Sand mold molding process parameter intelligent optimization method and system

Abstract

The present application relates to the field of parameter optimization, and particularly relates to a sand mold molding process parameter intelligent optimization method and system, the method comprising: collecting cavity geometry data and process parameters, and obtaining cross-section layer data along the sand shooting direction; constructing a filling resistance sequence according to the ratio of cross-section depth and effective filling width; constructing a resistance propagation correlation degree sequence in combination with the correlation coefficient of adjacent layer compactness average and the ratio of filling resistance index; dividing continuous level groups according to the resistance propagation correlation degree sequence and dividing particles into subgroups, establishing a hierarchical search topology, and sequentially activating downstream connections in the hierarchical search topology according to the convergence stability of the subgroups; executing a multi-objective particle swarm search under the constraint of the hierarchical search topology, collecting global Pareto frontier solution sets from each subgroup Pareto archive, and selecting the optimal process parameter combination to be sent to the molding machine PLC. The present application makes the information propagation sequence consistent with the physical time sequence of layer-by-layer filling of the sand flow, and improves the compactness of the high aspect ratio area.

Description

Technical Field

[0001] This invention relates to the field of parameter optimization, and in particular to a method and system for intelligent optimization of sand molding process parameters. Background Technology

[0002] Sand casting is one of the main forming processes for the mass production of mechanical parts. The molding process controls process parameters such as sand injection pressure, compaction time, and vibration frequency to ensure the molding sand achieves the specified compactness and permeability within the mold cavity, thereby guaranteeing the dimensional accuracy and internal quality of the castings. For compressor castings such as scroll plates and moving plates, the cavity contains spiral groove structures with a depth-to-width ratio of 4:1 or higher, resulting in drastic changes in wall thickness. Different regions respond significantly differently to process parameters, and improper parameter configuration will directly lead to insufficient compaction in certain areas, causing defects such as sand porosity and shrinkage, resulting in batch scrap. Therefore, how to accurately optimize process parameters for castings with complex cavities is a core requirement for casting companies to improve product qualification rates. Existing research has introduced multi-objective particle swarm optimization algorithms into the field of sand molding process parameter optimization, simultaneously optimizing multiple competing process objectives such as compaction and permeability by maintaining a Pareto archive.

[0003] Chinese patent document CN109558646B discloses a method for optimizing the forming sequence of multi-axis additive manufacturing, comprising: obtaining feature points of a triangular mesh model based on interactive selection; generating initial segmentation lines using a tracking projection algorithm; initially segmenting the model into multiple sub-blocks using a region growing algorithm based on the initial segmentation lines; initially sorting each sub-block according to the segmentation boundary information; using minimum support and step effect as optimization objective functions, using a multi-objective optimization algorithm to obtain the optimal forming direction of the sub-blocks; establishing an optimization segmentation plane through the initial segmentation lines and the optimal forming direction of the sub-blocks to smooth the sub-block segmentation boundaries; using optimization constraints as optimization objective functions, and under the premise of satisfying physical constraints, using a multi-objective optimization algorithm to optimize the multi-axis additive manufacturing forming sequence of the segmented sub-blocks. This method uses a multi-objective particle swarm optimization algorithm to solve for the optimal forming direction of the sub-blocks, employs a non-dominated solution sorting method and crowding distance calculation, and incorporates an elite strategy.

[0004] However, existing multi-objective particle swarm optimization algorithms employ fully connected or fixed-neighborhood search topologies, where all particles exchange information indiscriminately within the same topology. In the vortex cavity molding scenario, the sand-filling process exhibits significant layering: the shallow opening region fills first and reaches the target compaction level first, while the high aspect ratio spiral groove region, due to cross-sectional shrinkage and sand flow obstruction, experiences a delayed filling process, and its compaction formation mechanism differs fundamentally from that of the shallow region. The fixed search topology in existing technologies allows the parameter search for the corresponding deep region to receive global guidance information before the shallow region converges. This guidance information reflects the parameter preferences of the shallow region, which is inconsistent with the actual needs of the deep region. This leads to disordered search paths in the parameter space of the high aspect ratio region, and the final output parameter combination cannot simultaneously meet the compaction requirements of the local high aspect ratio region, resulting in continuous defects in the casting in that region. Summary of the Invention

[0005] To address the technical problem that existing multi-objective particle swarm optimization algorithms employ a fixed search topology, resulting in deep regions receiving inconsistent global guidance information before shallow regions have converged, thus failing to meet the compaction requirements of high aspect ratio regions, this invention provides an intelligent optimization method and system for sand mold molding process parameters.

