Stainless steel die state monitoring and process compensation method and system based on stamping force

Abstract

The invention discloses a stainless steel die state monitoring and process compensation method and system based on stamping force, and the method comprises the following steps: collecting multi-mode sensing data in a stamping process in a stamping period; processing the multi-modal perceptual data, and synchronously determining a predicted part quality state and a predicted equipment health state in a parallel manner; generating a set comprising a plurality of candidate compensation strategies based on the part quality state and the equipment health state; selecting and determining an optimal compensation strategy from the set of candidate compensation strategies based on a production target; and the optimal compensation strategy is applied to the subsequent stamping process. According to the method, the predicted part quality state and the predicted equipment health state are synchronously determined in parallel, the service life of the mold is remarkably prolonged based on the quality priority, health priority and balance strategy of the production target, and the shutdown risk and the comprehensive production cost caused by excessive wear or sudden failure of the mold are reduced.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of intelligent manufacturing, and particularly relates to a stainless steel die state monitoring and process compensation method and system based on stamping force. BACKGROUND

[0002] Stamping forming is a high-efficiency and low-cost batch production technology in modern manufacturing industry, and is widely used in the fields of automobiles, aerospace, electronic consumer goods and the like. The quality of a stamped part, such as dimensional accuracy, surface quality, presence or absence of cracking or wrinkling, is crucial to the performance of the final product. However, the stamping process is a complex elastic-plastic deformation process, and the result is highly sensitive to a variety of factors, including: mechanical property fluctuations of the sheet material, variations in lubrication conditions, gradual wear of the die, and changes in environmental temperature. In order to cope with these fluctuations, modern stamping production lines gradually shift from traditional open-loop control and post-quality detection to online monitoring and closed-loop process control based on sensors.

[0003] The stamping process control in the prior art mainly relies on offline die testing and open-loop control strategy. Before production, engineers determine a set of fixed process parameters through a large number of trial stamping and manual adjustment. In the production process, although some advanced systems introduce closed-loop feedback control and can collect machine speed, pressure, temperature and other data in real time through sensors, the control logic is usually reactive, that is, compensation is made after detecting the deviation. These methods are difficult to effectively cope with dynamic disturbances in the production process, such as material property fluctuations between raw material coils, changes in lubrication conditions or changes in environmental temperature. When new materials such as aluminum alloy and high-strength steel are used, the limitations of traditional methods become more prominent due to their narrower forming window. In addition, quality management tools such as statistical process control usually analyze after the fact and cannot prevent batch quality problems from occurring in real time. SUMMARY

[0004] To solve the problems raised in the background art, the application provides a stainless steel die state monitoring and process compensation method and system based on stamping force.

[0005] To achieve the above-mentioned purpose, the application provides the following technical scheme: a stainless steel die state monitoring and process compensation method based on stamping force, comprising the following steps:

[0006] S1, collecting multi-modal perception data in a stamping cycle;

[0007] S2, processing the multi-modal perception data to synchronously determine a predicted part quality state and a predicted equipment health state in a parallel manner;

[0008] S3, generating a set of candidate compensation strategies based on the part quality state and the equipment health state;

[0009] S4, selecting and determining an optimal compensation strategy from the set of candidate compensation strategies based on a production target;

[0010] S5, applying the optimal compensation strategy to the subsequent stamping process.

[0011] In a preferred embodiment of the present application, in the S1, the multi-modal sensing data includes stamping force, mold vibration, acoustic emission, mold temperature, sheet displacement, and optical detection data.

[0012] In a preferred embodiment of the present application, in the S2, the equipment health state body includes: real-time evaluation of the current wear rate of the mold, fatigue accumulation, and cracking risk.

[0013] In a preferred embodiment of the present application, in the S3, the set of candidate compensation strategies includes at least: a quality-first strategy with the primary goal of maximizing the part quality index; a health-first strategy with the primary goal of minimizing the deterioration rate of the equipment health state index; and a balanced strategy that balances between the part quality index and the equipment health state index.

[0014] In a preferred embodiment of the present application, in the S4, the selection and determination of the optimal compensation strategy from the set of candidate compensation strategies is specifically based on a digital twin model, which performs rapid simulation on each candidate compensation strategy to verify the comprehensive impact of part quality and equipment health loss, and selects the optimal compensation decision according to the production target.

[0015] The production target includes at least one of the quality-first mode, the health-first mode, and the balanced mode, and the system determines the optimal compensation strategy from the set of candidate compensation strategies according to the selected mode.

[0016] In a preferred embodiment of the present application, the following steps are further included:

[0017] When introducing new process conditions, a pre-constructed causal knowledge graph is used for reasoning to predict the causal effect of the new process conditions on the process result; and the prediction result of the causal effect is used to provide a set of initial parameters for a prediction model, which is used to determine the predicted part quality state and the predicted equipment health state in the S2, thereby shortening the training time required for the prediction model to reach a predetermined prediction accuracy under the new process conditions.

