A heat treatment strengthening process for an automobile aluminum alloy part after die casting

Abstract

The application discloses a heat treatment strengthening process for an automobile aluminum alloy part after die casting, and relates to the technical field of alloy die casting. The process comprises the following steps: obtaining a three-dimensional geometric model of a casting and defect detection data of a historical casting, calculating a three-dimensional wall thickness field of the casting, and extracting an engineering feature vector of the casting; performing feature region division on the three-dimensional geometric model and reconstructing the data structure of the defect detection data; performing double-channel adaptive fusion analysis, obtaining critical defect feature values of multiple feature regions, and determining a global critical temperature of the casting; performing process feasibility discrimination, and determining an upper limit of the holding time of the solid solution treatment; constructing a heat transfer proxy model, taking the global critical temperature and the upper limit of the holding time as constraints, performing Bayesian optimization search through the heat transfer proxy model, and determining heating process parameters; and outputting a solid solution heating curve corresponding to the heating process parameters, and performing heat treatment on the casting. The application effectively improves the adaptability and precision of heat treatment on complex die castings.

Description

Technical Field

[0001] This invention relates to the field of alloy die casting technology, and more specifically to a heat treatment strengthening process for automotive aluminum alloy parts after die casting. Background Technology

[0002] The trend of automotive lightweighting has driven the widespread application of integrated aluminum alloy die castings in automotive structural parts. After the aluminum alloy parts are die-cast, they need to be subjected to T6 heat treatment to enhance their mechanical properties. Traditional methods mainly rely on four types of heat treatment process design: solution treatment with unified process parameters, empirical stepped heating, mold flow simulation for defect prediction, and finite element heat transfer simulation.

[0003] However, the poor adaptability of unified process parameters makes complex structural parts prone to problems such as bubbling and scrapping or insufficient solution treatment; empirical stepped heating lacks precise quantitative basis, making it difficult to ensure heating temperature uniformity; defect data from mold flow simulation cannot be converted into heat treatment temperature constraints; finite element heat transfer simulation has high computational costs, lacks rapid prediction models, and has low process optimization efficiency. These factors result in existing technologies having insufficient adaptability and accuracy for the heat treatment of complex die-cast parts. Summary of the Invention

[0004] This invention provides a heat treatment strengthening process for die-cast automotive aluminum alloy parts, aiming to solve the technical problems of insufficient adaptability and accuracy of heat treatment for complex die-cast parts in the prior art.

[0005] In view of the above problems, the present invention provides a heat treatment strengthening process for automotive aluminum alloy parts after die casting, comprising: The three-dimensional geometric model of the casting and the defect detection data of historical castings are obtained. Based on the three-dimensional geometric model, the three-dimensional wall thickness field of the casting is calculated, and the engineering feature vector of the casting is extracted. By combining the three-dimensional wall thickness field and the engineering feature vector, the three-dimensional geometric model is divided into feature regions, and the defect detection data is reconstructed based on the feature region division results. Based on the data structure reconstruction results, a dual-channel adaptive fusion analysis is performed to obtain the critical defect feature values ​​of multiple feature regions. Combined with the prior alloy high-temperature mechanical parameters, the global critical temperature of the casting is determined. The feasibility of the process is determined based on the global critical temperature, and the upper limit of the heat preservation time for solution treatment is determined. A heat transfer proxy model is constructed, and the heating process parameters are determined by performing Bayesian optimization search through the heat transfer proxy model, with the global critical temperature and the upper limit of the holding time as constraints. Output the solution heating curve corresponding to the heating process parameters to perform heat treatment on the casting.

[0006] One or more technical solutions provided in this invention have at least the following technical effects or advantages: This invention provides a heat treatment strengthening process for die-cast automotive aluminum alloy parts. By sequentially completing three-dimensional geometric model and defect data processing, three-dimensional wall thickness field calculation, and engineering feature extraction, a precise data and feature foundation is laid for subsequent process optimization. Combining feature region division and defect data structure reconstruction, precise matching and adaptation of defect information to the casting structure is achieved. A dual-channel adaptive fusion analysis determines the global critical temperature, defining a safety boundary for the heat treatment process. Based on the critical temperature, process feasibility is assessed, and the upper limit of solution holding time is locked, avoiding the risks of solution blistering and overheating from the source. Based on a heat transfer proxy model combined with Bayesian optimization search, the optimal heating process parameters adapted to the casting characteristics are quickly obtained. Finally, a dedicated solution heating curve is output for heat treatment, effectively controlling heat treatment blistering defects in die-cast aluminum alloy parts and improving the heating temperature uniformity of complex structural parts, ensuring the full realization of the solution strengthening effect, and achieving personalized adaptation and efficient optimization of the heat treatment process. Attached Figure Description

[0007] Figure 1 A schematic diagram of a heat treatment strengthening process for die-cast automotive aluminum alloy parts provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the solution heating curve corresponding to the heating process parameters in a heat treatment strengthening process for die-cast automotive aluminum alloy parts provided in an embodiment of the present invention. Detailed Implementation

[0008] This invention provides a heat treatment strengthening process for die-cast automotive aluminum alloy parts, which addresses the technical problem of insufficient adaptability and precision in heat treatment of complex die-cast parts in existing technologies.

[0009] Examples, such as Figure 1 As shown, this invention provides a heat treatment strengthening process for die-cast automotive aluminum alloy parts, comprising: S1: Obtain the three-dimensional geometric model of the casting and the defect detection data of historical castings, calculate the three-dimensional wall thickness field of the casting based on the three-dimensional geometric model, and extract the engineering feature vector of the casting.

[0010] In this embodiment of the invention, a three-dimensional geometric model of the casting and defect detection data of historical castings are acquired. Based on the three-dimensional geometric model, the three-dimensional wall thickness field of the casting is calculated, and the engineering feature vector of the casting is extracted. The three-dimensional structure of automotive aluminum alloy die-cast parts is characterized by uneven wall thickness and complex deep cavity structures. After die casting, defects such as porosity and shrinkage cavities are easily generated inside, and the distribution of defects is strongly correlated with the casting wall thickness and structural features. In traditional processes, solution heat treatment processes are formulated solely based on manual experience or single geometric parameters, without quantitatively linking the three-dimensional wall thickness field, defect spatial location, and engineering features. This leads to problems such as blistering, overheating, or insufficient solution treatment. Furthermore, historical defect detection data is not structurally integrated with the three-dimensional geometric features, failing to provide accurate data support for process optimization. Therefore, it is necessary to first complete the feature analysis of the three-dimensional geometric model of the casting, the structured extraction of historical defect data, and the quantitative calculation of the three-dimensional wall thickness field and engineering feature vector, providing a data foundation for subsequent determination and optimization of critical process parameters.

[0011] Step S1 in the process provided in this embodiment of the invention includes: Obtain the three-dimensional geometric model of the casting; By using the historical inspection database of the target scenario and combining it with a preset traceability period, the defect inspection data is extracted. The defect inspection data includes at least the associated spatial location of the defect, dimensional features, heat treatment information of the corresponding casting, and inspection result labels. The three-dimensional geometric model is subjected to adaptive voxelization, and the wall thickness is calculated by traversing multiple voxels based on the voxelization result. The wall thickness calculation is implemented based on either the ray intersection method or the median transformation method. Based on the voxel mesh corresponding to the voxelization processing results, the wall thickness calculation results are stored in a structured manner to obtain the three-dimensional wall thickness field. The three-dimensional wall thickness field, the three-dimensional geometric model, and the defect detection data are jointly analyzed to extract the engineering feature vector. The engineering feature vector includes at least the global wall thickness feature, the deep cavity structure feature, the rib structure feature, the heat transfer efficiency feature, the thermal bottleneck feature, and the defect location feature.

[0012] First, the three-dimensional geometric model of the casting is obtained. The three-dimensional geometric model of the casting is three-dimensional CAD data representing the complete spatial structure of the automotive aluminum alloy die casting. It includes all geometric information such as the casting's shape, cavities, ribs, and wall thickness, and serves as the foundation for wall thickness calculation and feature analysis. The three-dimensional geometric model of the automotive aluminum alloy die casting to be processed is exported from the product design system. The model format is standard CAD format, retaining all structural features without simplification or omissions.

