A method and system for intelligent prediction and management of shrinkage porosity risk in cast iron parts

Abstract

This invention relates to the field of intelligent management technology in the casting process, specifically to an intelligent prediction and management method and system for shrinkage porosity risk in cast iron parts. First, it integrates a mechanistic model with a data-driven model—using the output of the mechanistic model as a benchmark evaluation value, and using the data-driven model to learn the deviation patterns between the benchmark and actual values. Second, it automatically triggers a feedforward control decision-making process based on the comprehensive evaluation results. Third, it uses causal analysis to identify the true causal relationship between various process factors and shrinkage porosity risk, rather than superficial correlation, and scientifically groups control measures based on the strength of causal influence, formulating differentiated adjustment schemes for each group. Finally, it integrates all the adjustment schemes into a comprehensive control scheme and transforms it into executable decision instructions, providing casting production managers with a systematic and traceable risk management tool, effectively reducing the shrinkage porosity defect rate in cast iron parts, and improving the quality of finished cast iron parts and production management efficiency.

Description

Technical Field

[0001] This invention relates to the field of intelligent management technology for casting processes, specifically to an intelligent prediction and management method and system for shrinkage porosity risk in cast iron parts. Background Technology

[0002] During the solidification process, cast iron parts are prone to shrinkage defects, especially at hot spots or in the last solidified areas, due to insufficient feeding of the molten iron during solidification. Shrinkage defects significantly reduce the density, mechanical properties (such as tensile strength and elongation), and sealing performance of cast iron parts, and in severe cases, lead to their complete scrapping, resulting in enormous waste of materials and energy. Therefore, accurately predicting shrinkage risk and developing reasonable control and management plans during cast iron production are of significant engineering value and economic benefit for improving the finished product qualification rate and reducing production costs.

[0003] Currently, foundry enterprises face the following technical and managerial challenges in managing shrinkage porosity risk:

[0004] First, existing methods for predicting shrinkage porosity risk mainly fall into two categories. One category is mechanistic modeling based on numerical simulation of the solidification process, which predicts shrinkage porosity tendency by solving heat transfer equations, flow equations, and feeding equations. These methods have clear physical meaning and interpretability, but many uncertainties exist in actual casting processes—such as fluctuations in the heat transfer coefficient of the mold interface, batch deviations in molten iron composition, and random fluctuations in pouring temperature. Mechanistic models struggle to accurately describe the combined impact of these complex factors on shrinkage porosity formation, leading to insufficient prediction accuracy. The other category is data-driven prediction methods, which use machine learning algorithms to establish a statistical mapping relationship between process parameters and shrinkage porosity defects. These methods are computationally fast, but are prone to overfitting when sample data is limited, lack physical constraints, and make it difficult to guarantee the reliability of the prediction results. Managers also find it difficult to understand the basis of these predictions, reducing the credibility of management decisions.

[0005] Secondly, existing quality management in the casting process largely adopts a "post-inspection + feedback adjustment" model. This means that after the cast iron parts have completely solidified, shrinkage defects are detected using non-destructive testing methods (such as ultrasonic testing and X-ray inspection), and then the process parameters for subsequent batches of cast iron parts are adjusted. This management model has two fundamental problems: first, it is "lagging"—it cannot effectively intervene in the currently solidifying cast iron parts, as defects have already formed and losses have occurred; second, it is "batch-based"—the control granularity is coarse, with parameter adjustments made on a batch-by-batch basis, ignoring the differentiated risks between different cast iron parts within the same batch or different areas of the same cast iron part. This leads to persistently high scrap rates and quality fluctuations.

[0006] Third, there are numerous control measures available to influence shrinkage porosity risk during the casting process, including pouring temperature adjustment, pouring speed control, mold cooling regime adjustment, and riser feeding pressure adjustment. These control measures belong to different process stages and are handled by different operational positions, resulting in fragmentation and a lack of coordination in traditional management models. Each control measure has a different impact path and a different time scale on shrinkage porosity risk (some respond quickly, while others have a significant lag). However, existing management methods typically employ a crude approach of "independent adjustment with a single measure" or "simultaneous adjustment with multiple measures of the same magnitude," failing to conduct differentiated and systematic collaborative management based on the response characteristics and impact intensity of different control measures. This leads to low efficiency in the utilization of control resources, offsetting control effects, and even producing side effects.

[0007] In summary, the existing technology lacks an intelligent management method for shrinkage porosity risk of cast iron parts that can systematically solve the above three major problems from a management perspective. That is, how to maintain the interpretability of the model while ensuring the accuracy of prediction, how to formulate and issue control decision instructions in advance before the shrinkage porosity risk is formed, and how to achieve systematic collaborative management based on the differentiated characteristics of different control methods. Summary of the Invention

[0008] In view of the above problems, this application provides an intelligent prediction and management method and system for shrinkage porosity risk of cast iron parts, which is used to solve the technical problems involved in the background art.

[0009] To achieve the above objectives, firstly, this application provides an intelligent prediction and management method for shrinkage porosity risk in cast iron parts, executed by a casting production management system, comprising the following steps:

[0010] Obtain multi-state information on the solidification process of cast iron parts;

[0011] Based on the multi-state information, a preset shrinkage risk assessment model is invoked to generate a baseline assessment value for shrinkage risk in each region of the cast iron part.

[0012] Using the deviation between the baseline assessment value of shrinkage risk and the actual detected value of shrinkage risk as the learning objective, and the temporal change characteristics of the multivariate state information as the input, a data-driven shrinkage risk deviation prediction model is trained. The prediction deviation output by the deviation prediction model is then fused with the baseline assessment value of shrinkage risk to generate a comprehensive assessment value of shrinkage risk.

[0013] The risk level of each region is determined based on the comprehensive assessment value of the shrinkage risk. For regions where the risk level exceeds the preset threshold, a feedforward control decision-making process is triggered. Based on the trend of shrinkage risk changes and the control effectiveness of each control measure, a preliminary control strategy for each control measure is formulated.

[0014] The causal influence of each process factor on shrinkage risk is determined by causal analysis. Based on the causal influence, the control measures are divided into at least two synergistic control groups, and a differentiated control plan is formulated for each group.

[0015] The adjustment plans of each group are integrated into a comprehensive control plan and transformed into decision-making instructions for each control measure, which are then issued and implemented.

