Industrial simulation parameter adaptive adjustment method based on AI intelligent agent

Abstract

The invention relates to the technical field of industrial simulation parameter optimization, and discloses an industrial simulation parameter adaptive adjustment method based on an AI agent, and the method comprises the steps: obtaining solidification process time sequence temperature field data, and generating a solidification leading edge evolution sequence; generating a feeding path evolution log by using a connectivity analysis algorithm; utilizing an inter-field coupling analysis algorithm to generate a flow-heat bidirectional coupling coefficient matrix; when the shrinkage cavity defect is detected, judging the precedence relation between the solidification completion moment and the feeding channel closing moment by utilizing a time sequence comparison algorithm, and generating a shrinkage cavity mechanism classification result; acquiring a corresponding coupling path selection decision based on the mechanism-coupling path mapping table; extracting an influence coefficient sub-matrix from the flow-heat bidirectional coupling coefficient matrix to generate a coupling parameter response predictor; and calculating a flow thermal coupling parameter adjustment scheme guided by a parameter adjustment amount generation mechanism by using a reverse solution algorithm. According to the invention, accurate classification and targeted parameter adjustment of the shrinkage mechanism are realized.

Description

Technical Field

[0001] This invention relates to the field of industrial simulation parameter optimization technology, and more specifically, to an adaptive adjustment method for industrial simulation parameters based on AI agents. Background Technology

[0002] During the solidification process of die casting, the formation of shrinkage cavities involves a complex thermo-fluid coupling mechanism. When a localized area of ​​molten metal in a casting solidifies and shrinks, if it cannot receive timely feeding from adjacent liquid phase regions, shrinkage cavities will form. From a physical mechanism perspective, the formation of shrinkage cavities can be attributed to two different causes: the first is improper solidification sequence leading to premature closure of the feeding channels, where the liquid phase region is surrounded by the solid phase and cannot receive feeding, known as sequential shrinkage cavities; the second is insufficient feeding pressure transmission, where although the feeding channels remain open, the pressure driving the molten metal to flow towards the shrinkage area is insufficient, known as driven shrinkage cavities.

[0003] Taking the automotive parts manufacturing industry as an example, a certain automotive parts manufacturer uses die casting to produce aluminum alloy engine brackets. The casting frequently exhibits shrinkage cavities at specific ribs, resulting in insufficient fatigue strength and a yield rate of only 65%. The traditional solution is for engineers to adjust several key process parameters (such as injection speed and mold temperature) based on experience, and then conduct multiple physical molding trials and CT scans. The entire process takes several weeks, is costly, and fails to completely solve the problem.

[0004] Engineers turned to simulation software for virtual experiments. They set initial process parameters, ran simulation calculations, and the results showed that high-temperature hotspots (areas prone to shrinkage cavities) did indeed exist in the region. Subsequently, the engineers manually modified the parameters (for example, increasing the flow rate of the local cooling channels) and ran the simulation again. However, this simulation-based approach is essentially still a cycle of "manual trial and error": engineers need to manually judge the direction and magnitude of parameter adjustments based on the simulation results, and re-run the simulation for verification after each adjustment, which is inefficient, and engineers may not be able to find the globally optimal combination of parameters.

[0005] Traditional methods for eliminating shrinkage cavities have the following shortcomings: First, shrinkage cavity diagnosis only focuses on location identification, detecting the presence of a shrinkage cavity in a specific spatial location, but failing to further distinguish whether the specific physical mechanism causing the shrinkage cavity is a sequential or driving factor. Due to a lack of understanding of the cause of shrinkage cavity formation, parameter adjustments lack specificity. Second, the fluid-thermal coupling parameter adjustment adopts a uniform strategy, failing to consider the differences in the most effective adjustment paths corresponding to different mechanism types: the root cause of sequential shrinkage cavity formation lies in the solidification sequence, and the focus should be on adjusting the temperature gradient to change the direction of the solidification front; the root cause of driving shrinkage cavity formation lies in insufficient pressure, and the focus should be on adjusting pressure parameters to enhance the feeding driving force. Third, the bidirectional coupling effect between the fluid field and the thermal field is ignored during the adjustment process. That is, flow velocity affects temperature distribution, and temperature affects the viscosity of the molten metal, thus affecting flow. Adjusting one physical field parameter may trigger adverse changes in the other, resulting in low overall optimization efficiency and unstable effects. Summary of the Invention

[0006] This invention provides an adaptive adjustment method for industrial simulation parameters based on AI intelligent agents, which solves the technical problems in related technologies, such as hole diagnosis only stopping at the location identification level and failing to distinguish the formation mechanism, using a uniform strategy for fluid-thermal coupling parameter adjustment without considering the differences in the most effective adjustment paths corresponding to different mechanism types, and ignoring the bidirectional coupling effect between the fluid field and the thermal field, resulting in low optimization efficiency.

[0007] This invention provides a method for adaptive adjustment of industrial simulation parameters based on AI agents, comprising the following steps: Acquire time-series temperature field data of the solidification process, and use the isothermal surface tracking algorithm to analyze the changes in the spatial positions of the liquidus and solidus lines over time to generate a solidification front evolution sequence. Based on the solidification front evolution sequence, the state of the feeding channel between each liquid phase region and the feeding source of the gate is evaluated using a connectivity analysis algorithm. The time and spatial coordinates of the channel closure at each location are recorded to generate a feeding path evolution log. Acquire flow field and temperature field data during the filling and solidification processes, and use the inter-field coupling analysis algorithm to calculate the influence coefficient of flow velocity on temperature distribution and the influence coefficient of temperature on the viscosity of molten metal, and generate a two-way fluid-thermal coupling coefficient matrix. When a shrinkage cavity defect is detected, the solidification completion time at the location of the shrinkage cavity is obtained, the closing time of the feeding channel at that location is extracted from the feeding path evolution log, and the time sequence comparison algorithm is used to determine the sequential relationship between the solidification completion time and the channel closing time, thereby generating a shrinkage cavity mechanism classification result. Based on the classification results of the hole shrinkage mechanism, the mechanism-coupling path mapping table is used to query and obtain the corresponding coupling path selection decision and the set of dominant parameters and subordinate parameters. Extract the influence coefficient submatrix of the dominant parameter on the subordinate parameter from the fluid-thermal bidirectional coupling coefficient matrix to generate a path-constrained coupling parameter response predictor; Obtain the current state deviation of the cavity location, use the reverse solution algorithm to calculate the adjustment amount of the dominant parameter and the compensation adjustment amount of the subordinate parameter required to eliminate the cavity, and generate a mechanism-oriented flow-heat coupling parameter adjustment scheme.

[0008] Furthermore, the step of using a time-series comparison algorithm to determine the chronological relationship between the solidification completion time and the channel closure time, and generating a classification result for the shrinkage cavity mechanism, includes: The moment when the temperature at the shrinkage cavity location drops below the solidification line temperature is retrieved from the solidification front evolution sequence and taken as the time when solidification at that location is completed. Query the closing time of the compensation channel at that position from the compensation path evolution log; If the closing time of the feeding channel is earlier than the solidification completion time, it indicates that the location has lost its feeding source when feeding is still needed, and is judged as a sequential shrinkage cavity; If the closing time of the feeding channel is later than or equal to the solidification completion time, it indicates that the feeding channel itself is not a bottleneck, and the cause of the shrinkage cavity is insufficient feeding driving force, which is judged as a driven shrinkage cavity.