[0013] For example, taking an automotive aluminum alloy battery tray casting as the implementation object, the complete three-dimensional geometric model of the battery tray is derived from the design end, including all structures such as thin-walled base plate, reinforcing ribs, mounting bosses, and deep cavities.

[0014] Secondly, the defect detection data is extracted from the historical inspection database of the target scenario, combined with a preset traceability period. This defect detection data includes at least the associated spatial location and dimensional characteristics of the defect, the heat treatment information of the corresponding casting, and the inspection result label. The historical inspection database is a structured database storing defects and heat treatment results of die-castings from the same batch, mold, and alloy. The traceability period refers to a preset historical data extraction time range used to filter valid historical inspection samples. The defect detection data includes the spatial location and dimensional characteristics of the defect, the heat treatment information of the corresponding casting, and the inspection result label, where the inspection result label is a Boolean value, with 0 representing no bubbling and 1 representing bubbling.

[0015] For example, for automotive aluminum alloy battery tray castings, the preset traceability period is the past 6 months. Defect data of battery trays produced by the same mold within this period are extracted from the historical inspection database, and a total of 300 sets of samples are extracted. Among them, the spatial location of the defects is (X: 52mm, Y: 130mm, Z: 18mm), the size characteristics are (equivalent radius 0.9mm), the heat treatment information is (solution temperature 500℃, holding time 60min), and the inspection result label is: 0 for 240 sets and 1 for 60 sets.

[0016] Next, the three-dimensional geometric model undergoes adaptive voxelization. Based on the voxelization result, wall thickness is calculated by traversing multiple voxels. This wall thickness calculation is achieved using either the ray intersection method or the median transformation method. Adaptive voxelization refers to the process of discretizing a continuous three-dimensional geometric model into three-dimensional mesh voxels, which can adjust the voxel accuracy to suit structural complexity. Wall thickness calculation refers to calculating the thickness at each voxel position of the casting, achieved using either the ray intersection method or the median transformation method. The ray intersection method can average the ray directions to improve robustness if there are minor disturbances. A single method or a combination of both methods can be selected depending on the complexity of the three-dimensional structure.

[0017] Specifically, adaptive voxelization is performed on the 3D geometric model to generate a voxel mesh; all voxels in the voxel mesh are traversed; either the ray intersection method or the median transformation method is used to calculate the wall thickness. For complex structures, both methods can be combined. The ray intersection method can be used to perform directional perturbation and take the average.

[0018] For example, the three-dimensional geometric model of the automotive aluminum alloy battery tray casting is adaptively voxelized to generate a uniform voxel mesh; all voxels are traversed. For the thin-walled bottom plate and the deep cavity region with complex structures, the ray intersection method is used to perform 8 perturbations on the ray direction and take the average to calculate the wall thickness; for the mounting boss and the thick region with regular structures, the central axis transformation method is used to calculate the wall thickness; the wall thickness of all voxels is calculated to obtain the thickness value of a single voxel.

[0019] Furthermore, based on the voxel mesh corresponding to the voxelization results, the wall thickness calculation results are structured and stored to obtain the three-dimensional wall thickness field. The three-dimensional wall thickness field refers to a global three-dimensional wall thickness distribution data field of the casting, formed by binding the wall thickness value of each voxel to its spatial location using the voxel mesh as a spatial index. Using the voxel mesh generated by adaptive voxelization as a framework, the calculated wall thickness results of each voxel are structured and stored according to voxel spatial coordinates, ultimately generating the three-dimensional wall thickness field of the casting.

[0020] For example, for automotive aluminum alloy battery tray castings, the wall thickness values ​​of each voxel are stored one by one using the spatial coordinates of the voxel grid as an index, and the three-dimensional wall thickness field of the battery tray is generated in a structured manner, clearly showing the full-domain wall thickness distribution of the thin-walled area of ​​the base plate, the rib area, and the thick-walled area of ​​the boss.

[0021] Finally, the three-dimensional wall thickness field, the three-dimensional geometric model, and the defect detection data are jointly analyzed to extract the engineering feature vector. This engineering feature vector includes at least the global wall thickness feature, deep cavity structure feature, rib structure feature, heat transfer efficiency feature, thermal bottleneck feature, and defect location feature. The three-dimensional wall thickness field, the three-dimensional geometric model, and the defect detection data are jointly analyzed, and the above six types of feature quantities are calculated one by one to form the engineering feature vector.

[0022] Among them, the engineering feature vector refers to a vector composed of the quantified features of casting geometry, defects, and heat transfer, which includes six types of feature quantities: global wall thickness feature quantity includes minimum wall thickness, maximum wall thickness, wall thickness ratio, average wall thickness, and wall thickness standard deviation; deep cavity structure feature quantity includes deep cavity volume ratio and maximum deep cavity depth; rib structure feature quantity includes minimum rib spacing; heat transfer efficiency feature quantity is the ratio of surface area to volume; thermal bottleneck feature quantity is the number of thermal bottleneck regions; and defect location feature quantity includes maximum defect distance from the surface and maximum defect equivalent radius.

[0023] For example, for automotive aluminum alloy battery tray castings, the following engineering feature vectors were extracted through joint analysis: Global wall thickness features: minimum wall thickness 3mm, maximum wall thickness 22mm, wall thickness ratio 7.33, average wall thickness 7.2mm, wall thickness standard deviation 4.1; Deep cavity structure features: deep cavity volume ratio 9.2%, maximum deep cavity depth 15mm; Rib structure features: minimum rib spacing 3.5mm; Heat transfer efficiency features: surface area to volume ratio 0.78mm². -1 Thermal bottleneck characteristics: number of thermal bottleneck regions: 3; Defect location characteristics: maximum defect distance from surface: 2.3 mm, maximum defect equivalent radius: 0.9 mm; finally forming a 12-dimensional standardized engineering feature vector.

[0024] In this embodiment of the invention, the acquisition of the three-dimensional geometric model of the casting, the accurate extraction of historical defect data, the adaptive voxelization and wall thickness calculation of the three-dimensional model, the generation of the three-dimensional wall thickness field and the extraction of engineering feature vectors were completed. The original geometric data and defect data of the casting were transformed into computable and quantifiable feature data, providing accurate basic data and feature support for subsequent feature region division, defect data reconstruction and critical temperature calculation, thus ensuring the accuracy and pertinence of the formulation of heat treatment process parameters from the source.

[0025] S2: Combining the three-dimensional wall thickness field and the engineering feature vector, the three-dimensional geometric model is divided into feature regions, and the defect detection data is reconstructed based on the feature region division results.

[0026] In this embodiment of the invention, the three-dimensional geometric model is divided into feature regions by combining the three-dimensional wall thickness field and the engineering feature vector, and the defect detection data is restructured based on the feature region division results. The wall thickness and structural characteristics of automotive aluminum alloy die-cast parts vary greatly at different locations, and the blistering risk and severity of defects in different regions such as deep cavities, ribs, and thin-wall transitions are completely different. Directly using global defect data would lead to confusion of defect characteristics in different regions, making it impossible to accurately determine the critical defect threshold for each region. Therefore, it is necessary to combine the three-dimensional wall thickness field and the engineering feature vector to divide feature regions, classify defects by region, and fit the defect probability distribution according to the heat treatment results, completing the structured reconstruction of the defect data and providing an accurate grouped and partitioned data foundation for subsequent calculation of critical defect feature values.

[0027] Step S2 in the process provided in this embodiment of the invention includes: Based on the wall thickness range and structural feature type, the three-dimensional geometric model is divided into multiple feature regions, wherein the structural feature types include deep cavity structure region, rib structure region and thin wall transition region; Each defect in the defect detection data is assigned to its corresponding feature region according to its spatial coordinates; The castings were divided into qualified and unqualified groups based on the heat treatment results. For each region in the qualified and unqualified groups, a parameterized feature probability distribution of historically detected defects within the region is fitted, wherein the parameterized feature probability distribution includes at least the distribution of equivalent radius, surface distance, radius-distance ratio, and defect density; The parameterized feature probability distribution is associated with the qualified and unqualified groups and output to obtain the data structure reconstruction result.