[0018] In one preferred embodiment of the present application, the causal knowledge graph is constructed based on a hybrid modeling process that integrates and fuses two different sources of information: professional knowledge based on physical mechanisms and engineering principles; and multi-modal perception data or historical process databases.

[0019] In one preferred embodiment of the present application, the following steps are further included: when it is detected that there is a persistent performance degradation in the stamping process, an automated causal root cause analysis is performed to generate a repair hypothesis about the causal root cause;

[0020] The repair hypothesis and its corresponding repair measures are virtually verified in the digital twin model; and when the virtual verification is successful, a repair action is actively performed on the physical stamping process, or a processing mode maintenance work order containing the virtual verification evidence is generated.

[0021] A stainless steel die state monitoring and process compensation system based on stamping, comprising:

[0022] A multi-modal sensor array for collecting multi-modal perception data of the stamping process;

[0023] A processing unit in communication with the sensor array, the processing unit being configured to perform:

[0024] A dual-state assessment engine for determining the predicted part quality state and the predicted equipment health state in parallel based on the real-time data;

[0025] A multi-objective decision control module for generating a set of candidate compensation strategies, each of which represents a different trade-off between the predicted part quality state and the predicted equipment health state, and selecting an optimal compensation strategy from them based on pre-set production objectives;

[0026] An actuator module for adjusting the parameters of the stamping process according to the optimal compensation strategy;

[0027] A digital twin module configured to receive the set of candidate compensation strategies and assist the multi-objective decision control module in selecting the optimal compensation strategy through simulation verification;

[0028] A causal adaptive module containing a causal knowledge graph configured to provide initialized information for the predicted part quality state and the predicted equipment health state in the dual-state assessment engine under new process conditions.

[0029] In one preferred embodiment of the present application, the following steps are further included:

[0030] The closed-loop self-healing module is configured to perform causal root cause analysis and virtually verify repair hypotheses using the digital twin module, and then autonomously perform repair actions or generate treatment mode maintenance instructions.

[0031] The present application solves the defects in the background art and has the following advantages:

[0032] 1. The present application synchronously and in parallel determines the predicted part quality state and the predicted equipment health state within one stamping cycle, and selects the optimal compensation strategy from multiple candidate strategies based on production targets, quality priority, health priority and balanced strategy, so that the system can quantitatively evaluate the two conflicting goals of part quality and die wear within milliseconds of each stamping cycle, and make strategic optimal compensation decisions through simulation verification of candidate strategies by the digital twin module. The present application ensures the current part quality and can also quantitatively and manage the health cost of the die. Compared with the existing technology which mostly adopts reactive control and only focuses on short-term quality at the expense of long-term equipment health, the present application overcomes the short-sighted logic of ensuring current product qualification at all costs, significantly extends the service life of the die under the premise of ensuring part quality, and reduces the risk of downtime and comprehensive production cost due to excessive die wear or sudden failure.

[0033] 2. The present application constructs a causal knowledge graph that integrates physical mechanisms and data-driven causal knowledge. When new process conditions are introduced, the system can use the graph for causal reasoning to predict the causal effects of new working conditions on process results and analyze the process. The system uses this mechanism-based prediction result to provide a set of optimized initial parameters for part quality and equipment health prediction models, thereby guiding the models to quickly converge under new conditions. The training time of the prediction model under new conditions is greatly shortened, realizing fast and robust adaptation with zero or few samples, overcoming the problem of poor adaptability due to reliance on fragile association learning when facing unknown conditions such as new materials or new dies, and the need for a large number of offline die testing and manual adjustments. The present application can shorten the traditional die testing time from hours to minutes and greatly reduce die testing waste.

[0034] 3. When the present application detects that the stamping process has a persistent performance degradation, it can automatically trigger causal root cause analysis and use the causal knowledge graph to trace back to generate repair hypotheses about the root cause. The system will safely virtually verify the repair hypothesis and its corresponding repair measures in the digital twin model, and only after successful verification will it actively perform repair actions, realizing automatic diagnosis and closed-loop self-repair of systematic root problems. Compared with the passive maintenance mode in the prior art, which is disconnected from diagnosis and execution and relies on manual post-analysis, the present application greatly improves maintenance efficiency and first repair success rate, ensuring the long-term stability and resilience of the production process. Attached Figure Description

[0035] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0036] Figure 1 This is a schematic diagram of the process of the present invention; Detailed Implementation

[0037] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. These drawings are simplified schematic diagrams, which are only used to illustrate the basic structure of the present invention and therefore only show the components relevant to the present invention.