[0009] Furthermore, the step of querying the mechanism-coupling path mapping table to obtain the corresponding coupling path selection decision and the set of dominant and subordinate parameters includes: If the classification result of the shrinkage mechanism is sequential shrinkage, it is mapped to the thermal field dominant adjustment path. The dominant parameter set under this path includes mold temperature, cooling channel water temperature and cooling channel flow rate, and the subordinate parameter set includes pressurization pressure and holding time. If the cavity shrinkage mechanism classification result is driven cavity shrinkage, it is mapped to the flow field dominant adjustment path. The dominant parameter set under this path includes pressurization pressure, pressurization switching position and holding time, and the subordinate parameter set includes mold temperature and cooling parameters.

[0010] Furthermore, the calculation of the influence coefficients of flow velocity on temperature distribution and temperature on the viscosity of molten metal using the inter-field coupling analysis algorithm includes: After applying a small perturbation to the flow field parameters, the coupled simulation is re-executed, the change in temperature field is recorded, and the ratio of the temperature change to the flow field perturbation is calculated as the influence coefficient of the flow field on the thermal field. After applying a small disturbance to the temperature field, the change in the flow field is recorded, and the ratio of the change in the flow field to the temperature disturbance is calculated as the influence coefficient of the thermal field on the flow field. The disturbance amount of the small disturbance is 0.1% to 1% of the current value of the parameter. If the current value of the parameter is zero, then 0.1% to 1% of the typical operating range of the parameter is taken as the disturbance amount. The thermal coupling coefficient matrix is ​​organized in the form of regional and temperature interval division, dividing the casting spatial domain into multiple regions and the temperature range into multiple intervals, and calculating the local coupling coefficient of each region in each temperature interval.

[0011] Further, the coupled parameter response predictor for generating path constraints includes: The control equations of fluid-thermal coupling are linearized at the current operating point, and the linear mapping relationship between state change and parameter increment is obtained by retaining the first-order terms. The linear mapping is achieved through matrix multiplication of the coupling coefficient submatrix and the dominant parameter increment vector, so that the dominant parameter increment is transmitted to each state variable of the flow-thermal field through the coupling coefficient submatrix to obtain the corresponding state change. The current operating point is determined by the flow field state value and temperature field state value at the moment when the shrinkage defect is detected.

[0012] Furthermore, the path-constrained coupled parameter response predictor is implemented using a feedforward neural network. The input layer of the feedforward neural network receives the dominant parameter adjustment vector, and the output layer outputs the flow-thermal field state change vector. The training process includes: The components of the dominant parameter adjustment vector and the components of the thermal field state change vector are preprocessed using the min-max normalization method. Using the parameter adjustment-state change records in historical simulation data as training samples, the Adam optimization algorithm is used for supervised learning. The loss function is the mean square error between the network-predicted fluid-thermal field state change and the actual fluid-thermal field state change recorded in the training samples. Training is considered converged when the change in the loss function value is less than a preset threshold after several consecutive training rounds.

[0013] Furthermore, the step of using the inverse solution algorithm to calculate the adjustment amount of the dominant parameter and the compensation adjustment amount of the subordinate parameter required to eliminate the shrinkage cavity includes: Obtain the current temperature and pressure field conditions at the location of the shrinkage cavity, and compare them with the target conditions to calculate the state deviation. The objective function is to minimize the residual between the predicted state change and the target deviation. The gradient descent method is used to solve for the adjustment of the dominant parameter. The step size factor is less than twice the reciprocal of the largest eigenvalue of the product of the transpose of the coupling coefficient submatrix and itself, so as to satisfy the convergence condition. Extract the submatrix of the influence of the dominant parameter on the control region of the subordinate parameter from the fluid-thermal bidirectional coupling coefficient matrix, and calculate the state change of the subordinate region caused by the adjustment of the dominant parameter. Extract the submatrix of the influence of subordinate parameters on its control region from the fluid-thermal bidirectional coupling coefficient matrix, and calculate the compensation adjustment amount of subordinate parameters so that the adjustment effect of subordinate parameters can offset the state change of subordinate region caused by the adjustment of dominant parameters.

[0014] Furthermore, for sequential shrinkage, the target state is determined as follows: the temperature at the shrinkage location when the channel is closed is set as the closing temperature, the target temperature is the sum of the liquidus temperature and the temperature margin, the temperature margin is 10% to 30% of the difference between the liquidus temperature and the solidus temperature, and the temperature field state deviation is the difference between the target temperature and the closing temperature. For driven shrinkage cavities, the target state is determined as follows: the shrinkage suction is equal to the product of the volume shrinkage rate of the molten metal and the volume and cooling rate of the local solidification region at the shrinkage cavity location, divided by the product of the effective cross-sectional area of ​​the feeding channel and the flow velocity of the molten metal; the pressure margin is equal to 32 times the product of the dynamic viscosity of the molten metal, the length of the feeding path, and the flow velocity of the molten metal in the feeding channel, divided by the square of the equivalent diameter of the feeding channel; the target state is that the feeding pressure at the shrinkage cavity location is greater than or equal to the sum of the shrinkage suction and the pressure margin.

[0015] Furthermore, it also includes: The thermal coupling parameter adjustment scheme is constrained to determine whether the adjustment amount of each parameter exceeds the equipment capacity range and process safety constraint range. For parameter adjustment amounts that exceed the range, the out-of-bounds parameter values ​​are replaced with the closest boundary values ​​using a projection algorithm, and the compensation amount of the associated parameters is recalculated to generate an executable parameter adjustment instruction. Apply the parameter adjustment instructions to the simulation environment, re-execute the solidification simulation calculation, and verify whether the shrinkage cavities have been eliminated; If the pinhole is eliminated, the input conditions and adjustment results of this adjustment are updated to the mechanism-coupling path mapping table as experience records; if the pinhole is not eliminated, the pinhole status is re-evaluated and a new round of parameter adjustment is carried out until the pinhole is eliminated or the maximum number of adjustment rounds is reached.

[0016] This invention provides an AI-based adaptive adjustment system for industrial simulation parameters, used to execute the aforementioned method, comprising: The solidification front evolution module is used to acquire time-series temperature field data of the solidification process, and uses the isothermal surface tracking algorithm to analyze the changes in the spatial position of the liquidus and solidus over time to generate a solidification front evolution sequence. The feeding path analysis module is used to evaluate the feeding channel status between each liquid phase region and the feeding source at the gate based on the solidification front evolution sequence and the connectivity analysis algorithm, record the time and spatial coordinates of channel closure at each location, and generate a feeding path evolution log. The fluid-thermal coupling analysis module is used to acquire flow field data and temperature field data during the filling and solidification processes. It uses the inter-field coupling analysis algorithm to calculate the influence coefficient of flow velocity on temperature distribution and the influence coefficient of temperature on the viscosity of molten metal, and generates a fluid-thermal bidirectional coupling coefficient matrix. The shrinkage cavity mechanism classification module is used to obtain the solidification completion time of the shrinkage cavity location when a shrinkage cavity defect is detected, extract the closing time of the feeding channel at that location from the feeding path evolution log, use a time sequence comparison algorithm to determine the chronological relationship between the solidification completion time and the channel closing time, and generate a shrinkage cavity mechanism classification result. The coupling path decision module is used to query the mechanism-coupling path mapping table based on the classification results of the hole shrinkage mechanism to obtain the corresponding coupling path selection decision and the set of dominant parameters and subordinate parameters. The response predictor building module is used to extract the influence coefficient submatrix of the dominant parameter on the subordinate parameter from the fluid-thermal bidirectional coupling coefficient matrix, and generate a path-constrained coupled parameter response predictor. The parameter adjustment scheme generation module is used to obtain the current state deviation of the shrinkage cavity position, use the reverse solving algorithm to calculate the adjustment amount of the dominant parameter and the compensation adjustment amount of the subordinate parameter required to eliminate the shrinkage cavity, and generate a mechanism-oriented flow-heat coupling parameter adjustment scheme.

