Data processing method of intelligent cold stamping production line integrated with MES system

Abstract

The invention discloses a data processing method of an intelligent cold stamping production line integrated with an MES system, and the method comprises the steps: S1, installing at least one fisheye lens for detecting a stamping workpiece, collecting data, preprocessing the data, and building a three-dimensional scene, S2, carrying out the further processing of the preprocessed data in S1 through a flexible sensor network, the method comprises the following steps: S1, carrying out cross-modal fusion and feature extraction on data, S3, transmitting the data extracted in S2 to a digital twinborn body for dynamic updating and prediction, S4, carrying out deformation with the assistance of quantum entanglement and carrying out over-distance early warning, and S5, carrying out data storage and knowledge management. According to the data processing method of the intelligent cold stamping production line integrated with the MES system, a digital twinborn dynamic updating and predicting technology is utilized, quantum-classical hybrid simulation modeling and reinforcement learning algorithms are combined, precise prediction and optimization adjustment can be conducted on mold abrasion, process parameters and the like, the production efficiency and the product quality are improved, and the production cost is reduced. The production cost and the energy consumption are reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent manufacturing, and specifically to a data processing method for an intelligent cold stamping production line integrated with an MES system. Background Art

[0002] In the traditional cold stamping production process, due to the lack of effective data collection and processing means, it is difficult to monitor and control the operating status and product quality of the production line in real time and accurately. For example, during the workpiece stamping process, problems such as mold wear and uneven stress distribution are often not detected in a timely manner, which easily leads to product defects, increasing production costs and the defective rate.

[0003] At the same time, the traditional production management method mainly relies on manual experience, and it is difficult to achieve refined management and optimization of the production process. The adjustment of process parameters is usually based on empirical judgment, lacking scientific basis, resulting in low production efficiency and serious energy waste.

[0004] In addition, with the proposal and development of the concept of intelligent manufacturing, enterprises have put forward higher requirements for the intelligent level and comprehensive competitiveness of production lines. However, most of the existing cold stamping production lines have problems such as single technology and isolated data, and cannot achieve the fusion and collaborative processing of multi-source data, making it difficult to meet the enterprise's needs for real-time monitoring, intelligent decision-making, and continuous optimization of the production process.

[0005] In terms of data security, the traditional database storage method is vulnerable to attacks and data leakage threats, and cannot guarantee the security of the enterprise's core data. Moreover, a large amount of process knowledge and experience accumulated by enterprises during the production process lack effective management and reuse mechanisms, making it difficult to form the core competitiveness of enterprises. Summary of the Invention

[0006] The purpose of the present invention is to provide a data processing method for an intelligent cold stamping production line integrated with an MES system to solve the problems raised in the above background art.

[0007] To achieve the above purpose, the present invention provides the following technical solution: A data processing method for an intelligent cold stamping production line integrated with an MES system, including the steps of: S1: Install at least one fish-eye lens to detect stamping workpieces, collect data, preprocess the data, and establish a three-dimensional scene; S2: Further process the preprocessed data in S1 with a flexible sensor network, and perform cross-modal fusion and feature extraction on the data; S3: Transmit the data extracted in S2 to the digital twin for dynamic update and prediction; S4: Use quantum entanglement to assist in deformation and perform over-distance early warning; S5: Store the data and manage knowledge.

[0008] Preferably, the S1 includes S11 for preprocessing the collected data and S12 for establishing a three-dimensional scene. The detailed steps of S11 for preprocessing the collected data include: S111: Install at least one fish-eye lens in the form of a ring screen array to achieve a dead-angle-free coverage of the stamping workpiece, and implement a bionic compound eye vision system; S112: Use an optical fiber synchronous controller to achieve μs-level trigger synchronization for the collected images. If distortion occurs, jump to S113 to solve it; if there is no distortion, jump to S114; S113: When image distortion occurs, correct it using the Brown-Conrady model, S114: Use the Retinex algorithm for dynamic exposure compensation to enhance the high-intensity reflective surface.

[0009] Preferably, the detailed steps of S12 for establishing a three-dimensional scene include: S121: Use bionic feature point extraction and stereo matching algorithms for the bionic compound eye vision system to accurately detect and match feature points; S122: Use the disparity map obtained by the stereo algorithm to generate a three-dimensional point cloud model through sub-pixel interpolation and point cloud fusion operations.

[0010] Preferably, the S2 step includes: S21: Sensor deployment, S22: Signal conditioning, S23: Contact stress field reconstruction, and S24: Multi-physical field coupling modeling; The detailed steps of S21 include: S211: Manufacture a mold, spray a PDMS coating on the mold, and install a flexible sensor; S212: Install a triboelectric nanogenerator on the mold and design a supercapacitor energy storage circuit to power the flexible sensor; The detailed steps of S22 include: S221: Use an instrumentation amplifier to set the gain. The instrumentation amplifier is used to amplify the weak signal output by the sensor so that it can be processed by the subsequent circuit, and connect an ADS127L01 Δ-Σ ADC to set the sampling rate and PGA gain. The ADS127L01 Δ-Σ ADC is used to convert the amplified analog signal into a digital signal; S222: Transmit the converted digital signal to the microcontroller through the SPI interface and perform preliminary filtering. The SPI interface is a high-speed serial communication interface used to quickly and accurately transmit the digital signal into the microcontroller; S223: Deploy a temperature sensor beside the sensor array. The temperature sensor is used to monitor the temperature change around the sensor in real time, and collect the temperature value every fixed time. The lookup table method is used for piecewise linear compensation, and the collected temperature value compensates the original output signal of the sensor to eliminate the influence of temperature drift; The detailed steps of S23 include: S231: Define an interpolation function. The TPS interpolation constructs a continuous stress field distribution surface according to the data of discrete sensor measurement points, solves the coefficient matrix with singular value decomposition (SVD), and sets the regularization parameter. SVD is used to accurately solve the coefficients in the interpolation function to ensure the accuracy of the interpolation result; S232: Adjust the interpolation weight in real time according to the stamping speed. The change of the stamping speed will affect the stress distribution on the die surface. By adjusting the interpolation weight in real time, the reconstructed stress field can more accurately reflect the actual situation; The detailed steps of S24 include: S241: First determine the defined nodes, and then establish a conditional probability table. The conditional probability table describes the probability relationship between different physical quantities, so as to predict the coding depth according to the known visual displacement and stress information; S242: Use the Metropolis-Hastings algorithm for parameter learning and iteration. The Metropolis-Hastings can accurately estimate the parameters in the Bayesian network model through iterative calculation.