[0038] This invention provides a method and system for monitoring the condition and compensating for the process of stainless steel dies based on stamping pressure, aiming to solve the problems in the existing technology regarding stamping pressure.

[0039] The fundamental problems of pressure process control include focusing only on short-term quality at the expense of long-term health, poor adaptability to new operating conditions, and a disconnect between diagnosis and execution.

[0040] The core idea of ​​this invention is to construct a self-evolving system with three logically progressive levels of capability:

[0041] Level 1: Real-time strategic decision-making. Overcoming the sacrifice of long-term health for short-term quality, in each stamping cycle, by evaluating the two conflicting objectives of part quality and equipment health in parallel, and using digital twins for arbitration, a strategically optimal compensatory decision is made that weighs the pros and cons.

[0042] The second level: cognitive engine adaptation. Overcoming poor adaptability to new working conditions, by constructing a causal knowledge graph that integrates physical mechanisms and data-driven approaches, when faced with unknown working conditions such as new materials and new molds, it no longer relies on fragile association learning, but performs causal reasoning to achieve rapid and robust adaptation to new working conditions with zero or few samples.

[0043] The third level: Resilient organism self-healing. Overcoming the disconnect between diagnosis and execution, it automatically triggers root cause analysis when long-term performance degradation is detected, and autonomously executes repairs or generates high-confidence maintenance work orders after successful verification.

[0044] like Figure 1 As shown, a method for monitoring the condition and compensating for the process of stainless steel dies based on stamping pressure is characterized by the following steps:

[0045] S1. Collect multimodal sensing data during the stamping process within one stamping cycle;

[0046] S2, processing the multi-modal sensing data to determine a predicted part quality status and a predicted equipment health status in a parallel manner;

[0047] S3, generating a set of candidate compensation strategies based on the part quality status and the equipment health status;

[0048] S4, selecting and determining an optimal compensation strategy from the set of candidate compensation strategies based on a production target;

[0049] S5, applying the optimal compensation strategy to a subsequent stamping process.

[0050] In the following, each step will be described in detail.

[0051] S1, collecting multi-modal sensing data during a stamping process within a stamping cycle;

[0052] A stamping cycle refers to the time required for a stamping machine to complete one complete, reciprocating stamping forming action.

[0053] Specifically, the data acquisition and sensing unit deploys multiple sensors on the stamping machine and the die, forming a multi-modal sensor array. The multi-modal sensing data specifically includes:

[0054] Stamping force data, using high-precision piezoelectric force sensors or high-frequency strain gauge force sensors, installed at the key load-bearing points of the four columns or the machine body of the stamping machine, for measuring the total tonnage and unbalanced load of the stamping, or directly embedded in the key load-bearing components inside the die, such as the upper die holder, the lower die holder, the bottom of the die insert, or the blank holder pin of the hydraulic cushion. The sampling frequency can be set to 50kHz or 100kHz to ensure the capture of high-frequency dynamic events such as material fracture or die impact during the stamping process.

[0055] Die vibration data, using high-frequency acceleration sensors, such as broadband piezoelectric accelerometers, installed near the die holder, guide column, or near the forming area of the die surface. The sampling frequency can be set to 50kHz or higher to capture the micro-vibration and impact signals of the die, which can effectively reflect the wear state, lubrication condition and abnormal impact of the die.

[0056] Acoustic emission data, using high-sensitivity acoustic emission sensors, directly attached to the die surface near areas where micro-cracks may occur, such as the die radius and the punch sharp corner. The sampling frequency can be set between 1MHz and 10MHz. The acoustic emission signals emitted by the acoustic emission sensors are transient elastic waves released when micro-cracks, friction or plastic deformation occur inside the material. By analyzing parameters such as ring count, energy, amplitude, etc. of the acoustic emission signals, the risk of material cracking and the expansion of micro-cracks in the die can be predicted at an early stage.

[0057] Mold temperature data, either embedded thermocouples or PT100 thermal resistors, or infrared thermal imagers, can acquire the surface temperature distribution of the mold and the sheet metal. They can be embedded in the mold at key friction areas such as drawbead, or under the corner radius of the punch, or use thermal imagers to target the working area of the mold. Temperature significantly affects the plastic forming performance of the material, especially hot forming or warm forming, and the effectiveness of lubricants. At the same time, it is also an important indicator for evaluating mold wear and fatigue.

[0058] Sheet metal displacement data, linear variable differential transformers, laser displacement sensors, or machine vision systems, can be set up in the mold measurement points to non-contact monitor the sheet metal edge in the drawing process. The amount of drawing is a key indicator to determine whether the material flow is uniform, whether there is a risk of wrinkling or cracking.