[0017] The beneficial effects of this invention are as follows: the AI ​​agent can select the correct flow-heat coupling adjustment path according to the formation mechanism of the shrinkage cavity, and accurately adjust the parameters to eliminate the shrinkage cavity defect while maintaining the balance of the flow-heat field. This solves the technical problems of inaccurate diagnosis, mismatched adjustment direction, and neglect of coupling effect in the shrinkage cavity elimination process, which leads to low optimization efficiency. Attached Figure Description

[0018] Figure 1 This is a flowchart of the AI-based adaptive adjustment method for industrial simulation parameters according to the present invention; Figure 2 This is a line graph of the temperature evolution curve of monitoring point P1 in this invention, showing the temperature evolution process of monitoring point P1 (located in the thick-walled region of the inner cavity of the shell) over time in step 100; Figure 3 This is a thermogram of the heat-flow coupling coefficient of the present invention, showing the heat-flow coupling coefficient matrix obtained by sensitivity analysis in step 300; Figure 4 This is a bar chart comparing the parameters before and after adjustment in this invention, showing the adjustment schemes for the mechanism-guided flow thermal coupling parameters generated in steps 700 and 800, and comparing the numerical changes of four key process parameters before and after adjustment. Figure 5 This invention presents a time-series relationship comparison graph, which illustrates the key time-series relationship changes in the classification of concave hole mechanisms. Figure 6 This is a line graph of the convergence curve of the reverse solution algorithm of the present invention, which shows the iterative convergence process of the reverse solution algorithm in step 700. Detailed Implementation

[0019] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, some features described in the examples may be combined in other examples.

[0020] This embodiment provides a method for adaptive adjustment of industrial simulation parameters based on an AI agent, the method comprising the following steps: Step 100: Obtain time-series temperature field data of the solidification process, and use the isothermal surface tracking algorithm to analyze the changes in the spatial positions of the liquidus and solidus lines over time to generate a solidification front evolution sequence.

[0021] During the die-casting solidification simulation, the simulation software outputs the temperature values ​​of each grid node in the casting at preset time intervals. The AI ​​agent obtains this time-series temperature field data from the simulation data interface. Using an isothermal surface tracking algorithm, with the liquidus temperature and solidus temperature as the target temperature values ​​for isothermal surface extraction, the positions of the liquidus and solidus isothermal surfaces in the casting spatial domain at each moment are calculated. The isothermal surface positions at each moment are organized in chronological order to form a solidification front evolution sequence describing how the solidification front advances from the casting surface to the interior.

[0022] Furthermore, the liquidus temperature and solidus temperature are inherent physical property parameters of cast alloy materials, which can be read from the material database of the simulation software. The material database stores the thermal physical property parameters of various cast alloys. For alloy materials not included in the database, the liquidus temperature and solidus temperature can be obtained by differential scanning calorimetry or calculated using phase diagram calculation software based on the alloy composition.

[0023] It should be noted that isothermal surface tracking algorithms refer to determining the location of a spatial surface with a specified temperature in discrete temperature field data through interpolation. For simulation data using structured meshes, the Marching Cubes algorithm can be used to extract the isothermal surface. The input to the Marching Cubes algorithm is the temperature value of each mesh node and the temperature value of the target isothermal surface, and the output is an isothermal surface mesh composed of triangular facets. For simulation data using unstructured meshes, the vertex coordinates of the isothermal surface can be extracted using element-based linear interpolation methods.

[0024] Taking the die casting of an aluminum alloy engine bracket as an example, this bracket is a key structural component produced by an automotive parts manufacturer. The casting dimensions are 280mm × 220mm × 150mm, and it is made of ADC12 aluminum alloy. The liquidus temperature of ADC12 alloy is read from the material database as 580°C, and the solidus temperature as 515°C. The simulation software outputs temperature field data at 0.05s intervals, and the AI ​​agent tracks the isothermal surface during the solidification stage after filling. Taking a monitoring point P1 (located in the thick-walled region of the bracket rib, with coordinates (142, 108, 76) mm) during solidification as an example, its temperature change over time is recorded as follows: Table 1 Temperature evolution data at monitoring point P1 According to the table above, the liquidus isotherm at position P1 passes through at 1.23 s (temperature drops to 580°C), and the solidus isotherm at 2.74 s (temperature drops to 515°C). The time of solidification completion at this position is... .

[0025] Step 200: Based on the solidification front evolution sequence, the connectivity analysis algorithm is used to evaluate the state of the feeding channel between each liquid phase region and the feeding source at the gate in real time, record the time and spatial coordinates of the channel closure at each location, and generate a feeding path evolution log.

[0026] The distribution of liquid phase regions at each time step is extracted from the solidification front evolution sequence. A liquid phase region is defined as a spatial domain with a temperature higher than the liquidus temperature. The gate location is marked as a feeding source node. A connectivity analysis algorithm is used to determine whether there are continuous liquid phase paths connecting each liquid phase region node in the casting to the feeding source node. When a node was connected to the feeding source in the previous time step but becomes disconnected in the current time step, the spatial coordinates of that node and the current time step are recorded as the closing event of the feeding channel at that location. All closing events are organized by time and space to form a feeding path evolution log.

[0027] It should be noted that connectivity analysis algorithms refer to determining whether a continuous path exists between two locations in a discrete spatial domain that satisfies specific conditions. In this embodiment, a breadth-first search algorithm based on graph theory can be used, treating each liquid phase region node as a vertex of the graph and the adjacency relationship between adjacent liquid phase nodes as edges. Traversing from the source node, the reachable node is the node connected to the source node.

[0028] For the aforementioned engine bracket, the gate location was set as the feeding source node (coordinates (0, 110, 75) mm). The AI ​​agent performed connectivity analysis at each time step, recording the connectivity status of the liquid phase path between location P1 and the gate. Due to the thin-walled transition region in the bracket rib area, this region solidified below the solidus temperature first at 1.85 s, causing the liquid connection path between location P1 and the gate to be severed. The key closure events recorded in the feeding path evolution log are as follows: Table 2 Evolution Log of P1 Position Compensation Path According to the table above, the closing time of the compensation channel at position P1 is... The closed position is located in the transition area of ​​the thin wall of the rib.

