A cold stamping quality control method based on multi-source data fusion and real-time optimization

Abstract

The present invention relates to the technical field of aluminum alloy cold stamping quality control, and in particular to a cold stamping quality control method based on multi-source data fusion and real-time optimization. The present invention inputs the fused feature space into a pre-trained LSTM model, associates the real-time stamping speed and holding time parameters through an attention mechanism, and realizes the joint prediction of rebound deviation and defect probability for the first time. The process parameter correction values ​​are dynamically generated based on the prediction results. Combined with the thermal expansion compensation algorithm and micron-level mold compensation, the quality control response speed reaches milliseconds, and the compensation accuracy is improved to ±2μm, which is more than 50% lower than the traditional PID control error.

Description

Technical Field

[0001] The present invention relates to the technical field of aluminum alloy cold stamping quality control, and in particular to a cold stamping quality control method based on multi-source data fusion and real-time optimization. Background Art

[0002] At present, aluminum alloy cold stamping technology is one of the core processes for lightweight manufacturing of new energy vehicles. Through a high-precision and high-efficiency stamping process, aluminum alloy sheets are processed into body structural parts (such as doors, hoods, battery pack shells, etc.), significantly reducing the weight of the entire vehicle and improving endurance and energy utilization efficiency.

[0003] The following technical problems exist in the existing technology:

[0004] 1. Existing technologies mainly rely on a single type of sensor (such as pressure or displacement sensors), with limited data acquisition dimensions. This makes it difficult to simultaneously obtain multimodal data such as mold pressure, displacement, temperature, and surface defects, affecting the comprehensive monitoring of the stamping process. In addition, traditional feature fusion methods use fixed weight distribution (such as equal-weighted averaging), lack a dynamic weight adjustment mechanism based on the material elastic modulus, and do not effectively reduce dimensionality. This leads to insufficient efficiency in extracting key information such as stamping energy characteristics and deformation models, and a high error rate.

[0005] 2. Existing technologies predict workpiece springback (the dimensional deviation caused by the elastic recovery of aluminum alloy sheet material after forming in traditional cold stamping processes, i.e., the maximum offset between the actual shape of the stamped part after demolding and the theoretical shape of the mold design) based on empirical formulas or finite element simulations. These methods are unable to dynamically correlate the real-time stamping speed and holding time parameters, resulting in low prediction accuracy and a defect missed detection rate exceeding 20%.

[0006] 3. The existing MES system does not effectively integrate material batches, process parameters and defect patterns. Quality problem tracing relies on manual cross-system data retrieval and lacks multi-dimensional correlation analysis capabilities. In addition, the existing MES system lacks the ability to correlate multi-source data. Process parameters, defect patterns and material batches are stored in a decentralized manner. Quality problem tracing relies on manual investigation, which is time-consuming.

[0007] Therefore, a cold stamping quality control method based on multi-source data fusion and real-time optimization is needed to solve the above problems. Summary of the Invention

[0008] The present invention provides a cold stamping quality control method based on multi-source data fusion and real-time optimization. The present invention inputs the fused feature space into a pre-trained LSTM model, associates the real-time stamping speed and holding time parameters through the attention mechanism, and realizes the joint prediction of rebound deviation and defect probability for the first time. The process parameter correction value is dynamically generated based on the prediction result. Combined with the thermal expansion compensation algorithm and micron-level mold compensation, the quality control response speed reaches millisecond level, and the compensation accuracy is improved to ±2μm, which is more than 50% lower than the traditional PID control error.

[0009] The technical solution adopted by the present invention to solve the above technical problems is: a cold stamping quality control method based on multi-source data fusion and real-time optimization, which realizes full-process closed-loop control of the production process through the MES system, including the following steps:

[0010] Step 1: The stamping equipment’s die pressure, hydraulic displacement, temperature, and workpiece surface image data are collected through a distributed sensor network and transmitted to the MES system after processing.

[0011] Step 2: Extract stamping energy characteristics through time-frequency analysis of pressure data, build a dynamic deformation model based on hydraulic displacement, calculate thermal compensation based on temperature, and extract defect characteristics through multi-scale morphological processing of images;

[0012] Step 3: Normalize the energy characteristics, deformation model, thermal compensation amount, and defect characteristics, generate weight coefficients based on the elastic modulus, and construct a three-dimensional fusion feature space through principal component analysis;

[0013] Step 4: Input the feature space into the pre-trained LSTM model, associate the stamping speed with the holding time, and output a quality prediction matrix containing the springback deviation and defect probability;

[0014] Step 5: When the springback deviation exceeds ±0.15mm and the defect probability is greater than 85%, the data is sent to the PLC via the bus to adjust the stamping speed and holding time parameters;

[0015] Step 6: Calculate the mold compensation amount based on the actual springback amount, and link the lubrication system to adjust the injection parameters in stages to perform micron-level compensation;

[0016] Step 7: After every 50 stampings, the LSTM parameters are updated using the laser measurement data. Incremental learning is initiated when the prediction error is greater than 0.1 mm for three consecutive times.

[0017] Step 8: Build a traceable knowledge base in the MES system to store batch data and form a closed-loop archive.

[0018] Furthermore, the distributed sensor network in step 1 includes:

[0019] A 16-point array piezoelectric pressure sensor group is arranged on the top of the mold cavity, with a sampling frequency of ≥5kHz and data transmission via the CAN bus protocol;

[0020] A sub-micron grating displacement sensor installed at the piston rod of the hydraulic cylinder;

[0021] The working surface of the stamping die is embedded with a 9-point infrared temperature measurement array, and the temperature data is calibrated using the time-of-flight algorithm;

[0022] Dual-light source polarization industrial cameras are installed on both sides of the conveyor line to synchronously trigger and collect images of 0.05mm level defects on the workpiece surface.

[0023] Furthermore, the step 2 of constructing a dynamic deformation model based on hydraulic displacement includes:

[0024] Step 2-1: Perform sliding window filtering on the hydraulic displacement data, and adjust the window length adaptively;

[0025] Step 2-2: Establish a material rebound prediction model based on the JC constitutive equation and calculate the theoretical deformation based on the elastic modulus data;

[0026] Step 2-3: Extract the energy spectrum characteristics of the pressure signal through wavelet packet decomposition and divide it into low-frequency stable area and high-frequency impact area;

[0027] Step 2-4: Perform grayscale projection analysis on the workpiece surface image and optimize the continuity of the defect boundary by combining morphological closing operation.

[0028] Furthermore, generating a weight coefficient based on the elastic modulus in step 3 includes:

[0029] Step 3-1: Query the elastic modulus value of the current batch of aluminum alloy from the MES system material database;

[0030] Step 3-2: Calculate the objective weight coefficient of the punching energy feature based on the entropy weight method;

[0031] Step 3-3: Generate material property correction factor based on elastic modulus;

[0032] Step 3-4: Reduce the four-dimensional features to three-dimensional space through principal component analysis, retaining more than 95% of the variance information.

[0033] Furthermore, the pre-training of the LSTM model in step 4 includes:

[0034] Step 4-1: Collect historical production data to build a training set and divide the input sequence into sliding time windows;

[0035] Step 4-2: Introduce the attention mechanism into the network structure to dynamically assign time weights to different features;

[0036] Step 4-3: Use the Bayesian optimization algorithm to adjust the number of LSTM hidden layer nodes;

[0037] Step 4-4: Perform weighted fusion of the real-time stamping speed parameters and the LSTM output gate.

[0038] Furthermore, adjusting the punching speed and holding time parameters in step 5 includes the following sub-steps:

[0039] Step 5-1: Calculate the stamping speed correction value based on the thermal expansion coefficient of aluminum alloy;

[0040] Step 5-2: Construct a holding time-temperature correlation curve based on historical process data, and dynamically adjust the holding time parameters using fuzzy control rules;

[0041] Step 5-3: Send the corrected parameters to the PLC controller via the PROFINET industrial bus protocol, ensuring that the response delay is less than 50ms;

[0042] Step 5-4: Generate parameter adjustment records in the MES system interface and associate the equipment operation log with the current process version number.