[0028] First, based on the wall thickness range and structural feature type, the three-dimensional geometric model is divided into multiple feature regions. These structural feature types include deep cavity structure regions, rib structure regions, and thin-walled transition regions. Feature regions are independent spatial areas defined according to the casting wall thickness distribution and structural functional attributes; they are the basic units for defect classification and risk assessment. Wall thickness ranges are segmented intervals set according to the casting thickness value, used to distinguish regions with different heat transfer characteristics, such as thick, medium, and thin walls. The structural feature types, including deep cavity structure regions, rib structure regions, and thin-walled transition regions, are characteristic structures of castings prone to blistering and uneven heat transfer.

[0029] Specifically, based on the wall thickness values ​​of the three-dimensional wall thickness field, a preset wall thickness interval threshold is established; combined with the deep cavity and rib structure features in the engineering feature vector, the deep cavity structure region, rib structure region, and thin wall transition region are marked and divided on the basis of the wall thickness interval; the spatial boundary of all feature regions is delineated to form a non-overlapping global feature partitioning result.

[0030] For example, for automotive aluminum alloy battery tray castings: preset wall thickness ranges: <5mm is the thin-walled area, 5-15mm is the medium-thick area, and >15mm is the thick area; combined with structural features, 6 feature areas are finally divided: thin-walled transition area, rib structure area, deep cavity structure area, medium-thick flange area, thick mounting boss area, and corner transition area; the three-dimensional spatial coordinate boundaries of each area are clearly defined to complete the full-domain feature division.

[0031] Next, each defect in the defect detection data is assigned to its corresponding feature region according to its spatial coordinates. All extracted defect detection data are traversed, and the three-dimensional spatial coordinates of each defect are extracted. The defect coordinates are compared one by one with the defined spatial boundaries of each feature region. Based on the coordinates, each defect is accurately assigned to its corresponding feature region, forming a regional defect dataset.

[0032] For example, for automotive aluminum alloy battery tray castings: traverse all defect data of 300 sets of samples, extract the coordinates of a single defect such as (X: 52mm, Y: 130mm, Z: 18mm); compare the boundary of the feature area to determine that the defect is located in the rib structure area; complete the classification of all defects, with 126 defects in the thin-walled transition area, 89 defects in the rib structure area, 42 defects in the deep cavity structure area, and the remaining defects in the other areas.

[0033] Next, based on the heat treatment results, the castings are divided into qualified and unqualified groups. The qualified group refers to the set of castings with a test result label of 0 after heat treatment. The unqualified group refers to the set of castings with a test result label of 1 after heat treatment. The test result labels of the defect detection data obtained from S1 are retrieved; according to the label value, castings with a label of 0 are assigned to the qualified group, and castings with a label of 1 are assigned to the unqualified group; thus forming two independent sets of casting samples, and retaining the corresponding defect data and heat treatment information for each group.

[0034] For example, regarding automotive aluminum alloy battery tray castings: out of 300 historical casting samples, 240 pieces with a test result label of 0 were classified into the qualified group, and 60 pieces with a label of 1 were classified into the unqualified group; at the same time, the defect data corresponding to each group were sorted out according to the regional division results.

[0035] Furthermore, for each region in both the qualified and unqualified groups, a parametric feature probability distribution of historically detected defects within that region is fitted. This parametric feature probability distribution includes at least the distributions of equivalent radius, surface distance, radius-distance ratio, and defect density. The parametric feature probability distribution refers to a quantifiable probability distribution model obtained by statistically fitting defect feature values, used to characterize the occurrence patterns of defect features. Specifically, the equivalent radius is the spherical equivalent size of the defect; the surface distance is the shortest distance from the defect to the outer surface of the casting; the radius-distance ratio is the ratio of the equivalent radius to the surface distance; and the defect density is the number of defects per unit volume.

[0036] Specifically, using the feature region as the unit and the qualified / unqualified group as the dimension, four types of feature values ​​are extracted for defects within each unit: equivalent radius, surface distance, radius-distance ratio, and defect density. The log-normal distribution / Weibull distribution is used to fit the four types of feature values ​​to obtain the parameterized probability distribution of each feature. It is ensured that the same feature region in both the qualified and unqualified groups independently completes the distribution fitting of the four types of features.

[0037] For example, for the rib structure region, the defect feature values ​​of the qualified group and the unqualified group are extracted respectively; the qualified group rib structure region is fitted to obtain: the equivalent radius conforms to the log-normal distribution and the radius-distance ratio is concentrated between 0.05 and 0.12; the unqualified group rib structure region has: the equivalent radius is larger and the radius-distance ratio is mostly greater than 0.15; the probability distribution of the four types of defect features of all 6 feature regions and two groups of samples is fitted.

[0038] Finally, the parameterized feature probability distribution is associated with the qualified and unqualified groups and output to obtain the data structure reconstruction result. The data structure reconstruction result refers to structured data containing feature regions, qualified / unqualified groups, and defect parameterized probability distributions. It is the result of reconstructing defect data from disordered to ordered and from global to partitioned.

[0039] Specifically, a three-level correlation mapping relationship is established between feature regions, qualified / unqualified groups, and parameterized probability distributions of four types of defects; the mapping results are stored and output as a structured array; and finally, a data structure reconstruction result that can be directly used for dual-channel adaptive fusion analysis is obtained.

[0040] For example, six feature regions are established, each associated with a qualified group and an unqualified group, and their equivalent radius, surface distance, radius-distance ratio, and defect density distribution are respectively bound to these regions. Structured reconstruction data is output, clearly indicating that the thin-walled transition region (the unqualified group) has the highest mean radius-distance ratio distribution, making it a high-risk area. This forms a complete data structure reconstruction result. One set of data structure reconstruction results is as follows: Feature region: Thin-walled transition region; Group: Unqualified group; Equivalent radius: Follows a log-normal distribution, mean 0.92 mm, standard deviation 0.31 mm; Surface distance: Follows a Weibull distribution, mean 1.85 mm, standard deviation 0.42 mm; Radius-distance ratio: Follows a log-normal distribution, mean 0.49, standard deviation 0.13; Defect density: Follows a Poisson distribution, mean 8.3 defects / cm². 3 Standard deviation 1.2 particles / cm 3 .

[0041] In this embodiment of the invention, precise partitioning management of the casting space is achieved through feature region division. Defects are classified and grouped by region and qualified / unqualified groups. The structured reconstruction of historical defect data is completed by fitting the parameterized probability distribution of defects. The disordered original defect data is transformed into standardized data with partitions, groups, and probability features, which accurately distinguishes the differences in defect features of different regions and different quality results. This provides a compliant, accurate, and directly callable data foundation for subsequent dual-channel adaptive fusion analysis and critical defect feature value extraction.

[0042] S3: Based on the data structure reconstruction results, perform dual-channel adaptive fusion analysis to obtain the critical defect feature values ​​of multiple feature regions, and combine them with the prior alloy high-temperature mechanical parameters to determine the global critical temperature of the casting.

[0043] In this embodiment of the invention, a dual-channel adaptive fusion analysis is performed based on the data structure reconstruction results to obtain critical defect feature values ​​for multiple feature regions. Combined with prior alloy high-temperature mechanical parameters, the global critical temperature of the casting is determined. For the partitioned and grouped defect data reconstructed by S2, relying solely on a single analysis method to determine critical defect feature values ​​is prone to bias. While ROC curves excel at determining classification thresholds and logistic regression excels at constructing a mapping relationship between defects and failure probabilities, a single method cannot simultaneously ensure both classification accuracy and probabilistic rationality. Furthermore, the critical defect features of each region cannot be directly used for heat treatment process constraints; they must be converted into temperature parameters based on the alloy's high-temperature mechanical properties. Moreover, the most conservative values ​​must be taken with a safety margin to fundamentally prevent solution blistering. Therefore, a dual-channel parallel analysis of ROC and logistic regression, along with adaptive weighted fusion, is required to obtain accurate critical defect feature values, which are then converted and used to determine the global critical temperature constraint.

[0044] Step S3 in the process provided in this embodiment of the invention includes: For each feature region, extract the qualified group and the unqualified group from the data structure reconstruction result respectively; Based on the defect data of the qualified group and the unqualified group, a dual-channel parallel analysis is performed by combining the ROC curve method and the logistic regression method, and the analysis results are fused through an adaptive fusion strategy to obtain the critical defect feature value of each feature region. Based on the critical defect characteristic values ​​of all feature regions, and combined with the high-temperature mechanical property parameters of the alloy, the global critical temperature constraint is determined.