[0059] Optical detection data, high-resolution industrial cameras or structured light scanners, integrated in the mold or downstream of the stamping line, are used for online detection of macro defects on the surface of the part, such as wrinkling, drawing, cracking, as a direct label or auxiliary input for part quality assessment in S2.

[0060] All different sensor data streams are synchronized by high-speed data acquisition cards. The raw data collected is converted into physical quantities in real time, and preliminary data preprocessing such as removing high-frequency noise, baseline correction, and data alignment is performed.

[0061] S2, process multi-modal perception data to determine a predicted part quality state and a predicted equipment health state in parallel.

[0062] This step is the core of the system's real-time strategic decision-making ability, aiming to overcome the short-sighted logic of guaranteeing the current product at all costs. Its subversive nature is that it no longer treats part quality and equipment health as two independent and time-scale different problems, but synchronously quantifies both in the millisecond time scale of each stamping cycle, providing complete decision-making basis for subsequent strategic trade-offs.

[0063] After the edge controller receives the multi-modal data frames collected by S1, the multi-modal data frames are sent into two independent, pre-trained prediction models in parallel, and the two models together constitute a double-state evaluation engine. A predicted part quality state represents the quality of the part being formed at the moment. A predicted equipment health state represents the health cost of the mold to form a part.

[0064] Cost.

[0065] Predicted part quality state refers to the real-time prediction of key quality indicators for the individual part that is currently being formed. Key quality indicators are specific physical metrics used to evaluate whether the part meets the design and functional requirements, which can include but are not limited to:

[0066] Geometric accuracy indicators: such as predicted springback angle or profile.

[0067] Material integrity indicators: such as predicted minimum wall thickness, thinning rate, or cracking risk.

[0068] Surface quality indicators: such as predicted wrinkle height or wrinkle risk.

[0069] Predicted part quality state is achieved by a specific module within the dual-state assessment engine in the system, which is the part quality prediction model. The part quality prediction model is a pre-trained supervised learning model. The supervised learning model can adopt a deep learning architecture, such as a deep neural network.

[0070] The input of the supervised learning model is the multi-modal perception data collected in step S1, which has been processed by feature engineering. The supervised learning model is constructed by learning the mapping relationship between a large amount of historical data, i.e. process features in history, and corresponding offline quality detection results.

[0071] In real-time operation, the supervised learning model predicts the numerical value of the key quality indicators according to the process features of the current cycle, which is the predicted part quality state defined in the invention.

[0072] Predicted equipment health state is a real-time, quantitative representation of the damage or cumulative damage that the die has suffered in the current stamping cycle. It is not a long-term offline health assessment, but a specific indication of the immediate quantification of the cost paid by the system for the current stamping action in the control loop of the stamping cycle. Predicted equipment health state can include but is not limited to wear rate, fatigue accumulation, and cracking risk.

[0073] Predicted equipment health state is achieved by a die health state assessment model in the dual-state assessment engine. The die health state assessment model is a hybrid model that combines physical mechanism-based methods and data-driven methods. The die health state assessment model can include a physical model based on domain knowledge, which is constructed by learning the relationship between process features and die health damage in historical data. Real health damage data can be obtained through periodic microscopic detection of the die, metallographic analysis, or historical maintenance records, such as die wear measurement after 10,000 strokes.

[0074] The input of the die health state assessment model is derived from the multi-modal perception data collected in step S1, focusing on features related to friction, impact, and temperature. The system inputs the features into both the die health state assessment model and the part quality prediction model in parallel, thereby determining the predicted values of both states synchronously within the same stamping cycle.

[0075] S3, based on the part quality state and the equipment health state, generating a set containing multiple candidate compensation strategies;

[0076] Each candidate compensation strategy represents a specific set of executable process parameters, such as blank holder force and stamping speed. Each candidate compensation strategy in the set containing multiple candidate compensation strategies achieves a different, optimal balance between the two conflicting objectives, i.e., the predicted part quality and the predicted equipment health.

[0077] The set of candidate compensation strategies at least includes: a quality-first strategy that prioritizes maximizing the part quality indicator, ensuring the highest part quality but accompanied by the highest die wear.

[0078] A health-first strategy that prioritizes minimizing the deterioration rate of the equipment health state indicator, achieving the lowest health loss of the die while ensuring that the part quality is still within the tolerance range.

[0079] A balanced strategy that achieves a balance between the part quality indicator and the equipment health state indicator.

[0080] The system employs an efficient multi-objective optimization algorithm to generate an approximate Pareto front in one run, which is the required set containing multiple candidate compensation strategies. Each strategy in the set containing multiple candidate compensation strategies is Pareto optimal, meaning that without sacrificing one objective, such as quality, the other objective, such as health, cannot be improved.

[0081] S4, based on a production target, selecting and determining an optimal compensation strategy from the set of candidate compensation strategies;

[0082] The production target specifically embodies a set of selectable operating modes, including but not limited to: a quality-first mode, also known as a rush mode.