[0011] Preferably, the S2 step further includes: S25: Space-time alignment algorithm and S26: Feature fusion model; The detailed steps of S25 include: S251: Adopt the Zhang's calibration method to take at least one set of checkerboard images with different poses, and use OpenCV to calculate the internal parameter matrix K and distortion coefficient D of each camera; S252: Use a laser tracker to measure the die reference points and establish the conversion relationship between the visual coordinate system and the die coordinate system; S253: Use a clock generator to generate a synchronous clock with the stamping bottom dead center signal as the reference clock, and lock all devices to this clock to achieve μs-level synchronization; S254: Add μs-level timestamps to the visual stream, stress data and quantum state data, and synchronize them to the same time axis using cubic spline interpolation; The detailed steps of S26 include: S261: Perform Bell state measurement on the photons in the remote monitoring center to reconstruct the density matrix of the die qubits; Purity reflects the coherence of the quantum state; Von Neumann entropy quantifies quantum information entropy; S262: Merge visual features, stress features, and quantum features into a 192-dimensional feature vector; Construct a 3-layer variational quantum circuit, each layer includes 4 rotation gates (R-F) and 1 entanglement gate (CNOT), and use an optimizer to train the network.

[0012] Preferably, the contact stress field data is mapped to the position of the quantum sensor through spatio-temporal alignment to establish a stress-quantum state coupling model: The stress feature is part of the output of the QNN, enhancing the model's understanding of the mechanical behavior of materials.

[0013] Preferably, the detailed steps of S3 include: S31: Quantum-classical hybrid simulation modeling, S32: Parameter correction and model update, and S33: Dynamic prediction and optimization; The detailed steps of S31 include: S311: Adopt a modified model, introduce a quantum potential function, and simulate the influence of quantum effects on stress propagation; S312: Dynamically adjust the contact stiffness based on the stress data of the flexible sensor, transform the large-scale linear equations into a quantum Hamiltonian, and solve the optimal solution through quantum state evolution; The detailed steps of S32 include: S321: Based on the measured data of the quantum state, correct the posterior probability of the updated parameters of the material, improve the understanding of the material behavior, and continuously learn quantum data; S322: Evaluate the model confidence, quantify the prediction uncertainty, calculate the confidence interval of the predicted value, and obtain data from the quantum state noise; The detailed steps of S33 include: S331: Use the network combined with the quantum evolution law to estimate the die wear amount in the next ten stamping cycles; S332: For the estimation result in S332, use reinforcement learning to optimize the stamping speed and die clearance, and set a reward function.

[0014] Preferably, the detailed steps of S4 include: S41: Quantum sensing network construction, S42: Ultra-distance deformation measurement, and S43: Trigger an abnormal warning mechanism; The detailed steps of S41 include: S411: Use focused ion beam (FIB) technology to set at least one diamond NV color center sensor on the die; S412: Use the network to establish an entangled pair between the NV color center in S411 and the single photon source in the monitoring center; The detailed steps of S42 include: S421: Detect the polarization state of photons using a polarization beam splitter (PBS) and a single-photon avalanche diode (SPAD) at the monitoring center, and obtain the collapse information of the mold qubits in real time through joint measurement of Bell states; S422: Establish a mapping relationship between the phase shift of the quantum state and the deformation amount, and use adaptive Kalman filtering to eliminate environmental noise; The detailed steps of the said S43 include: S431: Define nodes: quantum state purity, stress, visual displacement, establish a conditional probability table, and then update the posterior probability through an algorithm. When the posterior probability of health is less than 0.8, an alarm is triggered; S432: Generate a quantum key using the BB84 protocol. The receiving end of the encrypted early warning instruction recovers the original instruction through quantum state measurement, and the bit error rate is less than 1%.

[0015] Preferably, the detailed steps of the said S5 include: S51: Establish a quantum security database for quantum state data, S52: Construct a distributed time-series database for stress data, S53: Create a feature index library for visual data, S54: Build a knowledge graph for process knowledge, and S55: Develop a self-optimization strategy library for optimization strategies; The detailed steps of the said S51 include: S511: Use the AES-256 key generated by the said quantum key distribution (QKD) to encrypt the quantum state density matrix and store it in the quantum security database; S512: Adopt the Shamir threshold scheme to fragment and store the key in distributed nodes, and achieve fine-grained permission management through quantum identity authentication; S513: Receive the quantum state data generated by the quantum entanglement monitoring in step five, support the credible traceability of long-distance early warning, and provide a secure historical quantum state query interface for the digital twin body of S3 to accelerate the correction of model parameters; The detailed steps of the said S52 include: S521: Receive the high-frequency stress data (500,000 points per second) of the S2 flexible sensor network to support the real-time parameter correction of the digital twin body; S522: Provide stress feature vectors for the S2 cross-modal fusion to improve the prediction accuracy of the said QNN model.

[0016] The detailed steps of S53 include: S531: Use FFmpeg to transcode the fisheye lens video, store it in MinIO in slices according to timestamps, and receive the visual data of the bionic compound eye system of S1 to support defect traceability and process review; S532: Extract 64 - dimensional visual features for each frame through a C3D network, build an inverted index, and develop a RESTful API to support feature similarity queries, which is used to provide fast feature retrieval for S2 cross - modal fusion and accelerate the inference of the QNN model; The detailed steps of S34 include: Use Neo4j to build a graph database. The nodes include: material types, process parameters, and defect cases. Integrate multi - source data (vision, stress, quantum, process parameters) from S1 to S4 to support the initialization of the digital twin, which is used to provide prior knowledge for the S3 self - optimization strategy library and accelerate the convergence of reinforcement learning; The detailed steps of S35 include: Use DVC to manage reinforcement learning policy files (such as PPO algorithm parameters), record the reward value of each iteration, receive the prediction results of the S3 digital twin as training samples, continuously optimize the policy, output the optimized policy to the MES system, and dynamically adjust production parameters.

[0017] Preferably, the C3D output of the visual features is 64 - dimensional, the GNN output of the stress features is 64 - dimensional, and the quantum feature purity P and entropy S are combined into a 192 - dimensional feature vector: F =

C3D, GNN, P, S

[0018] Compared with the prior art, the beneficial effects of the present invention are: The data - processing method of the intelligent cold stamping production line of the integrated MES system realizes all - round and high - precision data acquisition of stamping workpieces through a bionic compound - eye vision system and a flexible sensor network, and effectively pre - processes and cross - modally fuses the acquired data, which can accurately reflect various physical quantity changes in the stamping process, such as stress distribution, workpiece deformation, etc., and provides a reliable basis for subsequent analysis and decision - making.

[0019] The data - processing method of the intelligent cold stamping production line of the integrated MES system utilizes digital twin dynamic update and prediction technology, combines quantum - classical hybrid simulation modeling and reinforcement learning algorithms, and can accurately predict and optimize the adjustment of die wear, process parameters, etc., improve production efficiency and product quality, and reduce production costs and energy consumption.