[0045] First, for each feature region, the qualified group and the unqualified group are extracted from the data structure reconstruction result. All feature regions divided by S2 are traversed; for each feature region, the four types of defect feature parameter data for the qualified group and the unqualified group are independently extracted from the data structure reconstruction result; after removing outliers, paired data of the qualified group dataset and the unqualified group dataset for each region are formed.

[0046] For example, for the six characteristic regions of the automotive aluminum alloy battery tray casting, the paired data of the thin-walled transition region is first extracted: the qualified group contains four types of characteristic parameters of 126 defects, and the unqualified group contains four types of characteristic parameters of 42 defects; the qualified / unqualified group data of the remaining five regions, such as the rib structure region and the deep cavity structure region, are extracted simultaneously.

[0047] Secondly, based on the defect data of the qualified and unqualified groups, a dual-channel parallel analysis is performed using the ROC curve method and logistic regression method. The analysis results are then fused through an adaptive fusion strategy to obtain the critical defect feature value for each feature region. The first analysis channel uses the ROC curve method, which classifies each defect feature parameter using qualified / unqualified as the classification label, finds the optimal classification threshold, and obtains the first channel critical value. The second analysis channel uses the logistic regression method, which establishes a probability mapping model between defect feature parameters and heat treatment blistering results, and calculates the second channel critical value according to a preset failure probability.

[0048] The adaptive weighted fusion strategy involves dynamically assigning weights based on the classification performance metrics of the two channels, such as AUC and accuracy, and then weighting and fusing the critical values ​​of the first and second channels to obtain the final critical defect feature value. Classification performance metrics are used to measure the reliability of channel analysis; the ROC channel uses the AUC value, and the logistic regression channel uses the goodness of fit, with higher values ​​carrying greater weight.

[0049] First, the first analysis channel uses the qualified group as negative samples and the unqualified group as positive samples. ROC curves are plotted for each of the four defect characteristic parameters: equivalent radius, surface distance, radius-distance ratio, and defect density. The maximum value of the Youden index corresponding to each parameter is calculated to determine the optimal classification threshold for each defect characteristic parameter, which is the critical value of the first channel. At the same time, the classification performance index AUC value of this channel is obtained.

[0050] Secondly, the second analysis channel uses four defect characteristic parameters as independent variables and the blistering label of the casting heat treatment as the dependent variable to construct a probability mapping model between the defect characteristic parameters and the heat treatment results. With a preset failure probability of 5% as the discrimination condition, the critical value of the second channel corresponding to each defect characteristic parameter is calculated, and the goodness of fit of the classification performance index of this channel is obtained.

[0051] Finally, the critical defect feature value is calculated using an adaptive weighted fusion formula, which is as follows: And satisfy the weight constraints ;in, = First channel AUC value / (First channel AUC value + Second channel goodness of fit) =1− .

[0052] For example, for the thin-walled transition region, the calculation is carried out using the risk factor radius-distance ratio as an example: the optimal classification threshold of the first channel is 0.38 after ROC analysis, that is, the critical value of the first channel is 0.38, and the corresponding classification performance index AUC=0.92; the critical value of the second channel is 0.41 after logistic regression analysis with a 5% failure probability, that is, the critical value of the second channel is 0.41, and the corresponding classification performance index goodness of fit=0.88. , ; Critical defect characteristic value after fusion Using the same method, the critical defect characteristic values ​​of the three parameters—equivalent radius, surface distance, and defect density—in the thin-walled transition region were calculated simultaneously, as well as the critical defect characteristic values ​​of all four parameters in the other five characteristic regions.

[0053] Finally, based on the critical defect characteristic values ​​of all feature regions and combined with the high-temperature mechanical property parameters of the alloy, the global critical temperature constraint is determined. The prior high-temperature mechanical property parameters of the alloy include preset experimental parameters such as yield strength, elastic modulus, and gas expansion coefficient at the solution treatment temperature of the aluminum alloy, used to convert defect characteristic values ​​into temperature values. The global critical temperature constraint refers to the minimum value of the converted temperature of critical defect characteristic values ​​in all feature regions, minus a preset safety margin, resulting in the final upper temperature limit, which is the safety boundary for solution treatment. The preset safety margin is a temperature allowance set to avoid process fluctuations, typically taken as 10~15℃.

[0054] Specifically, the critical defect characteristic values ​​of each feature region are substituted into the defect bubbling physical model, and combined with the prior alloy high-temperature mechanical parameters to calculate the critical temperature of each region; the minimum value among all region critical temperatures is taken; and the preset safety margin is subtracted to finally obtain the global critical temperature constraint.

[0055] For example, the converted critical temperatures of the six characteristic regions of the automotive aluminum alloy battery tray casting are 492℃, 505℃, 510℃, 508℃, 515℃, and 513℃, respectively. Taking the minimum value of 492℃ and deducting a preset safety margin of 10℃, the global critical temperature constraint is 482℃.

[0056] In this embodiment of the invention, the accuracy of determining critical defect characteristic values ​​is improved by using parallel analysis of ROC curves and logistic regression with adaptive weighted fusion, avoiding the bias of a single method. Furthermore, by combining the high-temperature mechanical parameters of the alloy, the defect characteristics are quantified into temperature parameters. The most conservative critical value is taken and a safety margin is reserved to determine the global critical temperature constraint. This provides a unique and safe hard temperature constraint for subsequent process feasibility assessment and determination of the upper limit of solution treatment and heat preservation time, and achieves precise control of bubbling risk from the defect mechanism level.

[0057] S4: Based on the global critical temperature, determine the feasibility of the process and the upper limit of the heat preservation time for solution treatment.

[0058] In this embodiment of the invention, the feasibility of the process is determined based on the global critical temperature, and the upper limit of the holding time for solution treatment is determined. The global critical temperature is the highest safe temperature at which the solution treatment of the casting does not cause blistering, and the preset effective lower limit of the solution temperature is the lowest temperature at which the aluminum alloy achieves solution strengthening. The two directly determine whether the solution treatment process has the conditions for safe implementation. If the global critical temperature is lower than the effective lower limit of the solution temperature, the solution temperature cannot meet the strengthening requirements, and forced treatment will result in ineffective solution treatment and a very high risk of blistering. If the process is feasible, the upper limit of the holding time should also be determined by combining the temperature difference, the high-temperature creep characteristics of the alloy, and the preset strain limit to avoid material creep, defect expansion, and blistering caused by prolonged high-temperature holding. Therefore, it is necessary to conduct a process feasibility assessment and accurately determine the upper limit of the solution holding time.

[0059] Step S4 in the process provided in this embodiment of the invention includes: The global critical temperature is compared with the preset effective lower limit of the solution temperature; If the global critical temperature is less than the lower limit of the effective temperature of the solution treatment, it is determined that the casting cannot be safely solution treated under the current defect conditions, and process feasibility warning information is output and the process optimization process is terminated. If the global critical temperature is greater than or equal to the lower limit of the effective solution temperature, the casting is determined to be suitable for solution treatment. Based on the temperature difference between the global critical temperature and the lower limit of the effective solution temperature, combined with the prior alloy high-temperature creep parameters and the preset target creep strain limit, the upper limit of the holding time for solution treatment is determined.

[0060] First, the global critical temperature is compared with the preset effective lower limit of the solution treatment temperature. The global critical temperature refers to the highest safe temperature at which the casting does not bubble during solution treatment, as determined in S3; that is, the global critical temperature constraint. The preset effective lower limit of the solution treatment temperature is the minimum process temperature required to achieve sufficient solution strengthening, determined by the aluminum alloy grade, and is a preset fixed parameter. The global critical temperature determined in S3 is retrieved, and the preset effective lower limit of the solution treatment temperature corresponding to the target aluminum alloy is extracted. The two temperature values ​​are then directly compared to clarify the relationship between them.

[0061] For example, for an automotive aluminum alloy battery tray casting, the alloy grade is AlSi9Cu3: the global critical temperature is 482℃, and the preset effective temperature lower limit of the solid solution is 470℃. By comparison, it is found that 482℃ > 470℃.