[0029] Step 300: Obtain real-time data of the flow field and temperature field during the filling and solidification process, and use the inter-field coupling analysis algorithm to calculate the influence coefficient of flow velocity on temperature distribution and the influence coefficient of temperature on the viscosity of molten metal, and generate a two-way flow-heat coupling coefficient matrix.

[0030] Flow field data (including the velocity distribution of the molten metal) and temperature field data during the filling and solidification processes are obtained from simulation software. Using an inter-field coupling analysis algorithm, the influence of flow velocity changes on temperature distribution changes is analyzed, as this influence is transmitted through convective heat transfer mechanisms. Simultaneously, the influence of temperature changes on molten metal viscosity changes is analyzed, as viscosity changes affect flow resistance. These influence relationships are quantified into a coefficient matrix, where matrix elements represent the degree of influence of a unit change in a parameter at one location on another parameter at another location, generating a bidirectional flow-heat coupling coefficient matrix.

[0031] It should be noted that the inter-field coupling analysis algorithm can be implemented using a sensitivity analysis method based on simulation data. Specifically, after applying a small perturbation to the flow field parameters, the coupling simulation is re-executed, the change in temperature field is recorded, and the ratio of the temperature change to the flow field perturbation is calculated as the influence coefficient of the flow field on the thermal field; similarly, after applying a small perturbation to the temperature field, the change in flow field is recorded, and the ratio of the flow field change to the temperature perturbation is calculated as the influence coefficient of the thermal field on the flow field.

[0032] Furthermore, the value of the small perturbation should meet the following conditions: the perturbation amount should be small enough relative to the current value of the parameter to ensure the effectiveness of the linear approximation, while it should be large enough to ensure that the change in response exceeds the numerical error of the simulation calculation. Specifically, the perturbation amount can be taken as 0.1% to 1% of the current value of the parameter. If the current value of the parameter is zero, then 0.1% to 1% of the typical operating range of the parameter should be taken as the perturbation amount.

[0033] In this embodiment, to more accurately capture the nonlinear characteristics of fluid-thermal coupling, the fluid-thermal bidirectional coupling coefficient matrix can be organized in the form of regional and temperature-interval division. The casting spatial domain is divided into multiple regions, and the temperature range is divided into multiple intervals. The local coupling coefficient of each region in each temperature interval is calculated, so that the fluid-thermal bidirectional coupling coefficient matrix can reflect the variation law of coupling strength with spatial position and temperature state.

[0034] For the aforementioned engine bracket, the casting space is divided into four regions: the gating region, the runner region, the thin-walled region (rib transition area), and the thick-walled region (rib main body area). Sensitivity analysis was performed, applying a 0.5 MPa perturbation to the boost pressure and a 5°C perturbation to the initial mold temperature, recording the changes in the temperature and flow fields in each region. Taking the coupling relationship between the thin-walled transition region and the thick-walled region as an example, the extracted local coupling coefficients are as follows: Table 3. Two-way coupling coefficient of heat flow (thin-walled region - thick-walled region) Based on the sensitivity analysis results, the influence coefficient of the flow field on the thermal field °C / (m / s), the coefficient of influence of the thermal field on the flow field (through viscosity change) (m / s) / °C indicates that within the solidification temperature range, an increase in temperature leads to a decrease in viscosity, which in turn slightly increases the flow rate.

[0035] Step 400: When a shrinkage cavity defect is detected, obtain the solidification completion time at the shrinkage cavity location, extract the closing time of the shrinkage channel at that location from the shrinkage path evolution log, use a time-series comparison algorithm to determine the order of solidification time and channel closing time, and generate shrinkage cavity mechanism classification results.

[0036] When simulation software or post-processing analysis detects a shrinkage cavity defect at a certain location in the casting, the spatial coordinates of that location are obtained. The moment when the temperature at that location drops below the solidification line temperature is retrieved from the solidification front evolution sequence and taken as the solidification completion time at that location. Query the closing time of the patching channel at this location from the patching path evolution log. The temporal comparison algorithm is used to compare the chronological relationship between two points in time: like This means that the feeding channel was closed before solidification was completed at that location, indicating that the location had lost its feeding source when feeding was still needed, and was judged as a sequential shrinkage cavity; like If the feeding channel remains unobstructed or simultaneously closed when solidification is complete at this location, it indicates that the feeding channel itself is not a bottleneck, and the cause of the shrinkage cavity is insufficient feeding driving force, which is determined to be a driven shrinkage cavity.

[0037] Furthermore, when When the closing time of the feeding channel coincides with the completion time of solidification, it is classified as a driven shrinkage cavity. The basis for this classification is that the feeding channel remains unobstructed throughout the solidification process until the moment of completion of solidification. If there is sufficient feeding driving pressure, the molten metal can flow into the shrinkage area to complete the feeding before solidification is completed. Therefore, in this case, the root cause of shrinkage cavity formation is still insufficient driving force rather than premature closure of the channel.

[0038] The judgment result is output as the classification result of the shrinkage mechanism.

[0039] After solidification, the simulation software detected a shrinkage cavity defect at location P1 (engine bracket rib area), with a volume of 2.3 mm³. This is the main defect location that caused the manufacturer's product yield to be only 65%. The AI ​​agent performs a timing comparison: obtaining the solidification completion time at this location from step 100. Obtain the closing time of the compensation channel at this location from step 200. .because This means that the feeding channel was closed 0.89 seconds before solidification at position P1. The shrinkage cavity at position P1 was determined to be a sequential shrinkage cavity, which was formed because the thin-walled transition area of ​​the rib solidified too early and cut off the feeding channel.

[0040] Step 500: Based on the classification results of the hole shrinkage mechanism, use the mechanism-coupling path mapping table to query and obtain the corresponding coupling path selection decision and the set of dominant parameters and subordinate parameters.

[0041] The mechanism-coupling path mapping table is a pre-established decision rule table that maps the types of concave cavity mechanisms to the most effective fluid-thermal coupling parameter adjustment paths. The concave cavity mechanism classification results can be used to query the mechanism-coupling path mapping table. If the classification result of the shrinkage mechanism is sequential shrinkage, it is mapped to the "thermal field dominant adjustment path". The dominant parameter set under this path includes parameters that affect the temperature gradient, such as mold temperature, cooling channel water temperature, and cooling channel flow rate. The subordinate parameter set includes flow field related parameters such as pressurization pressure and holding time. If the cavity shrinkage mechanism classification result is driven cavity shrinkage, it is mapped to the "flow field dominant adjustment path". The dominant parameter set under this path includes parameters that affect the shrinkage compensation driving force, such as pressurization pressure, pressurization switching position, and holding time. The subordinate parameter set includes thermal field related parameters such as mold temperature and cooling parameters.

[0042] Generate coupled path selection decisions and the corresponding sets of dominant and subordinate parameters.

[0043] It should be noted that the content of the mechanism-coupling path mapping table can be pre-defined based on domain knowledge, or it can be generated after statistical analysis of historical simulation data. For sequential shrinkage cavities, since the root cause is improper solidification sequence, changing the temperature gradient can change the direction and speed of the solidification front, thereby re-establishing a solidification sequence that is conducive to feeding. For driven shrinkage cavities, since the root cause is insufficient pressure, increasing the feeding pressure can enhance the force driving the molten metal to flow towards the shrinkage region.