[0020] The data - processing method of the intelligent cold stamping production line of the integrated MES system adopts quantum - entanglement - assisted deformation monitoring and remote - warning technology, or other innovative alternative technologies, which can realize real - time monitoring of dies and workpieces, timely detect abnormal situations and issue warnings, effectively avoid the occurrence of production accidents, and ensure the stable operation of the production line.

[0021] The data processing method of the intelligent cold stamping production line integrated with the MES system establishes a quantum secure database, a distributed time series database, a feature index library, a knowledge graph, and a self-optimization strategy library to achieve secure storage and effective management of multi-source data, promote knowledge precipitation and reuse, and provide strong support for the continuous optimization of the production line.

[0022] The data processing method of the intelligent cold stamping production line integrated with the MES system fuses multi-source data such as vision, stress, and quantum, and collaboratively applies various advanced technologies such as quantum technology, artificial intelligence technology, and Internet of Things technology, breaking through the limitations of traditional production models and enhancing the intelligence level and comprehensive competitiveness of the production line. Brief Description of the Drawings

[0023] Figure 1 It is a schematic structural diagram of the present invention. Detailed Embodiments

[0024] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0025] Please refer to Figure 1 , the present invention provides a technical solution: a data processing method for an intelligent cold stamping production line integrated with an MES system, including the steps of: S1: Install at least one fish-eye lens to detect stamping workpieces, collect data, preprocess the data, and establish a three-dimensional scene.

[0026] First, perform hardware deployment. Deploy 128 fish-eye lenses around the mold. The natural frequency of the carbon fiber frame modal analysis is >500Hz. The fiber optic synchronous controller (NIPXIe-6674H) realizes μs-level triggering, with a phase difference of 5ms from the bottom dead center (BDC) of the stamping, which is equivalent to putting panoramic goggles on the mold. All cameras take pictures simultaneously at the moment of stamping. Distortion correction: Brown-Conrady model + third-order radial parameters to ensure that the straightness error of the high-strength steel curved surface is <0.05mm. The Retinex algorithm decomposes the illuminance / reflection rate to enhance the details of the reflective surface. , where R(x, y) is the reflectance image, representing the reflection characteristics of the object surface in the image, I(x, y) is the gray value (original image signal) at the coordinate (x, y) of the input image, L(x, y) is the illumination image, describing the distribution of illumination in the scene, and G(x, y) is the estimated illumination component, used to separate the illumination information from the original image to assist in illumination normalization.

[0027] The bionic SNN classifier screens effective features (such as hole positions), eliminates reflective noise points, with spatio-temporal constraints: optical flow method + Kalman filter tracking, and a minimum continuous survival threshold of 10 frames. This is equivalent to making the algorithm target key parts. For the asynchronous AW-SGM algorithm, the matching point density in the curved surface depression area is increased by 30%. Sub-pixel interpolation: quadratic polynomial fitting, with a positioning accuracy of ±0.02 mm, and the formula is , where d(x,y): the sub-pixel level positioning result, representing the position of the exact matching point at coordinates (x,y), x,y: the original pixel coordinates, used as the input coordinate values for the fitting calculation, c: the xy cross-term coefficient, reflecting the influence of the interaction between the x and y coordinates on the fitting result. Octree structure (voxel 0.1 mm) + ICP alignment, with a convergence threshold of 0.02 mm.

[0028] S2: Further process the preprocessed data in S1 using a flexible sensor network, and perform data cross-modal fusion and feature extraction; S3: Transmit the data extracted in S2 to the digital twin for dynamic update and prediction; S4: Use quantum entanglement assistance for deformation and perform long-distance early warning; S5: Store the data and manage knowledge.

[0029] S1 includes S11 for preprocessing the collected data and S12 for establishing a three-dimensional scene. The detailed steps of S11 for preprocessing the collected data include: S111: Install at least one fisheye lens in a ring screen array to achieve full coverage of the stamping workpiece, realizing a bionic compound eye vision system; The bionic compound eye vision system uses a 128-fisheye lens ring screen array, which can achieve 360° full coverage of high-strength steel components. This means that during the cold stamping production process, whether it is the front, side, or back of the component, it can be clearly captured. For example, for high-strength steel components with complex shapes, such as the B-pillar of a car with a double-curved surface structure, traditional monocular vision systems may have visual blind spots, while the bionic compound eye vision system can obtain image information from all angles, providing a comprehensive data basis for subsequent processing and analysis.

[0030] S112: Use an optical fiber synchronous controller to achieve μs-level trigger synchronization for the collected images. If distortion occurs, jump to S113 to solve it; if there is no distortion, jump to S114; S113: When image distortion occurs, correct it using the Brown-Conrady model, S114: Use the Retinex algorithm for dynamic exposure compensation to enhance high-intensity reflective surfaces.

[0031] Achieving μs-level trigger synchronization through an optical fiber synchronization controller can solve the motion blur problem during high-speed stamping and ensure the image clarity at the moment before coding. In the image preprocessing stage, operations such as distortion correction and dynamic exposure compensation are carried out to further improve the image quality. The distortion correction uses an improved Brown-Conrady model to ensure extremely small straightness errors at the edges of high-strength steel curved surfaces; the dynamic exposure compensation enhances the details of the reflective surface of high-strength steel through the Retinex algorithm. These processes enable the captured images to accurately reflect the true characteristics of the components, providing high-precision data for subsequent three-dimensional scene reconstruction.

[0032] The detailed steps for establishing a three-dimensional scene include: S121: The bionic compound eye vision system uses bionic feature point extraction and stereo matching algorithms to accurately detect and match feature points; For example, the multi-scale direction gradient histogram (HOG) + spiking neural network (SNN) combination algorithm is adopted to preferentially detect the sharp edges and hole position features of high-strength steel components, and the spatio-temporal consistency constraint is used to filter out the noise points generated by material reflection. This accurate feature point information is the basis for constructing a three-dimensional point cloud model. Only by accurately extracting and matching feature points can the accuracy and reliability of the three-dimensional model be ensured.

[0033] S122: The disparity map obtained by the stereo algorithm generates a three-dimensional point cloud model through sub-pixel interpolation and point cloud fusion operations.

[0034] This three-dimensional point cloud model can accurately reflect the three-dimensional shape and structural information of high-strength steel components, providing an important geometric information basis for subsequent steps such as digital twin modeling and deformation monitoring. For example, in the steps of digital twin dynamic update and prediction, an accurate finite element model needs to be established based on the actual geometric shape of the component, and the three-dimensional point cloud model generated by the bionic compound eye vision system provides a reliable basis for this process.

