A method and system for analyzing pressure transducer data of high-strength continuous stamping parts for automobiles

Abstract

This application provides a method and system for analyzing pressure deformation data of high-strength continuously stamped parts for automobiles, relating to the field of stamping forming technology for automotive parts. It solves the technical problems in existing technologies where high-strength continuously stamped parts are prone to cracking, wrinkling, poor flatness, and die interference, and where pressure deformation stability and production safety are difficult to guarantee. The method includes: real-time acquisition of multi-source data from the stamping production line, including mechanical data of the stamping process, die state data, strip position images, feeding mechanism motion parameters, and material deformation image data; extraction of temporal characteristic parameters of the pressure deformation behavior of the part during continuous stamping, and inputting them into a pre-trained defect prediction model to obtain prediction results of the type and risk level of potential forming defects in the current stamping batch; dynamic generation of process adjustment suggestions through a process parameter optimization model, and feedback to the stamping production line control system. This application is used in the process of analyzing pressure deformation data of high-strength continuously stamped parts for automobiles.

Description

Technical Field

[0001] This application relates to the field of automotive parts stamping technology, and in particular to a method and system for analyzing pressure deformation data of high-strength continuous stamped parts for automobiles. Background Technology

[0002] With the trend towards lightweight and high-strength automotive components, high-strength steel is widely used in the continuous stamping production of key structural components such as hatchback door gas spring reinforcement plates and hinge reinforcement plates. These parts have significant undulations and require high flatness of welded surfaces. High-strength steel itself exhibits significant springback, and the complex working conditions of continuous stamping with multi-station collaboration and continuous material conveying make them prone to forming defects such as cracking, wrinkling, and excessive flatness during production. Traditional pressure deformation analysis is mostly based on monitoring single mechanical parameters or offline simulation, resulting in incomplete data dimensions, inaccurate feature extraction, and delayed defect prediction. Furthermore, adjustments to the feeding height can easily cause interference with the lower die, and there is a lack of effective control measures for the decreased pressure deformation stability caused by die wear. This makes it difficult to meet the real-time, accurate, and safe control requirements of continuous stamping. Therefore, there is an urgent need for a pressure deformation data analysis method for high-strength continuously stamped automotive parts that can predict forming defects and die interference risks in advance, dynamically optimize process parameters, and consider health status control. Summary of the Invention

[0003] This application provides a method and system for analyzing compression deformation data of high-strength continuous stamping parts for automobiles. It solves the technical problems in the prior art, such as cracking, wrinkling, poor flatness, and mold interference in high-strength continuous stamping parts due to the single data acquisition dimension, inaccurate extraction of compression deformation features, and interference due to feeding height. Furthermore, it addresses the difficulty in ensuring compression deformation stability and production safety.

[0004] To achieve the above objectives, this application adopts the following technical solution: In a first aspect, a method for analyzing compression deformation data of high-strength continuously stamped automotive parts is provided, comprising: S1: real-time acquisition of multi-source data from the stamping production line, wherein the multi-source data includes mechanical data of the stamping process, die state data, strip position images, feeding mechanism motion parameters, and material deformation image data; S2: based on the multi-source data, extracting temporal characteristic parameters of the compression deformation behavior of the parts during continuous stamping; S3: inputting the temporal characteristic parameters into a pre-trained defect prediction model to obtain prediction results of the type and risk level of potential forming defects in the current stamping batch; S4: based on the prediction results, dynamically generating process adjustment suggestions through a process parameter optimization model, and feeding the suggestions back to the stamping production line control system.

[0005] Based on the above technical solutions, the compression deformation data analysis method for high-strength continuous stamping parts for automobiles provided in this application overcomes the limitations of single-parameter detection by integrating multi-source acquisition modes of stamping mechanical data, die status data, strip position images, feeding mechanism motion parameters, and material deformation image data, achieving comprehensive coverage of compression deformation influencing factors. Furthermore, by extracting targeted time-series characteristic parameters such as dynamic time regularization distance, springback tendency index, and interference risk characteristics, it accurately characterizes the compression deformation behavior and potential key patterns of the parts. With the help of a pre-trained intelligent defect prediction model, it obtains the risk levels of defects such as cracking, wrinkling, flatness deviation, and die interference in advance, solving the problem of delayed defect prediction in traditional methods. Finally, through a process parameter optimization model constructed using reinforcement learning, combined with a constraint rule library of feeding height, flatness, and interference risk, it achieves multi-objective dynamic optimization, simultaneously combined with die health status assessment to form full-process control, ultimately ensuring the forming accuracy, production stability, and equipment safety of high-strength continuous stamping parts.

[0006] In conjunction with the first aspect above, in one possible implementation, the mechanical data in S1 includes, but is not limited to, the punching force F_i (t), the blank holder force F_b (t), and the sliding process s (t) of each station, where i = 1, 2, ..., n represent the station number and t is a time parameter; the mold state data includes the temperature T (t), vibration acceleration a (t), and wear monitoring value W (t) at key locations; the material deformation image data is acquired by a high-speed vision system deployed near the mold cavity and is used to characterize the material thickness change Δh and the surface strain field distribution ε.

[0007] In conjunction with the first aspect above, in one possible implementation, S1 further includes: real-time acquisition of the position image of the material strip in the mold and the motion parameters of the feeding mechanism; the time-series feature parameters further include the real-time gap feature parameter δ(t) of the material strip relative to the highest point of the lower mold, extracted based on the position image and motion parameters.

[0008] In conjunction with the first aspect mentioned above, in one possible implementation, the timing characteristic parameters include: the dynamic time warping distance between the punching force curve of a specific workstation calculated based on mechanical data and the standard curve. Local thinning rate gradient G calculated based on deformed image data; springback tendency index calculated based on multi-source data fusion: in, These are the weighting coefficients. Material yield strain; interference risk characteristics based on gap detection ,in, This is the safety gap threshold.

[0009] In conjunction with the first aspect above, in one possible implementation, in S3, the defect prediction model is a deep neural network model trained based on historical production data, whose input is the time-series feature parameters, and whose output includes at least the crack risk probability. Risk of wrinkling and the probability of flatness deviation and the probability of mold interference .