[0044] The AI ​​agent uses "sequential shrinkage" as the query key to search the mechanism-coupling path mapping table and obtains the "thermal field-dominated adjustment path" decision. The corresponding parameter set is divided into: the dominant parameter set includes the mold temperature corresponding to the thin-walled region. Water temperature in the thin-walled zone cooling channel Cooling water flow rate in thin-walled region The subordinate parameter set includes boost pressure. Holding time .

[0045] Step 600: Based on the coupled path selection decision, extract the influence coefficient submatrix of the dominant parameter on the subordinate parameter from the fluid-thermal bidirectional coupling coefficient matrix, calculate the joint response relationship of the dominant parameter change to the fluid-thermal field state using the coupled differential equation discretization method, and generate a path-constrained coupled parameter response predictor.

[0046] Based on the dominant parameter set and dependent parameter set determined in step 500, submatrices related to the dominant and dependent parameters are extracted from the fluid-thermal bidirectional coupling coefficient matrix generated in step 300. The coupling coefficient submatrices describe the intensity of the influence of changes in the dominant parameters on the dependent parameters and the action path of the convective thermal field state.

[0047] Using the discretization method of coupled differential equations, the incremental values ​​of the dominant parameters are established. Changes in the state of the heat flow field The mapping relationship between them. This mapping relationship can be expressed as: in, The extracted coupling coefficient submatrix, This is the response function obtained based on the discretization of the coupled differential equations. The specific calculation method for this response function is as follows: the governing equations of the fluid-thermal coupling are linearized at the current operating point, and after retaining the first-order terms, a linear mapping relationship between the state change and the parameter increment is obtained; this linear mapping is expressed through the coupling coefficient submatrix. With the dominant parameter increment vector Implementation of matrix multiplication operations.

[0048] Furthermore, the current operating point refers to the state point of the flow and thermal field at the time of linearization. The current operating point is determined by the flow field and temperature field state values ​​at the moment the shrinkage defect is detected, including the current values ​​of state variables such as temperature, velocity, and pressure at each location. The linearization process expands the Taylor series at the current operating point and retains the first-order terms, resulting in the following linear mapping: The matrix multiplication operation transmits the increment of the dominant parameter to each state variable in the thermal flow field through the coupling coefficient submatrix, thus obtaining the corresponding state change.

[0049] Furthermore, suppose the set of dominant parameters includes The parameters, including the state variables of the thermal flow field, are as follows: If there are multiple components, then the dominant parameter increment vector for 3D column vector, thermal field state change vector for 3D column vectors, coupling coefficient submatrix The dimension is This is to ensure dimension matching for matrix multiplication operations.

[0050] This mapping relationship is encapsulated as a path-constrained coupled parameter response predictor, which can predict changes in the fluid thermal field state based on the adjustment amount of the input dominant parameter.

[0051] In this embodiment, to improve prediction accuracy, the path-constrained coupled parameter response predictor can be implemented using a feedforward neural network. The input layer of this feedforward neural network receives the dominant parameter adjustment vector, and the output layer outputs the flow-thermal field state change vector.

[0052] Due to the dominant parameter adjustment vector The components may include different physical quantities such as temperature, pressure, flow rate, and time, with significant differences in their dimensions and numerical ranges; the vector of state change in the thermal flow field. The components in the vector include temperature and flow rate changes, which also exhibit dimensional inconsistencies. To eliminate the impact of these dimensional differences on neural network training and prediction, normalization preprocessing is required for both input and output data. Specifically, each component of the dominant parameter adjustment vector is scaled to a uniform numerical range using a min-max normalization method; the components of the convective thermal field state change vector are also processed using the min-max normalization method. The minimum and maximum values ​​of each parameter component required for normalization are statistically obtained from historical simulation data.

[0053] The aforementioned feedforward neural network is trained using supervised learning, with parameter adjustment-state change records from historical simulation data as training samples. The normalized parameter adjustments are used as input samples, and the corresponding normalized thermal field state changes are used as labels. The training process employs the Adam optimization algorithm, with the mean squared error loss function as the loss function, expressed as: in, The number of training samples. For the changes in the fluid-thermal field state predicted by the network, This represents the actual fluid thermal field state changes recorded in the training samples.

[0054] Furthermore, the convergence criterion for neural network training is: when the change in the loss function value over several consecutive training rounds is less than a preset threshold, the training is considered converged. Specifically, let the... The loss function value after training rounds is When satisfied Convergence is determined at a certain time, where the convergence threshold is... The value of is determined based on the magnitude of the state change, and can be taken as 1% to 5% of the standard deviation of the state change in the training samples.

[0055] During the prediction phase, the fluid-thermal field state change vector output by the neural network is a normalized value. It needs to be converted into a state change quantity with actual physical meaning through inverse normalization. Inverse normalization uses the opposite operation of normalization, and uses the minimum and maximum values ​​of each component recorded during the training phase to restore the normalized value to the numerical range of the original physical quantity.

[0056] Based on the dominant adjustment path of the thermal field, an influence coefficient submatrix is ​​extracted from the thermal-fluid coupling coefficient matrix to show the influence of the dominant parameters (mold temperature, cooling water temperature, cooling water flow rate) on the subordinate parameter control region (the action region of the booster system). The submatrix has the following dimensions: Where 4 represents the number of key state variables in the flow and heat field (temperature in the thin-walled region, temperature in the thick-walled region, flow velocity in the thin-walled region, and pressure in the thick-walled region), and 3 represents the number of dominant parameters. The extracted coupling coefficient submatrix The specific values ​​are: The rows correspond to the temperature changes in the thin-walled region (°C), the temperature changes in the thick-walled region (°C), the flow rate changes in the thin-walled region (m / s), and the pressure changes in the thick-walled region (MPa), respectively. The columns correspond to the mold temperature changes (°C), the cooling water temperature changes (°C), and the cooling water flow rate changes (L / min), respectively. The matrix coefficients correspond to the sensitivity analysis results in Table 3. For example, a cooling water temperature change of -3°C corresponds to a thin-walled region temperature change of -3.2°C, which translates to a unit coefficient of approximately 1.07.

[0057] Step 700: Obtain the current temperature and pressure field deviations at the location of the shrinkage cavity, input the deviation data into the path-constrained coupled parameter response predictor, and use the inverse solution algorithm to calculate the parameter adjustment amount required to eliminate the shrinkage cavity under the constraint of prioritizing the adjustment of the dominant parameter. At the same time, calculate the compensation adjustment amount of the subordinate parameter to maintain the flow-heat field balance, and generate a mechanism-oriented flow-heat coupled parameter adjustment scheme.

[0058] Obtain the current temperature and pressure field conditions at the location of the shrinkage cavity, compare them with the ideal state without shrinkage cavity (target state), and calculate the state deviation. The target state can be set according to process requirements. For example, for sequential shrinkage cavities, the target state is that the solidification completion time at the shrinkage cavity location is later than the closing time of the feeding channel; for driven shrinkage cavities, the target state is that the feeding pressure at the shrinkage cavity location is higher than the shrinkage suction.