[0062] Secondly, if the global critical temperature is lower than the lower limit of the effective solution temperature, it is determined that the casting cannot undergo safe solution treatment under the current defect conditions. A process feasibility warning is then output, and the process optimization process is terminated. The process feasibility warning indicates that the casting defect conditions do not meet the requirements for solution treatment, and includes reasons for infeasibility and suggestions for alternative processes. Termination of the process optimization process means stopping all subsequent steps, such as heat transfer proxy model construction and Bayesian optimization, and no longer generating solution heating process parameters.

[0063] Specifically, when the comparison result is that the global critical temperature is less than the lower limit of the effective temperature of the solution, the process is directly determined to be infeasible; an early warning message containing the defect risk and the reason for the temperature not meeting the standard is generated; the entire heat treatment process optimization process is immediately terminated, and it is recommended to use non-solution processes such as T5 aging.

[0064] For example, assuming the global critical temperature of the battery tray casting is 465℃, which is lower than the effective solution temperature limit of 470℃: it is determined that the casting cannot be safely solution treated under the current defect conditions; output warning information: the global critical temperature of 465℃ is lower than the effective solution temperature limit of 470℃, solution strengthening is ineffective and the risk of bubbling is extremely high, it is recommended to use T5 artificial aging process; terminate all subsequent process optimization steps.

[0065] Finally, if the global critical temperature is greater than or equal to the lower limit of the effective solution temperature, the casting is determined to be suitable for solution treatment. Based on the temperature difference between the global critical temperature and the lower limit of the effective solution temperature, combined with the prior alloy high-temperature creep parameters and the preset target creep strain limit, the upper limit of the holding time for solution treatment is determined.

[0066] The temperature difference, calculated as the global critical temperature minus the lower limit of the effective solution temperature, characterizes the safety margin of the solution temperature. The a priori high-temperature creep parameters for the alloy are experimentally determined preset parameters such as the creep coefficient and creep rate of the aluminum alloy under high-temperature solution conditions. The preset target creep strain limit refers to the maximum allowable strain value set to avoid creep deformation and defect expansion / blistering of the casting; it is a preset fixed parameter. The upper limit of holding time refers to the maximum time the casting can be held at the solution temperature; it is a time constraint for the solution process.

[0067] Specifically, calculate the temperature difference: ΔT = global critical temperature − lower limit of effective solution temperature; retrieve the high-temperature creep parameters of the aluminum alloy calibrated by prior experiments, as well as the preset target creep strain limit; calculate the upper limit of the holding time using the pre-calibrated analytical physical formula for creep strain, for example: In the formula: This is the upper limit of the heat treatment holding time; The preset target creep strain limit; The a priori high-temperature creep coefficient of the alloy; This represents the difference between the global critical temperature and the lower limit of the effective solution temperature; all parameters in the formula are experimentally calibrated prior parameters.

[0068] For example, for automotive aluminum alloy battery tray castings AlSi9Cu3, the temperature difference value Prior experimental calibration parameters: creep coefficient Preset target creep strain limit Substitute into the analytical formula to calculate: The upper limit of the heat preservation time for the solution treatment of the casting is 90 minutes.

[0069] In this embodiment of the invention, by comparing the global critical temperature with the lower limit of the effective temperature of solid solution, the feasibility of the solid solution process can be quickly determined, avoiding process attempts with ineffective solid solution and high blistering risk from the source. Under the premise that the process is feasible, the upper limit of the holding time is accurately determined by combining the temperature margin, alloy creep characteristics and strain limit, providing a time constraint for the subsequent optimization of heating process parameters, and taking into account both the solid solution strengthening effect and the risk control of high temperature creep and defect blistering.

[0070] S5: Construct a heat transfer proxy model, and use the global critical temperature and the upper limit of the holding time as constraints to perform Bayesian optimization search through the heat transfer proxy model to determine the heating process parameters.

[0071] In this embodiment of the invention, a heat transfer proxy model is constructed, and the heating process parameters are determined by Bayesian optimization search using the global critical temperature and the upper limit of the holding time as constraints. The three-dimensional wall thickness field of automotive aluminum alloy die-cast parts is high-dimensional data, making direct application to heat transfer calculations extremely complex. Traditional finite element heat transfer simulations are time-consuming and cannot meet the requirements of Bayesian optimization for multiple iterations and rapid evaluation. Simultaneously, the heating process parameters must match the casting structural characteristics and be constrained by the global critical temperature and the upper limit of the holding time to achieve optimal temperature uniformity while ensuring no bubbling. Therefore, a self-encoder is used to compress the three-dimensional wall thickness field features, construct a rapidly predictive heat transfer proxy model, and then Bayesian optimization is performed with safety parameters as constraints to accurately determine the heating process parameters suitable for the casting.

[0072] Step S5 in the process provided in this embodiment of the invention includes: The construction of the heat transfer proxy model includes: Construct and train an autoencoder-based feature compression layer, which is used to compress and reconstruct the three-dimensional wall thickness field; The heat transfer proxy model is constructed by concatenating the compressed reconstruction output of the feature compression layer with the engineering feature vector as the first input, the heating process parameters as the second input, and the key temperature index of the casting as the output. The heat transfer proxy model is iteratively trained by combining the heat treatment information and the detection result labels in the defect detection data.

[0073] First, an autoencoder-based feature compression layer is constructed and trained. This layer is used to compress and reconstruct the three-dimensional wall thickness field. The autoencoder-based feature compression layer is an unsupervised neural network structure consisting of an encoder and a decoder. The encoder reduces the high-dimensional three-dimensional wall thickness field to low-dimensional latent features, and the decoder reconstructs the low-dimensional features back to the original wall thickness field, achieving efficient compression of high-dimensional geometric features. Compression and reconstruction refers to the process of reducing and compressing the high-dimensional three-dimensional wall thickness field data, followed by reconstruction. This process is used to verify the integrity of the feature compression and preserve core heat transfer-related geometric features.

[0074] Specifically, an autoencoder network structure is constructed. The encoder uses a three-dimensional convolutional layer to extract the spatial features of the wall thickness field, and the decoder uses a three-dimensional deconvolutional layer to reconstruct the features. The three-dimensional wall thickness field of batch castings is used as training data, and the training objective is to minimize the mean square error between the original wall thickness field and the reconstructed wall thickness field to complete the training of the feature compression layer. After training, the encoder parameters are fixed and used as the feature compression module of the heat transfer proxy model.

[0075] For example, for automotive aluminum alloy battery tray castings: build an autoencoder with 3 layers of 3D convolutional encoder + 3 layers of 3D deconvolutional decoder; use 100 sets of 3D wall thickness fields of similar castings as training data, train until the reconstruction error is less than 5%; input the 3D wall thickness field of the battery tray into the trained feature compression layer, and compress the high-dimensional wall thickness field into a 32-dimensional low-dimensional compressed feature vector.

[0076] Secondly, the heat transfer proxy model is constructed by defining the concatenation result of the compressed reconstruction output of the feature compression layer and the engineering feature vector as the first input, the heating process parameters as the second input, and the key temperature index of the casting as the output. The first input is the low-dimensional compressed feature output by the autoencoder, which is fused with the 12-dimensional engineering feature vector to characterize the core geometry and defect features of the casting. The second input is the heating process parameters to be optimized, including the number of heating steps, the holding time of each step, and the heating rate. The key temperature index of the casting is the output of the heat transfer proxy model, including the final maximum temperature, the final minimum temperature, the final maximum temperature difference, and the solution treatment time. The heat transfer proxy model is a fast heat transfer prediction model built on neural networks, which can output the casting temperature index in milliseconds, replacing the time-consuming finite element simulation.

[0077] Specifically, the low-dimensional compressed feature vector is concatenated with the 12-dimensional engineering feature vector extracted by S1 to form the first input; the heating step temperature, holding time, and heating rate are defined as the second input; the final maximum temperature difference and temperature field standard deviation are defined as the model output, and a heat transfer proxy model in the form of a fully connected neural network is built.