[0006] In a first aspect, the present invention provides an intelligent optimization method for sand mold molding process parameters, employing the following technical solution: S1. Collect and preprocess cavity geometry data, historical process parameters, and corresponding compaction data to obtain cross-sectional layer data along the sand-filling direction; S2. Construct a filling retardation sequence based on the ratio of the depth of each cross-sectional layer from the cavity opening to the effective filling width. The filling retardation sequence includes multiple filling retardation indices; S3. Construct a retardation propagation correlation degree sequence based on the correlation coefficient of compaction of adjacent cross-sectional layers and the ratio of filling retardation indices; S4. Divide the particles into continuous hierarchical groups according to the retardation propagation correlation degree sequence, divide the particles into corresponding subgroups, and establish a hierarchical search with intra-group connections and initial inter-group disconnections. S5. Under the constraints of the hierarchical search topology, perform multi-objective particle swarm optimization, guide the downstream subgroups with the Pareto front solutions of the upstream subgroups according to the corresponding hindrance propagation correlation, construct Pareto archives for each subgroup, summarize the Pareto archives of each subgroup to obtain the global Pareto front solution set, select the modeling process parameter combination from the global Pareto front solution set, and send the modeling process parameter combination to the programmable logic controller of the modeling machine.

[0007] By combining cavity geometrical hindrance characteristics, the compactness correlation between adjacent cross-sectional layers, and particle swarm search topology, the search process can identify the differences in filling difficulty and parameter response between shallow open regions and deep high aspect ratio regions. Reliable optimization information is then progressively transmitted according to the actual filling sequence of sand flow from shallow to deep, thereby reducing the erroneous guidance of deep subgroups by parameter preferences generated before the shallow subgroups have converged, and mitigating the risk of disordered deep parameter search paths and premature convergence. Each subgroup maintains a Pareto archive and controls the cross-layer guidance intensity with hindrance propagation correlation, which also balances the competing objectives of minimum compactness of each cross-sectional layer and overall permeability. This method ensures that the compactness of high aspect ratio regions remains above the quality lower limit while avoiding a significant decrease in overall permeability. Therefore, it helps reduce sand porosity, shrinkage porosity, and batch scrap in complex castings such as vortex disks, improving the regional adaptability and production availability of process parameter optimization results.

[0008] Preferably, the process of collecting and preprocessing cavity geometric data, historical process parameters and corresponding compaction data includes: scanning the casting pattern to obtain three-dimensional point cloud data and performing noise reduction and surface reconstruction; reading the sand injection pressure, compaction time, vibration frequency and vibration duration; and screening out abnormal records; dividing the reconstructed closed three-dimensional model into equal intervals along the sand injection filling direction; and grouping the compaction measurement values ​​according to their respective cross-sectional layers, and outputting the cross-sectional layer data.

[0009] Preferably, constructing the filling retardation sequence includes: extracting the minimum channel width that sand can pass through in each cross-section layer from the cross-section layer data as the effective filling width, and extracting the distance of each cross-section layer from the cavity opening end as the cross-section depth; arranging the ratio of the cross-section depth to the corresponding effective filling width in order from near to far from the cavity opening end to obtain the filling retardation sequence.

[0010] The ratio of cross-sectional depth to minimum channel width is used to uniformly characterize the combined effects of sand flow distance and channel contraction on the filling process. It can quantitatively reflect the layer-by-layer changes in the degree of obstruction along the sand-shooting direction and identify the nonlinear obstruction jump positions that occur when the depth-to-width ratio jumps from 2:1 to 4:1. It can provide a basis corresponding to the physical structure of the cavity for the hierarchical division of the compaction step deterioration region and the layered search topology, reducing the subjectivity caused by relying on manual experience to divide the region.

[0011] Preferably, constructing the propagation retardation correlation sequence includes: determining the correlation coefficient of the average compaction of adjacent cross-sectional layers based on compaction data, combining the correlation coefficient with the ratio of the filling retardation index of the upstream and downstream cross-sectional layers in adjacent cross-sectional layers to obtain the propagation retardation correlation that characterizes the effective degree of transmission of the upstream filling state to the downstream, and arranging them in the order of cross-sectional layers to obtain the propagation retardation correlation sequence.

[0012] Simultaneously considering the actual mass correlation and filling retardation increase of adjacent cross-sectional layers, it is possible to distinguish between effective state transfer in the smooth transition region and transfer attenuation in the region of rapid cross-sectional contraction. When the downstream retardation index increases, the corresponding retardation propagation correlation is actively suppressed, thereby more accurately identifying the critical positions where shallow parameter information cannot effectively act on deep layers. This provides a quantitative basis for subsequent cutting off, weakening, or weighted recovery of information propagation between subgroups, reducing the possibility of mismatched guiding information entering the deep search space.

[0013] Preferably, dividing the continuous hierarchical groups includes: setting a threshold for the obstruction propagation correlation degree based on the mean of the obstruction propagation correlation degree sequence, marking positions below the obstruction propagation correlation degree threshold as boundary points, and dividing all cross-sectional layers into several continuous hierarchical groups based on the boundary points, so that the obstruction propagation correlation degree of adjacent cross-sectional layers in each continuous hierarchical group is not lower than the obstruction propagation correlation degree threshold, and outputting the continuous hierarchical groups.