[0010] In conjunction with the first aspect mentioned above, in one possible implementation, in S4, the process parameter optimization model is constructed using a reinforcement learning algorithm, with the objective function being the minimum defect prediction risk, expressed by the formula: in, This is an adjustable vector of process parameters, including feed height. Blank holder force setting value, etc. For the probability of each defect risk, These are the weighting coefficients. This is the penalty coefficient for parameter changes.

[0011] In conjunction with the first aspect mentioned above, in one possible implementation, the process parameter optimization model includes a pre-defined rule library concerning the relationship between feed height and product flatness, and mold interference risk. The constraints include: and When the optimization objective is to improve flatness and it is suggested to reduce the feeding height, the model simultaneously assesses the interference risk. When the risk exceeds the threshold, multi-objective optimization is initiated, and the highest safe feeding height that meets the flatness requirements is output.

[0012] In conjunction with the first aspect mentioned above, in one possible implementation, this method further includes: A mold health status assessment model is constructed, and the mold health index is calculated based on the collected mold status data and corresponding stroke information. It predicts the remaining service life of the device and triggers a preventative maintenance warning.

[0013] Secondly, a data system for high-strength stamped parts for automobiles is provided, comprising: a data acquisition module for real-time acquisition of multi-source data from the stamping process; a data processing and feature extraction module for synchronizing, filtering, and calculating feature parameters of the multi-source data; a defect prediction and diagnosis module, which incorporates the defect prediction model and outputs defect risk analysis results; a process optimization decision module, which incorporates the process parameter optimization model and generates and outputs process adjustment instructions; and a system interactive interface for displaying analysis results, early warning information, and receiving operator input.

[0014] In summary, in one possible implementation, the data acquisition module includes: force sensors and displacement sensors distributed at key locations on the press slide, mold backing plate, and punch and die; temperature sensors and vibration sensors mounted on the mold; a high-speed industrial camera and laser scanner arranged facing the mold cavity; and a visual positioning sensor for monitoring the position of the strip.

[0015] This application provides a data analysis method and system for pressure deformation of high-strength stamped parts for automobiles. It can solve the problems of cracking, wrinkling, and poor flatness caused by single data acquisition dimensions, inaccurate extraction of pressure deformation features, and delayed defect prediction in the production of high-strength continuous stamped parts in the prior art. It can also solve the problems of decreased pressure deformation stability and limited lifespan caused by mold interference hazards due to the lack of collaborative assessment of interference risks in feeding height. Through multi-source data acquisition, accurate feature extraction, intelligent defect prediction, multi-objective process optimization, and mold health management, it can achieve early avoidance of forming defects and interference risks, ensure pressure deformation stability and production safety, extend mold lifespan, and significantly improve the forming accuracy and production efficiency of high-strength steel parts.

[0016] It should be understood that the descriptions of technical features, technical solutions, beneficial effects, or similar language in this application do not imply that all features and advantages can be achieved in any single embodiment. Rather, it is understood that the description of a feature or beneficial effect means that a specific technical feature, technical solution, or beneficial effect is included in at least one embodiment. Therefore, the descriptions of technical features, technical solutions, or beneficial effects in this specification do not necessarily refer to the same embodiment. Furthermore, the technical features, technical solutions, and beneficial effects described in this embodiment can be combined in any suitable manner. Those skilled in the art will understand that embodiments can be implemented without one or more specific technical features, technical solutions, or beneficial effects of a particular embodiment. In other embodiments, additional technical features and beneficial effects may be identified in specific embodiments that do not embody all embodiments. Attached Figure Description

[0017] Figure 1 A flowchart illustrating a method for analyzing pressure transducer data of high-strength continuous stamped parts for automobiles, provided in this application embodiment; Figure 2 This is a system structure diagram of a pressure transducer data analysis system for high-strength continuous stamping parts for automobiles, provided in an embodiment of this application. Detailed Implementation

[0018] In the description of this application, unless otherwise stated, " / " means "or," for example, A / B can mean A or B. The "and / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, and B alone. Furthermore, "at least one" means one or more, and "multiple" means two or more. The terms "first," "second," etc., do not limit the quantity or order of execution, and "first," "second," etc., do not necessarily imply differences.

[0019] It should be noted that, in this application, the terms "exemplary" or "for example" are used to indicate that something is being described as an example, illustration, or illustration. Any embodiment or design described as "exemplary" or "for example" in this application should not be construed as being more preferred or advantageous than other embodiments or design solutions. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0020] Figure 1 A flowchart illustrating a method for analyzing pressure transformation data of high-strength continuous stamped parts for automobiles, as provided in this application embodiment, is shown below. Figure 1 As shown, the method includes: S1: Real-time acquisition of multi-source data from the stamping production line, including mechanical data of the stamping process, mold status data, and material deformation image data.

[0021] The mechanical data includes, but is not limited to, the impact force at each workstation. Blank pressure The sliding process s(t) is represented by i=1,2,......,n, where i represents the station number and t is the time parameter; the mold status data includes the temperature T(t), vibration acceleration a(t), and wear monitoring value W(t) at key locations; the material deformation image data is acquired by a high-speed vision system deployed near the mold cavity and is used to characterize the material thickness change Δh and the surface strain field distribution ε.

[0022] Specifically, mechanical data is collected using force and displacement sensors distributed at key locations on the press slide, die pad, and punch / cone, with a collection frequency of no less than 1kHz, to accurately capture the dynamic changes in force and displacement power within each stamping cycle. Temperature data is collected using contact temperature sensors, covering areas prone to heat generation such as the die cavity and die / cone cutting edges, with a detection range of 0-200℃. Vibration acceleration data is collected using piezoelectric vibration sensors to identify abnormal working conditions such as model loosening and off-center loading. Wear monitoring values ​​are obtained by periodically scanning the die / cone cutting edge contour using laser displacement sensors, and the wear is calculated by comparing it with the new die reference contour. Material deformation image data is acquired using a high-speed industrial camera and laser scanner deployed near the die cavity. The high-speed industrial camera has a frame rate of no less than 1000fps, and, in conjunction with a ring light source, clearly captures the surface morphology changes of the material during the stamping process. The laser scanner has an accuracy of ±0.01mm and is used to measure the material thickness change Δh and surface strain field distribution ε, providing intuitive image and quantitative data support for subsequent compression deformation feature extraction.