[0043] Furthermore, in step 6, the mold compensation amount is calculated based on the actual rebound amount, and the injection parameters are adjusted in stages by the linkage lubrication system. The micron-level compensation is performed, which includes the following sub-steps:

[0044] Step 6-1: Analyze the mold deformation trend based on the strain energy density and calculate the residual stress distribution;

[0045] Step 6-2: The mold fine-tuning mechanism is controlled by a servo motor, with a compensation accuracy of ±2μm;

[0046] Step 6-3: Classify the lubrication level according to the defect probability: when the defect probability is greater than 90%, the lubricant injection amount is double injection at level I; when the defect probability is between 70% and 90%, the lubricant injection amount is standard injection at level II; when the defect probability is less than 70%, the lubricant injection amount is reduced frequency injection at level III;

[0047] Step 6-4: Use the PID control algorithm to dynamically adjust the lubricant injection pressure to 0.2-0.8 MPa.

[0048] Furthermore, the incremental learning optimization step in step 7 includes:

[0049] Step 7-1: Construct a sliding time window to filter key historical data samples, including mold pressure curves, hydraulic displacement time series, temperature monitoring data, and defect image samples;

[0050] Step 7-2: Use Hampel filter to remove abnormal measurement data;

[0051] Step 7-3: Freeze the weights of the first three layers of the LSTM network and only update the output layer parameters;

[0052] Step 7-4: Reuse the feature extraction capabilities of the historical model through transfer learning technology.

[0053] Furthermore, step 8 of constructing a traceable knowledge base in the MES system includes the following sub-steps:

[0054] Step 8-1: Run the original data stream using the time series database storage device;

[0055] Step 8-2: Build a knowledge graph to associate material batch number, mold number and quality defect pattern;

[0056] Step 8-3: Deploy an OPC UA-based data interface to connect with the ERP system;

[0057] Step 8-4: Establish a SQL traceability query engine to support multi-dimensional root cause analysis.

[0058] The advantages of the present invention are:

[0059] 1. This invention uses a distributed sensor network (pressure, displacement, temperature, image) to collect multimodal data in real time. This data is combined with time-frequency analysis, dynamic deformation modeling, and morphological processing to overcome the limitations of a single data dimension. Furthermore, dynamic weight coefficients are generated based on the material's elastic modulus, and principal component analysis is used to construct a high-information-density three-dimensional fusion feature space. This solves the problem of feature redundancy and weight rigidity in traditional methods, improves the efficiency of extracting key quality features by over 40%, and provides precise input for subsequent prediction models.

[0060] 2. This invention integrates the feature space input into a pre-trained LSTM model and associates the real-time stamping speed and holding time parameters through an attention mechanism. This allows for the first joint prediction of springback deviation (±0.15mm accuracy) and defect probability (85% threshold). It also dynamically generates process parameter correction values ​​based on the prediction results. Combined with a thermal expansion compensation algorithm and micron-level mold compensation, this method enables a quality control response speed of milliseconds and an improvement in compensation accuracy to ±2μm, reducing the error by over 50% compared to traditional PID control.

[0061] 3. The present invention achieves closed-loop control of the entire process through the MES system. Laser measurement data is used to update the LSTM model after every 50 stampings. Incremental learning is triggered when the continuous prediction error is greater than 0.1mm, and transfer learning technology is combined to quickly adapt to changes in material parameters. At the same time, a knowledge graph is constructed to associate process parameters with defect patterns, supporting multi-dimensional root cause tracing, thereby increasing the efficiency of quality defect location by 70%, and the system's adaptability reaches industrial-grade continuous optimization standards. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0063] Figure 1 The present invention provides a flow chart of a cold stamping quality control method based on multi-source data fusion and real-time optimization. DETAILED DESCRIPTION

[0064] The technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0065] Example 1: Figure 1 The present invention provides a flow chart of a cold stamping quality control method based on multi-source data fusion and real-time optimization, such as Figure 1The cold stamping quality control method shown in the figure is based on multi-source data fusion and real-time optimization, which realizes the closed-loop control of the whole production process through the MES system, including the following steps: first, the mold pressure, hydraulic displacement, working temperature and workpiece surface image data of the aluminum alloy stamping equipment are collected in real time through the distributed sensor network, and the data format is unified through the industrial protocol gateway and directly transmitted to the MES system; the pressure data is subjected to time-frequency analysis to extract the stamping energy characteristics, a dynamic deformation model is constructed based on the hydraulic displacement data, the thermal compensation amount is calculated in combination with the working temperature data, and the workpiece surface image is subjected to multi-scale morphological processing to extract the defect characteristics; the stamping energy characteristics, deformation model, thermal compensation amount and defect characteristics are normalized, a dynamic weight coefficient is generated based on the elastic modulus of the material, and a three-dimensional fusion feature space is constructed through principal component analysis; the fusion feature space is input into the The pre-trained LSTM model is input, the real-time stamping speed and holding time parameters are associated, and a quality prediction matrix containing the springback deviation and defect probability is output; when the springback deviation exceeds ±0.15mm and the defect probability is greater than 85%, a stamping speed correction value is generated according to the thermal expansion compensation amount, and the process parameters are adjusted according to the holding time-temperature correlation curve, and the data is sent to the PLC for execution through the industrial bus; the mold compensation amount is calculated based on the actual springback amount after execution, and the lubrication system is linked to adjust the injection parameters according to the defect probability level to perform micron-level precision compensation; after completing 50 stampings, the LSTM model weight parameters are updated using laser 3D measurement data, and incremental learning optimization is initiated when the prediction error is greater than 0.1mm for three consecutive times; a traceable knowledge base is built in the MES system, and material parameters, process adjustment records and quality inspection data are stored according to production batches to form a closed-loop control file.

[0066] Among them, the MES system realizes the closed-loop control of the entire production process, which refers to the real-time monitoring and feedback adjustment of each link of the cold stamping process through the manufacturing execution system. Specifically, it can be achieved by using the industrial protocol gateway to unify the data format, issuing control instructions to the PLC for execution through the industrial bus protocol, and building a traceable knowledge base to store the entire process data. This feature realizes the dynamic adjustment of process parameters and the traceability of quality data, and solves the data island problem of the traditional system. Among them, the distributed sensor network collects the mold pressure, hydraulic displacement, working temperature and workpiece surface image data of the aluminum alloy stamping equipment in real time, which refers to a multimodal perception system composed of a piezoelectric pressure sensor group, a grating displacement sensor, an infrared temperature measurement array and a polarization industrial camera. It is specifically achieved by using the CAN bus protocol to transmit pressure data, the flight time algorithm to calibrate temperature data, and the dual light source synchronous triggering to collect surface defect images. This feature breaks through the limitation of the single sensor data dimension and improves the comprehensiveness of stamping process monitoring. Among them, time-frequency analysis to extract stamping energy characteristics refers to extracting spectral characteristics and analyzing energy distribution of pressure signals. This is achieved by using sliding window filtering and wavelet packet decomposition to divide low-frequency stable areas and high-frequency impact areas. This feature solves the problem of traditional time-domain analysis ignoring frequency-domain characteristics, which leads to energy assessment errors. Among them, dynamic deformation model construction refers to establishing a material rebound prediction model based on hydraulic displacement data. This is achieved by using the JC constitutive equation combined with elastic modulus data to calculate theoretical deformation variables and optimizing defect boundary continuity through grayscale projection analysis and morphological closing operations. This feature improves deformation prediction accuracy. Among them, principal component analysis to construct a three-dimensional fusion feature space refers to dimensionality reduction processing of stamping energy characteristics, deformation models, thermal compensation quantities, and defect characteristics. This is achieved by normalization processing to eliminate dimensional differences, calculating objective weight coefficients based on the entropy weight method, and retaining more than 95% of variance information. This feature solves the problem of multi-dimensional data redundancy and improves feature extraction efficiency. Among them, the output quality prediction matrix of the pre-trained LSTM model refers to the use of long-short-term memory neural networks to model time series features. Specifically, it adopts the attention mechanism to assign time weights, Bayesian optimization to adjust the number of hidden layer nodes, and associates the real-time stamping speed and holding time parameters to achieve this. This feature realizes the joint prediction of rebound deviation and defect probability. Among them, adjusting process parameters according to the holding time-temperature correlation curve refers to generating stamping speed correction values ​​according to the thermal expansion compensation amount. Specifically, it adopts fuzzy control rules to adjust time parameters and sends parameters to PLC execution through the PROFINET industrial bus protocol to achieve this. This feature realizes closed-loop control of process parameters with millisecond response speed. Among them, mold compensation calculation and micron-level precision compensation refer to the analysis of mold deformation trends based on actual rebound amount. Specifically, it adopts strain energy density to analyze residual stress distribution and servo motor control fine-tuning mechanism to perform ±2μm precision compensation to achieve this. This feature solves the problem of insufficient accuracy of traditional compensation mechanisms.Incremental learning optimization of the LSTM model involves using laser 3D measurement data to update model parameters. This is achieved by filtering key historical data samples with a sliding time window, freezing the weights of the first three layers and only updating the output layer parameters, and reusing feature extraction capabilities through transfer learning. This feature improves the model's adaptability to changes in material parameters. Construction of a traceable knowledge base involves storing the entire process data of production batches. This is achieved by using a time-series database to store raw data streams, constructing a knowledge graph to associate defect patterns with process parameters, and deploying an OPC UA interface to enable system interconnection. This feature supports multi-dimensional root cause tracing and rapid location of quality issues.