[0014] Based on the statistical level of the correlation degree of the obstruction propagation, the low correlation boundary point is identified so that the cross-sectional layers within the same continuous hierarchical group maintain a strong filling state transmission relationship, while the cross-sectional layers with significantly weakened transmission effect are classified into different hierarchical groups.

[0015] Preferably, establishing a hierarchical search topology with intra-group connections and initial inter-group disconnections includes: dividing the particles in the multi-target particle swarm into corresponding subgroups according to continuous hierarchical groups, establishing a fully connected hierarchical search topology structure within each subgroup, and ensuring that the hierarchical search topology connections between adjacent subgroups are disconnected during initialization, thereby obtaining a hierarchical search topology corresponding to the continuous hierarchical groups.

[0016] By maintaining sufficient information exchange within highly correlated regions and isolating adjacent hierarchical groups with significant differences in filling mechanisms during the initialization phase, each subgroup can prioritize searching parameter regions suitable for its own level's compactness requirements. This approach maintains the cooperative search efficiency of particles within a group while preventing deep subgroups from being interfered with by shallow parameter preferences when upstream results are still unstable. It also provides a dynamically adjustable hierarchical search topology for subsequently opening inter-layer connections according to convergence status.

[0017] Preferably, activating downstream connections in the hierarchical search topology includes: determining convergence stability based on the change in the standard deviation of particle fitness values ​​in the current iteration of the subgroup relative to the standard deviation of the subgroup in the initial iteration; when the convergence stability of the upstream subgroup meets a preset stability condition, or when the change in its convergence stability in multiple consecutive iterations meets a preset stagnation condition, activating the hierarchical search topology connection between the upstream subgroup and the adjacent downstream subgroup, and outputting a dynamically updated hierarchical search topology.

[0018] The change in fitness standard deviation relative to the initial state is used to determine whether the subgroup has formed a stable and reliable search result, and a stagnation trigger condition is set, so that the downstream hierarchical search topology connection can be adaptively activated according to the actual search process.

[0019] Preferably, constructing the Pareto archive for each subgroup includes: generating particles for each subgroup, such that the position vector of each particle corresponds to a parameter combination consisting of sand injection pressure, compaction time, vibration frequency, and vibration duration; obtaining the predicted compaction value of each cross-sectional layer and the predicted overall permeability of the molding sand corresponding to each parameter combination; using the minimum value among the predicted compaction values ​​of each cross-sectional layer and the predicted overall permeability of the molding sand as the maximization target, calculating the fitness of each particle; filtering non-dominated particles according to the Pareto dominance relationship between particles, and writing the non-dominated particles into the Pareto archive of the corresponding subgroup to obtain the Pareto archive for each subgroup.

[0020] Preferably, the process of aggregating the Pareto archives of each subgroup to obtain the global Pareto front solution set includes: when the downstream subgroup does not receive guidance information from the upstream subgroup, updating the corresponding particles based on their individual historical best positions and the guiding particles in the corresponding subgroup's Pareto archive, and updating the Pareto archive of the corresponding subgroup based on the Pareto dominance relationship of the updated particles; after the downstream hierarchical search topological connection is activated, adding the Pareto front solution of the upstream subgroup to the candidate set of guiding particles of the adjacent downstream subgroup, using the obstruction propagation correlation degree at the corresponding boundary position as the upstream guidance weight, and updating the particles and Pareto archives of the downstream subgroup based on the candidate set of guiding particles; after satisfying the preset termination condition, aggregating the updated Pareto archives of each subgroup to obtain the global Pareto front solution set.

[0021] Secondly, this invention provides an intelligent optimization system for sand mold molding process parameters, employing the following technical solution: The intelligent optimization system for sand molding process parameters includes a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement the intelligent optimization method for sand molding process parameters as described above.

[0022] The above-mentioned intelligent optimization method for sand molding process parameters is used to generate a computer program and store it in a memory so that it can be loaded and executed by a processor. Thus, a system can be made based on the memory and processor for easy use.

[0023] The present invention has the following technical effects: By constructing a filling resistance index, the cavity is divided into a hierarchical sequence with differentiated resistance levels along the filling direction. This allows the optimization process to perceive the differences in filling difficulty in each region rather than treating the cavity as a homogeneous whole, providing a quantitative basis for the hierarchical division of the hierarchical search topology based on the physical structure of the cavity. By constructing a resistance propagation correlation, the transmission intensity of the filling state between adjacent levels is quantitatively described, identifying the critical position where the transmission path is truncated. This ensures that the determination of the hierarchical boundaries of the hierarchical search topology is supported by physical facts rather than relying on artificial presets. Through a dynamic hierarchical search topology activation mechanism based on convergence stability, the information propagation order of each subgroup is kept consistent with the physical time sequence of the sand flow gradually filling from shallow to deep layers. This avoids deep subgroups receiving guidance information that deviates from actual needs before the shallow layer parameters are determined, ensuring that the deep high aspect ratio region obtains a targeted parameter search space. The improved multi-objective particle swarm optimization algorithm can output a combination of process parameters that simultaneously meet the requirements of global permeability and local high aspect ratio compactness. It solves the problems of disordered search paths and failure of the final parameter combination to take into account the compactness requirements of local areas under the existing fixed search topology structure, and reduces the defect rate of vortex cavity castings in high aspect ratio areas. Attached Figure Description

[0024] Figure 1 This is a flowchart of a method according to an embodiment of the present invention.