[0023] Furthermore, the multi-source data also includes: real-time acquisition of images showing the position of the material strip in the mold and the motion parameters of the feeding mechanism; the time-series feature parameters also include real-time gap feature parameters of the material strip relative to the highest point of the lower mold, extracted based on the position images and motion parameters. .

[0024] Specifically, a visual positioning sensor is installed at the feeding inlet to acquire relative position images of the material strip edge and the highest point of the lower die at a sampling frequency of no less than 500Hz; simultaneously, the displacement of the feeding mechanism is acquired in real time through a servo motor encoder. With speed of movement By combining camera calibration parameters with the kinematic model of the mechanism, the instantaneous vertical distance between the front end of the material strip and the highest point of the lower die is calculated in real time. ,in A sequence of location images. The time gap calculation function is based on geometric transformation and motion compensation; time gap characteristic parameters. The detection accuracy is ±0.1mm, which can dynamically reflect the changes in safety margin caused by material belt vibration, mold thermal deformation or equipment mechanical clearance accumulation, providing key input for subsequent interference risk warning and adaptive optimization of feeding height.

[0025] Based on this, by coordinating the deployment of various data acquisition devices, including force sensors, displacement sensors, temperature sensors, piezoelectric vibration sensors, laser displacement sensors, high-speed industrial cameras, laser scanners, and visual positioning sensors, at key locations such as the press, mold, and feeding port, multiple types of data acquisition devices are used to simultaneously acquire mechanical data, mold status data, material deformation image data, strip position images, and feeding mechanism motion parameters during the stamping process. This multi-dimensional and comprehensive data acquisition breaks through the limitations of traditional single-parameter monitoring in the stamping process, enabling comprehensive perception of the entire process status of high-strength steel continuous stamping. It also provides complete and accurate quantitative data support for subsequent pressure deformation feature extraction, defect prediction, process optimization, mold health assessment, and interference risk warning, avoiding analytical errors caused by missing monitoring dimensions or insufficient data accuracy, and ensuring the reliability and accuracy of subsequent technical links.

[0026] S2: Based on multi-source data, extract the temporal characteristic parameters of the part's pressure change behavior during continuous stamping.

[0027] The time-series characteristic parameters include: The dynamic time warping distance between the specific station punching force curve and the standard curve calculated based on mechanical data The local thinning rate gradient G is calculated based on deformed image data, and the springback tendency index is calculated based on multi-source data fusion. ,in, These are the weighting coefficients. Interference risk characteristics based on gap detection, representing the material's yield strain. ,in, This is the safety gap threshold.

[0028] Specifically, dynamic time warping distance The calculation steps are as follows: First, select key forming stations that significantly affect pressure change, such as flanging and punching. Based on the stamping pressure timing data of 500 qualified parts continuously produced during stable production line operation, obtain the stamping pressure timing curve of this station in one stamping cycle. Then, the arithmetic mean of all impact force values ​​corresponding to the same formation location is calculated using the formula: Retrieve the standard stamping pressure curve for the material grade and thickness corresponding to this workstation. Finally, a dynamic time warping algorithm is used to align and match the two curves, and the cumulative distance between the curves is calculated as the result. , The larger the value, the more obvious the deviation of the current pressure curve from the standard state, and the greater the risk of abnormal pressure deformation.

[0029] Based on this, by selecting key forming stations such as flanging and punching, a standard curve is fitted based on the historical stamping force data of qualified parts. Then, the dynamic time warping algorithm is used to calculate the actual punching force curve. Deviation distance from the standard curve This eliminates the problem of misalignment of the curve time axis caused by stamping cycle fluctuations, and enables precise comparison of the timing variation law of stamping force. It can intuitively quantify the degree of deviation between the actual stamping process and the standard working conditions. The greater the deviation, the higher the risk of pressure deformation anomaly. This provides the core characteristic indicators that can accurately reflect the abnormal mechanical parameters for the subsequent defect prediction model, and improves the sensitivity and accuracy of pressure deformation anomaly identification.

[0030] Specifically, the local thinning rate gradient G is calculated based on material deformation image data. First, the material thickness distribution data of the part before and after stamping is obtained by a laser scanner to calculate the local thinning rate at each sampling point. ,in, The original material is thick. The thickness of the material at the sampling point is used as the reference. Then, by selecting key areas such as the welding surface and ribs of the part, the ratio of the difference in thinning rate between adjacent sampling points to the spatial distance is calculated to obtain the local thinning rate gradient G. The larger the value of G, the more drastic the change in material thickness in the corresponding area, and the higher the risk of cracking.

[0031] Based on this, using the material thickness distribution data of parts before and after stamping collected by a laser scanner, the local thinning rate of each sampling point is calculated, and the local thinning rate gradient G of key areas is further extracted. The discrete data of material thickness change is transformed into a quantifiable gradient index, which can accurately capture the material thickness abrupt change characteristics of easily cracked parts such as weld surfaces and ribs. The magnitude of G value can directly characterize the severity of material thickness change in the corresponding area, thereby achieving accurate quantification of cracking risk during high-strength steel stamping. This provides core features that can reflect abnormal material deformation for subsequent defect prediction models, greatly improving the pertinence and accuracy of cracking defect prediction.

[0032] Specifically, the rebound tendency index Multi-source data fusion computing is employed, in which The maximum plastic strain time series value acquired by the data acquisition module. This corresponds to the yield strain of high-strength steel; The peak value of the punch pressure timing curve. The average stamping force within the same period; the weighting coefficients α and β range from 0.4 to 0.6, and α + β = 1, calibrated by fitting more than 500 sets of historical stamping data. The higher the value, the more pronounced the springback tendency of the part after stamping.