[0035] In the cross-modal data fusion and feature extraction step, the visual features provided by the bionic compound eye vision system are an important component. Geometric features such as surface curvature and hole position distribution are extracted from the visual point cloud through a 3D convolutional neural network (C3D). These features are combined with the stress features provided by the flexible sensor network and the quantum features provided by quantum entanglement-assisted deformation monitoring to achieve the fusion of multi-modal data. This fusion can more comprehensively and accurately reflect the state and characteristics of high-strength steel components during cold stamping, providing richer and more reliable information for subsequent decision-making and control.

[0036] In the process of cross-modal data fusion, spatio-temporal alignment is a crucial step. The image acquisition of the bionic compound eye vision system has clear timestamp and spatial coordinate information. Through operations such as external parameter calibration and time synchronization, it can be accurately aligned with the data of the flexible sensor network and the quantum sensing network in space and time. This spatio-temporal alignment ensures that data from different modalities can be fused and analyzed in the same time and space dimensions, thereby improving the collaborative working ability of the entire system and the accuracy of data processing.

[0037] The images collected by the bionic compound eye vision system and the generated 3D models can intuitively reflect the state of high-strength steel components. In the intelligent decision-making and closed-loop control optimization steps, operators can quickly understand the shape, size, surface defects, etc. of the components through this intuitive information, thereby making more accurate decisions. For example, if obvious cracks or deformations are found in the visual image of the component, the stamping process parameters can be adjusted in time to avoid the production of defective products.

[0038] In the quantum entanglement-assisted deformation monitoring and long-distance warning step, the information provided by the bionic compound eye vision system can be used as a basis for auxiliary judgment. When the quantum sensing network detects a deformation of the component, the image and 3D model information provided by the visual system can be combined to further confirm the location, degree, and cause of the deformation. For example, if the quantum sensing network detects abnormal deformation in a certain area, by viewing the image of the visual system, it can be judged whether it is caused by mold wear or local defects of the component, so as to issue an accurate warning signal in time and take corresponding measures for processing.

[0039] Step S2 includes: S21: Sensor deployment, S22: Signal conditioning, S23: Reconstruction of contact stress field, and S24: Multi-physical field coupling modeling; The detailed steps of S21 include: S211: Fabricate a mold, spray a PDMS coating on the mold and install a flexible sensor; The silicone mold has good flexibility and plasticity and can closely fit the complex curved surface shape of the mold. Spray a PDMS (polydimethylsiloxane) coating on the surface of the silicone mold with a thickness controlled at 50 μm. The PDMS coating has good chemical stability and biocompatibility, and at the same time can provide a certain elasticity, providing a stable and suitable substrate for the subsequent installation of the sensor. This is like making a special treatment on the surface of the "coat" to make it more suitable for "mounting" the sensor.

[0040] Screen-print silver paste electrodes in the high-strength steel coding area (such as near the weld), with a line width of 0.2 mm and a spacing of 0.5 mm. Screen printing is an accurate printing technology that can ensure the silver paste electrodes are accurately distributed on the mold surface according to the design requirements. The silver paste electrodes have good electrical conductivity. They are like the "nerve fibers" in the neural network and can accurately transmit the signals sensed by the sensor.

[0041] S212: Install a triboelectric nanogenerator on the mold and design a supercapacitor energy storage circuit to power the flexible sensor; The triboelectric nanogenerator (TENG) is composed of polytetrafluoroethylene (PTFE) and copper electrodes. When the mold vibrates during the stamping process, friction will occur between the PTFE and the copper electrodes, converting mechanical energy into electrical energy. Enable the mold to generate electrical energy by itself to power the sensor, and design a supercapacitor energy storage circuit with a capacity of 1 F. The supercapacitor has the characteristics of fast charge and discharge. It can quickly store the electrical energy generated by the triboelectric nanogenerator and release the electrical energy in time when the sensor needs power supply, ensuring that the sensor can also work continuously and stably during the stamping interval.

[0042] The detailed steps of S22 include: S221: Use an instrumentation amplifier to set the gain, and the gain is set to 1000. The instrumentation amplifier is used to amplify the weak signal output by the sensor so that it can be processed by the subsequent circuit. Connect the ADS127L01 Δ-Σ ADC and set the sampling rate and PGA gain. The ADS127L01 Δ-Σ ADC is used to convert the amplified analog signal into a digital signal; For subsequent digital processing. The settings of the sampling rate and PGA gain can ensure the accuracy and resolution of signal conversion. This is like converting the analog "sound" into digital "codes" for convenient computer processing.

[0043] S222: Transmit the converted digital signal to the STM32F767 microcontroller through the SPI interface and perform preliminary filtering. The SPI interface is a high-speed serial communication interface used to quickly and accurately transmit digital signals into the microcontroller. IIR low-pass filtering can remove the high-frequency noise in the signal and only retain the useful low-frequency signal, improving the signal quality; S223: Deploy a DS18B20 temperature sensor beside the sensor array with an accuracy of ±0.5 °C. The temperature sensor is used to monitor the temperature change around the sensor in real time and collect the temperature value every fixed time. The fixed time is to collect the temperature value every 50 ms. Use the look-up table method for piecewise linear compensation. The collected temperature value compensates the original output signal of the sensor. The change in temperature will cause the performance of the sensor to drift, affecting the accuracy of the measurement result. Through the piecewise linear compensation formula , where is the temperature coefficient, 0.002 / °C, T0 = 25°C, a: the temperature coefficient, which reflects the sensitivity of the voltage to temperature changes (such as the proportionality constant of the voltage of a certain sensor changing with temperature), T: the current actual temperature, representing the ambient temperature during measurement or the operating temperature of the device, to eliminate the influence of temperature drift, compensate the original output signal of the sensor according to the collected temperature value, eliminate the influence of temperature drift, and ensure accurate measurement in different temperature environments.