[0067] The core innovation of this invention lies in the synchronous acquisition of multi-source data in the stamping process through a distributed multimodal sensor network, the construction of a high-information-density fusion feature space combining dynamic weight allocation and principal component analysis, and the joint prediction of rebound amount and defect probability based on the LSTM model. Finally, the process parameters and mold compensation amount are closed-loop controlled by the MES system to form a full-process quality optimization system, which solves the problems of low feature extraction efficiency, insufficient prediction accuracy and system response delay of traditional methods.

[0068] The working process and principle of this invention involve real-time acquisition of multi-source data from aluminum alloy stamping equipment, including die pressure, hydraulic displacement, operating temperature, and workpiece surface images, through a distributed sensor network. This data is formatted and standardized via an industrial protocol gateway before being transmitted to the MES system. Time-frequency analysis is performed on the pressure data to extract stamping energy characteristics. A dynamic deformation model is constructed based on the hydraulic displacement data, and thermal compensation is calculated in combination with operating temperature data. Multi-scale morphological processing is then performed on the workpiece surface images to extract defect features. These features are normalized, and dynamic weight coefficients are generated based on the material elastic modulus. A three-dimensional fused feature space is constructed through principal component analysis. This fused feature space is input into a pre-trained LSTM model, which correlates the real-time stamping speed and dwell time parameters to output a quality prediction matrix. When the springback deviation or defect probability exceeds a threshold, the system generates a stamping speed correction value based on the thermal expansion compensation, adjusts the process parameters, and sends it to the PLC for execution. The die compensation is calculated based on the actual springback after execution, and the lubrication system is linked to adjust the injection parameters. After every 50 stampings, the system updates the LSTM model parameters using 3D laser measurement data, initiating incremental learning optimization as necessary. The MES system builds a traceable knowledge base and stores relevant data to form a closed-loop control archive.

[0069] As a preferred embodiment, the solution of the present invention is specifically implemented as follows:

[0070] First, a distributed sensor network was deployed on the cold stamping production line. A 16-point array of piezoelectric pressure sensors was placed on the top of the mold cavity, with a sampling frequency set to 5kHz and data transmitted via the CAN bus protocol. A submicron grating displacement sensor was installed on the hydraulic cylinder piston rod. A 9-point infrared temperature measurement array was embedded in the working surface of the stamping die, and temperature data was calibrated using a time-of-flight algorithm. Dual-light source polarization industrial cameras were installed on both sides of the conveyor line to synchronously trigger and capture images of the workpiece surface.

[0071] Secondly, the collected multi-source data is processed and features extracted. Wavelet packet decomposition is performed on the pressure data to extract energy spectrum features and delineate low-frequency stable regions from high-frequency impact regions. Sliding window filtering is performed on the hydraulic displacement data, with adaptive window length adjustment. A material rebound prediction model is established based on the JC constitutive equation, and theoretical deformation is calculated using elastic modulus data. Grayscale projection analysis is performed on the workpiece surface image, and morphological closing operations are used to optimize the continuity of defect boundaries.

[0072] The extracted features are then fused. The elastic modulus of the current batch of aluminum alloys is retrieved from the MES system's material database. The objective weight coefficient of the stamping energy feature is calculated using the entropy weight method. A material property correction factor is generated based on the elastic modulus. Principal component analysis is used to reduce the four-dimensional features to a three-dimensional space, retaining more than 95% of the variance information.

[0073] Next, the fused features are input into a pre-trained LSTM model for quality prediction. The LSTM model pre-training process includes: collecting historical production data to construct a training set, dividing the input sequence into sliding time windows; introducing an attention mechanism into the network structure to dynamically assign time weights to different features; using a Bayesian optimization algorithm to adjust the number of nodes in the LSTM hidden layer; and weighted fusion of the real-time stamping speed parameters with the LSTM output gate. When the predicted results show a springback deviation exceeding ±0.15mm and a defect probability greater than 85%, the system calculates a stamping speed correction value based on the thermal expansion coefficient. A holding time-temperature curve is constructed, and fuzzy control rules are used to adjust the time parameters. Parameters are transmitted to the PLC via the PROFINET industrial bus protocol, with response latency controlled within 50ms. The system then analyzes mold deformation trends based on strain energy density and calculates residual stress distribution. A servo motor controls the mold fine-tuning mechanism, achieving a compensation accuracy of ±2μm. Lubrication levels are classified based on defect probability, and the lubricant injection pressure is dynamically adjusted between 0.2 and 0.8MPa.

[0074] After every 50 stampings, the system updates the LSTM model using laser 3D measurement data. When the prediction error exceeds 0.1mm for three consecutive times, incremental learning optimization is initiated. A sliding time window is constructed to filter key historical data samples, and a Hampel filter is used to remove abnormal measurement data. The weights of the first three layers of the LSTM network are frozen, and only the output layer parameters are updated. Transfer learning technology is used to reuse the feature extraction capabilities of the historical model.

[0075] Finally, a traceability knowledge base was built within the MES system. A time-series database was used to store raw data streams. A knowledge graph was constructed to correlate material batch numbers, mold numbers, and quality defect patterns. An OPC UA-based data interface was deployed to connect with the ERP system. An SQL traceability query engine was established to support multi-dimensional root cause analysis.

[0076] Through the above scheme, the present invention realizes the real-time collection and fusion processing of multi-source data, improving the efficiency and accuracy of feature extraction. The LSTM model combines real-time process parameters for quality prediction, improving the prediction accuracy. The real-time process parameter adjustment and mold compensation mechanism based on the prediction results improve the response speed and accuracy of stamping quality control. Incremental learning and knowledge base construction realize the continuous optimization and quality traceability of the system. These technical means work together to effectively solve the problems of limited data dimension, low prediction accuracy, parameter adjustment lag, and quality traceability difficulties in traditional cold stamping quality control, and improve the quality stability and efficiency of cold stamping production.

[0077] The present invention further proposes a distributed sensor network including a 16-point array piezoelectric pressure sensor group arranged on the top of the mold cavity, with a sampling frequency of ≥5kHz, and data transmitted via the CAN bus protocol; a sub-micron grating displacement sensor installed on the hydraulic cylinder piston rod; a 9-point infrared temperature measurement array embedded on the working surface of the stamping mold, and a flight time algorithm is used to calibrate the temperature data; and dual-light source polarization industrial cameras are set up on both sides of the conveyor line for synchronously triggering the collection of 0.05mm level defect images on the workpiece surface.