[0025] Figure 2 This is a schematic diagram illustrating the actual operation effect of the hierarchical search topology activation mechanism of this invention.

[0026] Figure 3 This is a schematic diagram comparing the compactness of each cross-sectional layer corresponding to the output parameters of the hierarchical search topology scheme and the fixed search topology scheme. Detailed Implementation

[0027] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] This invention discloses an intelligent optimization method for sand molding process parameters, referring to... Figure 1 This includes the following steps: S1: Data Acquisition and Preprocessing.

[0029] A coordinate measuring machine (CMM) was used to scan the entire surface of the casting pattern, acquiring 3D point cloud data of the cavity. The point cloud sampling interval was set to 0.5 mm to ensure the integrity of geometric information in detailed areas such as spiral grooves. Simultaneously, historical process parameter records were read from the molding machine's PLC control system, including four core parameters: sand injection pressure, compaction time, vibration frequency, and vibration duration. A complete set of parameter values ​​was recorded for each mold run. The casting quality inspection results for the corresponding mold run were extracted from the production management system, including the compaction measurement values ​​at each inspection location and the final acceptance criteria for the casting. The data acquisition spanned over 300 mold runs from the most recent consecutive production runs, ensuring sufficient data for subsequent statistical analysis.

[0030] Noise reduction was performed on the 3D point cloud data, removing outliers that deviated from the fitted surface by more than three times the sampling interval. A closed 3D model of the cavity was then generated through surface reconstruction. A coordinate axis was established along the main direction of sand filling in the cavity. The cavity was divided into several cross-sectional layers along this axis at equal intervals, with each layer having a thickness of 2 mm, obtaining the contour geometric data of each cross-sectional layer. Outlier filtering was performed on historical process parameter data, removing records of parameter mutations caused by equipment failure or human intervention (i.e., records deviating from the parameter mean by more than three times the standard deviation), retaining valid data for subsequent analysis. The compaction measurements at each detection location were grouped according to their respective cross-sectional layers, establishing a statistical distribution of compaction for each cross-sectional layer.

[0031] S2: Construct a filling retardation sequence.

[0032] During sand-filling in a vortex cavity, the sand flow enters from the cavity opening and propagates deeper along the spiral groove. As depth increases, the cross-section gradually narrows. The sand flow front is subjected to the combined effects of sidewall friction and back pressure from the already deposited sand, resulting in a decrease in flow velocity and a lower packing density of sand particles in this region compared to the opening. This phenomenon is not a uniform decrease, but rather a step-like deterioration occurs at locations where the cross-sectional aspect ratio changes abruptly: when the aspect ratio jumps from 2:1 to 4:1, the effective filling cross-section of the sand flow shrinks sharply, and the degree of obstruction increases non-linearly. This step-like obstruction directly determines the upper limit of the final compaction of each layer. Even if the same pressure is applied in subsequent compaction processes, the increase in compaction in the deep, high aspect ratio region is lower than that in the shallow region due to insufficient initial packing density. Existing multi-objective particle swarm optimization methods place the parameter search of all regions within the same fixed topology, failing to distinguish the differences in filling difficulty between layers. A quantitative index that can characterize the degree of sand flow obstruction layer by layer is needed to provide a basis for the hierarchical division of the subsequent layered search topology.

[0033] From the geometric data of each cross-sectional layer obtained in step S1, extract the effective filling width and cross-sectional depth of each cross-sectional layer. The effective filling width is the minimum channel width that allows sand flow through the cross-sectional layer, and the cross-sectional depth is the distance of the cross-sectional layer from the cavity opening. Number each cross-sectional layer in order of distance from the opening from near to far, and denote it as the nth. The effective filling width of each cross-sectional layer is The cross-sectional depth is .

[0034] Definition of the first The filling resistance index for each cross-sectional layer is:

[0035] in, For the first The filling resistance index of each cross-sectional layer For the first The distance of each cross-sectional layer from the opening end of the cavity. For the first The effective filling width of each cross-sectional layer. The filling resistance indexes of each cross-sectional layer are arranged according to the layer number, forming a filling resistance sequence. ,in This represents the total number of cross-sectional layers.