[0016] Unlike existing technologies, the technical solution of this application is uniformly executed by the casting production management system, realizing a comprehensive upgrade of shrinkage porosity risk from post-detection to pre-prediction, from manual experience-based judgment to intelligent quantitative assessment, and from single control to collaborative management. First, by integrating the mechanistic model with a data-driven model—using the mechanistic model output as the benchmark assessment value and the data-driven model learning the deviation pattern between the benchmark and actual values—it retains the physical interpretability and operational adaptability of the mechanistic model while enabling intelligent compensation for nonlinear, time-varying, and uncertain factors that the mechanistic model cannot accurately describe, significantly improving the accuracy and robustness of shrinkage porosity risk assessment. Second, based on the comprehensive assessment results, a feedforward control decision-making process is automatically triggered, abandoning the passive mode of traditional feedback control that addresses defects first and then adjusts, achieving proactive intervention before shrinkage porosity risk forms. Furthermore, by employing causal analysis, the true causal relationship between various process factors and shrinkage porosity risk was identified, rather than merely a superficial correlation. Based on the strength of causal influence, control measures were scientifically grouped, and differentiated adjustment plans were formulated for each group. This approach enabled the rational allocation and precise deployment of control resources from a management perspective, overcoming the resource waste and ineffectiveness caused by the one-size-fits-all control methods of traditional approaches. Finally, the various adjustment plans were integrated into a comprehensive control plan and transformed into executable decision-making instructions. This provided casting production managers with a systematic and traceable risk management tool, effectively reducing the shrinkage porosity defect rate in cast iron parts and improving the quality of finished cast iron parts and production management efficiency.

[0017] As one embodiment of the present invention, after obtaining the multi-state information, the method further includes:

[0018] A dynamic association graph between state information is constructed based on the graph attention mechanism, and the association weight of each information node is calculated.

[0019] The feature representation of the original state information is reconstructed based on the association weights to generate an enhanced feature representation.

[0020] The key parameters of the shrinkage risk assessment model are corrected using the enhanced feature expression, and the corrected model output is used as the benchmark assessment value of the shrinkage risk.

[0021] As described above, by constructing a dynamic correlation graph between various state information through the graph attention mechanism, the complex correlation between process parameters in multi-source state information can be automatically discovered without relying on manual experience for feature selection and weight assignment, overcoming the management defects of strong subjectivity and poor consistency in traditional methods. Based on the correlation weight, the original state information is reconstructed, which allows the subsequent risk assessment model to focus on the truly salient features with management value. By using enhanced features to correct the key parameters of the mechanism model, dynamic matching between the mechanism model and actual working conditions is achieved, further improving the credibility and accuracy of the benchmark assessment value.

[0022] As one embodiment of the present invention, the method for constructing the shrinkage risk deviation prediction model is as follows:

[0023] A bidirectional memory neural network is used as the basic learning architecture to divide the solidification process of cast iron parts into multiple continuous stages according to time windows;

[0024] The difference between the current stage and the previous stage risk benchmark assessment value is used as the first input feature, and the normalized feature of the current stage multivariate state information is used as the second input feature. The two are combined and then input into the bidirectional memory neural network.

[0025] A global optimization algorithm is used to automatically optimize the network structure parameters and training hyperparameters of the bidirectional memory neural network.

[0026] As described above, using a bidirectional memory neural network as the basic learning architecture of the deviation prediction model can simultaneously capture the evolutionary patterns in both forward and backward directions during the solidification process of cast iron parts, thus fully utilizing the temporal information. Using the difference between adjacent stages of the mechanistic model as the first input feature is equivalent to injecting the model with the key information dimension of "change trend" in management decision-making, enabling the model to not only learn to assess the "current state" but also predict the "direction of change." The use of a global optimization algorithm to automatically optimize the network structure and training hyperparameters avoids the inefficient management mode of relying on repeated manual trials, thereby improving the automation level of model development and maintenance.

[0027] In one embodiment of the present invention, the preliminary control strategy is formulated in the feedforward control decision-making process as follows:

[0028] The contribution of each regulatory measure is equal to the product of its regulatory effectiveness coefficient and the amount of regulation. The change in the risk of easing in each region is equal to the sum of the contributions of all regulatory measures plus the residual term.

[0029] The optimization objective is to minimize the deviation between the change in the risk of easing in each region and the contribution of each regulatory measure. The optimization objective is solved by an orthogonal decomposition least squares strategy to obtain the preliminary adjustment amount of each regulatory measure and form the preliminary regulatory strategy.

[0030] As described above, the established feedforward control decision-making framework takes "minimizing deviation" as its optimization objective, transforming the formulation of control strategies into a quantifiable and computable mathematical management problem. The orthogonal decomposition least squares strategy is used to solve the problem, which can effectively address the complex situation where there may be correlations between the control effectiveness coefficients of various control measures, ensuring the uniqueness and numerical stability of the solution results. This provides rigorous mathematical support for the management decision-making process and avoids the management risks caused by relying on trial and error based on experience.

[0031] As one embodiment of the present invention, the regulation efficiency coefficient is continuously updated through an online learning strategy, and the update is based on: the deviation between the actual regulation effect and the expected effect, and the impact of the risk fluctuation of tightening in adjacent historical cycles on the regulation demand.

[0032] As described above, the online learning strategy enables the control efficiency coefficient to be continuously updated based on the feedback of the actual control effect, and the knowledge base of the management system to dynamically evolve with changes in equipment status, operating conditions, and external environment. This overcomes the shortcomings of the traditional static management system, which is characterized by "fixed models and inability to adapt," and achieves continuous improvement and spiraling upward development of the control management level.

[0033] As one embodiment of the present invention, the causal analysis method includes:

[0034] Based on historical data of shrinkage risk deviation and multivariate state information, a causal network diagram characterizing the causal relationship between process factors and shrinkage risk is constructed.

[0035] Intervention effect analysis was performed on the causal network diagram to calculate the causal effect value of each process factor on shrinkage risk, and the path with the causal effect value exceeding the set threshold was selected as the key causal path.

[0036] The intensity of the causal influence is determined based on the response sensitivity of each process factor on the critical causal path.

[0037] As described above, by constructing a causal network diagram and conducting intervention effect analysis, the true causal relationship between process factors and shrinkage risk can be identified from massive historical data. This effectively eliminates the interference of confounding factors and avoids management decision-making errors caused by misjudging correlation as causation. By screening key causal paths, it helps managers focus on the process links that are most decisive for shrinkage risk, thereby improving the pertinence and efficiency of risk management.