[0033] Based on this, a springback tendency index is constructed by integrating multi-source data such as the maximum plastic strain of the material and the peak and average values ​​of the impact force. Based on a large amount of historical stamping data, the weighting coefficients α and β were fitted and calibrated, and the originally independent deformation parameters were correlated with mechanical parameters to achieve accurate quantitative characterization of the springback tendency of high-strength steel during stamping. It can intuitively reflect the degree of springback risk after part stamping, solve the problem that traditional methods are difficult to comprehensively assess the factors affecting springback, and provide core characteristic indicators for subsequent defect prediction models that can accurately determine springback risk, thus improving the scientificity and accuracy of springback defect prediction.

[0034] Specifically, characteristics of interference risk Based on the material strip position image and motion parameter calculation The gap value between the material strip and the highest point of the lower die is collected in real time by the visual positioning sensor. The safety clearance threshold determined through interference analysis of the mold model is set to be no less than 3 mm; when When the current gap is within the safety threshold range, it indicates that there is no safety risk; when the gap is... When the value is greater than 0, it indicates that the gap is less than the safety threshold, and the larger the value, the higher the risk of interference.

[0035] Based on this, interference risk characteristics are calculated using the material strip position image and the motion parameters of the feeding mechanism. Through real-time gap values Mold safety clearance threshold The ratio relationship transforms the relative position of the strip and the lower die into a quantifiable risk indicator, which can intuitively reflect the interference risks caused by strip vibration, die thermal deformation, or accumulated mechanical clearance of the equipment. The size of the interference risk level can be accurately determined, which solves the problem that interference risk is difficult to quantify and warn in real time during traditional stamping processes. It provides a key basis for interference risk determination for subsequent process parameter optimization models and reduces the probability of collision between the die and the strip.

[0036] In summary, based on multi-source monitoring data of stamping, dynamic time-regulating distance is extracted in a targeted manner. Local thinning rate gradient G, rebound tendency index Interference risk characteristics Four core time-series characteristic parameters, through differentiated calculation logic, accurately quantify four types of pressure deformation anomalies: stamping pressure fluctuation, local material thinning, part springback tendency, and strip die interference. This not only solves the problems of discrete multi-source data, weak correlation, and difficulty in direct defect judgment in traditional stamping analysis, but also transforms the implicit pressure deformation risk in the continuous stamping process of high-strength steel into intuitive quantitative indicators, forming a feature system covering mechanical, deformation, and interference dimensions. This provides highly identifiable and highly correlated input basis for subsequent defect prediction models and process parameter optimization models, significantly improving the sensitivity of pressure deformation anomaly identification, the accuracy of defect prediction, and the pertinence of process optimization.

[0037] S3: Input the time-series feature parameters into the pre-trained defect prediction model to obtain the prediction results of the type and risk level of potential forming defects in the current stamping batch.

[0038] The defect prediction model is a deep neural network model trained based on historical production data. Its input is the time-series feature parameters, and its output includes at least the crack risk probability. Risk of wrinkling and the probability of flatness deviation and the probability of mold interference .

[0039] Specifically, the defect prediction model collects 1000 sets of historical stamping production data, covering production conditions of different grades of high-strength steel (340-780MPa) and different material thicknesses (1-3mm). Each set of data includes complete time-series characteristic parameters. The dataset includes the actual defect detection results, such as cracking, wrinkling, flatness deviation, and whether and how severe mold interference occurs. The dataset is normalized to eliminate the dimensional differences of different parameters. At the same time, the SMOTE algorithm is used to deal with the problem of sample imbalance to ensure that the proportion of various defect samples is balanced.

[0040] Leveraging the strong fitting ability of Long Short-Term Memory (LSTM) networks to time-series data, a network architecture consisting of an input layer, hidden layers, and an output layer was constructed. The input layer dimension matched the number of time-series parameters, the hidden layers were set to 2-3 layers with 64-128 neurons per layer, and the output layer used the Sigmoid activation function to output the risk probabilities of four types of defects. The Adam optimizer was employed, with the cross-entropy loss function as the objective function, and early stopping was used to prevent overfitting.

[0041] The pre-trained defect prediction model is embedded into the data processing system of the stamping production line. The prediction trigger condition is set to trigger batch prediction once every 50-100 stamped parts. During the prediction process, the temporal feature parameters of the current batch are extracted in real time and input into the model, and the model outputs... , , , Four risk probability values ​​are used to determine the defect risk level of the current batch based on the preset risk level threshold, and the prediction results are pushed to the process optimization decision module in real time.

[0042] Based on this, by inputting the multi-dimensional temporal feature parameters extracted by S2 into a pre-trained defect prediction model built on LSTM, the model is trained using historical production data covering multiple grades of high-strength steel and various material thicknesses. Through normalization processing and the SMOTE algorithm, the validity of the data and the balance of the samples are ensured. It can efficiently fit the correlation between stamping deformation behavior and defect occurrence. After the model is embedded in the production line data processing system, it can trigger real-time prediction according to set batches, outputting the risk probability values ​​of four types of defects: cracking, wrinkling, flatness deviation, and die interference, and determining the risk level. This not only breaks through the limitations of traditional post-event detection of stamping defects and realizes the early prediction of potential defects, but also provides a clear decision-making guide for subsequent process parameter optimization through quantified risk probabilities, significantly reducing the probability of batch scrap and improving the operational stability and intelligence level of the stamping production line.

[0043] S4: Based on the prediction results, process adjustment suggestions are dynamically generated through the process parameter optimization model, and the suggestions are fed back to the stamping production line control system.

[0044] The process parameter optimization model is constructed using a reinforcement learning algorithm, with the objective function being the minimum defect prediction risk, and employing the formula: ,in, This is an adjustable vector of process parameters, including feed height. Blank holder force setting value, etc. For the probability of each defect risk, These are the weighting coefficients. This is the penalty coefficient for parameter changes.

[0045] Specifically, the process parameter optimization model is based on a deep reinforcement learning (DRL) framework and employs the deep deterministic policy gradient (DDPG) algorithm to achieve continuous process parameter optimization decisions. The state space S is determined by the curve risk probability of the current stamping batch (…). , , , ), real-time time series characteristic parameters ( The components are determined based on the actual number of features and are normalized.