[0036] Filling resistance index The physical meaning of is the first The ratio of the depth the sand flow needs to traverse to the width of the passage at each cross-sectional layer. A larger value indicates greater resistance to the sand flow upon reaching that layer, lower initial packing density, and greater difficulty in subsequent compaction. The filling resistance sequence starts from... arrive The increasing trend reflects the physical fact that the degree of sand flow obstruction gradually intensifies from the opening end to the deep end. At the location where the cross-sectional depth-to-width ratio changes abruptly, the filling obstruction index of adjacent layers jumps. This jump point is the critical layer where compaction deteriorates sharply, providing a quantitative boundary for subsequent construction of the inter-layer propagation obstruction correlation and the division of particle subgroups.

[0037] S3: Construct the correlation degree of the obstruction of propagation.

[0038] As sand flows layer by layer into the cavity, the filling state of the shallow layer has a transferable effect on the filling effect of the adjacent deep layer: the bulk density of the shallow sand determines the inlet back pressure condition of the deep sand flow. When the shallow bulk density is high, the initial kinetic energy of the sand flow entering the deep layer is greater, and the compactness of the deep layer increases accordingly; conversely, when the shallow bulk density is loose, the kinetic energy of the sand flow entering the deep layer is insufficient, and the compactness of the deep layer further deteriorates. However, this interlayer transfer effect is not of equal intensity: in regions with gentle cross-sectional transitions, the transfer effect between adjacent layers is stronger, and adjustments to shallow layer parameters can effectively improve the deep layer; while at locations with abrupt cross-sectional contraction, the sudden narrowing of the channel cuts off the transfer path from the shallow layer to the deep layer, and the compactness of the deep region is almost independent of the parameter configuration of the shallow layer. This difference in transfer intensity determines the weight with which the subgroups of the particle swarm should exchange information: subgroups corresponding to adjacent layers with strong transfer effects should maintain a tighter information connection, while layers with weak transfer effects should cut off or weaken information transmission. The fixed search topology of existing multi-target particle swarm optimization cannot express the difference in the transmission intensity between levels. It is necessary to construct an index that can quantitatively describe the degree of attenuation of the transmission effect between adjacent levels.

[0039] The filling retardation sequence obtained from step S2 In the process, the filling resistance index values ​​of adjacent levels are extracted. and Simultaneously, from the statistical distribution data of the compactness of each cross-sectional layer in step S1, the first... The first cross-sectional layer and the first Correlation coefficient of the average compactness of each cross-sectional layer This characterizes the statistical correlation of compactness between adjacent layers in actual production.

[0040] Definition of the first The first cross-sectional layer and the first The degree of propagation hindrance between the cross-sectional layers is:

[0041] in, For the first The first cross-sectional layer and the first The degree of propagation hindrance between cross-sectional layers For the first The first cross-sectional layer and the first The correlation coefficient of the average compactness of each cross-sectional layer For the first The filling resistance index of each cross-sectional layer For the first The filling hindrance index of each cross-sectional layer. The hindrance propagation correlation between adjacent layers is arranged in hierarchical order to form a hindrance propagation correlation sequence. .

[0042] Blocking the propagation correlation The physical meaning of is: the first The change in the filling state of the first cross-sectional layer can be transmitted to the second. The effectiveness of each cross-sectional layer. When the degree of resistance in deeper layers is much greater than that in shallower layers. Much larger ,ratio Approaching zero, hindering the propagation correlation A compressed state indicates that the transmission path is truncated at that location due to a sharp contraction of the cross-section, and the compactness of the deeper regions is independent of the parameter configuration of the shallower layers. When the degree of resistance in adjacent layers is similar... and quite, Approaching 1, the degree of obstruction of propagation correlation The correlation coefficient remains at the mean of compactness. This indicates that shallow states can be effectively transmitted to deeper states. The position where a low value appears in the propagation correlation sequence is the hierarchical boundary of the hierarchical search topology: information propagation between particle subgroups on both sides of this position should be weakened or cut off, so that each can independently search for parameter regions that are suitable for the compactness requirements of its own level.

[0043] S4: Dynamically reconstruct the particle search topology.

[0044] In the standard implementation of multi-objective particle swarm optimization (PSO) algorithms, all particles share the same fully connected or fixed annular neighborhood search topology. When the optimization object is the vortex cavity shaping parameters, the response mechanisms of each cross-sectional level to the process parameters are fundamentally different: the shallow opening region is mainly controlled by sand injection pressure and vibration frequency, while the compaction of the deep high aspect ratio region is also constrained by the sand flow hindrance effect. The same set of parameters may produce opposite effects at different levels. If the deep subgroup obtains guidance information from the global archive before the shallow subgroup has converged, the parameter adjustment direction pointed to by this guidance information is determined based on the shallow response, which will mislead the deep subgroup from its actual optimal region. The physical process of sand flow filling is completed sequentially from shallow to deep. The information propagation order of the hierarchical search topology should correspond to this physical time sequence: after the shallow subgroup searches and converges first, its results are used as the boundary condition input for the deep subgroup. It is necessary to construct a state index that can determine whether each subgroup meets the downstream propagation conditions, as the triggering basis for the dynamic activation of the hierarchical search topology connection.