[0083] A health-first mode, also known as a high-cost die mode.

[0084] A balanced mode, also known as a regular production mode.

[0085] Selecting and determining an optimal compensation strategy from the set of candidate compensation strategies specifically means that the set of candidate compensation strategies is submitted to a digital twin module. The digital twin module is configured as a decision arbitrator, performing two key functions to assist decision-making:

[0086] Safety verification: The digital twin module simulates each candidate compensation strategy to predict extreme physical phenomena that are difficult for the multi-modal sensor array to capture, such as instantaneous tearing of material, severe local wrinkling, or equipment overload. When a candidate compensation strategy leads to extreme physical phenomena in simulation, it is immediately eliminated from the candidate set, even if its predicted outcome in step S3 is qualified.

[0087] Result refinement and ranking: Physical simulation can provide more accurate prediction results than pure data models, thus allowing for a more refined ranking of candidate strategies in terms of their true quality and health impact.

[0088] After verification, refinement, and ranking by the digital twin module, the system performs a final multi-criteria decision on the remaining, safe candidate compensation strategies:

[0089] Objective translation: The system translates pre-defined production objectives, such as rush mode or balance mode, into a dynamic cost function or uses a weighted sum approach. Weight distribution: The system assigns different weights to quality and health objectives based on the selected mode. For example, in rush mode,

[0090]

[0091] In balance mode,

[0092] where, represents the part quality state, represents the equipment health state.

[0093] The system calculates a comprehensive decision score for each candidate strategy using the weighted function and selects the strategy with the highest score as the optimal compensation strategy.

[0094] S5, apply the optimal compensation strategy to subsequent stamping processes.

[0095] Based on the prediction results, process conditions are adjusted proactively before the start or during the next stamping cycle to prevent potential quality deviations and excessive health loss. After determining the unique optimal compensation strategy in step S4, the system enters the execution phase. The edge controller sends specific process parameter instructions corresponding to the optimal strategy to the PLC of the stamping machine and related actuator modules.

[0096] The above steps S1 to S5 describe the real-time decision-making closed loop of the system under stable working conditions. However, when the production conditions change fundamentally, for example, a completely new brand of stainless steel material is replaced, the performance of the prediction model trained based on historical data will decrease significantly. To solve this problem, the present application introduces a higher level of cognitive ability, namely an adaptive framework based on causal reasoning.

[0097] A stamping-based stainless steel die state monitoring and process compensation method further comprises the following steps:

[0098] When introducing a new process condition, a pre-constructed causal knowledge graph is used for reasoning to predict the causal effect of the new process condition on the process result; and the prediction result of the causal effect is used to provide a set of initial parameters for the prediction model, which is used to determine the predicted part quality state and the predicted equipment health state in step S2, thereby shortening the training time required for the prediction model to reach a predetermined prediction accuracy under the new process condition.

[0099] The pre-constructed causal knowledge graph is not a conventional knowledge graph that simply stores data relationships, but a combination of professional knowledge based on physical mechanisms and engineering principles in the system and multi-modal perception data or historical process databases. The pre-constructed causal knowledge graph is constructed based on a directed acyclic graph, and the definitions of its nodes and edges are as follows:

[0100] Nodes represent key variables in the stamping process, including but not limited to: material properties such as yield strength, thickness, adjustable process parameters such as blank holder force, speed, friction coefficient, and final process results, i.e. part quality and equipment health loss.

[0101] Edges represent directed and direct causal relationships between node variables, rather than simple associations. For example, an edge explicitly represents that an increase in material yield strength leads to an increase in springback angle.

[0102] When a new process condition is introduced, for example, the system receives a notification through MES that the next batch will use a completely new high-strength steel with a nominal yield strength 15% higher than the historical benchmark. Instead of looking for similar working conditions in the historical database for knowledge transfer, the system inputs the properties of the new working condition as the corresponding nodes in the causal knowledge graph. Causal reasoning is performed in the causal knowledge graph, i.e. information propagation along the causal paths in the causal knowledge graph.

[0103] Through reasoning, the system derives a mechanism-based prediction of the causal effect of the new process condition on the process result without any new samples. For example, the causal knowledge graph will reason that because the yield strength has increased by 15%, the springback angle is expected to increase by about 8-12% and the die wear rate will accelerate by about 5% without changing the existing process parameters.

[0104] By predicting causal reasoning, without consuming any physical trial, predict what direction and approximate magnitude the introduction of a new working condition will have on the downstream key quality indicators and die health loss. The prediction results of the causal effects generated by reasoning will serve as a large amount of physical prior knowledge. Physical prior knowledge is used to guide and initialize the part quality prediction model and die health assessment model in step S2, so that it can quickly converge, thereby overcoming the vulnerability of traditional correlation-based learning.