[0046] The motion space A is an adjustable process parameter vector θ, specifically including the feeding height. Set value of blank holder force at each station Continuous parameters, where The adjustment range is [ The blank holder force is adjustable within the range of [50kN, 500kN] to ensure that the parameters are adjusted to meet the physical constraints of the production line equipment.

[0047] In the objective function middle, As a penalty item for defect risk, The weighting coefficients for each defect risk are determined based on the degree of defect impact: Cracking risk weight. =0.4, Flatness deviation risk weight =0.3, wrinkle risk weight =0.2, Interference Risk Weight =0.1, and satisfies =1; For parameter change penalty term, This is a penalty coefficient, ranging from 0.01 to 0.05, used to prevent drastic fluctuations in process parameters from causing production line instability. The parameter adjustment amount is the squared Euclidean distance between the current parameter adjustment and the parameter value of the previous period. The optimization objective of the model is to minimize... This means simultaneously reducing defect risk and stabilizing process parameters.

[0048] The training mode combines offline simulation and online fine-tuning. First, a stamping process simulation environment is built based on historical production data. Different combinations of process parameters are input to simulate changes in defect risk and complete the offline pre-training of the model. After training, the model is deployed to the actual production line. Real-time data is collected through online interaction to continuously fine-tune the strategy grid parameters. An experience backtracking pool is introduced to store state-action-reward samples. Randomly sampled samples are used to construct the network and value network. The value network is used to evaluate the long-term benefits of the current action, and the strategy network is used to output the optimal process parameter vector θ.

[0049] The optimal process parameter vector θ output by the model needs to be verified against the preset constraint rules. After the constraints of feeding height and interference risk are met, it is converted into specific control commands and sent to the servo feeding system and blank holder force control system of the production line to realize the dynamic adjustment of process parameters. At the same time, the changes in defect risk after adjustment are recorded as training data for the next iteration.

[0050] In summary, based on the risk probability output by the defect prediction model, a process parameter optimization scheme with the goal of minimizing defect risk is constructed through a deep reinforcement learning model. Under the premise of taking into account the stability of process parameters, adjustment instructions for parameters such as feeding height and blank holder force are dynamically generated and sent to the production line control system. This achieves adaptive optimization of stamping process parameters, reduces the occurrence rate of forming defects, and improves the intelligent control level and stable operation capability of the production line.

[0051] Furthermore, the process parameter optimization model includes a pre-defined rule library relating the feeding height to product flatness and mold interference risk. The constraints include: and When the optimization objective is to improve flatness and it is suggested to reduce the feeding height, the model simultaneously assesses the interference risk. When the risk exceeds the threshold, multi-objective optimization is initiated, and the highest safe feeding height that meets the flatness requirements is output.

[0052] Specifically, the constraint rule base is constructed based on historical process experimental data and 3D interference simulation results. It uses corresponding data of product flatness monitoring values ​​and real-time mold clearance monitoring values ​​under different feeding heights to obtain the feeding height-flatness correlation formula through multiple linear regression fitting. ,in Here, k is the flatness deviation value, and k is the fitting coefficient. The reference feeding height is used; simultaneously, interference simulation is performed based on the 3D model of the mold to determine the minimum safe clearance at different material strip positions. and the physical limits of the feeding height [ The associated formulas, constraint thresholds, and corresponding working condition labels are entered into the rule base to form a set of constraint parameters that can be dynamically called.

[0053] When the model's core optimization objective is to improve flatness, it prioritizes outputting values ​​that minimize flatness deviation. Feeding height adjustment scheme for materials below the acceptable threshold; synchronously retrieve gap constraints from the rule base. If so, the feeding height and the corresponding edge clamping force adjustment value will be output directly; The system was deemed to have exceeded the interference risk threshold, triggering a multi-objective optimization process.

[0054] Multi-objective optimization takes meeting the flatness requirement as a hard constraint and maximizing the feeding height as the optimization objective, and reconstructs the objective function as follows: The model solves this constrained optimization problem using the Particle Swarm Optimization (PSO) algorithm, traversing the parameter points within the feasible region of the feeding height and selecting all points that meet the flatness requirements. The maximum value is selected as the optimal solution to ensure precise control of product flatness while avoiding mold interference risks.

[0055] Based on this, by pre-setting a constraint rule library based on historical process experimental data and mold 3D interference simulation results in the process parameter optimization model, the correlation and constraint conditions between feeding height and product flatness and mold interference risk are clarified. When the model adjusts the feeding height with the goal of improving flatness, the mold interference risk can be evaluated simultaneously. When the interference risk exceeds the standard, a multi-objective optimization process is initiated. The particle swarm optimization algorithm is used to solve the maximum safe feeding height that meets the flatness qualification requirements. This not only solves the contradiction between improving flatness and mold interference risk in traditional process adjustment, but also achieves the dual goals of precise flatness control and mold safety protection, avoiding mold collision accidents caused by blindly reducing the feeding height, and further improving the safety and reliability of process parameter optimization.

[0056] Furthermore, a mold health status assessment model is constructed. Based on the collected mold status data and corresponding stroke information, the mold health index is calculated, and its remaining service life is predicted, triggering a preventive maintenance warning.

[0057] Specifically, key state parameters throughout the mold's entire lifecycle are selected as evaluation indicators, including: steady-state temperature values ​​at key locations of the mold. Root mean square value of vibration acceleration Wear of the die edges (W), and the number of abnormal shutdowns during the stamping process. Based on historical fault data, weighting coefficients are assigned to each indicator, such as wear amount weighting. Vibration acceleration weight Temperature weighting =0.2, Weight of abnormal downtime The total weight is 1.

[0058] First, each indicator is normalized to eliminate dimensional differences, resulting in a health score for each indicator (range [0,1]): Wear and tear health score ,in, The maximum permissible wear of the die cutting edge is set based on the dimensional accuracy requirements of the part; vibration acceleration key value. ,in The vibration acceleration reference value for the new mold. The critical value of vibration acceleration before mold failure; temperature health score. ,in To ensure the safe operating temperature of the mold, The critical temperature for failure; health score for abnormal shutdown. ,in The maximum number of abnormal downtimes allowed in a single month; secondly, the mold health index is calculated by weighted summation. , The value range is [0,1]. The closer the value is to 1, the better the mold condition; the closer the value is to 0, the more severe the mold deterioration.