[0059] Furthermore, for the target state of sequential shrinkage, the time relationship needs to be converted into temperature field state deviation for optimization. Let the shrinkage location be at the moment the channel closes. The temperature is The goal is to make that position in To maintain the liquid state and keep the feeding channels open at all times, the target temperature should be higher than the liquidus temperature. The temperature field state deviation can be quantified as: in, As a temperature margin, the value is taken as 10% to 30% of the difference between the liquidus temperature and the solidus temperature, to ensure that there is still sufficient liquid phase feeding capacity at the shrinkage cavity when the channel is closed.

[0060] Furthermore, for the target state of driven shrinkage cavity, the relationship between the feeding pressure and the shrinkage suction can be quantified as follows: in, The compensating pressure at the location of the shrinkage cavity. This is the suction force generated by the shrinkage of molten metal during solidification. The pressure margin is determined based on the casting wall thickness and the length of the feeding path. The greater the wall thickness or the longer the feeding path, the greater the required pressure margin.

[0061] Furthermore, pressure margin The calculation formula is: in, This represents the dynamic viscosity of the molten metal within its solidification temperature range. This is the length of the feeding path from the gate to the location of the shrinkage cavity. To compensate for the flow velocity of the molten metal in the feeding channel, This is the equivalent diameter of the feeding channel. This formula, based on the principle of laminar flow pressure drop calculation in circular pipes, represents the flow resistance that molten metal must overcome when flowing through the feeding path.

[0062] Furthermore, the contraction suction force The calculation formula is: in, The volume shrinkage rate of the molten metal. This represents the volume of the locally solidified region at the location of the shrinkage cavity. The cooling rate at this location, To compensate for the effective cross-sectional area of ​​the channel, This represents the flow velocity of the molten metal in the feeding channel. All of these parameters can be extracted from the calculation results of the simulation software.

[0063] State deviation After normalization, the coupled parameter response predictor of the input path constraint is used to calculate the dominant parameter adjustment that can eliminate the bias using an inverse solution algorithm. The parameter adjustment values ​​obtained by inverse solving are numerical values ​​in the normalized space, and need to be converted into parameter adjustment values ​​with actual physical units through inverse normalization.

[0064] Furthermore, the objective of the inverse solution is to minimize the residual between the predicted state change and the target deviation, and its objective function is: When using the gradient descent method, the parameter update formula is: in, Step size factor For the number of iterations, Indicates transpose. This is the transpose of the coupling coefficient submatrix.

[0065] Furthermore, step size factor The value of should satisfy the convergence condition, and its range is: in, For matrix The largest eigenvalue. The convergence condition guarantees that the objective function value of the gradient descent method monotonically decreases and converges to the optimal solution during the iteration process. In practical applications, Can be taken as To achieve a faster convergence speed.

[0066] After obtaining the adjustment amount of the dominant parameter, the influence of the adjustment of the dominant parameter on the control region of the subordinate parameter is calculated based on the bidirectional coupling coefficient matrix of heat flow. The subordinate parameter is then used for compensation adjustment to offset the influence and maintain the overall balance of the heat flow field.

[0067] Furthermore, the calculation method for the subordinate parameter compensation adjustment is as follows: extract the influence submatrix of the dominant parameter on the control region of the subordinate parameter from the fluid-thermal bidirectional coupling coefficient matrix. Calculate the change in the state of the subordinate region caused by the adjustment of the dominant parameter. Then, the influence submatrix of subordinate parameters on its control region is extracted from the heat-fluid coupling coefficient matrix. Then the subordinate parameter compensation adjustment amount By solving This allows the adjustment of subordinate parameters to offset the changes in the state of subordinate regions caused by the adjustment of dominant parameters.

[0068] By combining the adjustment amount of the dominant parameter and the compensation adjustment amount of the subordinate parameter, a mechanism-oriented fluid-thermal coupling parameter adjustment scheme is generated.

[0069] For the sequential shrinkage cavity at position P1, the target state is to ensure that the thin-walled transition region completes solidification at position P1 at the specified time. The thin-walled region remains liquid at the moment the channel is closed. The temperature is 512°C, which is below the liquidus temperature of 580°C. Temperature margin is set. (Assuming the temperature difference between the liquidus and solidus is 65°C, which is 20%), then the target temperature deviation is: This means that the temperature of the thin-walled region needs to be increased by 81°C at 1.85s to maintain the liquid state. The target state deviation vector... The input is a reverse algorithm, which performs 12 gradient descent iterations (step size factor). The convergence yields the dominant parameter adjustment amount. This means that the mold temperature in the thin-walled area is increased by 18.5°C, the cooling water temperature is increased by 12.0°C (using higher temperature circulating water to slow down cooling), and the cooling water flow rate is reduced by 2.5L / min.

[0070] Based on the fluid-thermal coupling coefficient matrix, the impact of adjusting the dominant parameter on the control region of the subordinate parameter is calculated. Since adjusting the dominant parameter alters the temperature field distribution, it leads to changes in the viscosity of the molten metal, thus affecting the pressure transmission efficiency of the flow field. Therefore, the effective feeding pressure in the thick-walled region needs to be compensated. The subordinate parameter pressurization pressure also needs to be adjusted accordingly. To maintain the stability of the feeding driving force in the thick-walled region, the final fluid-thermal coupling parameter adjustment scheme is as follows: Table 4 Mechanism-guided adjustment schemes for heat-fluid coupling parameters Based on steps 100 to 700 of this application, the following steps are also included: Step 800: Perform constraint checks on the convection-thermal coupling parameter adjustment scheme, and use the projection algorithm to limit the adjustment amount of each parameter within the range of equipment capacity and process safety, and generate executable parameter adjustment instructions.

[0071] Obtain the parameter capabilities of the die-casting equipment (such as maximum boost pressure, mold temperature adjustment range, cooling water flow rate range, etc.) and process safety constraints (such as maximum mold temperature limit, pressure fluctuation limit, etc.). Perform constraint checks on the adjustment amounts of each parameter in the fluid-thermal coupling parameter adjustment scheme generated in step 700 to determine whether they exceed the allowable range.

[0072] For parameter adjustments that exceed the acceptable range, a projection algorithm is used to project them to the feasible region boundary. The input to the projection algorithm is the original parameter adjustment vector and the upper and lower bound constraints of each parameter; the output is the parameter adjustment vector that satisfies the constraints. The projection algorithm replaces the out-of-bounds parameter values ​​with the closest boundary values, while recalculating the compensation amounts of other related parameters to maintain the overall consistency of the adjustment scheme. An executable parameter adjustment instruction that satisfies all constraints is then generated.

[0073] Furthermore, let the initial adjustment amount of a certain dominant parameter be... Its upper bound constraint is The lower bound constraint is Adjustment amount after projection Determined according to the following rules: If but ,like but ,otherwise ,in This is the index for the dominant parameter. When the dominant parameter is projected onto the boundary, a difference arises between its actual adjustment and the original adjustment. The gap in the change of the flow-heat field state caused by this difference needs to be compensated by the subordinate parameters. The compensation amount is calculated by multiplying the corresponding column vector in the flow-heat bidirectional coupling coefficient matrix with the difference.