[0105] The prediction results obtained by the causal reasoning in the previous step are injected into the prediction model as physical prior knowledge. The specific technical process is as follows:

[0106] Parameter initialization, the system discards random weight selection and adjusts it specifically. According to the results of causal reasoning, the system adjusts the weights or bias terms in the prediction model related to the rebound output. The system can also adjust the model structure, such as increasing the weights of related network layers or neurons, to make it more plastic when learning new working conditions.

[0107] Guiding loss function, the system takes the prediction results of causal effects as soft constraints and injects them into the loss function of model training.

[0108] The method of deeply integrating causal prior knowledge enables the prediction action in step S2 to start learning with an initial state very close to the optimal solution of the new working condition. The system only needs very little actual physical trial stamping data to quickly converge to a high-precision steady state and complete the self-adaptation of the new working condition, for example, from the traditional trial of 2 hours and 50 pieces of waste to automatic adaptation of 1 minute and 0 pieces of waste.

[0109] The above-mentioned system can cope with real-time fluctuations and working condition switching. However, in the long run, the system may encounter some gradual and continuous performance degradation problems that cannot be solved by periodic compensation, such as abnormal wear of the die and gradual clogging of the lubrication system. A new step is introduced to overcome existing continuous performance degradation problems.

[0110] A stainless steel die state monitoring and process compensation method based on stamping, further comprising the following steps:

[0111] When detecting that the stamping process has a continuous performance degradation, performing an automated causal root cause analysis to generate a repair hypothesis about the causal root cause; virtually verifying the repair hypothesis and the repair measures corresponding thereto in the digital twin model; and when the virtual verification is successful, actively performing a repair action on the physical stamping process, or generating a processing mode maintenance work order containing virtual verification evidence.

[0112] Persistent performance degradation is a specific system state, characterized by a slow, progressive, and unintended negative trend, which is not caused by regular material batch fluctuations, but by a deeper, systemic root cause. For example, progressive die wear, lubrication system clogging. The manifestations of persistent performance degradation include, but are not limited to:

[0113] Slow deterioration of quality indicators, for example, scrap rate is still slowly rising.

[0114] Persistent increase in compensation efforts, for example, the average blank holder force compensation required by the system to maintain the same quality target has increased by 15% continuously and slowly.

[0115] Once the above degradation trend is identified, the system automatically triggers a closed-loop self-healing module for causal root cause analysis. The closed-loop self-healing module uses the causal knowledge graph containing the knowledge of the entire system built in the previous section to perform backward reasoning, taking the observed effect as evidence, and tracing back all possible causal paths that lead to this effect in the causal knowledge graph. According to the probability or strength of the path, a set of repair hypotheses about the causal root cause is generated in order of possibility. For example:

[0116] H1 (probability 60%): Lubrication system 3 nozzle flow decreased by 20%, causing increased friction in the northeast corner of the die.

[0117] H2 (probability 30%): The clearance of No. 2 guide pillar increased by $0.05mm$ due to wear, causing a slight misalignment during die closing, increasing the forming resistance.

[0118] H3 (probability 10%): The material supplier changed the rolling process, causing a slight systematic change in the friction coefficient of the material that was not detected.

[0119] Virtual verification of repair hypotheses in the digital twin model is the key and safety guarantee of the self-healing capability of the invention. The system does not directly take action or issue an alarm based on the hypothesis with the highest confidence (H1), but starts a virtual hypothesis verification program, using the digital twin model as a safe causal sandbox. Specifically:

[0120] Problem recurrence verification, the system issues a simulation task to the digital twin model: simulate the scenario of a 20% decrease in the flow rate of lubrication system 3 nozzle under current production conditions. Compare the performance indicator trend of the simulation output, for example, the change curve of the stamping force compensation in the simulation with the observed data trend. If the two are highly matched, for example, the correlation is greater than 0.9, then hypothesis H1 is preliminarily confirmed. If the matching degree is low, then continue to verify H2.

[0121] Repair measure validation, assuming H1 is successfully reproduced, the system proceeds to the second step simulation: in the above reproduced failure scenario, simulate its corresponding repair measure, i.e. restore the normal flow of nozzle 3. The system will observe whether this repair measure can make the simulation predicted performance indicators (punch force compensation) return to the normal baseline, and will not cause other unexpected negative effects, for example, cause wrinkles in other areas.

[0122] When a repair hypothesis and its corresponding repair measure are virtually verified as effective and safe in the digital twin, the system enters the execution phase. Depending on the nature of the problem and the pre-set system permissions, it will take one of the two actions:

[0123] Proactively perform a repair action on the physical stamping process: if the repair measure is within the scope of the system's autonomous control permissions, for example, the output pressure of the lubrication pump is set too low because of the flow reduction of nozzle 3, the system will automatically issue an instruction to the PLC to adjust the output pressure of the lubrication pump, and continuously monitor the repair effect, forming a complete diagnosis-repair closed loop.