[0045] Based on the propagation hindrance correlation sequence obtained in step S3 Determine the hierarchical boundaries of the hierarchical search topology. Set a threshold for hindering propagation correlation. To block the propagation correlation sequence with a degree lower than The location is marked as the dividing point, and all Each cross-sectional layer is divided into several consecutive hierarchical groups. The cross-sectional layer retardation propagation correlation degree within each hierarchical group is... The above describes a hierarchical search topology where particles within a group maintain full connectivity; the hierarchical search topology connections between adjacent hierarchical groups are initially disconnected and will only be activated once the upstream hierarchical group meets the activation conditions. (Threshold for hindering propagation correlation) The value is taken as 0.5 times the mean of the delayed propagation correlation sequence. The basis for this value is that the delayed propagation correlation is lower than half of the mean at the location where the cross section contracts sharply, and the physical transmission effect at this point is no longer sufficient to support effective information sharing.

[0046] The particles in a multi-objective particle swarm are divided into corresponding subgroups according to their hierarchical groups, denoted as the first subgroup. The subgroup is Definition of the first The convergence stability of the subgroup is:

[0047] in, For the first The convergence stability of a subgroup For the first The standard deviation of the fitness values ​​of all particles in the current iteration of the subgroup For the first The standard deviation of the fitness values ​​of all particles at the initial iteration of each subgroup.

[0048] When the Convergence stability of subgroups When the value reaches 0.8 or higher, the subgroup is determined to have entered a stable convergence state, and its connection with the downstream subgroup is activated. Hierarchical search topological connections between subgroups. After activation, the first... The Pareto front solutions of each subgroup are used as guiding information to be passed to the first... Subgroup, number The particles in each subgroup will update their velocity by the first... The leading edge solutions of each subgroup are included in the candidate set of guiding particles. If the first... If, in 20 consecutive iterations of a subgroup, the change in convergence stability between adjacent iterations is less than 0.01, then the subgroup is considered to have stalled in the search, and its connection with the downstream subgroup is forcibly activated. The hierarchical search topology connections between subgroups. The 20-round window is based on the following: in particle swarm optimization, the commonly used window length for determining convergence stagnation is 2 to 5 times the number of parameter dimensions. In this scheme, the process parameter dimensions are 4, so 5 times is used, resulting in 20 rounds. A change of 0.01 corresponds to the relative change rate of the fitness standard deviation falling to within 1%, indicating that the subgroup no longer produces new search progress within that window.

[0049] Convergence stability The physical meaning of is: the first The degree of dispersion of particles within a subgroup in the parameter space is the convergence ratio relative to the initial state. The closer this value is to 1, the more concentrated the particles within the subgroup are within a narrow parameter region, and the process parameters corresponding to this hierarchical group tend to be determined. An activation threshold of 0.8 corresponds to a fitness standard deviation decreasing to 20% of the initial value, indicating that 80% of the search uncertainty within this subgroup has been eliminated, and its output Pareto front solution has sufficient reliability to serve as the boundary condition for downstream subgroups. The dynamic activation mechanism of hierarchical search topology connections ensures that the information propagation order is consistent with the physical time sequence of the sand flow gradually filling from shallow to deep layers: shallow layer parameters are determined first, then reliable starting conditions are provided for deep layer searches, avoiding the deep subgroup receiving invalid guiding information when the shallow layer is still uncertain. When a subgroup cannot reach the convergence threshold for a long time due to the expansion of the multi-objective front, the stagnation criterion ensures that downstream subgroups can still obtain upstream input within a finite number of iterations, preventing the algorithm from degenerating into an unguided search at the downstream level.

[0050] like Figure 2 As shown, the actual operating effect of the hierarchical search topology activation mechanism of the present invention is demonstrated: the shallow subgroups reach the convergence stability activation threshold first, and the convergence speed of the deep subgroups is significantly accelerated after the hierarchical search topology connections are opened in sequence. The timing of each subgroup reaching stable convergence is consistent with the physical order of cavity sand flow filling.

[0051] S5: Perform a multi-objective particle swarm search under hierarchical search topology constraints and output the optimal parameter combination.

[0052] Multi-objective particle swarm optimization (PSO) iterative search is performed under hierarchical search topological constraints. During the initialization phase, particles are randomly generated for each subgroup. The position vector of each particle corresponds to a set of molding process parameters, namely, sand-shooting pressure, compaction time, vibration frequency, and vibration duration. The velocity vector is randomly initialized. An independent Pareto archive is established for each subgroup to store non-dominated solutions discovered during the subgroup's search process.