[0078] The 16-point piezoelectric pressure sensor array at the top of the mold cavity is arranged in an equidistant circular pattern, covering the mold working surface. Each sensor is spaced 1 / 8 the mold diameter to ensure the integrity of the pressure field distribution. The piezoelectric sensor array works with the CAN bus protocol to achieve multi-channel parallel transmission, with a data packet interval of ≤0.2ms to prevent high-frequency signal distortion. The submicron grating displacement sensor at the hydraulic cylinder piston rod uses a magnetic scale to maintain a 0.3-0.5mm spacing between the scale and the readhead, achieving a resolution of 0.1μm, capable of capturing sudden displacement changes during the rapid rebound phase. The 9-point infrared temperature measurement array on the stamping die working surface utilizes an isothermal zone partitioning strategy, with each measurement point covering an area of ​​no more than 30mm². A time-of-flight algorithm calculates the time difference between infrared pulse emission and reception to maintain temperature calibration error within ±1.5°C. A dual-light source polarization industrial camera forms a 30° cross-illumination angle on both sides of the conveyor line. The trigger signal is synchronized with the press stroke switch, and the exposure time is set to 1 / 120 of the stamping cycle, ensuring an error of less than 2ms between image acquisition and workpiece arrival.

[0079] Specifically, the piezoelectric pressure sensor array overcomes the localized limitations of single-point detection through a multi-point layout. At a sampling frequency of 5kHz, it can capture transient pressure fluctuations lasting ≥0.2ms. The high-resolution design of the grating displacement sensor enables precise quantification of micron-level displacement changes in the hydraulic cylinder, providing fundamental data for dynamic deformation models. The spatial density of the infrared temperature measurement array, combined with a time-of-flight algorithm, effectively eliminates the effects of the mold surface oxide layer on emissivity, providing real-time output of the corrected work surface temperature distribution. A dual-light source polarization camera uses cross-illumination to eliminate interference from metal surface reflections, and a synchronized triggering mechanism ensures a complete image of the workpiece surface during the stamping cycle. These sensors transmit data via a hybrid of multiple protocols, such as CAN bus and PROFINET, enabling timestamp alignment within the MES system to form a multi-dimensional data stream. For example, during a stamping stroke, when the pressure sensor array captures a transient impact force of 12.5kN, the grating displacement sensor simultaneously records a 0.8mm displacement jump, the infrared temperature measurement array detects an 8°C temperature rise on the mold work surface, and the polarization camera captures an image of a 0.06mm microcrack on the workpiece edge. The spatiotemporal alignment of multi-source data provides precise input for subsequent feature fusion, reducing the error of punch energy feature extraction to less than 3%.

[0080] As a preferred embodiment, the solution of the present invention is specifically implemented as follows:

[0081] The distributed sensor network includes a 16-point array of piezoelectric pressure sensors located at the top of the mold cavity, sampling at a 5kHz frequency and transmitting data via the CAN bus protocol. A submicron grating displacement sensor is installed on the hydraulic cylinder piston rod. A 9-point infrared temperature measurement array is embedded in the working surface of the stamping die, calibrating temperature data using a time-of-flight algorithm. Dual-light source polarization industrial cameras are installed on both sides of the conveyor line to synchronously trigger and capture images of defects on the workpiece surface as small as 0.05mm.

[0082] Specifically, a 16-point array of piezoelectric pressure sensors is arranged in a 4x4 matrix, covering the critical area at the top of the mold cavity. Each pressure sensor has a measurement range of 0-1000 MPa and linearity better than 0.1% FS. The submicron grating displacement sensor has a resolution of 0.1 μm and a measurement range of 0-500 mm. The 9-point infrared temperature measurement array uses thermopile sensors and has a temperature measurement range of 0-300°C with an accuracy of ±1°C. The dual-light source polarization industrial camera has a resolution of 2048x2048 pixels and a frame rate of up to 200 fps.

[0083] Furthermore, the CAN bus utilizes the high-speed CAN 2.0B protocol with a baud rate of 1Mbps. A time-of-flight algorithm calibrates temperature data by measuring the propagation time of infrared light in air, eliminating the influence of environmental factors. The dual-light source polarization industrial camera utilizes a polarizing filter and a cross-polarized light source to enhance the contrast of surface defects.

[0084] Through the above-mentioned technical solutions, the present invention achieves high-precision, synchronous acquisition of multi-source data. The comprehensive acquisition of pressure, displacement, temperature, and image data lays the foundation for subsequent multimodal data fusion. The high sampling frequency and precise measurement range ensure the timeliness and accuracy of the data. The CAN bus protocol and time-of-flight algorithm improve the reliability of data transmission and the accuracy of temperature measurement. Dual-light source polarization imaging technology enhances the detection of surface defects. These improvements effectively enhance the comprehensive monitoring capabilities of the cold stamping process and provide comprehensive and accurate data support for quality control.

[0085] The present invention further proposes the construction of a dynamic deformation model, which specifically includes the following steps: performing sliding window filtering on the hydraulic displacement data and adaptively adjusting the window length; establishing a material rebound prediction model based on the JC constitutive equation, and calculating the theoretical deformation variable in combination with the elastic modulus data; extracting the energy spectrum characteristics of the pressure signal through wavelet packet decomposition, and dividing the low-frequency stable area and the high-frequency impact area; performing grayscale projection analysis on the workpiece surface image, and optimizing the continuity of the defect boundary in combination with morphological closing operations.

[0086] Among them, the window length of the sliding window filter is automatically adjusted according to the spectral characteristics of the hydraulic displacement signal. When the change of the signal frequency component is detected to exceed the set threshold, the window length is dynamically adjusted in the range of 100-500ms with a step size of 50ms; the JC constitutive equation introduces the strain rate sensitivity coefficient and the temperature softening effect parameter, and combines the material elastic modulus to calculate the theoretical deformation variable. The model input includes the stamping speed, holding time and measured temperature data; the wavelet packet decomposition uses the db4 wavelet basis function to perform a 5-layer decomposition, dividing the pressure signal energy spectrum into a low-frequency stable area of ​​0-400Hz and a high-frequency impact area of ​​400-2000Hz; the grayscale projection analysis generates a projection histogram along the X / Y axis direction of the workpiece surface image, and the boundary is optimized using a 3×3 circular structure element through morphological closing operation.

[0087] Specifically, the hydraulic displacement signal is filtered through an adaptive sliding window, effectively suppressing high-frequency noise while preserving true deformation characteristics. Dynamic adjustment of the window length is achieved through real-time calculation of the signal's power spectral density, triggering a change in window length when the main frequency band fluctuates by more than ±10%. A deformation model based on the JC constitutive equation introduces the strain hardening exponent n and the strain rate sensitivity coefficient m, combined with measured data on the elastic modulus E, to calculate the theoretical deformation variable ΔL=ƒ(σ,ε˙,T), where σ is the equivalent stress, ε˙ is the strain rate, and T is the mold temperature. After wavelet packet decomposition of the pressure signal, the energy fraction of the low-frequency stable region is used to assess the stability of the stamping process. When the energy peak in the high-frequency impact region exceeds a threshold, a machine status warning is triggered. After grayscale projection analysis of the workpiece surface image locates defect areas, a morphological closing operation is performed to fill in the broken pixels at the boundary, increasing the degree of closure of the defect contour to over 95%, providing continuous boundary data for subsequent feature extraction. Through the coordinated execution of the above steps, the prediction error of the dynamic deformation model is reduced to within 0.05mm, the accuracy of high-frequency impact feature recognition is increased to 98%, and the defect boundary positioning accuracy reaches ±2 pixels.

[0088] As a preferred embodiment, the solution of the present invention is specifically implemented as follows:

[0089] The construction of the dynamic deformation model involves the following steps: Sliding window filtering of the hydraulic displacement data and adaptive adjustment of the window length. Specifically, the hydraulic displacement data is preprocessed using the Kalman filter algorithm. The sliding window length is dynamically adjusted based on the data fluctuation amplitude. The window length is shortened when the fluctuation amplitude is large and extended when the fluctuation amplitude is small, balancing the filtering effect with real-time performance.