[0078] For example, for automotive aluminum alloy battery tray castings: the 32-dimensional three-dimensional wall thickness field compression features output by the autoencoder are concatenated with the 12-dimensional engineering feature vector extracted by S1 to form a 44-dimensional first input; three levels of heating process parameters are defined, including three levels of stepped temperature, three levels of holding time, and two levels of heating rate, totaling eight parameters as the second input; the 44-dimensional first input and the eight-dimensional second input are concatenated to form a 52-dimensional total model input, and a fully connected neural network heat transfer proxy model is constructed. The specific structure is as follows: 52-dimensional neurons in the input layer, 128-dimensional neurons in the first hidden layer with ReLU activation function, 256-dimensional neurons in the second hidden layer with ReLU activation function, 128-dimensional neurons in the third hidden layer with ReLU activation function, and the output layer includes four-dimensional outputs: final maximum temperature difference, final maximum temperature, final minimum temperature, and solution treatment time. The activation function of the output layer is a linear function.

[0079] Finally, the heat transfer proxy model is iteratively trained by combining the heat treatment information and detection result labels from the defect detection data. The training dataset is derived from the historical defect detection data extracted in S1. Historical heat treatment information, detection result labels, and corresponding measured temperature indicators of the castings are extracted to construct a high-quality training dataset covering the entire process parameter domain. The first and second inputs corresponding to the historical samples are used as the training inputs of the heat transfer proxy model, and the measured temperature indicators are used as the model training labels. The Adam optimizer is used for iterative training, and the mean squared error (MSE) is used as the model loss function. When the model loss value does not decrease significantly for 20 consecutive rounds and the prediction error reaches a preset threshold, the model is considered to have converged, and the heat transfer proxy model training is completed.

[0080] For example, 300 sets of data on automotive aluminum alloy battery tray castings covering different structural features and process parameters are extracted from the historical inspection database of S1. These data include heat treatment parameters, bubbling labels, and measured temperature indicators. This amount of data can fully cover the process space and meet the training requirements of a small neural network. The first and second inputs of each set of samples are used as model inputs, and the measured maximum final temperature difference is used as the training label. The Adam optimizer is used, and the model is iteratively trained for 100 rounds with mean squared error as the loss function. When the model loss value tends to stabilize for 20 consecutive rounds and the prediction error of the maximum final temperature difference is controlled within ±1.5℃, the model is considered to have converged, and a mature heat transfer proxy model that can be used for rapid optimization and evaluation is obtained.

[0081] Among them, the heating process parameters are determined by performing Bayesian optimization search through the heat transfer proxy model, using the global critical temperature and the upper limit of the holding time as constraints, including: The number of heating steps is determined based on the engineering feature vector. Based on the heating step number, an optimization problem is defined, wherein the optimization problem uses the holding time and heating rate corresponding to the heating step number as optimization variables, and minimizes the final maximum temperature difference output by the heat transfer proxy model as the optimization objective; By combining the global critical temperature, the upper limit of the heat preservation time, and the heating control characteristics of the target heating equipment, constraints are determined, and a Bayesian optimization search is performed on the optimization problem to obtain the heating process parameters.

[0082] First, the number of heating steps is determined based on the engineering feature vector. The number of heating steps refers to the number of stages of segmented heating during the solution heating process, determined according to the degree of difference in casting wall thickness; the greater the difference in wall thickness, the more steps are required. The wall thickness ratio parameter in the engineering feature vector is retrieved, and the number of heating steps is matched according to preset rules: wall thickness ratio < 3 is level 1, 3 ≤ wall thickness ratio < 6 is level 2, 6 ≤ wall thickness ratio < 10 is level 3, and wall thickness ratio ≥ 10 is level 4.

[0083] For example, for an automotive aluminum alloy battery tray casting, the wall thickness ratio in the engineering feature vector is 7.33, which conforms to the rule of 6≤wall thickness ratio<10, and the number of heating steps is determined to be 3.

[0084] Secondly, based on the number of heating steps, an optimization problem is defined, wherein the optimization problem uses the holding time and heating rate corresponding to the heating steps as optimization variables, and minimizes the maximum final temperature difference output by the heat transfer surrogate model as the optimization objective. The optimization variables refer to the process parameters to be optimized in Bayesian optimization, namely the holding time of each heating step and the heating rate between steps. The optimization objective is to minimize the maximum final temperature difference of the casting and improve temperature uniformity through the prediction of the heat transfer surrogate model.

[0085] For example, the optimization variable corresponding to the three heating stages is determined as follows: Stage 1 heat preservation time. Level 2 insulation time Level 3 insulation time The heating rate from level 1 to level 2 Level 2 to Level 3 heating rate ; Define the optimization objective function: , This represents the final maximum temperature difference output by the heat transfer proxy model. Therefore, the optimization variable is... The optimization objective is to minimize the final maximum temperature difference output by the heat transfer proxy model. .

[0086] Finally, by combining the global critical temperature, the upper limit of the holding time, and the heating control characteristics of the target heating equipment, constraints are determined, and a Bayesian optimization search is performed on the optimization problem to obtain the heating process parameters. The constraints serve as the parameter boundaries for Bayesian optimization, ensuring that the optimization results meet safety requirements and equipment execution capabilities. The heating equipment control characteristics refer to inherent parameters of the heat treatment furnace, such as the upper limit of the heating rate, temperature control accuracy, and minimum holding time. The Bayesian optimization search is a highly efficient optimization algorithm based on a probabilistic surrogate model, rapidly finding the optimal process parameters within the constraint space with a small number of iterations.

[0087] Specifically, Latin hypercube sampling is used to generate initial optimization samples within the constraint space. These samples are then input into a trained heat transfer surrogate model to obtain the corresponding objective function values. A Gaussian process is used as the probabilistic model for Bayesian optimization. The sampling points are iteratively selected and the optimal variables are searched by improving the acquisition function based on expectations. When the convergence condition is met, the search stops and the optimal holding time and heating rate are output. The convergence condition is: the improvement in the optimal objective function value after 10 consecutive iterations is less than 0.1℃, or the number of iterations reaches a preset maximum of 100 to form the final heating process parameters. The constraints for this optimization include: temperature constraints: heating temperature at each stage ≤ global critical temperature; time constraints: final stage holding time ≤ upper limit of holding time, total holding time conforms to production cycle time; equipment constraints: heating rate ≤ maximum equipment heating rate, holding time ≥ minimum equipment control time.

[0088] For example, for automotive aluminum alloy battery tray castings, the optimized constraint is: a 3-level temperature gradient. All temperatures are ≤482℃; the time constraint is the third-level heat preservation time. ≤90min; Equipment constraint is heating rate ≤5℃ / min. The Bayesian optimization was set to a maximum of 100 iterations, with the convergence condition being that the maximum temperature difference improvement in the optimal final state is <0.1℃ for 10 consecutive iterations. The optimization process stopped after meeting the convergence condition, ultimately yielding the optimal heating process parameters. , , , , , , .

[0089] In this embodiment of the invention, a high-dimensional three-dimensional wall thickness field feature compression is completed by an autoencoder, and a heat transfer proxy model with millisecond-level prediction is constructed to replace the time-consuming finite element heat transfer simulation and improve optimization efficiency. Based on engineering features, the matching heating step number is determined, and the optimization objective is to minimize the final maximum temperature difference. Bayesian optimization is carried out in combination with triple constraints of temperature, time and equipment to accurately obtain the optimal heating process parameters that take into account both safety and temperature uniformity.

[0090] S6: Output the solution heating curve corresponding to the heating process parameters to perform heat treatment on the casting.

[0091] In this embodiment of the invention, the solution heating curve corresponding to the heating process parameters is output for heat treatment of the casting. In traditional fixed-step heating processes, even after parameter optimization, significant temperature differences still exist between thick and thin-walled sections of complex aluminum alloy castings, easily leading to insufficient local solution treatment or overheating. Introducing sinusoidal micro-oscillations during the holding stage can enhance convection within the furnace through dynamic temperature disturbance, improving the temperature uniformity of the casting. However, the improvement effect of micro-oscillations needs to be verified first, and then the micro-oscillation parameters need to be optimized in conjunction with safety and equipment constraints to avoid micro-oscillations causing temperature deviations or equipment malfunction. Therefore, it is necessary to verify the effect and optimize the parameters to output a suitable solution heating curve that balances temperature uniformity, safety, and equipment feasibility.