[0038] As one embodiment of the present invention, the method of dividing the control measures into at least two synergistic control groups based on the intensity of causal influence, and formulating differentiated control schemes for each group, specifically:

[0039] Based on the response characteristics of each control measure, control measures that meet the rapid response condition are classified into the rapid response control group, and the rest are classified into the steady-state compensation control group.

[0040] For the rapid response control group, the weight adjustment range is calculated based on the causal influence intensity and the current shrinkage risk deviation, and the control weight is dynamically adjusted based on the adjustment range to generate the first control scheme;

[0041] For the steady-state compensation control group, the shrinkage risk deviation is accumulated and statistically analyzed within a set time interval, and a second adjustment scheme is generated based on the accumulated results and the preset compensation coefficient.

[0042] As described above, dividing the control measures into a rapid response group and a steady-state compensation group based on their response characteristics reflects the concept of differentiated management: the rapid response group is responsible for suppressing sudden risk fluctuations, which is equivalent to the "emergency response" mechanism of management; the steady-state compensation group is responsible for eliminating long-term accumulated deviations, which is equivalent to the "continuous improvement" mechanism of management; the two types of control groups work together to take into account both the response speed and adjustment accuracy of the management system, and achieve the hierarchical management effect of "balancing emergency and normal situations, and emphasizing both speed and stability".

[0043] As one embodiment of the present invention, the execution flow of the orthogonal decomposition least squares strategy is as follows:

[0044] The regulation efficiency coefficient matrix of each regulation measure is orthogonally transformed to make the transformed regulation efficiency coefficient vectors independent of each other.

[0045] Based on the transformed coefficient matrix and the current change in the risk of contraction, the optimization objective is solved by least squares estimation to obtain a unique numerical solution for the initial adjustment amount of each control measure.

[0046] As described above, orthogonalization eliminates the coupling effect between various control measures, enabling management decisions to clearly identify the independent contribution of each control measure. The least squares estimation method based on the orthogonalized coefficient matrix can greatly simplify the solution process and improve computational efficiency, providing a feasible technical path for real-time decision-making in management systems and ensuring the output of executable control schemes within a limited time.

[0047] As one embodiment of the present invention, the multi-state information includes at least the temperature field information of the cast iron part, the material composition information of the cast iron part, the pouring process information, and the heat exchange information of the mold. The multi-state information is collected through a cross-process information integration platform.

[0048] The cross-process information integration platform collects the status information of the smelting process, casting process, and solidification and cooling process in a unified manner according to the time axis, forming a full-process status information database for cast iron parts, which can be called by the shrinkage porosity risk deviation prediction model.

[0049] As described above, by integrating the status information of each process, including smelting, casting, and solidification cooling, through a cross-process information integration platform, the information silos of the entire casting process are broken down from the management level. This enables the risk assessment model to make comprehensive judgments by utilizing multi-source information from the entire process, avoiding decision-making biases caused by information fragmentation. The establishment of a full-process status information database provides managers with complete data asset accumulation, supporting subsequent knowledge mining and management strategy optimization.

[0050] To achieve the above objectives, in a second aspect, this application provides an intelligent prediction and management system for shrinkage porosity risk in cast iron parts, comprising:

[0051] The information integration module is used to acquire multi-dimensional state information of the solidification process of cast iron parts;

[0052] The risk assessment engine is used to call a pre-set shrinkage risk assessment model to generate a shrinkage risk benchmark assessment value, and run a data-driven shrinkage risk deviation prediction model to generate a prediction deviation. The prediction deviation is then fused with the benchmark assessment value to output a comprehensive shrinkage risk assessment value.

[0053] The control decision engine is used to trigger a feedforward control decision process based on the comprehensive assessment value of shrinkage risk, formulate preliminary control strategies for each control measure based on the trend of shrinkage risk changes and the control effectiveness of each control measure; and determine the causal influence intensity of each process factor by running a sequential causal analysis model, divide the control measures into at least two coordinated control groups according to the causal influence intensity and formulate adjustment schemes for each group, integrate the adjustment schemes of each group into a comprehensive control scheme and convert it into decision command outputs for each control measure.

[0054] Unlike existing technologies, the technical solution of this application constructs a closed-loop management system through the coordinated operation of an information integration module, a risk assessment engine, and a control decision engine. This system enables the management of shrinkage porosity risk in cast iron parts to be upgraded from manual reliance to intelligent control, from passive response to proactive prevention and control, and from single-point control to collaborative management. The system has clearly defined functions for each module, standardized interfaces, and good scalability and cross-production line promotion capabilities, providing a feasible system solution for casting enterprises to achieve digital transformation in quality management.

[0055] The above description of the invention is merely an overview of the technical solution of this application. In order to enable those skilled in the art to better understand the technical solution of this application and to implement it based on the description and drawings, and to make the above-mentioned objectives and other objectives, features and advantages of this application easier to understand, the following description is provided in conjunction with the specific embodiments and drawings of this application. Attached Figure Description

[0056] The accompanying drawings are only used to illustrate the principles, implementation methods, applications, features, and effects of specific embodiments of this application and other related content, and should not be considered as limitations on this application.

[0057] In the accompanying drawings of the instruction manual:

[0058] Figure 1 This is a flowchart illustrating the steps of an intelligent prediction and management method for shrinkage porosity risk in cast iron parts according to this application.

[0059] Figure 2 This is a structural block diagram of an intelligent prediction and management system for shrinkage porosity risk of cast iron parts according to this application;

[0060] The reference numerals used in the above figures are explained as follows:

[0061] 1. Information integration module;

[0062] 2. Risk assessment engine;

[0063] 3. Regulate the decision-making engine. Detailed Implementation

[0064] To illustrate the possible application scenarios, technical principles, implementable specific solutions, and achievable objectives and effects of this application in detail, the following description, in conjunction with the listed specific embodiments and accompanying drawings, provides a detailed explanation. The embodiments described herein are merely illustrative of the technical solutions of this application and are therefore intended to limit the scope of protection of this application.

[0065] In this document, the term "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The term "embodiment" appearing in various places throughout the specification does not necessarily refer to the same embodiment, nor does it specifically limit its independence or connection with other embodiments. In principle, in this application, as long as there are no technical contradictions or conflicts, the technical features mentioned in each embodiment can be combined in any way to form corresponding implementable technical solutions.