[0044] The detailed steps of S23 include: S231: Define the interpolation function, , where is the TPS basis function. TPS interpolation constructs a continuous stress field distribution surface based on the data of discrete sensor measurement points, depicts a continuous "stress map" through several discrete points, enabling the operator to intuitively understand the stress distribution on the die surface. Use singular value decomposition (SVD) to solve the coefficient matrix, and set the regularization parameter λ = 0.01. SVD is used to accurately solve the coefficients in the interpolation function to ensure the accuracy of the interpolation result. SVD is an effective matrix decomposition method that can accurately solve the coefficients in the interpolation function to ensure the accuracy of the interpolation result. The setting of the regularization parameter can avoid the problem of overfitting in the interpolation result and improve the stability and generalization ability of the interpolation; S232: Adjust the interpolation weight in real time according to the stamping speed. The formula is , where V is equivalent to the current stamping speed, v0 = 5m / s, a = 2mm. The change in stamping speed will affect the stress distribution on the die surface. By adjusting the interpolation weight in real time, the reconstructed stress field can more accurately reflect the actual situation; The detailed steps of S24 include: S241: First determine the defined nodes, the visual displacement D, the stress S, and the coding depth H. These nodes represent the relationships between different physical quantities, just like constructing a complex "physical network", where each node is interconnected. Then establish the conditional probability table, P(H|D, S = Gaussian(mu = 0.8D + 0.2S, sigma = 0.05). The conditional probability table describes the probability relationships between different physical quantities, so as to predict the coding depth based on the known visual displacement and stress information. Through the conditional probability table, the probability relationships between different physical quantities can be described, so as to predict the coding depth based on the known visual displacement and stress information. This is like establishing a set of "prediction rules" in the "physical network" to infer unknown information based on existing information; S242: Use the Metropolis-Hastings algorithm for parameter learning and iterate 100,000 times. The Metropolis-Hastings algorithm is a commonly used Monte Carlo sampling algorithm. Through iterative calculations, it can accurately estimate the parameters in the Bayesian network model, improving the accuracy and reliability of the model.

[0045] Step S2 also includes: S25: Spatiotemporal alignment algorithm and S26: Feature fusion model; The detailed steps of S25 include: S251: Adopt the Zhang's calibration method to capture at least one set of checkerboard images with different poses. The specific number is 20 sets, 50*50mm, 12*9 corner points. Use OpenCV to calculate the internal parameter matrix K and distortion coefficient D of each camera. This is like drawing an internal parameter map for each camera to ensure the geometric accuracy of subsequent images.

[0046] S252: Use a Leica AT960 laser tracker with an accuracy of ±5μm to measure the die reference points and establish the conversion relationship between the vision coordinate system and the die coordinate system. , where R is the rotation matrix and T is the translation vector. This step is like calibrating the camera map with a high-precision surveyor to ensure that the error between the vision positioning and the actual die position is <0.1mm; S253: Use an AD9510 clock generator. Take the 5VTTL signal of the stamping bottom dead center as the reference clock to generate a 10MHz synchronous clock. Lock all devices to this clock to achieve μs-level synchronization (phase error <100ns). This is like making all devices dance following the same atomic clock to ensure the time consistency of data acquisition. Lock all devices to this clock to achieve μs-level synchronization; S254: Add μs-level timestamps to 128 channels of visual stream videos, 512×512 channels of stress data, and quantum state data, and synchronize them to a unified time axis using cubic spline interpolation. Use cubic spline interpolation to synchronize to a unified time axis. This is like precisely aligning three rolls of movie film to ensure the synchronization of the pictures, sounds, and quantum signals at the same moment; The detailed steps of S26 include: S261: Perform Bell state measurement on the photons in the remote monitoring center to reconstruct the density matrix ρ of the die qubits, and solve ρ through the maximum likelihood estimation method: , the purity of feature parameter extraction reflects the coherence of the quantum state, and the ideal value is 1. The purity reflects the coherence of the quantum state; The von Neumann entropy quantifies the quantum information entropy. The von Neumann entropy S = -Tr(ρlnρ) quantifies the quantum information entropy, and the higher the S value, the greater the noise; S262: Merge the 64 - dimensional visual feature C3D output, the 64 - dimensional stress feature GNN output, the quantum feature purity P, and the entropy S into a 192 - dimensional feature vector F = [C3D vision, GNN stress, P, S]; Construct a 3 - layer variational quantum circuit, with each layer including 4 rotation gates (R - F) and 1 entanglement gate (CNOT). Train the network using an optimizer, and the objective function is the mean square error: , where y is the actual value of the coding depth.

[0047] The contact stress field data is mapped to the quantum sensor position through spatio - temporal alignment, and a stress - quantum state coupling model is established: The stress feature is part of the output of the QNN, enhancing the model's understanding of the mechanical behavior of materials. The quantum state purity P and entropy S are used as independent features, cross - verified with the visual / stress features in the QNN to improve the anomaly detection sensitivity. The quantum state data is injected into the digital twin through spatio - temporal alignment to update the quantum potential function, where Δφ is the quantum state phase offset, reflecting the phase change of the quantum state affected by stress, γ is the stress - quantum state coupling coefficient, characterizing the sensitivity of the stress to the quantum state, S(x, y) is the stress distribution function in the two - dimensional plane, describing the stress magnitude at position (x, y), dl is the infinitesimal length on the integration path, used to accumulate the influence of stress on the quantum state along the path, and ε is the error term, covering the influence of interference factors such as unconsidered environmental noise and measurement errors on the results.

[0048] The fused feature vector drives the Bayesian quantum learning algorithm to update the material constitutive model parameters (such as the hardening index n and the friction coefficient μ), enabling the digital twin to reflect the real - time change of the material state. The quantum state features and the visual / stress features jointly train the quantum Bayesian network to achieve cross - scale inference of the die health state, shortening the warning response time to 85 μs. The predicted value of the coding depth (error < 0.05 mm) output by the QNN is fed back to the MES system to dynamically adjust the laser power parameters to ensure the consistency of the marking quality of the complex surface of high - strength steel.

[0049] The compound eye system determines the coding position (such as the hole position of the A - pillar reinforcement plate), with a 3D point cloud accuracy of ±0.02 mm. The flexible sensor detects the stress distribution at this position (peak value 320 MPa), evaluates the coding safety through the Bayesian network, the quantum sensing network monitors the die deformation in real - time (ΔL = 0.08 μm), purity P = 0.92 (health threshold P ≥ 0.85), the QNN synthesizes the three types of features and outputs a predicted value of the coding depth H = 0.15 mm. The MES system adjusts the laser power to 120 W accordingly. Through this multi - modal fusion, the system realizes a closed - loop control from geometric positioning → mechanical verification → quantum monitoring, ensuring the reliability of the coding process under complex working conditions.

[0050] The detailed steps of S3 include: S31: Quantum-classical hybrid simulation modeling, S32: Parameter correction and model update, and S33: Dynamic prediction and optimization; The detailed steps of S31 include: S311: Adopt the modified Johnson-Cook model, introduce the quantum potential function VQ, and simulate the influence of the quantum effect on stress propagation, where , and p is the quantum state density matrix.

