

[0187] When the anomalies of the local thickness anomaly zone and the accelerating expanding crack are eliminated, record the final adjustment parameters, such as the pressure adjustment value of the hot pressing device and the rate adjustment value of the stretching device, and the verification result, such as the elimination situation of the thickness anomaly zone and the suppression situation of the crack expansion, into the historical parameter library. These records can provide reference for subsequent gold foil processing and help handle similar problems. At the same time, update the local feature detection rule of the surface defect recognition model according to the experience and data in this processing process, so that the model can more accurately identify and handle the defects on the gold foil surface. For example, store the final adjustment parameters and the verification result into the historical parameter library, and modify and optimize the local feature detection rule of the surface defect recognition model according to the data in the processing process.

[0188] Figure 2 The following is a schematic diagram of the hardware entity of a control processing system provided by an embodiment of the present invention, as Figure 2 shown. The hardware entity of the control processing system 1000 includes: a processor 1001 and a memory 1002. Among them, the memory 1002 stores a computer program that can run on the processor 1001, and when the processor 1001 executes the program, it implements the steps in the method of any of the above embodiments.

Claims

1. A control processing method based on gold foil processing, characterized in that: The method comprises: Synchronously collecting lattice structure characteristic data of the target material in the pre-processing stage through a multi-channel sensor, wherein the lattice structure characteristic data includes a grain size distribution map and a grain boundary density parameter; According to the peak interval in the grain size distribution map, the target material is divided into at least two sub-material clusters, and a corresponding initial hot-pressing gradient range and an extension rate reference value are assigned to each sub-material cluster; Extract historical hot-pressing strength compensation parameters and extension rate correction coefficients matching each sub-material cluster from a preset parameter library; Performing time-domain filtering on the historical heat-pressure intensity compensation parameters to generate an optimized heat-pressure intensity gradient sequence corresponding to each sub-material cluster; Performing frequency domain smoothing processing on the extension rate correction coefficient to generate an optimized extension rate distribution curve corresponding to each sub-material cluster; According to the optimized hot pressing strength gradient sequence and optimized extension rate distribution curve of each sub-material cluster, an initial processing parameter set is generated by fusion; Determining a first control model adapted to the target material according to the hot pressing intensity gradient sequence in the initial processing parameter set, wherein the first control model includes a dynamic mapping relationship between hot pressing intensity and thickness change; Determining a second control model adapted to the target material according to the extension rate distribution curve in the initial processing parameter set, wherein the second control model includes a nonlinear association rule between the extension rate and the surface uniformity; Based on the real-time thickness deviation data and surface uniformity fluctuation data during the gold foil processing, activating the target model in the first control model or the second control model, and generating dynamic adjustment instructions through the target model; The dynamic adjustment instruction is input into the actuator of the gold foil processing equipment to correct the pressure gradient of the hot pressing device or the rate distribution of the stretching device.

2. The method according to claim 1, characterized in that The determining of the first control model and the second control model includes: Inputting the optimized heat pressure intensity gradient sequence into a preset recursive neural network model, iteratively updating the hidden layer weights through multi-cycle training, and generating a heat pressure intensity compensation function in the first control model, wherein the heat pressure intensity compensation function is a target model in the first control model; Inputting the optimized extension rate distribution curve into a convolutional neural network model, adjusting the convolution kernel parameters through a back propagation algorithm, and generating an extension rate correction mapping table in the second control model; the extension rate correction mapping table is a target model in the second control model; After the recursive neural network model and the convolutional neural network model converge, respectively extracting key feature vectors in the hot pressure strength compensation function and the extension rate correction mapping table; Performing similarity matching between the key feature vector and the lattice structure feature data collected in real time to verify the compatibility of the first control model and the second control model; When the similarity matching result is lower than a preset threshold, the training cycles of the recursive neural network model and the convolutional neural network model are readjusted until the key feature vector meets the adaptability condition.

3. The method according to claim 2, characterized in that Based on the real-time thickness deviation data and surface uniformity fluctuation data in the gold foil processing process, activating the target model in the first control model or the second control model, and generating dynamic adjustment instructions through the target model, including: Real-time monitoring of the thickness deviation trend line and surface uniformity oscillation amplitude during gold foil processing; When the thickness deviation trend line exceeds a first warning threshold and the surface uniformity oscillation amplitude is within a preset stable range, activating a hot pressing intensity compensation function in the first control model; Generating a dynamic pressure adjustment instruction of the hot pressing device according to the gradient compensation rule in the hot pressing intensity compensation function; When the surface uniformity oscillation amplitude exceeds a second warning threshold and the thickness deviation trend line is in a preset stable interval, activating an extension rate correction mapping table in the second control model; Generating a dynamic rate adjustment instruction for the extension device according to the rate distribution rule in the extension rate correction mapping table; If the thickness deviation trend line and the surface uniformity oscillation amplitude both exceed the warning threshold, the dynamic pressure adjustment instruction generated by the first control model is executed first, and the adjustment instruction of the second control model is delayed until the hot pressing device completes the correction operation.