[0066] Unless otherwise defined, the technical terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the use of related terms herein is merely for the purpose of describing particular embodiments and is not intended to limit this application.

[0067] In the description of this application, the term "and / or" is used to describe the logical relationship between objects, indicating that three relationships can exist. For example, X and / or Y means: X exists, Y exists, and X and Y exist simultaneously. Additionally, the character " / " in this document generally indicates that the preceding and following objects have an "or" logical relationship.

[0068] In this application, terms such as “first” and “second” are used only to distinguish one entity or operation from another, and do not necessarily require or imply any actual quantity, hierarchy or order relationship between these entities or operations.

[0069] Unless otherwise specified, the use of terms such as “comprising,” “including,” “having,” or other similar expressions in this application is intended to cover non-exclusive inclusion, which does not exclude the presence of additional elements in a process, method, or product that includes the stated elements, such that a process, method, or product that includes a list of elements may include not only those defined elements but also other elements not expressly listed, or elements inherent to such a process, method, or product.

[0070] In this application, expressions such as "greater than", "less than", and "exceeding" are understood to exclude the stated number; expressions such as "above", "below", and "within" are understood to include the stated number. Furthermore, in the description of the embodiments of this application, "multiple" means two or more (including two), and similar expressions related to "multiple" are also understood in this way, such as "multiple groups" and "multiple times", unless otherwise explicitly specified.

[0071] In the description of the embodiments of this application, the space-related expressions used, such as "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "vertical," "top," "bottom," "inner," "outer," "clockwise," "counterclockwise," "axial," "radial," and "circumferential," indicate the orientation or positional relationship based on the orientation or positional relationship shown in the specific embodiments or drawings. They are only for the purpose of describing the specific embodiments of this application or for the reader's understanding, and do not indicate or imply that the device or component referred to must have a specific position, a specific orientation, or be constructed or operated in a specific orientation. Therefore, they should not be construed as limitations on the embodiments of this application.

[0072] Unless otherwise expressly specified or limited, the terms "installation," "connection," "linking," "fixing," and "setting," as used in the description of the embodiments of this application, should be interpreted broadly. For example, "connection" can be a fixed connection, a detachable connection, or an integral setting; it can be a mechanical connection, an electrical connection, or a communication connection; it can be a direct connection or an indirect connection through an intermediate medium; it can be the internal connection of two components or the interaction between two components. For those skilled in the art to which this application pertains, the specific meaning of the above terms in the embodiments of this application can be understood according to the specific circumstances.

[0073] This embodiment uses the quality management system of a cast iron casting production line as an application scenario. This production line mainly produces automotive engine bracket-type cast iron parts made of ductile iron (grade QT500-7), with a maximum wall thickness of approximately 60mm and a minimum wall thickness of approximately 12mm. The pouring temperature range is 1380℃~1420℃, and the mold is resin sand casting. The production line has already established preliminary information infrastructure, including data acquisition terminals for each process and an industrial network.

[0074] Step 1: Information integration and acquisition of diverse status information.

[0075] like Figure 1 As shown in the figure, the intelligent prediction and management method for shrinkage porosity risk of cast iron parts provided in this embodiment is uniformly executed by the casting production management system.

[0076] When the cast iron part is poured and begins to solidify, the information acquisition process is initiated. The management system connects the sensor network and detection equipment of each process through the information integration module to acquire multi-dimensional state information of the current solidification process of the cast iron part. In this embodiment, the multi-dimensional state information specifically includes the following four categories:

[0077] (1) Temperature field information of cast iron parts: A total of 16 K-type armored thermocouples were arranged in the hot spot area, thin-walled area and near the gating and riser system of the cast iron parts. The temperature measurement range was 0~1600℃, the accuracy class was I, and the sampling frequency was 10Hz. The temperature field information reflects the temperature distribution and its rate of change in each area of ​​the cast iron parts during the solidification process. It is the most critical state information for judging the risk of shrinkage porosity.

[0078] (2) Material composition information of cast iron parts: Before casting, the composition of the molten iron sample is analyzed by direct reading spectrometer to obtain the percentage content of elements such as carbon, silicon, manganese, phosphorus, sulfur, and magnesium. Material composition information directly affects the solidification mode and shrinkage porosity. Cast iron parts with different composition systems exhibit significantly different shrinkage porosity sensitivities.

[0079] (3) Pouring process information: including pouring temperature (measured online by an immersion thermocouple), pouring time (recorded by a timer), and pouring speed (measured by a flow meter at the pouring port). The pouring process parameters determine the initial thermal state and filling flow behavior of the molten iron when it enters the mold.

[0080] (4) Heat exchange information of the mold: The heat flux density distribution at the interface between the cast iron part and the mold is measured by a heat flux density sensor embedded in the mold wall. This information reflects the cooling capacity of the mold for the solidification process of the cast iron part.

[0081] The aforementioned diverse status information is transmitted in real time to a cross-process information integration platform via industrial Ethernet. This platform aligns the status information of the smelting, casting, and solidification / cooling processes according to timestamps, forming a complete status information database for cast iron parts, providing a comprehensive data foundation for subsequent risk assessment model calls.

[0082] Step 2: Calling up the risk assessment model and generating benchmark assessment values.

[0083] After obtaining multi-dimensional status information, the management system calls the preset shrinkage risk assessment model to generate the baseline assessment value of shrinkage risk for each region of the cast iron part.

[0084] In this embodiment, the shrinkage risk assessment model is a computational model based on solidification physics mechanisms, comprising three sub-modules:

[0085] (1) Temperature field evolution module: Based on the Fourier heat conduction equation and considering the release of latent heat of solidification, the temperature field of the solidification process of the cast iron is numerically solved using the finite difference method to obtain the temperature distribution and cooling rate of each region of the cast iron at different solidification times. The boundary conditions of this module are determined by the heat exchange information of the mold and the ambient temperature information.

[0086] (2) Feeding Channel Evaluation Module: Based on the feeding flow theory at the solidification front, the module calculates the degree of blockage of the feeding channel by dendrite overlap and eutectic growth, and outputs the feeding channel coefficient (value range 0~1, the larger the value, the smoother the feeding channel). The calculation of this coefficient takes into account the ratio of temperature gradient to solidification rate and the degree of compositional segregation of the alloy.