[0124] Generate a treatment method maintenance work order containing virtual verification evidence: if the repair requires human intervention, for example, the physical blockage of nozzle 3 requires manual cleaning or replacement, the system will automatically generate a treatment method maintenance work order in the enterprise maintenance management system. The treatment method maintenance work order is a high-confidence action instruction, which contains damaged objects and is accompanied by virtual verification evidence, such as simulation reproduction comparison chart, repair preview result, detailed explanation of the reason why the damaged object is damaged, and specific quantitative improvement expected after repairing the damaged object (for example: after repair, the punch force compensation is expected to decrease by 15%, the mold wear rate in this area is expected to decrease by 30%).

[0125] The present application also provides a stainless steel die state monitoring and process compensation system based on punch force, comprising:

[0126] Multi-modal sensor array: physically deployed on the stamping equipment and die, including force sensors, vibration sensors, acoustic emission sensors, temperature sensors, displacement sensors and optical sensors. Responsible for real-time, synchronous acquisition of multi-dimensional physical signals of the stamping process, and digitization and preliminary processing through high-speed data acquisition module.

[0127] Processing unit: is the computing core of the system, which is communicatively connected with the sensor array. It is a distributed computing entity, whose functions are realized by multiple software modules deployed on the edge and cloud. The processing unit is configured to perform the following functions:

[0128] Dual-state assessment engine: Responsible for executing step S2, it comprises two parallel, lightweight inference models: a part quality prediction model and a mold health assessment model. This engine is typically deployed on an edge controller to meet millisecond-level real-time prediction requirements, receiving real-time data from a sensor array and instantly outputting the predicted part quality status and the predicted equipment health status.

[0129] The multi-objective decision control module is responsible for executing steps S3 and S4. It has a built-in multi-objective optimization algorithm to generate a Pareto calculus containing multiple candidate compensation strategies based on the output of the dual-state evaluation engine.

[0130] The optimal set. Each strategy represents a different trade-off between part quality and equipment health. It also includes decision logic that helps select the optimal compensation strategy based on preset production targets received from the MES system. This module is also deployed on the edge controller to ensure real-time decision-making.

[0131] The digital twin module is a key decision support and verification module, containing a high-fidelity physical simulation model and playing a dual role:

[0132] In the real-time decision-making loop, a set of candidate strategies generated by the multi-objective decision control module is received, and the optimal compensation strategy is selected through rapid simulation verification.

[0133] In the long-term self-healing cycle, the repair hypotheses generated by causal root cause analysis are received and these hypotheses are virtually verified as a causal sandbox.

[0134] The causal adaptive module is the core of the system's ability to quickly learn from new operating conditions. It includes a causal knowledge graph built and maintained in the cloud. When a new operating condition is detected, this module is configured to perform causal reasoning on the causal knowledge graph, predict the impact of the new operating condition, and use these reasoning results to provide initialization information for the prediction model in the dual-state evaluation engine, thereby achieving rapid adaptation.

[0135] Closed-loop self-healing module: It is configured to continuously monitor long-term performance degradation trends, trigger causal root cause analysis when degradation is detected, generate a set of repair hypotheses using a causal knowledge graph, call the digital twin module to virtually verify the repair hypotheses, and autonomously execute repair actions after successful verification by issuing instructions to the edge controller or generating treatment maintenance instructions containing detailed verification evidence.

[0136] The actuator module is the physical execution end of the system, responsible for physically adjusting the parameters of the stamping process according to the optimal compensation strategy output by the processing unit. It includes all programmable control units on the stamping press, such as the multi-point blank holder force control system, the programmable lubrication system, and the slide motion curve controller of the servo press.

[0137] In the description of the application, it is to be understood that the terms "first", "second", "third" and the like, merely identify features belonging to distinct categories and do not imply a relative importance or a specific order of sequence. It is to be understood that a reference to a feature "comprising" a means also discloses features "consisting of" the means and features "consisting essentially of" the means. Hence, use of the term "comprising" in the description of a feature does not exclude other features but allows inclusion of additional features but does not require the presence of such additional features. Also, use of the term "comprising" as used in the claims does not exclude any features but allows inclusion of additional features but does not require the presence of such additional features.

[0138] In the description of the application, the terms "one embodiment", "some embodiments", "an embodiment", "exemplary embodiment", "specific embodiment" or "some examples" are used to describe particular features, structures, materials or characteristics included in at least one embodiment or example of the application. The illustrative description of the above terms in the description of the application does not necessarily indicate that the described features, structures, materials or characteristics are essential to the application.