4. The method according to claim 3, characterized in that The step of inputting the dynamic adjustment instruction to the actuator of the gold foil processing equipment to correct the pressure gradient of the hot pressing device or the rate distribution of the stretching device in real time includes: According to the gradient compensation value in the dynamic pressure adjustment instruction, the instantaneous pressure value of each pressure zone in the hot pressing device is adjusted in stages; After each pressure adjustment, real-time feedback data of the thickness of the gold foil is collected and deviation calculation is performed with the predicted value of the hot pressure strength compensation function; If the deviation value continues to decrease, the current pressure gradient adjustment direction is maintained until the target thickness range is reached; If the deviation value shows an expanding trend, the pressure gradient is adjusted in the opposite direction and the parameter iteration of the thermal pressure intensity compensation function is re-triggered; Dynamically adjust the transmission speed of the extension device in different sections according to the speed distribution parameters in the dynamic speed adjustment instruction; During the rate adjustment process, the optical sensor is used to detect the change in the uniformity of the gold foil surface in real time, and the correction coefficient in the extension rate correction mapping table is dynamically updated according to the detection result.

5. The method according to claim 4, characterized in that The method also includes a feedback calibration phase after the dynamic adjustment instruction is executed, including the following operations: Collect the final thickness distribution data and surface uniformity detection map after gold foil processing; Comparing the final thickness distribution data with the predicted thickness range of the first control model to generate a hot pressing model error signal; Performing overlap analysis on the surface uniformity detection map and the expected uniformity template of the second control model to generate an extended model error signal; adjusting a gradient compensation rule in the first control model according to a system error component in the thermal pressure model error signal; Optimizing the extension rate correction coefficient in the second control model according to the random error component in the extension model error signal; The adjusted gradient compensation rule and extension rate correction coefficient are synchronously updated to the historical parameter library of the initial processing parameter set.

6. The method according to claim 5, characterized in that The process of adjusting the gradient compensation rule and the extension optimization rate correction coefficient includes: Identifying a steady-state deviation component in the thermal pressure model error signal, and calculating a steady-state compensation amount by a sliding window averaging method; adjusting the reference pressure gradient in the thermal pressure intensity compensation function according to the steady-state compensation amount; Identifying high frequency oscillation components in the extended model error signal and extracting effective correction components through a low pass filter; adjusting the rate distribution weight in the extended rate correction mapping table according to the effective correction component; The updated hot pressure strength compensation function and elongation rate correction mapping table replace the original model parameters.

7. The method according to claim 1, characterized in that The method also includes performing the following pre-processing operations before the device is started, including: Parse the lattice symmetry identifier and the atomic spacing encoding segment in the lattice type code corresponding to the target material to generate a lattice feature description vector; Inputting the lattice feature description vector into a preloaded material property database for multi-dimensional similarity search, and screening out a candidate reference template set whose similarity with the current lattice type code exceeds a matching threshold; The following operations are performed on each candidate reference template in the candidate reference template set: extracting the average thickness deviation value and surface uniformity compliance rate in its historical application records, and calculating the comprehensive process efficiency score; Prioritizing the candidate reference template set according to the comprehensive process efficiency score, and selecting the candidate reference template with the highest score as the optimal reference hot pressing gradient template; Extracting a set of extension rate reference curves that are process-related to the optimal benchmark hot-pressing gradient template from a material property database; Verify the synergy between rate distribution and thermal pressure gradient for each curve in the extension rate reference curve set, remove abnormal curves with rate mutation or thermal pressure conflict, and generate an optimized extension rate reference curve set; According to the historical processing stability index of each optimized extension rate reference curve in the optimized extension rate reference curve set, the optimized extension rate reference curve with the smallest fluctuation coefficient is selected as the optimal extension rate reference curve; The optimal reference hot pressing gradient template and the optimal extension rate reference curve are aligned in process timing to generate a reference configuration of an initial processing parameter set.

8. The method according to claim 7, characterized in that The verifying the synergy between the rate distribution and the thermal pressure gradient for each curve in the extension rate reference curve set includes: Extracting rate distribution data of a single curve to be verified from the extension rate reference curve set, and synchronously acquiring a thermal pressure gradient sequence in the optimal reference thermal pressure gradient template that is time-aligned with the curve to be verified; According to the phase nodes of the thermal pressure gradient sequence, a corresponding coordinated timestamp identifier is marked on the curve to be verified, and a rate distribution curve with a timing mark is generated; Traversing each coordinated timestamp identifier in the rate distribution curve with timing marks, detecting whether the rate change slope in the adjacent time interval exceeds a preset mutation threshold, and identifying abnormal time intervals with rate mutations; When an abnormal time interval is identified, the thermal pressure gradient value of the corresponding time node in the optimal reference thermal pressure gradient template is traced back to determine whether the pressure fluctuation direction of the thermal pressure gradient value is reversely interfering with the rate mutation direction; If the reverse interference intensity exceeds the conflict threshold, the abnormal time interval is marked as a pressure conflict area, and the starting position and duration of the pressure conflict area in the velocity distribution curve are calculated; According to the starting position and duration of the pressure conflict area, a rate correction interval is defined on the curve to be verified, and an alternative rate smoothing segment is generated based on the correction records of similar conflicts in the historical parameter library; Inserting the alternative rate smoothing segment into the rate correction interval to generate an optimized extended rate curve, and deleting abnormal data points in the original rate distribution curve that overlap with the pressure conflict area; Performing integrity check on all optimized extension rate curves that have completed the replacement insertion operation in the extension rate reference curve set, and removing residual curves that still contain uncorrected conflict areas; The optimized extension rate curves that have passed the verification are merged into the optimized extension rate reference curve set.

9. A control processing system, comprising a memory and a processor, wherein the memory stores a computer program that can be run on the processor, characterized in that: When the processor executes the program, the steps in the method according to any one of claims 1 to 8 are implemented.

Images