[0087] (3) Shrinkage porosity determination module: Based on the thermodynamic and kinetic conditions of shrinkage porosity formation, combined with the pressure distribution and feeding capacity of each region of the cast iron part, the nucleation probability of shrinkage porosity and the volume fraction of shrinkage porosity are calculated.

[0088] The three sub-modules are coupled and solved together. Based on the temperature field evolution results of each region of the cast iron part, the feeding channel state and shrinkage porosity tendency are calculated, and finally, the shrinkage porosity risk mechanism assessment value of each region of the cast iron part at the current moment is output. ( Indicates the first (The cast iron part is discretized into approximately 5000 finite difference grid cells, with each grid cell serving as an independent evaluation region).

[0089] After obtaining the mechanism assessment value, it is configured as the baseline assessment value for shrinkage risk. This benchmark assessment represents a quantitative estimate of shrinkage risk based on physical mechanisms and has good physical interpretability.

[0090] Preferably, before invoking the risk assessment model for loosening, the management system first performs a feature reconstruction step based on a graph attention mechanism. The specific method is as follows:

[0091] First, the management system categorizes the acquired multi-dimensional state information into multiple information sets based on physical attributes. In this embodiment, these are categorized into a thermal property information set (including thermal conductivity, specific heat capacity, density, etc.), a process information set (including pouring temperature, pouring speed, mold temperature, etc.), and a geometric information set (including cast iron part wall thickness, modulus, etc.).

[0092] Then, a dynamic association graph is constructed based on a graph attention mechanism to connect the various state information. In this dynamic association graph, different types of process parameters serve as different types of graph nodes, and the edges between nodes represent the information relationships between the parameters. The attention score between each parameter node is calculated using the attention mechanism, which quantifies the association strength between each pair of parameters, thereby constructing a dynamic association matrix that reflects the coupling relationships of multiple parameters.

[0093] The association weights (two dimensions: associativity centrality and eigenvector centrality) of each information node are calculated based on the dynamic association matrix. These association weights are then used to reconstruct the feature representation of the original state information in a weighted manner, thereby generating an enhanced feature representation.

[0094] Finally, the key boundary parameters (such as the heat transfer coefficient at the mold interface and the latent heat release rate) of the shrinkage porosity risk assessment model are corrected using enhanced feature representation. The corrected model output value is then used as the benchmark assessment value for shrinkage porosity risk. This step enables the benchmark assessment value to be dynamically and adaptively adjusted based on actual operating conditions.

[0095] Step 3: Training the risk deviation prediction model and generating comprehensive evaluation values.

[0096] The management system uses the deviation between the baseline assessment value and the actual detection value of shrinkage risk as the learning objective, and the time-series change characteristics of multi-dimensional state information as input to train a data-driven shrinkage risk deviation prediction model. Then, the predicted deviation output by the deviation prediction model is fused with the baseline assessment value to generate a comprehensive shrinkage risk assessment value.

[0097] In this embodiment, the shrinkage risk deviation prediction model is built based on a bidirectional memory neural network (BiLSTM) learning architecture. The specific construction method is as follows:

[0098] First, the management system will analyze the entire solidification process of the cast iron parts according to the time step. The second is divided into multiple consecutive stages, each stage corresponding to a time window in the solidification process of the cast iron part, totaling approximately 400 consecutive stages.

[0099] Then, for each stage Construct the following two input features:

[0100] (1) First input feature - current stage risk benchmark assessment value Compared with the previous stage's baseline assessment value of risk reduction The difference This difference reflects the changing trend of risk between adjacent stages and is key management information for determining whether the risk is "intensifying" or "mitigating".

[0101] (2) Second input feature - the normalized feature vector of multivariate state information in the current stage, including the normalized values ​​of parameters such as real-time temperature value, temperature change rate, pouring speed, and heat flux density of each temperature measuring point.

[0102] The first and second input features are concatenated and combined, and then input into a bidirectional memory neural network. The bidirectional memory neural network contains a forward propagation layer and a backward propagation layer, which can simultaneously capture the temporal dependencies of the solidification process from both forward and backward directions.

[0103] A global optimization algorithm—differential evolution—was used to automatically optimize the network structure parameters (number of hidden layer nodes) and training hyperparameters (maximum number of iterations, learning rate) of the bidirectional memory neural network. Using the root mean square error (RMSE) as the fitness evaluation function, the optimal hyperparameter combination was obtained after approximately 200 iterations: 72 hidden nodes in the forward layer, 68 hidden nodes in the backward layer, a maximum of 150 iterations, and a learning rate of 0.0012.

[0104] The root mean square error (RMSE) is a performance metric used in the model evaluation phase, and its calculation formula is as follows:

[0105] ;

[0106] in, The total number of samples in the validation set, For the first The actual detected value of the risk of shrinkage in each sample. The first output of the deviation prediction model The prediction bias of a sample.

[0107] The actual measured value of shrinkage porosity risk refers to the quantitative risk value obtained by detecting shrinkage porosity defects in various areas of the cast iron part after it has completely solidified, using offline non-destructive testing methods (including ultrasonic testing and X-ray digital imaging). This measured value is used to construct the labeled sample set required for model training, namely, a paired sample database of "state information - measured shrinkage porosity risk value" for historical cast iron parts. After the model is put into online operation, the management system does not need to obtain this measured value in real time; it only needs to use the already trained deviation prediction model for forward inference. In other words, this measured value only exists in the data labeling stage of offline training, not as a real-time input in the online operation stage.

[0108] The deviation prediction model was trained using historical casting production data (containing approximately 2,000 sets of cast iron samples). The training employed a backpropagation algorithm over time, and training was stopped when the evaluation metrics on the validation set no longer improved.

[0109] After training, the model is put into online operation. For the current stage of the cast iron part, the combined feature vector is input into the trained deviation prediction model, and the model outputs the predicted deviation. Then, the management system integrates the prediction deviation with the baseline assessment value of shrinkage risk (in this embodiment, additive integration is used, i.e., ...). ), generate a comprehensive risk assessment value for shrinkage. .

[0110] The advantage of this comprehensive evaluation value lies in that it retains the physical baseline of the mechanistic model (ensuring interpretability) while intelligently compensating for systematic biases in the mechanistic model using a data-driven model (improving accuracy). In actual production, the prediction accuracy of the comprehensive evaluation value is significantly improved compared to using either the mechanistic model or the data-driven model alone.