[0139] Although the embodiments of the present application have been shown and described above, it is to be understood that the above-described embodiments are merely exemplary, and are not to be taken as limiting the present application. Accordingly, various modifications, alterations, replacements and variations can be made to the above-described embodiments within the scope of the present application by those skilled in the art.

Claims

1. A method for monitoring the state of a stainless steel die based on the stamping force and process compensation, characterized by, The method comprises the following steps: S1, collecting multi-modal sensing data during a stamping cycle; S2, processing the multi-modal sensing data to determine a predicted part quality state and a predicted equipment health state in a parallel manner; S3, generating a set of candidate compensation strategies based on the part quality state and the equipment health state; S4, selecting and determining an optimal compensation strategy from the set of candidate compensation strategies based on a production target; S5, applying the optimal compensation strategy to subsequent stamping processes.

2. A method for monitoring the state of a stainless steel die based on stamping force and process compensation according to claim 1, characterized in that: In the S1, the multi-modal sensing data includes stamping force, die vibration, acoustic emission, die temperature, sheet displacement, and optical detection data.

3. The stamping-based stainless steel die condition monitoring and process compensation method according to claim 1, characterized in that: In the S2, the equipment health state includes real-time evaluation of the current wear rate, fatigue accumulation, and cracking risk of the die.

4. The stamping-based stainless steel die condition monitoring and process compensation method according to claim 1, characterized in that: In the S3, the set of candidate compensation strategies includes at least a quality-first strategy that prioritizes maximizing part quality indicators, a health-first strategy that prioritizes minimizing the deterioration rate of equipment health state indicators, and a balanced strategy that balances the part quality indicators and the equipment health state indicators.

5. The stamping-based stainless steel die condition monitoring and process compensation method according to claim 1, characterized in that: In the S4, the optimal compensation strategy is selected and determined from the set of candidate compensation strategies by simulating each candidate compensation strategy based on a digital twin model to verify the comprehensive impact of part quality and equipment health loss, and selecting the optimal compensation decision based on the production target. The production target includes at least one of a quality-first mode, a health-first mode, and a balanced mode, and the system determines the optimal compensation strategy from the set of candidate compensation strategies according to the selected mode.

6. A method of stamping force based stainless steel die condition monitoring and process compensation as claimed in claim 1 wherein, Further comprising the following steps: When a new process condition is introduced, a pre-constructed causal knowledge graph is used for reasoning to predict the causal effect of the new process condition on the process result, and the predicted result of the causal effect is used to provide a set of initial parameters for a prediction model used to determine the predicted part quality state and the predicted equipment health state in the S2, thereby shortening the training time required for the prediction model to reach a predetermined prediction accuracy under the new process condition.

7. A method of stamping force based stainless steel die condition monitoring and process compensation as claimed in claim 6 wherein: The causal knowledge graph is constructed based on a hybrid modeling process that integrates and fuses two different sources of information: professional knowledge based on physical mechanisms and engineering principles; and multi-modal sensing data or historical process databases.

8. A punch force based stainless steel die condition monitoring and process compensation method as claimed in claim 1, wherein, Further comprising the following steps: When persistent performance degradation in the stamping process is detected, an automated causal root cause analysis is performed to generate a repair hypothesis about the causal root cause; The repair hypothesis and its corresponding repair measures are virtually verified in the digital twin model, and when the virtual verification is successful, a repair action is actively performed on the physical stamping process, or a processing mode maintenance work order containing evidence of the virtual verification is generated.

9. A punch force based stainless steel die condition monitoring and process compensation system, characterized in that, It includes: A multi-modal sensor array for collecting multi-modal sensing data during a stamping process; a processing unit communicatively coupled to the sensor array, the processing unit configured to perform: a dual-state assessment engine configured to determine, in parallel, the predicted part quality state and the predicted equipment health state based on the real-time data; a multi-objective decision control module configured to generate a set of candidate compensation strategies each representing a different trade-off between the predicted part quality state and the predicted equipment health state, and to select an optimal compensation strategy from the set based on pre-defined production objectives; an actuator module configured to adjust parameters of a stamping process according to the optimal compensation strategy; a digital twin module configured to receive the set of candidate compensation strategies and to assist the multi-objective decision control module in selecting the optimal compensation strategy through simulation verification; a causal adaptation module comprising a causal knowledge graph configured to provide initialization information for the predicted part quality state and the predicted equipment health state in the dual-state assessment engine under new process conditions.

10. A punch force based stainless steel die condition monitoring and process compensation system as claimed in claim 9, wherein, further comprising: a closed-loop self-healing module configured to perform causal root cause analysis and to virtually verify repair hypotheses using the digital twin module, and then to autonomously perform repair actions or to generate treatment mode maintenance instructions.

Images