[0092] Step S6 in the process provided in this embodiment of the invention includes: Based on the heating process parameters and combined with the heating control characteristics of the target heating equipment, a preset sinusoidal micro-oscillation is superimposed during the heat preservation stage; The heat transfer proxy model calculates the final temperature field with and without sinusoidal micro-oscillations, and calculates the maximum improvement in the final temperature difference accordingly. If the improvement in the final maximum temperature difference is greater than or equal to the preset improvement threshold, the micro-oscillation is determined to be effective. The amplitude and period of the sinusoidal micro-oscillation are used as optimization variables. Based on the Bayesian optimization method, the optimal micro-oscillation parameters are searched for in the heat preservation stage, and the solution heating curve of the step-micro-oscillation is output. If the improvement in the final maximum temperature difference is less than the preset improvement threshold, the micro-oscillation is deemed invalid, and the solution heating curve of the stepped heating is directly output based on the heating process parameters.

[0093] The method of searching for optimal micro-oscillation parameters during the heat preservation stage based on Bayesian optimization also includes: Set a set of constraints including upper limit of amplitude, lower limit of amplitude, upper limit of period, and lower limit of period.

[0094] First, based on the heating process parameters and the heating control characteristics of the target heating equipment, a preset sinusoidal micro-oscillation is superimposed during the heat preservation stage. The heating control characteristics of the target heating equipment refer to the inherent parameters of the heat treatment furnace, including temperature control accuracy. Temperature response time Maximum / minimum executable heating rate, etc. Preset sinusoidal micro-oscillation refers to the initial sinusoidal temperature disturbance superimposed during the highest temperature holding stage to verify the effect of micro-oscillation. A fixed amplitude and period are preset for preliminary evaluation of the effect.

[0095] Specifically, the heating process parameters obtained from S5 are retrieved to determine the temperature, holding time, and heating rate of each stage. Based on the control characteristics of the target heating equipment, a preset sinusoidal micro-oscillation is superimposed during the highest temperature holding stage. The superposition method for the micro-oscillation is as follows: with the temperature of the highest temperature holding stage as the center temperature, the amplitude is set according to a preset value. ,cycle A sinusoidal temperature curve is generated as the initial control curve for the heat preservation stage.

[0096] For example, the heating process parameters obtained in S5 are: , , , , , , The control characteristics of the target heating equipment are: temperature control accuracy. Temperature response time In the third stage of heat preservation, a preset sinusoidal micro-oscillation is superimposed, with an initial amplitude... Initial period An initial micro-oscillation heat preservation curve is generated.

[0097] Secondly, the heat transfer proxy model calculates the final temperature field under sinusoidal micro-oscillation and non-sinusoidal micro-oscillation conditions, and correspondingly calculates the improvement in the maximum final temperature difference. The final temperature field without sinusoidal micro-oscillation refers to the temperature distribution field after the casting is held at room temperature when only a stepped heating process is used; the corresponding maximum final temperature difference is denoted as... This refers to the maximum temperature difference between the thick and thin sections. The final-state temperature field with sinusoidal micro-oscillations refers to the temperature distribution field of the casting after the heat treatment is completed, following the superposition of preset sinusoidal micro-oscillations; the corresponding maximum final-state temperature difference is denoted as... The improvement in the final maximum temperature difference characterizes the effect of micro-oscillations on improving temperature uniformity, and is defined as follows: This refers to the difference between the maximum final temperature difference with and without oscillation. The greater the improvement, the better the micro-oscillation effect.

[0098] Specifically, the step heating parameters without sinusoidal micro-oscillations are input into the heat transfer surrogate model trained by S5 to calculate the oscillatory final-state temperature field and extract the maximum final-state temperature difference. The heating parameters, superimposed with preset sinusoidal micro-oscillations, are input into the heat transfer proxy model to calculate the oscillating final-state temperature field, and the maximum final-state temperature difference is extracted. ; Calculate the maximum improvement in final temperature difference This is used for subsequent effect evaluation.

[0099] For example, by inputting the step heating parameters without micro-oscillations into the heat transfer surrogate model, the final temperature field without oscillations and the maximum temperature difference in the final state are obtained. The heating parameters, superimposed with preset micro-oscillations, are input into the heat transfer proxy model to obtain the oscillating final-state temperature field and the maximum final-state temperature difference. ; Calculate the maximum improvement in final temperature difference .

[0100] Furthermore, if the improvement in the final maximum temperature difference is greater than or equal to a preset improvement threshold, the micro-oscillation is deemed effective. Using the amplitude and period of the sinusoidal micro-oscillation as optimization variables, the optimal micro-oscillation parameters for the heat preservation stage are searched based on a Bayesian optimization method, and the solution heating curve of the step-micro-oscillation is output. The preset improvement threshold is the minimum improvement amount required to determine whether the micro-oscillation is effective; it is a preset process threshold, for example, set to 1℃. That is, the micro-oscillation must reduce the final maximum temperature difference by at least 1℃ to be considered effective. The micro-oscillation optimization variables are the two parameters of the sinusoidal micro-oscillation, namely the amplitude... With period .

[0101] The process of searching for optimal micro-oscillation parameters during the heat preservation stage based on Bayesian optimization includes setting a set of constraints, including upper and lower limits for amplitude, and upper and lower limits for period. The constraint set refers to the parameter boundary constraints set to ensure the safety of the micro-oscillations and the executability of the equipment: Upper limit for amplitude: That is, the oscillation peak does not exceed the foaming constraint temperature, where Global critical temperature; lower limit of amplitude: That is, the amplitude must be greater than the furnace control precision, otherwise the furnace cannot operate; lower limit of cycle: That is, the oscillation period must be greater than twice the furnace response time; otherwise, the furnace temperature cannot follow the set oscillation. Maximum period: If the value exceeds this, the oscillation frequency becomes too low, degenerating into a simple temperature step and losing its oscillation significance.

[0102] Specifically, the improvement amount ΔΔT of the final maximum temperature difference is compared with the preset improvement threshold. If ΔΔT ≥ the threshold, the micro-oscillation is deemed effective, and the micro-oscillation parameter optimization process begins. The amplitude of the sinusoidal micro-oscillation is used as the reference value. With period To optimize the variables, the optimization objective is set as minimizing the maximum final temperature difference; a set of constraints is introduced, and the optimal value is searched based on the Bayesian optimization method. and The optimized convergence condition is: the improvement in the maximum final temperature difference is less than 0.1℃ after 10 consecutive iterations, or the number of iterations reaches a preset maximum of 100; combining the optimal micro-oscillation parameters and the step heating parameters, such as Figure 2 As shown, a complete step-micro-oscillation solid solution heating curve is generated and output to the furnace control system.

[0103] For example, the preset improvement threshold is 1°C. The micro-oscillation was deemed valid; the optimization variable was set as follows. and The optimization objective is to minimize the maximum final temperature difference; constraints are introduced as follows: , , , Through Bayesian optimization, the optimal micro-oscillation parameters were obtained by satisfying the convergence condition that the improvement amount after 10 consecutive iterations is <0.1℃. , By combining the stepped heating parameters and the optimal micro-oscillation parameters, a complete stepped-micro-oscillation solution heating curve is generated. In the third-stage heat preservation stage, sinusoidal micro-oscillation is performed with an amplitude of 5℃ and a period of 10min, centered at 470℃.

[0104] Conversely, if the improvement in the final maximum temperature difference is less than a preset improvement threshold, the micro-oscillation is deemed invalid, and the solution heating curve for stepped heating is directly output based on the heating process parameters. The improvement in the final maximum temperature difference ΔΔT is compared with the preset improvement threshold. If ΔΔT < the threshold, the micro-oscillation is deemed invalid and cannot effectively improve temperature uniformity. The stepped heating process parameters obtained in S5 are directly used to generate a step heating solution curve without micro-oscillation, which is then output to the furnace control system.

[0105] For example, assuming the maximum temperature difference improvement in the final state is ΔΔT = 0.5℃ < the preset threshold of 1℃, the micro-oscillation is deemed invalid; the stepped heating parameters obtained from S5 are directly used to generate a solid solution curve with fixed stepped heating, which is then output to the furnace control system.