[0090] A material rebound prediction model was established based on the JC constitutive equation, and theoretical deformation was calculated using elastic modulus data. Furthermore, the Johnson-Cook constitutive equation, a well-known technical method, was used to describe the mechanical behavior of aluminum alloy materials at high strain rates. Combined with real-time elastic modulus data, theoretical deformation was calculated using finite element analysis. Wavelet packet decomposition was used to extract the energy spectrum characteristics of the pressure signal, dividing it into a low-frequency stable region and a high-frequency impact region. The pressure signal was subjected to a five-layer wavelet packet decomposition using the db4 wavelet basis to extract the energy characteristics of each frequency band. Based on the energy distribution, the signal was divided into a low-frequency stable region and a high-frequency impact region. Grayscale projection analysis was performed on the workpiece surface image, and morphological closing operations were used to optimize the continuity of defect boundaries. Thus, the workpiece surface image was first grayscaled, followed by horizontal and vertical grayscale projection analysis to identify potential defect areas. Morphological closing operations were then used to optimize the defect boundaries using a 3x3 structuring element to improve boundary continuity.

[0091] Through the above technical solution, the present invention realizes adaptive filtering processing of hydraulic displacement data and improves data quality. The material rebound prediction model established based on the JC constitutive equation can accurately describe the mechanical behavior of aluminum alloy materials during the stamping process. The wavelet packet decomposition method effectively extracts the energy spectrum characteristics of the pressure signal, providing a reliable basis for subsequent analysis. Grayscale projection analysis combined with morphological operations optimizes the defect boundary identification effect. The comprehensive application of these technical means significantly improves the accuracy and reliability of the dynamic deformation model, providing strong support for cold stamping quality control.

[0092] The present invention further proposes that the generation of a dynamic weight coefficient includes the following sub-steps: querying the elastic modulus value of the current batch of aluminum alloy from the MES system material database; calculating the objective weight coefficient of the stamping energy characteristic based on the entropy weight method; generating a material property correction factor according to the elastic modulus; and reducing the four-dimensional feature to a three-dimensional space through principal component analysis, retaining more than 95% of the variance information.

[0093] Among them, the elastic modulus query operation is implemented through the material batch tracking module built into the MES system. The database stores the measured elastic modulus value range of aluminum alloy plates, and the data update frequency is synchronized with the production batch. During the entropy weight method calculation process, the pressure time-frequency characteristics, the displacement dynamic model output value, the thermal compensation amount, and the defect area parameters are constructed into an evaluation matrix. The discrete degree of each characteristic indicator is calculated through information entropy to generate the initial weight coefficient. The material property correction factor is dynamically adjusted according to the deviation ratio between the measured elastic modulus value and the standard value. When the elastic modulus is higher than the standard value, the correction factor increases the displacement model weight according to a linear relationship; when the elastic modulus is lower than the standard value, the thermal compensation amount weight is increased. During the principal component analysis process, the four-dimensional feature matrix is ​​orthogonally transformed into a three-dimensional space vector, and the cumulative variance contribution rate threshold is used to control the amount of information retained after dimensionality reduction.

[0094] Specifically, within the MES system's material database, the elastic modulus values ​​for the current batch of aluminum alloys are stored in XML format, including the smelting furnace number, rolling batch, and third-party inspection report number. The query operation uses SQL statements to match the production order number with the material inventory record to obtain the actual measured elastic modulus value in real time. When calculating using the entropy weight method, each indicator of the stamping energy characteristic matrix is ​​normalized to a value in the range of 0-1. During the information entropy calculation process, the natural logarithm is used to calculate the entropy value of each indicator, and the objective weight coefficient is determined by the difference between the entropy value and 1. The material property correction factor is generated using the elastic modulus ratio function. For example, when the measured elastic modulus is 70GPa and the standard value is 69GPa, the correction factor increases the displacement model weight by 1.4%. When performing principal component analysis, the eigenvalues ​​of the covariance matrix are arranged in descending order, and the first three principal components are selected so that the cumulative variance contribution rate reaches the 95% threshold. The coordinates of the three-dimensional eigenvector after dimensionality reduction are calculated by linear combination of the eigenvector and the original data. This process ensures that the dynamic weight distribution of different material batches not only reflects the objective laws of the data, but also combines the physical properties of the materials, so that the fused feature space eliminates redundant dimensions while retaining key quality information.

[0095] As a preferred embodiment, the solution of the present invention is specifically implemented as follows:

[0096] The generation of dynamic weight coefficients includes the following sub-steps:

[0097] Query the elastic modulus value of the current batch of aluminum alloy from the MES system's material database. Specifically, use an SQL query to extract the elastic modulus value of the aluminum alloy used in the current production batch from the MES system's material database. For example, the elastic modulus of 6061-T6 aluminum alloy is approximately 68.9 GPa.

[0098] The objective weight coefficient of the stamping energy feature is calculated based on the entropy weight method. Furthermore, the entropy weight method is used to process multidimensional data such as pressure and displacement collected during the stamping process to calculate the information entropy of each feature, thereby obtaining the objective weight coefficient. The greater the information entropy, the greater the uncertainty of the feature and the smaller its weight coefficient.

[0099] Generate a material property correction factor based on the elastic modulus. Substitute the queried elastic modulus value into a preset correction formula to generate a material property correction factor used to adjust the weight. For example, the correction factor can be calculated as 1 + (E-E0) / E0, where E is the elastic modulus of the current batch of materials and E0 is the baseline elastic modulus.

[0100] Principal component analysis (PCA) reduces the four-dimensional features to three-dimensional space, retaining more than 95% of the variance information. Specifically, the four-dimensional feature data, including stamping energy characteristics, deformation model, thermal compensation, and defect characteristics, are input into the PCA algorithm. By calculating the eigenvalues ​​and eigenvectors, the top three principal components with a cumulative contribution rate of more than 95% are selected to achieve dimensionality reduction.

[0101] Through the above technical solution, the present invention achieves dynamic weight adjustment based on the material elastic modulus, improving the accuracy of feature fusion. Furthermore, principal component analysis effectively reduces data dimensionality, retains key information, and improves subsequent processing efficiency. This approach overcomes the limitations of traditional fixed weight allocation methods, enabling fused features to better reflect material properties and processing conditions, providing a more reliable data foundation for cold stamping quality control.

[0102] The present invention further proposes a pre-training method for the LSTM model, which includes the following steps: collecting historical production data to construct a training set and dividing the input sequence according to a sliding time window; introducing an attention mechanism into the network structure to dynamically assign time weights to different features; using a Bayesian optimization algorithm to adjust the number of nodes in the LSTM hidden layer; and performing weighted fusion of real-time stamping speed parameters and the LSTM output gate.

[0103] Among them, the historical data training set adopts a sliding time window mechanism, and the window length is dynamically adjusted according to the stamping rhythm; the attention mechanism realizes feature weight distribution by calculating the time step correlation matrix; Bayesian optimization uses prediction error as the optimization objective function to search for the optimal number of hidden layer nodes; the output gate weighted fusion uses a gating mechanism to linearly superimpose real-time process parameters with the model output.

[0104] Specifically, historical production data is divided into sliding windows of 50 time steps based on the stamping cycle, with each window containing a sequence of pressure, displacement, temperature, and image features. The attention mechanism generates a weight distribution by calculating the cosine similarity of each time step, giving higher weight to the hydraulic displacement mutation feature. Bayesian optimization determined the number of hidden layer nodes to be 128 in 20 iterations, reducing the training error by 12% compared to the fixed structure. During the output gate fusion stage, the real-time stamping speed parameters are matrix-multiplied with the LSTM hidden state to form a joint speed-state vector, and the information transfer ratio is controlled by the Sigmoid function. This fusion mechanism reduces the standard deviation of the model's prediction error under stamping speed mutation conditions from 0.08mm to 0.05mm.