[0111] Step 4: Risk level determination and triggering of feedforward control decision-making process.

[0112] The management system is based on a comprehensive assessment of the risk of shrinkage. Determine the shrinkage porosity risk level in different areas of the cast iron part. This embodiment sets three risk levels:

[0113] The mathematical expression for risk level determination is:

[0114] ;

[0115] in, The threshold for medium risk. The threshold value is designated as the high-risk threshold. The threshold value is determined based on the statistical correspondence between historical shrinkage defect detection data and mechanistic model calculations for the production line. Specifically, ROC curve analysis was performed on the calculated shrinkage risk mechanism values ​​of at least 200 historical cast iron parts and their corresponding X-ray shrinkage levels. The threshold value corresponding to the maximum Youden index was selected as the medium-risk threshold. and high risk threshold .

[0116] For risk levels exceeding a preset threshold (in this embodiment, the preset threshold is medium risk, i.e.) In areas where shrinkage defects actually occur, the management system automatically triggers a feedforward control decision-making process. The core management idea of ​​this process is: at the current moment, based on the risk status and changing trends reflected by the comprehensive assessment value, to formulate adjustment plans for various control measures in advance before the shrinkage defects actually form, thereby realizing a management model transformation from "post-event remediation" to "pre-event control".

[0117] In the feedforward control decision-making process, the initial control strategy is formulated as follows:

[0118] The contribution of each regulatory measure to the risk of easing is equal to the product of its regulatory effectiveness coefficient and the amount of regulation. That is, the contribution of the j-th regulatory measure to the ith region is... The change in the risk of monetary tightening in each region equals the sum of the contributions of all regulatory measures plus the residual term:

[0119] ;

[0120] in, For the first The change in the risk of shrinkage in each region (calculated from the change in the comprehensive assessment value within adjacent control periods). The total number of control measures participating in feedforward control decision-making (in this embodiment) This includes four methods: adjusting pouring temperature, adjusting pouring speed, adjusting mold cooling system, and adjusting riser feeding pressure. For the first The first regulatory measure in the The regulation efficiency coefficient of each region For the first The adjustment quantity to be solved for each control measure This is the residual term.

[0121] Regulation efficiency coefficient The initial value is obtained through regression analysis of process experimental data. Preferably, this coefficient is continuously updated through an online learning strategy—the management system compares the deviation between the actual and expected effects of each control decision, and dynamically adjusts the control efficiency coefficient by combining the impact of easing risk fluctuations on control demand within adjacent historical cycles, so that the management system's knowledge base can continuously evolve with changes in operating conditions.

[0122] The regulation efficiency coefficient The physical meaning of is: the first When the adjustment amount of each regulatory tool changes by one unit, at the first... The change in the comprehensive assessment value of shrinkage porosity risk caused by each area of ​​the cast iron part is quantified as "shrinkage porosity risk unit / process adjustment unit". The initial value of the control efficiency coefficient is obtained as follows:

[0123] ;

[0124] This coefficient is obtained by fitting partial derivatives to historical process adjustment experimental data. In actual operation, this coefficient is continuously updated through an online learning strategy.

[0125] The optimization objective of the management system is to minimize the deviation between the changes in the risk of easing in each region and the contribution of each regulatory measure, that is:

[0126] ;

[0127] in, This refers to the number of cast iron areas where the risk level exceeds a preset threshold.

[0128] The management system employs an orthogonal decomposition least squares strategy to solve the aforementioned optimization objective. The specific execution process is as follows: First, a Gram-Schmidt orthogonalization transformation is performed on the control efficiency coefficient matrix of each control measure, ensuring that the transformed control efficiency coefficient vectors satisfy the linear independence condition where the inner product is zero, thereby eliminating the coupling effect between the control measures. Then, based on the orthogonally transformed coefficient matrix and the current change in the risk of tightening, the normal equation is solved using least squares estimation to obtain a unique numerical solution for the initial adjustment amount of each control measure, forming the initial control strategy.

[0129] Step 5: Causal analysis and formulation of differentiated synergistic regulation schemes.

[0130] Based on the preliminary control strategy generated by the feedforward control decision-making process, the management system further uses causal analysis to determine the causal influence of each process factor on the risk of shrinkage. According to the causal influence, the control measures are divided into at least two coordinated control groups, and differentiated control plans are formulated for each group.

[0131] The specific execution flow of the above causal analysis method is as follows:

[0132] First, the management system constructs a causal network diagram representing the causal relationship between process factors and shrinkage risk based on historically accumulated shrinkage risk deviation data and multivariate state information. The causal network diagram is a directed acyclic graph (DAG), where nodes represent variables in the solidification process (such as casting temperature, cooling rate, and shrinkage risk), and directed edges represent the causal direction between variables. This embodiment uses a constraint-based causal structure learning algorithm (PC algorithm) to construct the causal network diagram, determines the causal direction between nodes through conditional independence testing, and sets the significance level to 0.05.

[0133] After the causal network diagram is constructed, a do-calculus intervention operation is performed on it to calculate the causal effect value of each process factor node on the shrinkage risk outcome node. This value quantifies the degree to which the shrinkage risk changes when a certain process factor is artificially altered. Paths with causal effect values ​​exceeding a set threshold (0.5 in this embodiment) are selected as key causal paths and sorted according to their causal effect values. The response sensitivity of each process factor on the key causal paths is calculated, and this sensitivity is used as a numerical indicator to quantify the strength of the causal influence of each process factor.

[0134] After obtaining the causal influence strength of each process factor, the management system groups the control measures according to this strength. The grouping is based on the response characteristics of each control measure (the time required from issuing the control command to the actual production of the control effect):

[0135] Control measures that meet the fast response condition (response time less than 2 seconds in this embodiment) are classified into the fast response control group. In this embodiment, pouring temperature regulation and pouring speed regulation belong to this group (response time approximately 0.5~1 seconds).

[0136] The remaining control measures are classified into the steady-state compensation control group. In this embodiment, the adjustment of the mold cooling system and the regulation of the riser feeding pressure belong to this group (response time is about 3 to 5 seconds).