[0059] Based on the collected mold condition data and corresponding stroke counts, a health index-stroke deterioration curve is fitted. N represents the cumulative number of stamping cycles. A nonlinear regression algorithm is used to establish the degradation equation: ,in The initial health index for the new mold, where k is the cracking rate coefficient fitted from historical data; for example, setting a health index warning threshold. =0.3, when the calculated real-time value is 0.3. At that time, the model uses the degradation equation to inversely estimate the remaining service life. : ,in for The corresponding cumulative number of stamping times This represents the current cumulative number of stamping operations for the mold.

[0060] The health assessment model sets up a three-level early warning mechanism, based on... The numerical value automatically triggers corresponding warnings. For example, a level 1 warning indicates a good health status, where 0.7 < 0.7. ≤1 indicates only routine inspection and alert; Level 2 warning indicates slight degradation, 0.3 < ≤0.7 indicates a need for calculation and maintenance; a Level 3 warning indicates severe degradation. If the value is ≤0.3, an emergency maintenance is required. The machine should be stopped immediately and vulnerable parts of the mold replaced to avoid batch scrap or mold chipping accidents. The warning information will be pushed to the system interface at the same time, and maintenance suggestions and historical maintenance records will be recorded to form a closed loop for mold health management.

[0061] Based on this, a mold health status assessment model is constructed, integrating multi-dimensional status parameters such as mold temperature, vibration acceleration, cutting edge wear, and number of abnormal shutdowns. After normalization and weighted summation, the mold health index is calculated. Based on long-term data, it fits the degradation curve to predict the remaining service life, and with a three-level early warning mechanism, it realizes maintenance reminders and closed-loop management. It breaks through the limitations of traditional mold maintenance, which relies on experience and has strong lag. It can accurately capture the mold degradation trend, avoid batch scrap, equipment damage and downtime losses caused by mold failure in advance, extend the service life of the mold, ensure the continuous and stable operation of the stamping production line, and improve the overall production efficiency and cost control capabilities.

[0062] In summary, this method achieves end-to-end status awareness through multi-dimensional, high-precision, and multi-source data acquisition. It quantifies implicit pressure deformation risks by selectively extracting various time-series feature parameters, predicts potential defects in advance using an LSTM model, dynamically optimizes process parameters through a deep reinforcement learning model, and resolves the conflict between flatness improvement and mold interference using a pre-defined constraint rule base and multi-objective optimization. Furthermore, it combines a mold health status assessment model to predict degradation and implement closed-loop maintenance, forming a complete technical chain of "data acquisition - feature extraction - defect prediction - process optimization - mold maintenance." This solves the problems of insufficient precision in pressure deformation feature extraction and limited data acquisition dimensions in the existing high-strength continuous stamping parts production process. The system addresses issues such as part cracking, wrinkling, and non-flatness due to the lag in defect prediction. It also mitigates industry pain points such as mold interference risks, decreased pressure deformation stability, and limited mold life caused by the lack of coordinated assessment of feeding height adjustments. Through comprehensive multi-source data collection, precise extraction of pressure deformation characteristics, intelligent prediction of forming defects, multi-objective optimization of process parameters, and full-cycle management of mold health, the system achieves early identification and avoidance of forming defects and mold interference risks. This ensures the stability and safety of the stamping process, extends mold life, and significantly improves the forming accuracy of high-strength steel stamped parts and the overall production efficiency of the production line.

[0063] Figure 2 This application provides a pressure transformation data analysis system for high-strength continuous stamping parts for automobiles, such as... Figure 2 As shown, the system includes: The data acquisition module is used to acquire multi-source data of the stamping process in real time.

[0064] Furthermore, the data acquisition module includes: force sensors and displacement sensors distributed at key positions of the press slide, mold backing plate, and punch and die; temperature sensors and vibration sensors mounted on the mold; a high-speed industrial camera and laser scanner arranged facing the mold cavity; and a visual positioning sensor for monitoring the position of the strip.

[0065] Specifically, force sensors and displacement sensors are arranged at key force-bearing positions on the press slide, mold pad, and punch and die, acquiring the punching force of each station in real time at a sampling frequency of not less than 1kHz. Blank pressure The system accurately captures the dynamic changes in internal force and displacement during the gliding process (s(t)). A contact-type temperature sensor is installed in the mold cavity, die edges, and other heat-prone areas, covering a detection range of 0-200℃. It collects the temperature T(t) at key mold locations in real time, reflecting the mold's thermal equilibrium state. A piezoelectric vibration sensor, fixed to the mold body, collects the root mean square value of vibration acceleration a(t) to identify abnormal conditions such as mold loosening and uneven loading. A high-speed industrial camera and laser scanner are positioned facing the mold cavity, with a camera frame rate of at least 1000fps, working in conjunction with a ring light source to clearly capture the material's movement during the stamping process. The surface morphology changes are measured by a laser scanner with an accuracy of ±0.01mm to measure the material thickness distribution of the parts before and after stamping, and to calculate the material thickness change Δh and the surface strain field distribution ε. A visual positioning sensor is installed at the feeding inlet and collects images of the relative positions of the material strip edge and the highest point of the lower die at a sampling frequency of not less than 500Hz. At the same time, the displacement L(t) and motion speed v(t) of the feeding mechanism are obtained by the servo motor encoder. The real-time gap parameter δ(t) of the material strip relative to the highest point of the lower die is calculated through geometric transformation and motion compensation function. The detection accuracy reaches ±0.1mm, and the changes in safety margin caused by material strip jitter and die thermal deformation are dynamically fed back.

[0066] In summary, the data acquisition module integrates force sensors, displacement sensors, temperature sensors, vibration sensors, high-speed industrial cameras, laser scanners, and visual positioning sensors, and is strategically deployed at key locations such as the press, die, and feeding mechanism. This enables the simultaneous acquisition of stamping mechanical data, die status data, material deformation image data, and strip position parameters. It ensures the accuracy and frequency requirements of each data type and constructs a multi-dimensional data acquisition system covering the entire stamping process. This overcomes the limitations of traditional single-parameter monitoring and provides comprehensive, accurate, and timely basic data support for subsequent pressure deformation feature extraction, defect prediction, process optimization, and die health assessment. It avoids subsequent analysis errors caused by missing data dimensions or insufficient accuracy, thus improving the reliability and accuracy of the entire pressure deformation data analysis system.