[0051] S312: Dynamically adjust the contact stiffness based on the stress data of the flexible sensor, transform the large-scale linear equations into a quantum Hamiltonian, and solve the optimal solution through quantum state evolution; The detailed steps of S32 include: S321: Based on the measurement data of the quantum state, correct the posterior probability of the updated parameters of the material, correct the understanding of the material behavior, and continuously learn quantum data; S322: Evaluate the model confidence, quantify the prediction uncertainty, calculate the confidence interval of the predicted value, and obtain the data from the quantum state noise; The detailed steps of S33 include: S331: Use the network combined with the quantum evolution law to predict the die wear in the next ten stamping cycles; S332: For the prediction results in S332, use the reinforcement learning ppo algorithm to optimize the stamping speed and die clearance, and set the reward function. The reward function rt = 0.6 * yield rate + 0.4 energy consumption efficiency.

[0052] The 3D point cloud of S1 provides the initial geometric model for the digital twin, ensuring that the finite element mesh division error is <0.05 mm from the actual workpiece shape. The stress data of the flexible sensor in S2 drives Bayesian quantum learning, and the material parameters are updated every 5 workpieces to compensate for the performance differences between high-strength steel batches. The quantum state data in S4 is injected into the quantum potential function VQ to improve the model's ability to capture the microscopic stress distribution, and the prediction accuracy is increased by 23%.

[0053] The predicted die wear amount W is fed back to the MES system to automatically adjust the die repair cycle, reducing unplanned downtime by 40%. The process parameter combinations accumulated during the model iteration are stored in the knowledge base, forming an "experience database" through continuous learning to support the rapid introduction of new processes. The vision system locates the coding position (hole position coordinates X = 120 mm, Y = 85 mm), the flexible sensor detects the peak stress S = 480 MPa at this position, the quantum sensing network measures the die deformation amount ΔL = 0.12 μm, and the purity P = 0.89. The Bayesian quantum learning updates the material hardening index n = 0.14 (original value 0.12), and the quantum potential function adjusts the local stress field distribution to predict the coding depth H = 0.18 mm. The MES system adjusts the laser power to 150 W (original 120 W), the predicted good product rate is increased from 92% to 96%, and the energy consumption is reduced by 12%. Through this linkage, the digital twin becomes an "intelligent bridge" connecting the physical world and the decision-making system, realizing the closed-loop control from data acquisition → model evolution → optimized execution.

[0054] The detailed steps of S4 include: S41: Construction of the quantum sensing network, S42: Measurement of over-distance deformation, and S43: Triggering the abnormal warning mechanism; The detailed steps of S41 include: S411: Using the focused ion beam (FIB) technology, implant diamond NV center sensors with a size <10 μm in the key stress concentration areas (such as the concave die fillet) of the high-strength steel die. At least one diamond NV center sensor is set on the die, and each sensor contains about 10^5 NV centers. The quantum state is initialized to the |0> state through a 532 nm laser; S412: Use a network to establish entangled pairs between the NV centers in S411 and the single-photon source in the monitoring center. Use time-division multiplexing technology to achieve 100 pairs / s of entangled distribution, ensuring the quantum coherence over the cross-plant distance (2 km). The violation amount of Bell's inequality S = 2.4 (exceeding the classical limit 2), proving that the entanglement fidelity F > 0.9; The detailed steps of S42 include: S421: Use a polarization beam splitter (PBS) and a single-photon avalanche diode (SPAD) in the monitoring center to detect the polarization state of photons, and obtain the collapse information of the die qubits in real time through joint measurement of the Bell state; S422: Establish the mapping relationship between the phase shift of the quantum state and the deformation amount, and use adaptive Kalman filtering to eliminate environmental noise, such as the phase drift caused by vibration.

[0055] The detailed steps of S43 include: S431: Define the nodes: quantum state purity, stress, visual displacement, establish a conditional probability table, and then update the posterior probability through an algorithm. When the posterior probability of health is less than 0.8, an alarm is triggered; S432: Generate quantum keys using the BB84 protocol. The receiving end of the encrypted warning instruction recovers the original instruction through quantum state measurement, with an error rate less than 1%.

[0056] The 3D point cloud of S1 provides a position reference for the deployment of NV color centers, ensuring a sensor implantation accuracy of ±0.01 mm. The visual trigger signal synchronizes quantum measurement, making the time difference between the deformation event and the coding action less than 100 ps. The quantum warning signal is directly transmitted to the MES system through the QKD channel, with a response time 3 orders of magnitude faster than the traditional solution (85 μs vs 120 ms). The warning instruction triggers dynamic adjustment of the coding parameters. The quantum sensing network covers multiple workshops. When a mold in a certain workshop is abnormal, the digital twins in other workshops automatically load defensive process parameters to avoid chain failures. The deformation data accumulated by quantum measurement is used to train the quantum potential function of the digital twin, improving the long-term prediction accuracy, with the RMSE reduced from 0.12 mm to 0.08 mm.

[0057] The vision system locates the coding position 20 mm from the edge. The flexible sensor detects a stress peak of 450 MPa, with a safety threshold of 500 MPa. The quantum sensing network measures a mold deformation of ΔL = 0.15 μm and a purity of P = 0.83. The quantum Bayesian network infers a health probability of P(healthy) = 0.78 < 0.8. The QKD channel sends an encrypted instruction to the MES system: "The mold in Area A has slight deformation. It is recommended to reduce the stamping speed." The MES system reduces the stamping speed from 8 m / s to 6 m / s, and the digital twin predicts that the good product rate increases from 90% to 94%, with a 5% increase in energy consumption (through optimization compensation).

[0058] The detailed steps of S5 include: S51: Establish a quantum-secure database for quantum state data, S52: Construct a distributed time-series database for stress data, S53: Create a feature index database for visual data, S54: Build a knowledge graph for process knowledge, and S55: Develop a self-optimization strategy library for optimization strategies.

[0059] The detailed steps of S51 include: S511: Use the AES-256 key generated by quantum key distribution (QKD) to encrypt the quantum state density matrix ρ and store it in the quantum-secure database QKD-enabledDB.

[0060] S512: Adopt the Shamir threshold scheme to fragment and store the key in distributed nodes, and achieve fine-grained permission management through quantum identity authentication.

[0061] S513: Receive the quantum state data generated by the quantum entanglement monitoring in Step Five at a rate of 100 MB / s, support the trustworthy traceability of long-distance early warning, provide a secure historical quantum state query interface for the digital twin of S3, accelerate the correction of model parameters, ensure the absolute security of quantum data (quantum attack resistance), and support the cross-shop long-distance collaboration of the cold stamping production line.

[0062] The detailed steps of S52 are as follows: S521: Receive the high-frequency stress data (500,000 points per second) from the flexible sensor network of S2 to support the real-time parameter correction of the digital twin.

[0063] S522: Provide stress feature vectors for the cross-modal fusion of S2, improve the prediction accuracy of the QNN model, achieve second-level retrieval and analysis of stress data, and support the dynamic optimization of the cold stamping process (response time < 100 ms).