[0137] After grouping, the management system develops differentiated adjustment plans for each group:

[0138] For the rapid response control group, the management system calculates the weight adjustment range based on the strength of the causal impact and the current risk deviation of tightening, and dynamically adjusts the control weights based on this adjustment range to generate the first control plan. Specifically, the control measure with a stronger causal impact has a larger weight adjustment range and plays a stronger leading role in the control. This strategy enables the management system to respond quickly to sudden risk fluctuations and perform an "emergency management" function.

[0139] For the steady-state compensation control group, the management system accumulates and statistically analyzes the shrinkage risk deviation within a set time interval (10 seconds in this embodiment), and generates a second adjustment scheme based on the accumulated results and a preset compensation coefficient. This strategy enables the management system to smoothly eliminate long-term accumulated deviations and perform a "continuous improvement" function.

[0140] The two control schemes work together to balance the response speed and adjustment accuracy of the management system.

[0141] Step 6: Integration of comprehensive control plans and issuance of decision-making instructions.

[0142] The management system integrates the first adjustment scheme of the rapid response control group and the second adjustment scheme of the steady-state compensation control group into a comprehensive control scheme. The integration method is as follows: the adjustment amounts of each control measure are directly summed, because the two control groups target different control measures, thus avoiding any issue of duplicate adjustments.

[0143] After the comprehensive control plan is completed, the management system converts it into decision instructions for each control measure. In this embodiment, the generated decision instructions include:

[0144] Pouring process: Pouring temperature increased by 8℃;

[0145] Pouring process: Pouring speed reduced by 5%;

[0146] Cooling process: The cooling air volume for the mold is reduced by 12%;

[0147] Feeding process: The riser feeding pressure increases by 0.3 MPa.

[0148] The aforementioned decision instructions are sent to the execution terminals (PLC controllers or operator stations) of each process through the output interface of the management system, and the on-site operators or automated actuators perform the corresponding adjustment operations according to the instructions.

[0149] After each management cycle (2 seconds in this embodiment), the management system automatically repeats the process from step one to step six above to achieve real-time online assessment and continuous feedforward management of shrinkage risk.

[0150] To verify the management effectiveness of this embodiment, a comparative experiment was conducted on the production line of this embodiment. Three management batches with the same material, cast iron part structure, and process conditions were selected, each batch containing 50 cast iron parts:

[0151] The first batch adopted the traditional management model – “post-inspection + feedback adjustment” (after the cast iron parts were solidified, shrinkage defects were found through ultrasonic testing, and then the process parameters of subsequent batches of cast iron parts were adjusted).

[0152] The second batch adopted the management system of the present invention, but only enabled the fusion prediction function (risk intelligent assessment part), and did not enable the feedforward control decision and causal grouping management functions;

[0153] The third batch adopts the complete management scheme of this invention (fusion prediction + feedforward regulation + causal grouping collaborative management).

[0154] The experimental results are shown in Table 1 below:

[0155]

[0156] Table 1

[0157] The test results in Table 1 above show that after adopting the complete management solution of this embodiment:

[0158] The consistency between the comprehensive risk assessment value and the X-ray detection results reached 95.8%, which is a significant improvement compared to using the mechanism model alone (about 82%) and the data-driven model alone (about 88%). At the same time, managers can trace the physical basis of the prediction results through the benchmark assessment value, overcoming the management trust problem of the "black box model".

[0159] The control and response mechanism has shifted from "post-event remediation with a 30-second delay" to "pre-event control with a 2-second advance notice," achieving a model upgrade from passive to proactive management.

[0160] Through causal analysis and group-based regulation, the synergistic efficiency of various regulatory measures has been improved by about 40% compared with extensive unified regulation, demonstrating the value of "precision management".

[0161] The generation of decision instructions no longer relies on the personal experience of operators, but is based on systematic intelligent evaluation and optimization calculation, ensuring the consistency of management decisions among different shifts and different operators.

[0162] The above test results fully verify the practical application value and management effectiveness of this embodiment in the field of casting shrinkage risk management.

[0163] like Figure 2 As shown, this embodiment also provides an intelligent prediction and management system for shrinkage porosity risk in cast iron parts, including:

[0164] Information integration module 1 is used to acquire multi-state information of the solidification process of cast iron parts;

[0165] Risk assessment engine 2 is used to call a preset shrinkage risk assessment model to generate a shrinkage risk benchmark assessment value, and run a data-driven shrinkage risk deviation prediction model to generate a prediction deviation. The prediction deviation is then fused with the benchmark assessment value to output a comprehensive shrinkage risk assessment value.

[0166] The control decision engine 3 is used to trigger the feedforward control decision process based on the comprehensive assessment value of shrinkage risk, formulate preliminary control strategies for each control measure based on the trend of shrinkage risk changes and the control effectiveness of each control measure; and determine the causal influence intensity of each process factor by running the sequential causal analysis model, divide the control measures into at least two coordinated control groups according to the causal influence intensity and formulate adjustment schemes for each group, integrate the adjustment schemes of each group into a comprehensive control scheme and convert it into decision command outputs for each control measure.

[0167] Unlike existing technologies, the technical solution in this embodiment constructs a closed-loop management system that integrates information perception, intelligent assessment, strategy generation, and command issuance through the coordinated operation of an information integration module, a risk assessment engine, and a control decision engine. This achieves a comprehensive upgrade in the management of shrinkage porosity risk in cast iron parts, moving from manual reliance to intelligent control, from passive response to proactive prevention and control, and from single-point control to collaborative management. The system's modules have clearly defined functions and standardized interfaces, possessing excellent scalability and cross-production line promotion capabilities, providing a feasible system solution for foundry enterprises to achieve digital transformation in quality management.

[0168] Those skilled in the art will understand that although some embodiments herein include certain features included in other embodiments but not others, combinations of features from different embodiments are intended to be within the scope of this application and form different embodiments. For example, in the claims, any of the claimed embodiments can be used in any combination.

[0169] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and not to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. These modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application, and they should all be covered within the scope of the claims and specification of this application. In particular, as long as there is no structural conflict, the various technical features mentioned in the embodiments can be combined in any way. This application is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.