[0105] As a preferred embodiment, the solution of the present invention is specifically implemented as follows:

[0106] The pre-training of the LSTM model consists of the following steps:

[0107] First, historical production data was collected to construct a training set, and the input sequences were divided into sliding time windows. Specifically, cold stamping production data from the past six months was extracted from the MES system database, including multi-dimensional time series such as pressure curves, displacement time series, and temperature changes. Using a 30-second sliding time window, the training samples were divided into 5-second steps, generating approximately 100,000 training sequences.

[0108] The attention mechanism is then introduced into the network structure to dynamically assign temporal weights to different features. Specifically, an attention layer is added after the LSTM unit output to calculate weight coefficients at different time steps. The attention weights are normalized using a softmax function to achieve adaptive weighting of features at key time points.

[0109] Furthermore, the Bayesian optimization algorithm is used to adjust the number of LSTM hidden layer nodes. First, the search space is defined as 32 to 256 nodes, and the Gaussian process regression model is set as the proxy model. Through iterative optimization, the model performance of different numbers of nodes is evaluated on the validation set, and finally the optimal number of hidden layer nodes is determined to be 128. The present invention can perform weighted fusion of the real-time stamping speed parameter and the LSTM output gate. The specific operation is to perform element-level multiplication fusion with the activation value of the LSTM output gate after linear transformation of the stamping speed value at the current moment. In this way, the real-time process parameter information is introduced into the prediction process of the network, thereby improving the adaptability of the model to dynamic working conditions.

[0110] Through the above-mentioned technical solution, the present invention fully utilizes historical production data to construct a high-quality training set, extracts key temporal features through an attention mechanism, and combines Bayesian optimization to determine the optimal network structure. Furthermore, by integrating real-time process parameters into the prediction process, the LSTM model's modeling capabilities and prediction accuracy for dynamic changes in the cold stamping process are enhanced. This method effectively addresses the difficulties of traditional models in capturing long-term dependencies and insufficient feature extraction at key time points, providing a reliable data foundation for subsequent quality prediction and process parameter optimization.

[0111] The present invention further proposes that adjusting process parameters according to the holding time-temperature correlation curve includes the following sub-steps: calculating a stamping speed correction value based on the thermal expansion coefficient; constructing a holding time-temperature correlation curve and adjusting time parameters using fuzzy control rules; sending parameters to the PLC via the PROFINET industrial bus protocol with a response delay of less than 50ms; generating a parameter adjustment record on the MES interface and associating the equipment operation log with the process version number.

[0112] Among them, the stamping speed correction value is dynamically calculated based on the material thermal expansion coefficient and the measured temperature deviation, with a correction range of 0.15-0.8mm / s; the holding time-temperature correlation curve is constructed using the piecewise linear interpolation method, combined with fuzzy control rules to process the nonlinear mapping relationship between temperature and time; the PROFINET protocol uses time-sensitive network technology to ensure real-time parameter transmission, and the data frame structure contains a timestamp check field; parameter adjustment records are stored in XML format, and the associated fields include device ID, operator number, and process version hash value.

[0113] Specifically, the speed correction is determined by multiplying the thermal expansion coefficient by the actual value measured by the temperature sensor, so that the stamping speed adapts to the thermal deformation trend of the mold. The fuzzy control rule quantifies the temperature deviation and rate of change into five fuzzy levels, and uses the weighted average method to defuzzify the output time adjustment value to solve the nonlinear problem of the temperature-time relationship. The PROFINET protocol adopts a periodic synchronous transmission mode, sending a control instruction every 8ms, and the timestamp verification mechanism eliminates the impact of network jitter. The parameter adjustment record and the equipment operation log are bidirectionally bound through the transaction ID. The MES system automatically generates a version number and stores it in association with the material batch. As a result, the closed-loop adjustment from temperature monitoring to process parameters is completed within 0.2 seconds. The parameter adjustment record can support version traceability query based on three dimensions: time, equipment, and operator.

[0114] As a preferred embodiment, the solution of the present invention is specifically implemented as follows:

[0115] Calculate the stamping speed correction value based on the thermal expansion coefficient. First, obtain the thermal expansion coefficient of the current workpiece material. Then, combined with the real-time measured mold temperature, calculate the material's thermal expansion. Based on the deviation between the thermal expansion and the target size, the stamping speed correction value is determined using a preset speed-size relationship curve. Construct a holding time-temperature correlation curve, and use fuzzy control rules to adjust the time parameters. Through historical production data analysis, a correlation curve between holding time and mold temperature is established. Using the current mold temperature as input, after fuzzification processing, the holding time adjustment value is inferred based on the preset fuzzy rule base.

[0116] Parameters are sent to the PLC via the PROFINET industrial bus protocol, with a response delay of 50ms. Leveraging the real-time capabilities of the PROFINET protocol, the revised punch speed and hold time parameters are packaged into data frames and transmitted to the PLC controller via Industrial Ethernet. Upon receiving the parameters, the PLC immediately executes the update, keeping the response delay of the entire process within 50ms.

[0117] Generate parameter adjustment records on the MES interface, linking the equipment operation log and process version number. After each parameter adjustment, the MES system automatically generates an adjustment record containing information such as the adjustment time, parameter values ​​before and after the adjustment, and the reason for the adjustment. This record is also stored in association with the current equipment operation log and process version number for subsequent traceability and analysis.

[0118] Through the above-mentioned technical solution, the present invention achieves real-time and precise adjustment of stamping process parameters. Based on thermal expansion compensation and fuzzy control, the stamping speed and holding time are dynamically optimized, improving workpiece forming accuracy. The use of a high-speed industrial bus ensures timely parameter adjustment, minimizing production delays. A comprehensive parameter adjustment record and version management mechanism provide reliable data support for quality issue analysis, effectively enhancing the quality control level of the cold stamping process.

[0119] The present invention further proposes a method for calculating mold compensation, including the following steps: performing sliding window filtering on hydraulic displacement data with adaptive window length adjustment; establishing a material rebound prediction model based on the JC constitutive equation and calculating theoretical deformation using elastic modulus data; extracting the energy spectrum characteristics of the pressure signal through wavelet packet decomposition to divide it into a low-frequency stable region and a high-frequency impact region; performing grayscale projection analysis on the workpiece surface image and optimizing the continuity of defect boundaries using morphological closing operations; analyzing mold deformation trends based on strain energy density and calculating residual stress distribution; controlling the mold fine-tuning mechanism with a servo motor, achieving a compensation accuracy of ±2μm; and classifying lubrication levels based on defect probability: when the defect probability is greater than 90%, the lubricant injection rate is doubled at Level I; when the defect rate is between 70% and 90%, the lubricant injection rate is standard at Level II; and when the defect probability is less than 70%, the lubricant injection rate is reduced at Level III; and a PID control algorithm is used to dynamically adjust the lubricant injection pressure to 0.2-0.8MPa.

[0120] Among them, the mold deformation trend is determined based on strain energy density analysis, and the residual stress distribution data is obtained through finite element simulation calculation; the servo motor drives the mold fine-tuning mechanism through a ball screw, and the displacement resolution reaches 0.1μm; the lubrication level is divided according to the statistical distribution model of the defect probability, and the probability threshold is determined through historical production data training; the injection pressure adjustment adopts closed-loop PID control, and the control cycle is synchronized with the mold compensation action.

[0121] Specifically, mold compensation calculations first determine the mold deformation trend through strain energy density analysis. Residual stress distribution data is obtained through finite element simulation and used to guide the servo motor's compensation direction. The servo motor drives the mold fine-tuning mechanism via a ball screw mechanism, achieving a displacement resolution of 0.1μm and a compensation accuracy of ±2μm. The lubrication system automatically switches operating modes based on the online defect probability calculation: when the defect probability exceeds 90%, double injection mode is triggered, increasing the injection volume to 200% of the baseline value; when the defect probability is between 70% and 90%, the standard injection volume is maintained; and when the probability is below 70%, reduced-frequency injection mode is adopted, reducing the injection frequency by 50%. The injection pressure is dynamically adjusted within the range of 0.2-0.8MPa by a PID controller. Real-time feedback from a pressure sensor forms a closed-loop control loop. The control cycle is synchronized with the mold compensation action, ensuring the timing of lubricant supply and mechanical compensation. This solution, through the synergistic effect of graded lubrication control and dynamic pressure regulation, reduces lubricant consumption by 30% while maintaining consistent surface quality after compensation.