[0074] For the aforementioned parameter adjustment scheme, equipment capacity and process safety constraints were checked. The die-casting equipment has a mold temperature adjustment range of 150°C to 220°C, a cooling water temperature range of 20°C to 45°C, a cooling water flow rate range of 5L / min to 25L / min, and a booster pressure range of 30MPa to 55MPa. After inspection, the adjusted thin-walled zone mold temperature (198.5°C), cooling water temperature (37.0°C), cooling water flow rate (12.5L / min), and booster pressure (45.35MPa) are all within the allowable range and require no projection correction. An executable parameter adjustment command was generated: "Thin-walled zone mold temperature setting: 198.5°C; thin-walled zone cooling channel circulating water temperature setting: 37.0°C; thin-walled zone cooling water flow rate setting: 12.5L / min; booster pressure setting: 45.35MPa".

[0075] Step 900: Apply the parameter adjustment command to the simulation environment, re-execute the solidification simulation calculation, verify whether the shrinkage cavities have been eliminated, update the adjustment experience to the mechanism-coupling path mapping table, and output the shrinkage-cavity-free solidification result after mechanism-guided coupling adjustment.

[0076] The parameter adjustment command generated in step 800 is sent to the simulation software to update the corresponding process parameter settings. The solidification process simulation calculation is re-executed to obtain the adjusted solidification results. Shrinkage cavity detection is performed on the adjusted solidification results to verify whether the defects at the original shrinkage cavity locations have been eliminated.

[0077] If the shrinkage cavity is eliminated, the input conditions (shrinkage cavity mechanism type, initial state deviation) and adjustment results (parameter adjustment amount, adjustment effect) of this adjustment are updated to the mechanism-coupling path mapping table as an experience record to optimize subsequent adjustment decisions for similar shrinkage cavities. The output is the shrinkage-cavity-free solidification result after mechanism-guided coupling adjustment for engineers to confirm.

[0078] If the crater is not completely eliminated, return to step 400 to reassess the crater status and perform a new round of parameter adjustments until the crater is eliminated or the preset maximum number of adjustment rounds is reached.

[0079] Furthermore, the maximum number of adjustment rounds is set based on the following: the maximum number of adjustment rounds should be greater than or equal to the sum of the number of possible transformations of the hole-shrinking mechanism and the number of iterations required for parameter convergence. Since the same hole-shrinking location may change from sequential hole-shrinking to driven hole-shrinking or vice versa during parameter adjustment, each mechanism type requires at least one round of adjustment. Therefore, the lower bound of the maximum number of adjustment rounds is 2. At the same time, considering the convergence characteristics of the inverse solution algorithm, the maximum number of adjustment rounds can be set to a number that is several times the dimension of the parameter space to ensure that the algorithm has enough adjustment space to reach a feasible solution.

[0080] The parameter adjustment commands were applied to the simulation software, and the solidification simulation was re-executed. After adjustment, the solidification rate of the thin-walled transition region slowed down, and the closing time of the feeding channel was delayed from 1.85s to 2.92s. The solidification completion time at position P1 was 2.78s, which met the requirements. Under the given conditions, the feeding channel remained unobstructed when solidification was complete at position P1. The shrinkage cavity detection results showed that the shrinkage cavity defect at position P1 had been eliminated. A comparison of key timing parameters before and after adjustment is as follows: Table 5 Comparison of key timing sequences before and after parameter adjustment The AI ​​agent updated the experience record of this adjustment to the mechanism-coupling path mapping table: "Sequential shrinkage cavity, caused by premature solidification of the thin-walled area, thermal field-dominated adjustment is effective, mold temperature increased by 18.5°C and subordinate parameter pressure compensation +0.35MPa, single round adjustment successfully eliminated it."

[0081] It is understood that data preprocessing methods known to those skilled in the art include data cleaning, data transformation, and data reduction. Data transformation includes type conversion and normalization and standardization. Although the dimensions and types of data were omitted in the description of the preceding embodiments, data preprocessing is a technical knowledge known to those skilled in the art and a prerequisite step in data processing. Therefore, the previously described well-known data preprocessing steps were not described independently.

[0082] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.

Claims

1. A method for adaptive adjustment of industrial simulation parameters based on AI intelligent agents, characterized in that, Includes the following steps: Acquire time-series temperature field data of the solidification process, and use the isothermal surface tracking algorithm to analyze the changes in the spatial positions of the liquidus and solidus lines over time to generate a solidification front evolution sequence. Based on the solidification front evolution sequence, the state of the feeding channel between each liquid phase region and the feeding source of the gate is evaluated using a connectivity analysis algorithm. The time and spatial coordinates of the channel closure at each location are recorded to generate a feeding path evolution log. Acquire flow field and temperature field data during the filling and solidification processes, and use the inter-field coupling analysis algorithm to calculate the influence coefficient of flow velocity on temperature distribution and the influence coefficient of temperature on the viscosity of molten metal, and generate a two-way fluid-thermal coupling coefficient matrix. When a shrinkage cavity defect is detected, the solidification completion time at the location of the shrinkage cavity is obtained, the closing time of the feeding channel at that location is extracted from the feeding path evolution log, and the time sequence comparison algorithm is used to determine the sequential relationship between the solidification completion time and the channel closing time, thereby generating a shrinkage cavity mechanism classification result. Based on the classification results of the hole shrinkage mechanism, the mechanism-coupling path mapping table is used to query and obtain the corresponding coupling path selection decision and the set of dominant parameters and subordinate parameters. Extract the influence coefficient submatrix of the dominant parameter on the subordinate parameter from the fluid-thermal bidirectional coupling coefficient matrix to generate a path-constrained coupling parameter response predictor; Obtain the current state deviation of the cavity location, use the reverse solution algorithm to calculate the adjustment amount of the dominant parameter and the compensation adjustment amount of the subordinate parameter required to eliminate the cavity, and generate a mechanism-oriented flow-heat coupling parameter adjustment scheme.

2. The method according to claim 1, characterized in that, The step of using a time-series comparison algorithm to determine the chronological relationship between the solidification completion time and the channel closure time, and generating a classification result for shrinkage cavity mechanisms, includes: The moment when the temperature at the shrinkage cavity location drops below the solidification line temperature is retrieved from the solidification front evolution sequence and taken as the time when solidification at that location is completed. Query the closing time of the compensation channel at that position from the compensation path evolution log; If the closing time of the feeding channel is earlier than the solidification completion time, it indicates that the location has lost its feeding source when feeding is still needed, and is judged as a sequential shrinkage cavity; If the closing time of the feeding channel is later than or equal to the solidification completion time, it indicates that the feeding channel itself is not a bottleneck, and the cause of the shrinkage cavity is insufficient feeding driving force, which is judged as a driven shrinkage cavity.

3. The method according to claim 2, characterized in that, The process of querying the mechanism-coupling path mapping table to obtain the corresponding coupling path selection decision and the set of dominant and subordinate parameters includes: If the classification result of the shrinkage mechanism is sequential shrinkage, it is mapped to the thermal field dominant adjustment path. The dominant parameter set under this path includes mold temperature, cooling channel water temperature and cooling channel flow rate, and the subordinate parameter set includes pressurization pressure and holding time. If the cavity shrinkage mechanism classification result is driven cavity shrinkage, it is mapped to the flow field dominant adjustment path. The dominant parameter set under this path includes pressurization pressure, pressurization switching position and holding time, and the subordinate parameter set includes mold temperature and cooling parameters.