[0067] The data processing and feature extraction module is used to synchronize, filter, and calculate feature parameters of multi-source data.

[0068] Specifically, the data processing and feature extraction module first performs time synchronization and noise filtering on the mechanical data, mold status data, material deformation image data, strip position image, and feeding mechanism motion parameters transmitted by the data acquisition module. This eliminates errors caused by data transmission delays and environmental interference, ensuring the temporal consistency and validity of the multi-source data. Secondly, based on the preprocessed multi-source data, the module calculates the core temporal characteristic parameters representing the pressure change behavior of the stamping process according to a preset algorithm. This includes using a dynamic time warping algorithm to calculate the deviation distance between the actual stamping pressure curve and the standard curve at a specific station. Calculate the local thinning rate gradient G in key areas based on material thickness distribution data; calculate the springback tendency index by integrating maximum plastic strain and impact force parameters. Based on the real-time gap value of the material strip and the safety threshold, the interference risk characteristics are derived. Finally, a set of highly recognizable quantitative feature parameters that can be directly used for defect prediction is output, providing a reliable input for the accurate calculation of subsequent models.

[0069] In summary, the data processing and feature extraction module effectively eliminates errors caused by data transmission delays and environmental interference by performing time synchronization and noise filtering preprocessing on multi-source acquired data, ensuring the temporal consistency and validity of multi-source data. Then, it specifically calculates the dynamic time warping distance. Local thinning rate gradient G, rebound tendency index Interference risk characteristics Four types of core quantitative feature parameters transform discrete and complex raw monitoring data into a highly recognizable set of feature parameters, solving the problem that traditional stamping data is difficult to use directly for defect judgment. This provides reliable input for the accurate calculation of subsequent defect prediction models, and greatly improves the sensitivity of pressure transformer anomaly identification and the accuracy of defect prediction.

[0070] The defect prediction and diagnosis module has a built-in defect prediction model and is used to output defect risk analysis results.

[0071] Specifically, the defect prediction and diagnosis module incorporates an LSTM deep neural network model trained on historical data from various grades of high-strength steel and different material thicknesses. This model, optimized through normalization and the SMOTE algorithm, ensures sample balance and data validity. It can also receive outputs from the data processing and feature extraction modules. G , The system uses four core time-series characteristic parameters to quickly calculate and output the risk probability values ​​of four types of defects: cracking, wrinkling, flatness deviation, and mold interference. At the same time, it determines the defect risk level of the current stamping batch based on preset thresholds and pushes the quantified risk analysis results to the process optimization decision-making stage in real time, so as to realize the early prediction of potential defects and provide clear decision guidance for subsequent process parameter adjustments.

[0072] The process optimization decision module has a built-in process parameter optimization model, which is used to generate and output process adjustment instructions.

[0073] Specifically, the process optimization decision module incorporates a process parameter optimization model built on the Deep Deterministic Strategy Gradient (DDPG) algorithm. This model takes minimizing defect prediction risk and process parameter fluctuations as its objective function. It incorporates the probabilities of four types of defect risks (cracking, wrinkling, flatness deviation, and mold interference) output by the defect prediction and diagnosis module, as well as the temporal feature parameters output by the data processing and feature extraction module, into the state space. It uses continuous parameters that conform to the physical constraints of the equipment, such as feeding height and blank holder force at each station, as the action space. It first completes offline simulation pre-training based on historical production data, and then deploys it to the actual production line to continuously fine-tune the strategy parameters through online interaction. At the same time, it calls the pre-set constraint rule library to verify the matching of feeding height and mold interference risk. When the flatness optimization objective conflicts with the interference risk threshold, it starts the particle swarm optimization algorithm to perform multi-objective optimization. Finally, it outputs the optimal process parameter vector that meets the production constraints and converts it into control commands to be issued to the servo feeding system and blank holder force control system of the production line. Simultaneously, it records the defect risk change data after parameter adjustment for model iterative optimization.

[0074] In summary, the process optimization decision module incorporates a process parameter optimization model built on a deep deterministic strategy gradient algorithm. By integrating defect risk probability and temporal characteristic parameters to construct a state space, and using parameters such as feeding height and blank holder force as the action space, it achieves adaptive optimization of process parameters through a combination of offline pre-training and online fine-tuning. At the same time, relying on a pre-set constraint rule library and particle swarm optimization algorithm, it effectively balances the requirements for improving product flatness and controlling mold interference risks. The output of optimal process parameter commands can directly drive the production line to perform precise control, which not only solves the problems of traditional process adjustment relying on experience and response lag, but also significantly reduces the occurrence rate of forming defects and avoids production line instability caused by drastic parameter fluctuations, further improving the intelligence and precision level of high-strength steel continuous stamping production.

[0075] The system's interactive interface is used to display analysis results, early warning information, and receive input from operators.

[0076] In summary, the precise deployment of multiple types of sensors in the data acquisition module enables synchronous acquisition of multi-source data throughout the stamping process. The data processing and feature extraction module transforms discrete raw data into a highly recognizable set of pressure-transformer characteristic parameters. The defect prediction and diagnosis module relies on the LSTM model to predict defects in advance. The process optimization decision module outputs optimal process adjustment instructions based on deep reinforcement learning and multi-objective optimization. Combined with the system's interactive interface, human-machine collaborative management is achieved. This effectively solves the problems of cracking, wrinkling, and poor flatness caused by single data acquisition dimensions, inaccurate pressure-transformer feature extraction, and delayed defect prediction in the production of high-strength continuous stamped parts in existing technologies. It also addresses the pain points of decreased pressure-transformer stability and limited lifespan caused by mold interference hazards due to the lack of collaborative assessment of interference risks in feeding height. Through multi-source data acquisition, precise feature extraction, intelligent defect prediction, multi-objective process optimization, and mold health management, the forming defects and interference risks are avoided in advance, ensuring pressure-transformer stability and production safety, extending mold lifespan, and significantly improving the forming accuracy and production efficiency of high-strength steel parts.