[0122] Specifically, the lubrication levels are divided into: Defect suppression: The amount of lubricant injection directly affects the contact state between the material and the mold: High defect probability (>90%): Double injection (Level I) can form a thicker oil film to reduce the risk of surface scratches and material tearing. Low defect probability (<70%): Reduced frequency injection (Level III) avoids material sliding deviation caused by excessively thick oil film and prevents secondary deviation in rebound amount. Association with compensation amount: When the defect probability is high (such as material cracking), the workpiece stiffness distribution changes and the mold compensation amount needs to be recalibrated. PID pressure adjustment: Dynamic matching compensation amount: The injection pressure (0.2-0.8MPa) is controlled in real time through the PID algorithm to ensure that the lubricant flow is synchronized with the compensation action: High pressure (0.8MPa): Rapidly form a lubricating film and cooperate with the mold for rapid fine-tuning. Low pressure (0.2MPa): Minimal lubrication avoids interference with the mold-material contact state after high-precision compensation.

[0123] The above steps establish a closed-loop linkage between mechanical and process compensation. The forward control chain: die compensation (mechanical) → correcting geometric deviations → reducing springback; lubrication adjustment (process) → suppressing surface defects → stabilizing material mechanical properties → improving compensation calculation accuracy. The reverse feedback chain: if insufficient lubrication leads to defects (such as scratches), the actual workpiece stiffness changes, springback prediction becomes invalid, and a model update is required. This overcomes the limitations of a single compensation method. By combining Level I lubrication with die compensation, springback control accuracy can be improved from ±0.15mm to ±0.05mm. This method enables precise calculation and execution of die compensation. Strain energy density analysis accurately identifies high-stress areas in the die, providing a precise basis for compensation. A servo motor-driven fine-tuning mechanism achieves high-precision compensation of ±2μm. Dynamic adjustment of lubrication level based on defect probability effectively reduces surface defects caused by friction. A PID algorithm controls injection pressure to ensure uniform lubricant distribution. These measures collectively improve the stability of the cold stamping process and product quality.

[0124] The present invention further proposes a method for building a traceable knowledge base in an MES system, comprising the following steps: using a time-series database storage device to run the original data stream; constructing a knowledge graph to associate material batch numbers, mold numbers, and quality defect patterns; deploying an OPC UA-based data interface to achieve interconnection with the ERP system; and establishing an SQL traceability query engine to support multi-dimensional root cause analysis.

[0125] Among them, the time series database adopts a distributed architecture to store tens of thousands of sensor data streams per second, with a data compression rate of no less than 70%; the knowledge graph establishes a triple relationship model through the graph database, and the nodes contain material physical properties and process parameters; the OPC UA interface realizes two-way communication of 2,000 data per second; the SQL engine supports time series correlation queries, with a response time of less than 3 seconds.

[0126] Specifically, the raw data stream of equipment operation is connected to the time series database through the industrial protocol gateway, and a timestamp partition storage strategy is adopted, in which pressure data is stored at a sampling rate of 5kHz and temperature data is archived at intervals of 0.1 seconds. When constructing the knowledge graph, the batch number and mold number of the aluminum alloy material are used as entity nodes, and relationship edges are established through the quality defect pattern. At the same time, the elastic modulus parameters are associated with historical process adjustment records. A secure communication channel is configured based on the OPC UA data interface to achieve real-time synchronization of production work order data between the MES system and the ERP system, and the transmission delay is controlled within 50ms. The SQL traceability query engine has a built-in time window function, which can perform cross-data table association queries. For example, it automatically associates the pressure curve and temperature monitoring values ​​of the corresponding time period through the time of defect occurrence, thereby achieving multi-dimensional root cause positioning of quality problems and improving defect analysis efficiency by 70%.

[0127] As a preferred embodiment, the solution of the present invention is specifically implemented as follows: When constructing a traceable knowledge base in the manufacturing execution system, the original data stream is first run using a time series database storage device. The data stream records the mold pressure curve, hydraulic displacement time series, and temperature monitoring data of the stamping equipment in a time series format. Furthermore, a knowledge graph is constructed through a graph database, and a triple relationship is established between the material batch number, mold number, and quality defect pattern, wherein the material batch number is associated with the material elastic modulus parameter, the mold number is associated with the historical compensation record, and the quality defect pattern classification stores the morphological characteristic data of cracks, wrinkles, and scratches. Subsequently, a data interface module based on the OPC UA protocol is deployed to achieve interconnection between the manufacturing execution system and the enterprise resource planning system, and the equipment operating parameters and production order information are mapped through a unified address space model. Finally, a structured query language traceability engine is established to support multi-dimensional combined queries based on time range, defect type, and material batch. The traceability query engine optimizes query response efficiency through precompiled statements and automatically generates stamping process parameter adjustment suggestions through the associated root cause analysis algorithm.

[0128] Through the above-mentioned technical solution, the present invention effectively solves the problem of data silos in the quality traceability process, achieving efficient cross-system data association and rapid retrieval. The construction of the knowledge graph organically links material properties, mold status, and defect patterns. The structured query language engine quickly locates abnormal batches through multi-dimensional combination conditions. The OPC UA interface breaks down the barriers between production data and management systems, thereby forming a complete quality traceability chain and significantly improving the accuracy and efficiency of process anomaly analysis.

[0129] The present invention further proposes a method for building a traceable knowledge base in the MES system, including using a time-series database storage device to run the original data stream, building a knowledge graph to associate material batch numbers, mold numbers and quality defect patterns, deploying an OPC UA-based data interface to achieve interconnection with the ERP system, and establishing an SQL traceability query engine to support multi-dimensional root cause analysis.

[0130] The time series database uses a circular buffer structure to compress and store raw pressure, displacement, and temperature waveform data using millisecond-level timestamps, with a compression rate of less than 80%. The knowledge graph establishes triple relationships through the graph database, probabilistically associating stamping defect patterns with material elastic modulus deviation values ​​and the number of mold fine-tuning attempts. The OPC UA interface is configured with a bidirectional communication protocol to automatically map process parameter adjustment records to ERP production work orders. The SQL traceability query engine has a built-in multi-condition indexing algorithm that supports combined queries by time interval, equipment number, and defect type, with a response time of less than 2 seconds.

[0131] Specifically, the pressure and displacement signals generated by equipment operation are stored in a time series database in the form of a time series. Each record contains a timestamp, equipment ID, and original value, and the storage period is set to continuously roll over 12 months. The knowledge graph is constructed using the Neo4j graph database. Nodes include mold numbers, material batch numbers, and defect type codes, and edge attributes record association strength coefficients. When querying for a specific defect pattern, the engine automatically traverses the associated nodes to generate a causal chain. The OPC UA interface establishes a secure channel between MES and ERP. Process parameter adjustment instructions are encapsulated as UA node information and bidirectionally bound to the material code and process version number in the ERP work order. The SQL traceability engine uses a columnar storage structure and establishes a B+ tree index for quality inspection data. When executing joint queries, it optimizes the IO load through predicate push-down technology, achieving a second-level response for 500,000 records. This method reduces the time it takes to locate the root cause of stamping defects from 30 minutes to less than 5 minutes, and improves the data association accuracy to 98%.