[0053] The fitness of each particle is evaluated by two objective functions: the first objective function is the minimum predicted compaction value of each cross-sectional layer under the given parameter combination, and the second objective function is the predicted overall permeability of the molding sand under the given parameter combination. Both objective functions are maximized, and the quality of particles is determined by Pareto dominance, without weighted merging. For any two particles... and , if the particle The value of the first objective function is greater than or equal to that of the particle. Furthermore, the value of the second objective function is also greater than or equal to that of the particle. At the same time, its value on at least one objective function is strictly greater than that of the particle. Then it is called a particle. Dominant Particle . If particles With particles If two particles are mutually non-dominant, meaning neither particle is superior to the other in both objective functions, then both are considered non-dominant particles. All non-dominant particles discovered by each subgroup in each iteration constitute the Pareto non-dominant solution set for that subgroup and are stored in a separate Pareto archive, serving as the foundation for subsequent candidate sets of guiding particles. All non-dominant particles are not weighted and merged; their quality is determined solely by the aforementioned Pareto dominance relationship.

[0054] After the iteration begins, each subgroup independently updates its particle position and velocity. For subgroups that have not yet received upstream guidance information, the particle velocity update only references the historical best position of individuals within the subgroup and the guiding particles in the subgroup's Pareto archive. After each iteration, the convergence stability of each subgroup is calculated to determine whether the activation condition is met. When the upstream subgroup meets the activation condition, its Pareto front solution is passed to the candidate set of guiding particles for the downstream subgroup. In subsequent iterations, the particles in the downstream subgroup simultaneously refer to both the subgroup's archive and the upstream front solution for velocity updates. The guidance weight of the upstream front solution for the downstream particle velocity update is directly taken as the hindrance propagation correlation degree at the corresponding boundary position. The value, that is, in the social learning term of the particle velocity update formula, is... It is multiplied into the upstream guiding component as a weighting coefficient.

[0055] The iteration continues until the termination condition is met: the convergence stability of all subgroups reaches 0.8 or higher, or the total number of iterations reaches a preset upper limit. After the iteration terminates, the Pareto archives of all subgroups are summarized and merged into a global Pareto front solution set.

[0056] Each solution in the global Pareto front solution set corresponds to a complete combination of molding process parameters. Each parameter combination has undergone a search process under hierarchical search topological constraints, taking into account the compaction requirements of both shallow opening regions and deep high aspect ratio regions. When selecting the final solution from the global Pareto front solution set, the minimum compaction value of each cross-sectional layer is used as the screening criterion: the parameter combination with the highest minimum compaction value across all cross-sectional layers is selected first, ensuring that the compaction of high aspect ratio regions is not lower than the lower limit specified in the casting quality standard. If multiple parameter combinations meet the lower limit requirement for compaction, the one with the highest permeability is selected as the final output solution.

[0057] The final output combination of process parameters is directly sent to the molding machine's PLC control system to guide subsequent batches of production. The expected compactness distribution of each cross-sectional layer corresponding to this parameter combination is simultaneously output for verification by process engineers. During production, the compactness of the castings is randomly checked every fixed number of mold runs. The test results are compared with the expected distribution. When the deviation exceeds the allowable range, a new round of optimization iteration is triggered, updating the process parameters.

[0058] like Figure 3 As shown, the comparison of the compactness of each cross-sectional layer corresponding to the output parameters of the layered search topology scheme and the fixed search topology scheme of the present invention is demonstrated: the compactness of the layered search topology scheme in the high aspect ratio region is significantly higher than that of the fixed search topology scheme, the compactness of all cross-sectional layers remains above the lower limit standard, and the global air permeability does not show a significant decrease.

[0059] This invention also discloses an intelligent optimization system for sand molding process parameters, including a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, the intelligent optimization method for sand molding process parameters according to this invention is implemented.

[0060] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for intelligent optimization of sand mold molding process parameters, characterized in that, Including the following steps: S1. Collect and preprocess cavity geometry data, historical process parameters, and corresponding compaction data to obtain cross-sectional layer data along the sand-filling direction; S2. Construct a filling retardation sequence based on the ratio of the depth of each cross-sectional layer from the cavity opening to the effective filling width. The filling retardation sequence includes multiple filling retardation indices; S3. Construct a retardation propagation correlation degree sequence based on the correlation coefficient of compaction of adjacent cross-sectional layers and the ratio of filling retardation indices; S4. Divide the particles into continuous hierarchical groups according to the retardation propagation correlation degree sequence, divide the particles into corresponding subgroups, and establish a hierarchical search with intra-group connections and initial inter-group disconnections. S5. Under the constraints of the hierarchical search topology, perform multi-objective particle swarm optimization, guide the downstream subgroups with the Pareto front solutions of the upstream subgroups according to the corresponding hindrance propagation correlation, construct Pareto archives for each subgroup, summarize the Pareto archives of each subgroup to obtain the global Pareto front solution set, select the modeling process parameter combination from the global Pareto front solution set, and send the modeling process parameter combination to the programmable logic controller of the modeling machine.