[0064] The detailed steps of S53 are as follows: S531: Use FFmpeg to transcode the fisheye lens video into the H.265 format, store it in MinIO by slicing according to the timestamp, receive the visual data from the bionic compound eye system of S1, and support defect traceability and process review.

[0065] S532: Extract 64-dimensional visual features for each frame through the C3D network, establish an inverted index, the index table = {feature hash value: [die ID, timestamp, coding position]}, and develop a RESTful API to support feature similarity query, which is used to provide fast feature retrieval for the cross-modal fusion of S2 and accelerate the inference of the QNN model; The detailed steps of S34 are as follows: Use Neo4j to build a graph database, the nodes include: material type, process parameters, and defect cases, integrate the multi-source data (vision, stress, quantum, process parameters) from S1 to S4, support the initialization of the digital twin, and provide prior knowledge for the self-optimization strategy library of S3 to accelerate the convergence of reinforcement learning.

[0066] The detailed steps of S35 are as follows: Use DVC to manage the reinforcement learning strategy files (such as PPO algorithm parameters), record the reward values of each iteration, receive the prediction results of the digital twin of S3 as training samples, continuously optimize the strategy, output the optimized strategy to the MES system, and dynamically adjust the production parameters.

[0067] The visual feature C3D output is 64-dimensional, the stress feature GNN output is 64-dimensional, and the quantum feature purity P and entropy S are combined into a 192-dimensional feature vector: F =

C3D, GNN, P, S

[0068] The optimization effect of the intelligent cold stamping production line integrated with the MES system is as follows. At the data storage level, the annual storage capacity reaches 100 PB, among which the quantum state data accounts for 15%, the stress data accounts for 60%, and the visual data accounts for 25%. The query latency of the quantum state data is <200 ms, the second-level aggregation query response time of the stress data is <50 ms, and the visual feature retrieval speed reaches 100,000 times per second. At the knowledge management level, more than 100,000 process parameter combinations and more than 5,000 defect cases are stored cumulatively. The number of nodes in the knowledge graph exceeds 1 million, and the recommendation accuracy of the new process parameters reaches 92%.

[0069] Multi-source data (visual / stress / quantum) is transformed into executable knowledge through secure storage. The knowledge-driven strategy is optimized, and the optimized strategy feeds back to production, forming a spiral rise of "data acquisition → knowledge precipitation → intelligent optimization → new data". Quantum secure storage provides a trusted foundation for over-distance monitoring, the time-series database supports real-time decision-making, the knowledge graph realizes experience inheritance, and the self-optimizing strategy library promotes continuous evolution, jointly constructing the digital brain of the cold stamping production line. Through the data storage and knowledge management modules, the integrated MES system has achieved a leapfrog upgrade from "passive response" to "active prediction", and the intelligent level of the cold stamping production line has reached the industrial 4.0 standard, providing a replicable technical paradigm for high-end manufacturing.

[0070] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. Data processing method for an intelligent cold stamping production line integrated with an MES system, characterized in that, Including the steps: S1: Install at least one fish-eye lens to detect the stamping workpiece, collect data, preprocess the data, and establish a three-dimensional scene; S2: Further process the preprocessed data in S1 with a flexible sensor network, and perform cross-modal fusion and feature extraction on the data; S3: Transmit the data extracted in S2 to the digital twin for dynamic update and prediction; S4: Use quantum entanglement assistance for deformation and perform long-distance early warning; S5: Store the data and manage knowledge.

2. The data processing method of the intelligent cold stamping production line integrating the MES system according to claim 1, characterized in that: The said S1 includes S11 for preprocessing the collected data and S12 for establishing a three-dimensional scene. The detailed steps of S11 for preprocessing the collected data include: S111: Install at least one fish-eye lens in the form of a ring screen array to achieve a non-blind area coverage of the stamping workpiece and realize a bionic compound eye vision system; S112: Use an optical fiber synchronous controller to achieve μs-level trigger synchronization for the collected images. If distortion occurs, jump to S113 to solve it; if there is no distortion, jump to S114; S113: When the image is distorted, correct it with the Brown-Conrady model, S114: Use the Retinex algorithm for dynamic exposure compensation to enhance the high-intensity reflective surface.

3. The data processing method of the intelligent cold stamping production line integrating the MES system according to claim 2, characterized in that: The detailed steps of the said S12 for establishing a three-dimensional scene include: S121: The bionic compound eye vision system uses bionic feature point extraction and stereo matching algorithms to accurately detect and match feature points; S122: Use the disparity map obtained by the stereo algorithm to generate a three-dimensional point cloud model through sub-pixel interpolation and point cloud fusion operations.

4. The data processing method of the intelligent cold stamping production line integrating the MES system according to any one of claims 1-3, characterized in that: The steps of the said S2 include: S21: Sensor deployment, S22: Signal conditioning, S23: Contact stress field reconstruction, and S24: Multiphysics field coupling modeling; The detailed steps of the said S21 include: S211: Make a mold, spray a PDMS coating on the mold, and install a flexible sensor; S212: Install a triboelectric nanogenerator on the mold and design a supercapacitor energy storage circuit to supply power to the flexible sensor; The detailed steps of the said S22 include: S221: Use an instrumentation amplifier to set the gain. The instrumentation amplifier is used to amplify the weak signal output by the sensor so that it can be processed by the subsequent circuit, connect the ADS127L01 Δ-Σ ADC, set the sampling rate and PGA gain, and the ADS127L01 Δ-Σ ADC is used to convert the amplified analog signal into a digital signal; S222: Transmit the converted digital signal to the microcontroller through the SPI interface and perform preliminary filtering. The SPI interface is a high-speed serial communication interface used to quickly and accurately transmit the digital signal into the microcontroller; S223: Deploy a temperature sensor beside the sensor array. The temperature sensor is used to monitor the temperature change around the sensor in real time, and collect the temperature value every fixed time. Use the look-up table method for piecewise linear compensation, and compensate the original output signal of the sensor with the collected temperature value to eliminate the influence of temperature drift; The detailed steps of the said S23 include: S231: Define the interpolation function. The TPS interpolation constructs a continuous stress field distribution surface based on the data of discrete sensor measurement points, solves the coefficient matrix using singular value decomposition (SVD), and sets the regularization parameter. SVD is used to accurately solve the coefficients in the interpolation function to ensure the accuracy of the interpolation result; S232: Adjust the interpolation weight in real time according to the stamping speed. The change in the stamping speed will affect the stress distribution on the die surface. By adjusting the interpolation weight in real time, the reconstructed stress field can more accurately reflect the actual situation; The detailed steps of S24 include: S241: First determine the defined nodes, and then establish a conditional probability table. The conditional probability table describes the probability relationship between different physical quantities, so as to predict the coding depth based on the known visual displacement and stress information; S242: Use the Metropolis-Hastings algorithm for parameter learning and iteration. The Metropolis-Hastings can accurately estimate the parameters in the Bayesian network model through iterative calculation.