4. The method according to claim 1, characterized in that, The calculation of the influence coefficients of flow velocity on temperature distribution and temperature on the viscosity of molten metal using the inter-field coupling analysis algorithm includes: After applying a small perturbation to the flow field parameters, the coupled simulation is re-executed, the change in temperature field is recorded, and the ratio of the temperature change to the flow field perturbation is calculated as the influence coefficient of the flow field on the thermal field. After applying a small disturbance to the temperature field, the change in the flow field is recorded, and the ratio of the change in the flow field to the temperature disturbance is calculated as the influence coefficient of the thermal field on the flow field. The disturbance amount of the small disturbance is 0.1% to 1% of the current value of the parameter. If the current value of the parameter is zero, then 0.1% to 1% of the typical operating range of the parameter is taken as the disturbance amount. The thermal coupling coefficient matrix is ​​organized in the form of regional and temperature interval division, dividing the casting spatial domain into multiple regions and the temperature range into multiple intervals, and calculating the local coupling coefficient of each region in each temperature interval.

5. The method according to claim 1, characterized in that, The coupled parameter response predictor for generating path constraints includes: The control equations of fluid-thermal coupling are linearized at the current operating point, and the linear mapping relationship between state change and parameter increment is obtained by retaining the first-order terms. The linear mapping is achieved through matrix multiplication of the coupling coefficient submatrix and the dominant parameter increment vector, so that the dominant parameter increment is transmitted to each state variable of the flow-thermal field through the coupling coefficient submatrix to obtain the corresponding state change. The current operating point is determined by the flow field state value and temperature field state value at the moment when the shrinkage defect is detected.

6. The method according to claim 5, characterized in that, The path-constrained coupled parameter response predictor is implemented using a feedforward neural network. The input layer of the feedforward neural network receives the dominant parameter adjustment vector, and the output layer outputs the flow-thermal field state change vector. The training process includes: The components of the dominant parameter adjustment vector and the components of the thermal field state change vector are preprocessed using the min-max normalization method. Using the parameter adjustment-state change records in historical simulation data as training samples, the Adam optimization algorithm is used for supervised learning. The loss function is the mean square error between the network-predicted fluid-thermal field state change and the actual fluid-thermal field state change recorded in the training samples. Training is considered converged when the change in the loss function value is less than a preset threshold after several consecutive training rounds.

7. The method according to claim 1, characterized in that, The calculation of the dominant parameter adjustment and subordinate parameter compensation adjustment required to eliminate shrinkage cavities using the inverse solution algorithm includes: Obtain the current temperature and pressure field conditions at the location of the shrinkage cavity, and compare them with the target conditions to calculate the state deviation. The objective function is to minimize the residual between the predicted state change and the target deviation. The gradient descent method is used to solve for the adjustment of the dominant parameter. The step size factor is less than twice the reciprocal of the largest eigenvalue of the product of the transpose of the coupling coefficient submatrix and itself, so as to satisfy the convergence condition. Extract the submatrix of the influence of the dominant parameter on the control region of the subordinate parameter from the fluid-thermal bidirectional coupling coefficient matrix, and calculate the state change of the subordinate region caused by the adjustment of the dominant parameter. Extract the submatrix of the influence of subordinate parameters on its control region from the fluid-thermal bidirectional coupling coefficient matrix, and calculate the compensation adjustment amount of subordinate parameters so that the adjustment effect of subordinate parameters can offset the state change of subordinate region caused by the adjustment of dominant parameters.

8. The method according to claim 7, characterized in that, For sequential shrinkage, the target state is determined as follows: the temperature at the shrinkage location when the channel is closed is the closing temperature, the target temperature is the sum of the liquidus temperature and the temperature margin, the temperature margin is 10% to 30% of the difference between the liquidus temperature and the solidus temperature, and the temperature field state deviation is the difference between the target temperature and the closing temperature. For driven shrinkage cavities, the target state is determined as follows: the shrinkage suction is equal to the product of the volume shrinkage rate of the molten metal and the volume and cooling rate of the local solidification region at the shrinkage cavity location, divided by the product of the effective cross-sectional area of ​​the feeding channel and the flow velocity of the molten metal; the pressure margin is equal to 32 times the product of the dynamic viscosity of the molten metal, the length of the feeding path, and the flow velocity of the molten metal in the feeding channel, divided by the square of the equivalent diameter of the feeding channel; the target state is that the feeding pressure at the shrinkage cavity location is greater than or equal to the sum of the shrinkage suction and the pressure margin.

9. The method according to any one of claims 1 to 8, characterized in that, Also includes: The thermal coupling parameter adjustment scheme is constrained to determine whether the adjustment amount of each parameter exceeds the equipment capacity range and process safety constraint range. For parameter adjustment amounts that exceed the range, the out-of-bounds parameter values ​​are replaced with the closest boundary values ​​using a projection algorithm, and the compensation amount of the associated parameters is recalculated to generate an executable parameter adjustment instruction. Apply the parameter adjustment instructions to the simulation environment, re-execute the solidification simulation calculation, and verify whether the shrinkage cavities have been eliminated; If the pinhole is eliminated, the input conditions and adjustment results of this adjustment are updated to the mechanism-coupling path mapping table as experience records; if the pinhole is not eliminated, the pinhole status is re-evaluated and a new round of parameter adjustment is carried out until the pinhole is eliminated or the maximum number of adjustment rounds is reached.

10. An industrial simulation parameter adaptive adjustment system based on an AI agent, used to execute the method according to any one of claims 1 to 9, characterized in that, include: The solidification front evolution module is used to acquire time-series temperature field data of the solidification process, and uses the isothermal surface tracking algorithm to analyze the changes in the spatial position of the liquidus and solidus over time to generate a solidification front evolution sequence. The feeding path analysis module is used to evaluate the feeding channel status between each liquid phase region and the feeding source at the gate based on the solidification front evolution sequence and the connectivity analysis algorithm, record the time and spatial coordinates of channel closure at each location, and generate a feeding path evolution log. The fluid-thermal coupling analysis module is used to acquire flow field data and temperature field data during the filling and solidification processes. It uses the inter-field coupling analysis algorithm to calculate the influence coefficient of flow velocity on temperature distribution and the influence coefficient of temperature on the viscosity of molten metal, and generates a fluid-thermal bidirectional coupling coefficient matrix. The shrinkage cavity mechanism classification module is used to obtain the solidification completion time of the shrinkage cavity location when a shrinkage cavity defect is detected, extract the closing time of the feeding channel at that location from the feeding path evolution log, use a time sequence comparison algorithm to determine the chronological relationship between the solidification completion time and the channel closing time, and generate a shrinkage cavity mechanism classification result. The coupling path decision module is used to query the mechanism-coupling path mapping table based on the classification results of the hole shrinkage mechanism to obtain the corresponding coupling path selection decision and the set of dominant parameters and subordinate parameters. The response predictor building module is used to extract the influence coefficient submatrix of the dominant parameter on the subordinate parameter from the fluid-thermal bidirectional coupling coefficient matrix, and generate a path-constrained coupled parameter response predictor. The parameter adjustment scheme generation module is used to obtain the current state deviation of the shrinkage cavity position, use the reverse solving algorithm to calculate the adjustment amount of the dominant parameter and the compensation adjustment amount of the subordinate parameter required to eliminate the shrinkage cavity, and generate a mechanism-oriented flow-heat coupling parameter adjustment scheme.

Images