[0132] As a preferred embodiment, the solution of the present invention is specifically implemented as follows: a storage architecture based on a time series database is configured in the MES system, and an SQL query engine with multi-threaded processing capabilities is deployed. The engine has a built-in data index module and establishes an association relationship between the material batch number and the process parameter table. Specifically, a defect pattern dimension table is created in the data table design, which includes the stamping batch number, mold number, and sensor data fingerprint fields, and is bidirectionally associated with the quality inspection result table through foreign key constraints. When the query engine receives a composite condition query request including a time range, a material elastic modulus interval, and a mold pressure peak, it automatically parses the query conditions and generates an execution plan, and quickly locates matching records through parallel scanning indexes. For example, when an edge crack defect is detected in a batch of workpieces, the operator inputs the timestamp range, mold number, and temperature fluctuation threshold, and the query engine jointly scans the process parameter history table and the defect image feature table to generate an associated result set including the material batch number, the holding time adjustment record, and the corresponding defect probability.

[0133] Through the above technical solution, the present invention realizes cross-table correlation query between process parameter adjustment records and quality inspection data, solves the problem of low efficiency of manual cross-system data retrieval, and enables the root cause analysis of quality problems to be quickly located to specific material batches or equipment operating parameter abnormalities based on multi-dimensional data association, effectively shortening the time period for defect tracing.

[0134] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A cold stamping quality control method based on multi-source data fusion and real-time optimization, characterized in that: The MES system is used to achieve closed-loop control of the entire production process, including the following steps: Step 1: The stamping equipment’s die pressure, hydraulic displacement, temperature, and workpiece surface image data are collected through a distributed sensor network and transmitted to the MES system after processing. Step 2: Extract stamping energy characteristics through time-frequency analysis of pressure data, build a dynamic deformation model based on hydraulic displacement, calculate thermal compensation based on temperature, and extract defect characteristics through multi-scale morphological processing of images; Step 3: Normalize the energy characteristics, deformation model, thermal compensation amount, and defect characteristics, generate dynamic weight coefficients based on the elastic modulus, and construct a three-dimensional fusion feature space through principal component analysis; Step 3-1: Query the elastic modulus value of the current batch of aluminum alloy from the MES system material database; Step 3-2: Calculate the objective weight coefficient of the punching energy feature based on the entropy weight method; Step 3-3: Generate a material property correction factor based on the elastic modulus. The material property correction factor is dynamically adjusted based on the deviation ratio between the measured elastic modulus and the standard value. When the elastic modulus is higher than the standard value, the correction factor linearly increases the displacement model weight in the objective weight coefficient; when the elastic modulus is lower than the standard value, the thermal compensation weight in the objective weight coefficient is increased. Step 3-4: Reduce the dimensionality of the four-dimensional features to three-dimensional space through principal component analysis, retaining more than 95% of the variance information; Step 3-4-1: Input the stamping energy characteristics, deformation model, thermal compensation amount and defect characteristic data into the principal component analysis algorithm. By calculating the eigenvalues ​​and eigenvectors, the first three principal components with a cumulative contribution rate of more than 95% are selected to achieve dimensionality reduction. Step 4: Input the feature space into the pre-trained LSTM model, associate the stamping speed with the holding time, and output a quality prediction matrix containing the springback deviation and defect probability; Step 5: When the springback deviation exceeds ±0.15mm and the defect probability is greater than 85%, the data is sent to the PLC via the bus to adjust the stamping speed and holding time parameters; Step 6: Calculate the mold compensation amount based on the actual springback amount, and link the lubrication system to adjust the injection parameters in stages to perform micron-level compensation; Step 7: After every 50 stampings, the LSTM parameters are updated using the laser measurement data. Incremental learning is initiated when the prediction error is greater than 0.1 mm for three consecutive times. Step 8: Build a traceable knowledge base in the MES system to store batch data and form a closed-loop archive.

2. The cold stamping quality control method based on multi-source data fusion and real-time optimization according to claim 1 is characterized in that: The distributed sensor network in step 1 includes: A 16-point array piezoelectric pressure sensor group is arranged on the top of the mold cavity, with a sampling frequency of ≥5kHz and data transmission via the CAN bus protocol; A sub-micron grating displacement sensor installed at the piston rod of the hydraulic cylinder; The working surface of the stamping die is embedded with a 9-point infrared temperature measurement array, and the temperature data is calibrated using the time-of-flight algorithm; Dual-light source polarization industrial cameras are installed on both sides of the conveyor line to synchronously trigger and collect images of 0.05mm level defects on the workpiece surface.

3. The cold stamping quality control method based on multi-source data fusion and real-time optimization according to claim 1 is characterized in that: Constructing a dynamic deformation model based on hydraulic displacement in step 2 includes: Step 2-1: Perform sliding window filtering on the hydraulic displacement data, and adjust the window length adaptively; Step 2-2: Establish a material rebound prediction model based on the JC constitutive equation and calculate the theoretical deformation based on the elastic modulus data; Step 2-3: Extract the energy spectrum characteristics of the pressure signal through wavelet packet decomposition and divide it into low-frequency stable area and high-frequency impact area; Step 2-4: Perform grayscale projection analysis on the workpiece surface image and optimize the continuity of the defect boundary by combining morphological closing operation.

4. The cold stamping quality control method based on multi-source data fusion and real-time optimization according to claim 1 is characterized in that: The pre-training of the LSTM model in step 1 includes: Step 4-1: Collect historical production data to build a training set and divide the input sequence into sliding time windows; Step 4-2: Introduce the attention mechanism into the network structure to dynamically assign time weights to different features; Step 4-3: Use the Bayesian optimization algorithm to adjust the number of LSTM hidden layer nodes; Step 4-4: Perform weighted fusion of the real-time stamping speed parameters and the LSTM output gate.

5. The cold stamping quality control method based on multi-source data fusion and real-time optimization according to claim 1 is characterized in that: Adjusting the punching speed and holding time parameters in step 5 includes the following sub-steps: Step 5-1: Calculate the stamping speed correction value based on the thermal expansion coefficient of aluminum alloy; Step 5-2: Construct a holding time-temperature correlation curve based on historical process data, and dynamically adjust the holding time parameters using fuzzy control rules; Step 5-3: Send the corrected parameters to the PLC controller via the PROFINET industrial bus protocol, ensuring that the response delay is less than 50ms; Step 5-4: Generate parameter adjustment records in the MES system interface and associate the equipment operation log with the current process version number.

6. The cold stamping quality control method based on multi-source data fusion and real-time optimization according to claim 1 is characterized in that: In step 6, the mold compensation amount is calculated based on the actual rebound amount, and the injection parameters are adjusted in stages by the linkage lubrication system. The micron-level compensation is performed, which includes the following sub-steps: Step 6-1: Analyze the mold deformation trend based on the strain energy density and calculate the residual stress distribution; Step 6-2: The mold fine-tuning mechanism is controlled by a servo motor, with a compensation accuracy of ±2μm; Step 6-3: Classify the lubrication level according to the defect probability: when the defect probability is greater than 90%, the lubricant injection amount is double injection at level I; when the defect probability is between 70% and 90%, the lubricant injection amount is standard injection at level II; when the defect probability is less than 70%, the lubricant injection amount is reduced frequency injection at level III; Step 6-4: Use the PID control algorithm to dynamically adjust the lubricant injection pressure to 0.2-0.8 MPa.

7. The cold stamping quality control method based on multi-source data fusion and real-time optimization according to claim 1 is characterized in that: The incremental learning step in step 7 is implemented by transfer learning technology, including: Step 7-1: Construct a sliding time window to filter key historical data samples, including mold pressure curves, hydraulic displacement time series, temperature monitoring data, and defect image samples; Step 7-2: Use Hampel filter to remove abnormal measurement data; Step 7-3: Freeze the weights of the first three layers of the LSTM network and only update the output layer parameters; Step 7-4: Reuse the feature extraction capabilities of the historical model through transfer learning technology.

8. The cold stamping quality control method based on multi-source data fusion and real-time optimization according to claim 1 is characterized in that: Step 8 of constructing a traceable knowledge base in the MES system includes the following sub-steps: Step 8-1: Run the original data stream using the time series database storage device; Step 8-2: Build a knowledge graph to associate material batch number, mold number and quality defect pattern; Step 8-3: Deploy an OPC UA-based data interface to connect with the ERP system; Step 8-4: Establish a SQL traceability query engine to support multi-dimensional root cause analysis.

Images