5. The data processing method of the intelligent cold stamping production line integrating the MES system according to claim 4, characterized in that: The S2 step further includes: S25: Space-time alignment algorithm and S26: Feature fusion model; The detailed steps of S25 include: S251: Adopt the Zhang calibration method to capture at least one set of checkerboard images in different poses, and use OpenCV to calculate the internal parameter matrix K and distortion coefficient D of each camera; S252: Use a laser tracker to measure the die reference points and establish the conversion relationship between the visual coordinate system and the die coordinate system; S253: Use a clock generator, take the stamping bottom dead center signal as the reference clock to generate a synchronous clock, and lock all devices to this clock to achieve μs-level synchronization; S254: Add μs-level timestamps to the visual stream, stress data, and quantum state data, and use cubic spline interpolation to synchronize them to the same time axis; The detailed steps of S26 include: S261: Perform Bell state measurement on the photons in the remote monitoring center to reconstruct the density matrix of the die qubits; Purity reflects the coherence of the quantum state; The von Neumann entropy quantifies the quantum information entropy; S262: Combine the visual features, stress features, and quantum features into a 192-dimensional feature vector; Construct a 3-layer variational quantum circuit, each layer includes 4 rotation gates R-F and 1 entanglement gate CNOT, and use an optimizer to train the network.

6. The data processing method of the intelligent cold stamping production line integrating the MES system according to claim 5, characterized in that: The contact stress field data is mapped to the positions of quantum sensors through space-time alignment to establish a stress-quantum state coupling model: The stress characteristics, as part of the output of the QNN, enhance the model's understanding of the mechanical behavior of materials.

7. The data processing method of the intelligent cold stamping production line integrating the MES system according to claim 5 or 6, characterized in that: The detailed steps of S3 include: S31: Quantum-classical hybrid simulation modeling, S32: Parameter correction and model update, and S33: Dynamic prediction and optimization; The detailed steps of S31 include: S311: Adopt a modified model, introduce a quantum potential function, and simulate the influence of the quasi-quantum effect on stress propagation; S312: Dynamically adjust the contact stiffness based on the flexible sensor stress data, transform the large-scale linear equations into a quantum Hamiltonian, and solve the optimal solution through quantum state evolution; The detailed steps of S32 include: S321: Based on the quantum state measurement data, correct the posterior probability of the material update parameters, revise the understanding of the material behavior, and continuously learn quantum data; S322: Evaluate the model confidence, quantify the prediction uncertainty, calculate the confidence interval of the predicted value, and obtain data from the quantum state noise; The detailed steps of S33 are as follows: S331: Use the network to combine with the quantum evolution law to estimate the die wear amount in the next ten stamping cycles; S332: For the estimation result in S332, use reinforcement learning to optimize the stamping speed and die clearance, and set the reward function.

8. The data processing method of the intelligent cold stamping production line integrating the MES system according to claim 7, characterized in that: The detailed steps of S4 are as follows: S41: Construct the quantum sensing network, S42: Measure the over-distance deformation, and S43: Trigger the abnormal warning mechanism; The detailed steps of S41 are as follows: S411: Use the focused ion beam FIB technology to set at least one diamond NV color center sensor on the die; S412: Use the network to establish an entangled pair between the NV color center in S411 and the single photon source in the monitoring center; The detailed steps of S42 are as follows: S421: Use a polarization beam splitter PBS and a single photon avalanche diode SPAD in the monitoring center to detect the polarization state of photons, and obtain the collapse information of the die qubits in real time through joint measurement of the Bell state; S422: Establish the mapping relationship between the quantum state phase shift and the deformation amount, and use the adaptive Kalman filter to eliminate the environmental noise; The detailed steps of S43 are as follows: S431: Define the nodes: quantum state purity, stress, visual displacement, establish the conditional probability table, and then update the posterior probability through the algorithm. When the posterior probability is less than 0.8, trigger an alarm; S432: Use the BB84 protocol to generate quantum keys, encrypt the warning instruction, and the receiving end recovers the original instruction through quantum state measurement, with the bit error rate less than 1%.

9. The data processing method of the intelligent cold stamping production line integrating the MES system according to claim 8, characterized in that: The detailed steps of S5 are as follows: S51: Establish a quantum security database for quantum state data, S52: Construct a distributed time series database for stress data, S53: Create a feature index library for visual data, S54: Build a knowledge graph for process knowledge, and S55: Develop a self-optimization strategy library for optimization strategies; The detailed steps of S51 are as follows: S511: Use the AES-256 key generated by the quantum key distribution QKD to encrypt the quantum state density matrix and store it in the quantum security database; S512: Adopt the Shamir threshold scheme to fragment and store the key in distributed nodes, and achieve fine-grained permission management through quantum identity authentication; S513: Receive the quantum state data generated by the quantum entanglement monitoring in step five, support the credible traceability of the over-distance warning, provide a secure historical quantum state query interface for the digital twin of S3, and accelerate the model parameter correction; The detailed steps of S52 are as follows: S521: Receive the high-frequency stress data of the S2 flexible sensor network to support the real-time parameter correction of the digital twin; S522: Provide stress feature vectors for the S2 cross-modal fusion to improve the prediction accuracy of the QNN model; The detailed steps of S53 are as follows: S531: Use FFmpeg to transcode the fisheye lens video, store it in MinIO in slices according to the time stamp, and receive the visual data of the S1 bionic compound eye system to support defect traceability and process review; S532: Extract 64-dimensional visual features for each frame through the C3D network, build an inverted index, and develop a RESTful API to support feature similarity queries, which is used to provide fast feature retrieval for S2 cross-modal fusion and accelerate the inference of the QNN model; The detailed steps of S34 include: Use Neo4j to build a graph database. The nodes include: material type, process parameters, and defect cases. Integrate multi-source data from S1 to S4 to support the initialization of the digital twin, which is used to provide prior knowledge for the S3 self-optimization strategy library and accelerate the convergence of reinforcement learning; The detailed steps of S35 include: Use DVC to manage the reinforcement learning strategy files, record the reward values for each iteration, receive the prediction results of the S3 digital twin as training samples, continuously optimize the strategy, output the optimized strategy to the MES system, and dynamically adjust the production parameters.

10. The data processing method of the intelligent cold stamping production line integrating the MES system according to claim 5, characterized in that: The visual feature C3D output is 64-dimensional, the stress feature GNN output is 64-dimensional, and the quantum feature purity P and entropy S are merged into a 192-dimensional feature vector: F = 【C3D, GNN, P, S】.

Images