2. The intelligent optimization method for sand mold molding process parameters according to claim 1, characterized in that, The process of collecting and preprocessing cavity geometry data, historical process parameters and corresponding compaction data includes: scanning the casting pattern to obtain three-dimensional point cloud data and performing noise reduction and surface reconstruction; reading sand injection pressure, compaction time, vibration frequency and vibration duration; and screening out abnormal records; dividing the reconstructed closed three-dimensional model into equal intervals along the sand injection filling direction; and grouping the compaction measurement values ​​according to their respective cross-sectional layers, and outputting the cross-sectional layer data.

3. The intelligent optimization method for sand mold molding process parameters according to claim 1, characterized in that, The construction of the filling retardation sequence includes: extracting the minimum channel width that sand can pass through in each cross-section layer from the cross-section layer data as the effective filling width, and extracting the distance of each cross-section layer from the cavity opening end as the cross-section depth; arranging the ratio of the cross-section depth to the corresponding effective filling width in order from near to far from the cavity opening end to obtain the filling retardation sequence.

4. The intelligent optimization method for sand mold molding process parameters according to claim 1, characterized in that, The construction of the propagation retardation correlation sequence includes: determining the correlation coefficient of the mean compaction of adjacent cross-sectional layers based on compaction data; combining the correlation coefficient with the ratio of the filling retardation index of the upstream and downstream cross-sectional layers in adjacent cross-sectional layers to obtain the propagation retardation correlation that characterizes the effective transmission of the upstream filling state to the downstream; and arranging them in the order of cross-sectional layers to obtain the propagation retardation correlation sequence.

5. The intelligent optimization method for sand mold molding process parameters according to claim 1, characterized in that, The process of dividing the continuous hierarchical groups includes: setting a threshold for the obstruction propagation correlation degree based on the mean of the obstruction propagation correlation degree sequence; marking positions below the obstruction propagation correlation degree threshold as boundary points; and dividing all cross-sectional layers into several continuous hierarchical groups based on the boundary points, so that the obstruction propagation correlation degree of adjacent cross-sectional layers within each continuous hierarchical group is not lower than the obstruction propagation correlation degree threshold, and outputting the continuous hierarchical groups.

6. The intelligent optimization method for sand mold molding process parameters according to claim 1, characterized in that, Establishing a hierarchical search topology with intra-group connectivity and initial inter-group disconnection includes: dividing the particles in the multi-target particle swarm into corresponding subgroups according to continuous hierarchical groups, establishing a fully connected hierarchical search topology within each subgroup, and ensuring that the hierarchical search topology connections between adjacent subgroups are disconnected during initialization, thus obtaining a hierarchical search topology corresponding to the continuous hierarchical groups.

7. The intelligent optimization method for sand mold molding process parameters according to claim 1, characterized in that, Activating downstream connections in the hierarchical search topology involves: determining convergence stability based on the change in the standard deviation of particle fitness values ​​in the current iteration of the subgroup relative to the standard deviation of the subgroup in the initial iteration; when the convergence stability of the upstream subgroup meets a preset stability condition, or when the change in its convergence stability in multiple consecutive iterations meets a preset stagnation condition, activating the hierarchical search topology connection between the upstream subgroup and the adjacent downstream subgroup, and outputting a dynamically updated hierarchical search topology.

8. The intelligent optimization method for sand mold molding process parameters according to claim 1, characterized in that, Constructing the Pareto archive for each subgroup includes: generating particles for each subgroup, such that the position vector of each particle corresponds to a parameter combination consisting of sand injection pressure, compaction time, vibration frequency, and vibration duration; obtaining the predicted compaction values ​​of each cross-sectional layer and the predicted overall permeability of the molding sand corresponding to each parameter combination; using the minimum value among the predicted compaction values ​​of each cross-sectional layer and the predicted overall permeability of the molding sand as the maximization target, calculating the fitness of each particle; filtering non-dominated particles according to the Pareto dominance relationship between particles, and writing the non-dominated particles into the Pareto archive of the corresponding subgroup, thus obtaining the Pareto archive for each subgroup.

9. The intelligent optimization method for sand mold molding process parameters according to claim 1, characterized in that, The process of aggregating the Pareto archives of each subgroup to obtain the global Pareto front solution set includes: when a downstream subgroup does not receive guidance information from an upstream subgroup, updating the corresponding particles based on their individual historical best positions and the guiding particles in the corresponding subgroup's Pareto archive, and updating the Pareto archive of the corresponding subgroup based on the Pareto dominance of the updated particles; after the downstream hierarchical search topological connection is activated, adding the Pareto front solution of the upstream subgroup to the candidate set of guiding particles of the adjacent downstream subgroups, using the obstruction propagation correlation degree at the corresponding boundary position as the upstream guidance weight, and updating the particles and Pareto archives of the downstream subgroups based on the candidate set of guiding particles; after satisfying the preset termination condition, aggregating the updated Pareto archives of each subgroup to obtain the global Pareto front solution set.

10. A sand mold molding process parameter intelligent optimization system, characterized in that, include: The processor and memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the intelligent optimization method for sand molding process parameters according to any one of claims 1-9 is implemented.

Images