Claims

1. A method for intelligent prediction and management of shrinkage risk of cast iron pieces, characterized in that, This process, executed by the casting production management system, includes the following steps: Obtain multi-state information on the solidification process of cast iron parts; Based on the multi-state information, a preset shrinkage risk assessment model is invoked to generate a baseline assessment value for shrinkage risk in each region of the cast iron part. Using the deviation between the baseline assessment value of shrinkage risk and the actual detected value of shrinkage risk as the learning objective, and the temporal change characteristics of the multivariate state information as the input, a data-driven shrinkage risk deviation prediction model is trained. The prediction deviation output by the deviation prediction model is then fused with the baseline assessment value of shrinkage risk to generate a comprehensive assessment value of shrinkage risk. The risk level of each region is determined based on the comprehensive assessment value of the shrinkage risk. For regions where the risk level exceeds the preset threshold, a feedforward control decision-making process is triggered. Based on the trend of shrinkage risk changes and the control effectiveness of each control measure, a preliminary control strategy for each control measure is formulated. The causal influence of each process factor on shrinkage risk is determined by causal analysis. Based on the causal influence, the control measures are divided into at least two synergistic control groups, and a differentiated control plan is formulated for each group. The adjustment plans of each group are integrated into a comprehensive control plan and transformed into decision-making instructions for each control measure, which are then issued and implemented.

2. The method of claim 1, wherein the method further comprises: After obtaining the multi-state information, it also includes: A dynamic association graph between state information is constructed based on the graph attention mechanism, and the association weight of each information node is calculated. The feature representation of the original state information is reconstructed based on the association weights to generate an enhanced feature representation. The key parameters of the shrinkage risk assessment model are corrected using the enhanced feature expression, and the corrected model output is used as the benchmark assessment value of the shrinkage risk.

3. The method of claim 1, wherein the method further comprises: The method for constructing the shrinkage risk deviation prediction model is as follows: A bidirectional memory neural network is used as the basic learning architecture to divide the solidification process of cast iron parts into multiple continuous stages according to time windows; The difference between the current stage and the previous stage risk benchmark assessment value is used as the first input feature, and the normalized feature of the current stage multivariate state information is used as the second input feature. The two are combined and then input into the bidirectional memory neural network. A global optimization algorithm is used to automatically optimize the network structure parameters and training hyperparameters of the bidirectional memory neural network.

4. The intelligent prediction and management of the shrinkage risk of cast iron pieces method according to claim 1, characterized in that, In the feedforward control decision-making process, the preliminary control strategy is formulated as follows: The contribution of each regulatory measure is equal to the product of its regulatory effectiveness coefficient and the amount of regulation. The change in the risk of easing in each region is equal to the sum of the contributions of all regulatory measures plus the residual term. The optimization objective is to minimize the deviation between the change in the risk of easing in each region and the contribution of each regulatory measure. The optimization objective is solved by an orthogonal decomposition least squares strategy to obtain the preliminary adjustment amount of each regulatory measure and form the preliminary regulatory strategy.

5. The method of claim 4, wherein the method further comprises: The regulation efficiency coefficient is continuously updated through an online learning strategy. The update is based on the deviation between the actual regulation effect and the expected effect, as well as the impact of the risk fluctuation of tightening in adjacent historical cycles on the regulation demand.

6. The intelligent prediction and management method for shrinkage porosity risk of cast iron parts according to claim 1, characterized in that, The causal analysis method includes: Based on historical data of shrinkage risk deviation and multivariate state information, a causal network diagram characterizing the causal relationship between process factors and shrinkage risk is constructed. Intervention effect analysis was performed on the causal network diagram to calculate the causal effect value of each process factor on shrinkage risk, and the path with the causal effect value exceeding the set threshold was selected as the key causal path. The intensity of the causal influence is determined based on the response sensitivity of each process factor on the critical causal path.

7. The intelligent prediction and management method for shrinkage porosity risk of cast iron parts according to claim 6, characterized in that, The intensity of the causal influence divides the regulatory measures into at least two coordinated regulatory groups, and formulates differentiated regulatory schemes for each group, specifically: Based on the response characteristics of each control measure, control measures that meet the rapid response condition are classified into the rapid response control group, and the rest are classified into the steady-state compensation control group. For the rapid response control group, the weight adjustment range is calculated based on the causal influence intensity and the current shrinkage risk deviation, and the control weight is dynamically adjusted based on the adjustment range to generate the first control scheme; For the steady-state compensation control group, the shrinkage risk deviation is accumulated and statistically analyzed within a set time interval, and a second adjustment scheme is generated based on the accumulated results and the preset compensation coefficient.

8. The intelligent prediction and management method for shrinkage porosity risk of cast iron parts according to claim 4, characterized in that, The execution flow of the orthogonal decomposition least squares strategy is as follows: The regulation efficiency coefficient matrix of each regulation measure is orthogonally transformed to make the transformed regulation efficiency coefficient vectors independent of each other. Based on the transformed coefficient matrix and the current change in the risk of contraction, the optimization objective is solved by least squares estimation to obtain a unique numerical solution for the initial adjustment amount of each control measure.

9. The intelligent prediction and management method for shrinkage porosity risk of cast iron parts according to claim 1, characterized in that, The multi-dimensional state information includes at least the temperature field information of the cast iron part, the material composition information of the cast iron part, the pouring process information, and the heat exchange information of the mold. The multi-dimensional state information is collected through a cross-process information integration platform. The cross-process information integration platform collects the status information of the smelting process, casting process, and solidification and cooling process in a unified manner according to the time axis, forming a full-process status information database for cast iron parts, which can be called by the shrinkage porosity risk deviation prediction model.

10. A smart prediction and management system for shrinkage porosity risk in cast iron parts, characterized in that, include: The information integration module is used to acquire multi-dimensional state information of the solidification process of cast iron parts; The risk assessment engine is used to call a pre-set shrinkage risk assessment model to generate a shrinkage risk benchmark assessment value, and run a data-driven shrinkage risk deviation prediction model to generate a prediction deviation. The prediction deviation is then fused with the benchmark assessment value to output a comprehensive shrinkage risk assessment value. The regulation decision engine is used to trigger the feedforward regulation decision process based on the comprehensive assessment value of the shrinkage risk, and to formulate the preliminary regulation strategy of each regulation measure based on the trend of shrinkage risk change and the regulation effectiveness of each regulation measure. Furthermore, the runtime causal analysis model determines the causal influence intensity of each process factor, divides the control measures into at least two coordinated control groups based on the causal influence intensity, formulates control schemes for each group, integrates the control schemes of each group into a comprehensive control scheme, and transforms it into decision command outputs for each control measure.

Images