[0077] It should be noted that, in addition to the specific embodiments described above, those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Although the description of the present invention is presented in conjunction with preferred embodiments, this does not mean that the features of the invention are limited to those embodiments. On the contrary, the purpose of describing the invention in conjunction with embodiments is to cover other options or modifications that may be derived based on the claims of the present invention. To provide a deep understanding of the invention, many specific details are included in the above description, and the invention may also be implemented without using these details. Furthermore, to avoid confusion or obscuring the focus of the invention, some specific details will be omitted in the description. It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0078] It should be noted that in this specification, similar reference numerals and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0079] In the description of this embodiment, it should be noted that the terms "upper," "lower," "inner," "bottom," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship commonly used when the product of the invention is in use. They are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. The terms "first," "second," etc., are only used to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0080] In the description of this embodiment, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set up," "connected," and "linked" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this embodiment based on the specific circumstances.

[0081] While the present invention has been illustrated and described with reference to certain preferred embodiments, those skilled in the art should understand that the above description is a further detailed explanation of the invention in conjunction with specific embodiments, and should not be construed as limiting the specific implementation of the invention to these descriptions. Various changes in form and detail can be made by those skilled in the art, including several simple deductions or substitutions, without departing from the spirit and scope of the invention.

Claims

1. A method for analyzing pressure transformation data of high-strength continuous stamped parts for automobiles, characterized in that, The method includes: S1: Real-time acquisition of multi-source data from the stamping production line, including mechanical data of the stamping process, die status data, strip position image, feeding mechanism motion parameters, and material deformation image data; S2: Based on the multi-source data, extract the time-series characteristic parameters of the part's pressure deformation behavior during continuous stamping; S3: Input the time-series feature parameters into the pre-trained defect prediction model to obtain the prediction results of the type and risk level of potential forming defects in the current stamping batch; S4: Based on the prediction results, process adjustment suggestions are dynamically generated through the process parameter optimization model, and the suggestions are fed back to the stamping production line control system.

2. The method for analyzing pressure transformation data of high-strength continuous stamped parts for automobiles according to claim 1, characterized in that, In S1, the mechanical data includes, but is not limited to, the impact force at each workstation. Blank pressure The sliding process s(t) is represented by i=1,2,......,n, where i represents the workstation number and t is the time parameter. The mold status data includes the temperature T(t), vibration acceleration a(t), and wear monitoring value W(t) at key locations; The material deformation image data is acquired by a high-speed vision system deployed near the mold cavity and is used to characterize the material thickness variation Δh and the surface strain field distribution ε.

3. The method for analyzing pressure transformation data of high-strength continuous stamped parts for automobiles according to claim 1, characterized in that, S1 further includes: real-time acquisition of the position image of the material strip in the mold and the motion parameters of the feeding mechanism; the time-series feature parameters further include the real-time gap feature parameter δ(t) of the material strip relative to the highest point of the lower mold, extracted based on the position image and motion parameters.

4. The method for analyzing pressure transformation data of high-strength continuous stamped parts for automobiles according to claim 1, characterized in that, The time-series feature parameters include: The dynamic time warping distance between the specific station punching force curve and the standard curve calculated based on mechanical data ; The local thinning rate gradient G is calculated based on deformed image data; Rebound tendency index calculated based on multi-source data fusion: in, These are the weighting coefficients. The yield strain of the material; Interference risk characteristics based on gap detection ,in, This is the safety gap threshold.

5. The method for analyzing pressure transformation data of high-strength continuous stamped parts for automobiles according to claim 1, characterized in that, In step S3, the defect prediction model is a deep neural network model trained based on historical production data. Its input is the time-series feature parameters, and its output includes at least the crack risk probability. Risk of wrinkling and the probability of flatness deviation and the probability of mold interference .

6. The method for analyzing pressure transformation data of high-strength continuous stamped parts for automobiles according to claim 1, characterized in that, In step S4, the process parameter optimization model is constructed using a reinforcement learning algorithm, with the objective function being the minimum defect prediction risk, and employing the formula: in, This is an adjustable vector of process parameters, including feed height. Blank holder force setting value, etc. For the probability of each defect risk, These are the weighting coefficients. This is the penalty coefficient for parameter changes.

7. The method for analyzing pressure transformation data of high-strength continuous stamped parts for automobiles according to claim 6, characterized in that, The process parameter optimization model has a pre-defined rule library regarding the relationship between feeding height and product flatness, and mold interference risk. The constraints include: and ; When the optimization objective is to improve flatness and suggest reducing the feeding height, the model simultaneously assesses the interference risk. When the risk exceeds the threshold, multi-objective optimization is initiated, and the highest safe feeding height that meets the flatness requirements is output.

8. The method for analyzing pressure transformation data of high-strength continuous stamped parts for automobiles according to claim 1, characterized in that, The method further includes: A mold health status assessment model is constructed, and the mold health index is calculated based on the collected mold status data and corresponding stroke information. It predicts the remaining service life of the device and triggers a preventative maintenance warning.

9. A pressure transducer data analysis system for high-strength continuous stamping parts for automobiles, characterized in that, include: The data acquisition module is used to acquire multi-source data of the stamping process in real time; The data processing and feature extraction module is used for synchronizing, filtering, and calculating feature parameters of multi-source data; The defect prediction and diagnosis module has the defect prediction model built in, and is used to output defect risk analysis results; The process optimization decision module has the built-in process parameter optimization model, which is used to generate and output process adjustment instructions; The system's interactive interface is used to display analysis results, early warning information, and receive input from operators.

10. A pressure transducer data analysis system for high-strength continuous stamping parts for automobiles according to claim 9, characterized in that, The data acquisition module includes: Force and displacement sensors are distributed in key positions of the press slide, mold pad, and punch and die; Temperature sensors and vibration sensors are installed on the mold; High-speed industrial cameras and laser scanners designed for mold cavity layout; Visual positioning sensors used to monitor the position of the conveyor belt.

Images