[0106] In this embodiment of the invention, a preset micro-oscillation effect verification is used to distinguish between effective and ineffective micro-oscillation scenarios: in the effective scenario, the micro-oscillation parameters are optimized in combination with safety and equipment constraints to generate a stepped-micro-oscillation solution heating curve. Dynamic temperature disturbances enhance convection within the furnace, further improving the temperature uniformity of the casting. In the ineffective scenario, the stepped heating curve is directly output to avoid the control complexity caused by ineffective micro-oscillations. The final output heating curve takes into account temperature uniformity, bubbling safety, and equipment feasibility, providing a precise and reliable control basis for subsequent heat treatment.

[0107] The process provided in this embodiment of the invention further includes: Based on the actual heat treatment test results of the casting after performing the solution heating curve, update the heat treatment information and test result labels in the defect detection data; Based on preset data update trigger conditions, update trigger judgment is performed on the updated defect detection data; If the data update triggering condition is met, the heating process parameters are updated based on the updated defect detection data; If the data update triggering condition is not met, the current heating process parameters will remain unchanged.

[0108] First, based on the actual heat treatment test results of the casting after executing the solution heating curve, update the heat treatment information and test result labels in the defect detection data. The actual heat treatment test results refer to the measured data obtained through non-destructive testing and temperature detection after the casting executes the solution heating curve output by S6, including whether the casting blisters, the actual heat treatment process parameters executed, the measured final temperature field, and the defect distribution.

[0109] Specifically, after the casting completes solution heat treatment, it undergoes comprehensive testing. The actual solution heating curve parameters, test results, measured defect parameters, and final temperature data are recorded. The initial defect detection database is retrieved, and the measured heat treatment information, test result labels, and related data are supplemented and entered according to the database's established format to complete the real-time database update. The updated database is then cleaned to remove abnormal data, ensuring data integrity and accuracy.

[0110] For example, 10 castings were heat-treated according to the stepped-micro-oscillation solution heating curve output by S6. After testing, the actual execution parameters were found to be: , , , , , , The micro-oscillation amplitude was 5℃ and the period was 10min; the test results showed that none of the 10 pieces had bubbling, and the maximum temperature difference in the final state was 4.8℃; the 10 sets of measured heat treatment parameters, test labels, and temperature difference data were added to the defect detection database to complete the database update; the data was cleaned, one set of abnormal temperature difference data caused by temporary equipment failure was removed, and nine sets of valid data were retained.

[0111] Secondly, based on preset data update trigger conditions, the updated defect detection data is subjected to update trigger judgment. The preset data update trigger conditions refer to pre-defined threshold conditions for determining whether the heating process parameters need to be updated. These conditions are formulated in conjunction with data increments, prediction errors, and changes in the defect rate to ensure the necessity and rationality of the update. Update trigger judgment involves comparing the actual situation of the updated defect detection data with the preset trigger conditions to determine whether the process parameter update process needs to be initiated.

[0112] Specifically, the preset data update trigger conditions are retrieved: Trigger condition 1: The amount of newly added valid detection data after the update is ≥10 sets, ensuring that the sample size is sufficient to support the model update; Trigger condition 2: Based on the updated data, the prediction error of the heat transfer proxy model is ≥2℃; Trigger condition 3: The foaming rate of the casting after the update increases by ≥5% compared to the previous update; Statistical analysis is performed on the updated defect detection data to determine whether any of the above trigger conditions are met; The judgment result is output: If any trigger condition is met, it is determined that an update is required; if none are met, it is determined that an update is not required.

[0113] For example, continuing with the casting of automotive aluminum alloy battery trays: Preset trigger conditions: ≥10 new valid data sets, prediction error ≥2℃, and foaming rate increase ≥5%; Statistical update of data: 9 new valid data sets, trigger condition 1 is not met; the prediction error of the heat transfer proxy model based on the new data is 1.2℃, trigger condition 2 is not met; the foaming rate is still 0%, trigger condition 3 is not met; Judgment result: None of the preset trigger conditions are met, and it is determined that no process parameter update is required.

[0114] Furthermore, if the data update triggering condition is met, the heating process parameters are updated based on the updated defect detection data. Process parameter update refers to re-executing the heat transfer proxy model training and Bayesian optimization process of S5 based on the updated defect detection data to obtain new heating process parameters adapted to the current production conditions, replacing the original parameters. Update logic: The original process optimization framework is used, only the training data is replaced with the updated defect detection data, ensuring the consistency and accuracy of parameter updates.

[0115] Specifically, if the triggering conditions are met, the updated defect detection data is retrieved, and the S5 process is re-executed: the heat transfer proxy model is retrained; based on the updated model, combined with the global critical temperature and the upper limit of the holding time, Bayesian optimization is carried out again; new heating process parameters are obtained; the original heating process parameters are replaced, the new parameters are used for the subsequent solution heat treatment of the casting, and the process parameter file is updated simultaneously.

[0116] For example, assuming that 15 more castings are subsequently heat-treated, adding 15 sets of valid data, bringing the total to 24 sets, the trigger condition of ≥10 sets of new valid data is met: The updated defect detection database is retrieved, and the heat transfer proxy model is retrained until the prediction error is ≤1.5℃; based on the new model, combined with a global critical temperature of 482℃ and a holding time limit of 90 min, Bayesian optimization is performed again; new heating process parameters are obtained: , , , , , , , The micro-oscillation amplitude is 4.5℃ and the period is 12min; the original parameters are replaced for subsequent production, and the process file is updated.

[0117] Conversely, if the data update triggering condition is not met, the current heating process parameters remain unchanged. If the update triggering judgment result indicates that no update is needed, it is clear that the current heating process parameters still meet production requirements; the original heating process parameters obtained in S5 are continued to be used to perform subsequent solution heat treatment of the casting; the actual heat treatment test results of the casting are continuously collected to accumulate data and prepare for the next data update triggering judgment.

[0118] For example, for automotive aluminum alloy battery tray castings, nine sets of valid data were previously added, but none of the triggering conditions were met: it was determined that the current heating process parameters were still suitable for production; the parameters were continued to be used to perform subsequent casting heat treatment, while the test data of each batch of castings was continuously collected until the triggering conditions were met, and then the parameters were updated.

[0119] In this embodiment of the invention, defect detection data is dynamically updated through feedback of actual detection results. The necessity of updating process parameters is determined by combining preset trigger conditions, avoiding process fluctuations caused by blind updates. When the trigger conditions are met, the process parameters are re-optimized based on the updated data to adapt to changes in production conditions and improve long-term production stability. When the conditions are not met, the parameters remain unchanged to ensure production continuity. The entire process forms a closed loop of execution-detection-update-optimization, further improving the reliability of solution heat treatment for aluminum alloy die castings, reducing the risk of blistering, and ensuring product quality consistency.

[0120] Through the specific implementation methods described above, the embodiments of the present invention achieve the following technical effects: This invention provides a heat treatment strengthening process for die-cast automotive aluminum alloy parts. First, the three-dimensional geometry and defect data of the casting are transformed into standardized engineering feature vectors. Then, combined with feature region division, the defect data is structurally reconstructed. Using ROC and logistic regression dual-channel adaptive fusion analysis, the critical defect feature values ​​of each region are accurately determined. Based on the alloy's high-temperature mechanical parameters, a global critical temperature constraint is calculated. Simultaneously, the feasibility of the solution treatment process and the upper limit of the holding time are set. On this basis, a lightweight heat transfer proxy model is constructed, and Bayesian optimization is used to quickly solve for the optimal stepped heating process parameters. Then, through sinusoidal micro-oscillation effect verification and parameter optimization, a solution heating curve that balances temperature uniformity and equipment feasibility is output. The entire process achieves precise control from feature quantification, risk assessment, safety boundary determination to process parameter optimization and heating curve generation. This fundamentally solves the problems of blistering, uneven solution treatment, and temperature over-control caused by insufficient consideration of geometric and defect features in traditional solution heat treatment of aluminum alloy die-cast parts. While ensuring the solution strengthening effect, it reduces the risk of heat treatment failure and improves process adaptability and production stability.

[0121] It should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A heat treatment strengthening process for die-cast automotive aluminum alloy parts, characterized in that, include: The three-dimensional geometric model of the casting and the defect detection data of historical castings are obtained. Based on the three-dimensional geometric model, the three-dimensional wall thickness field of the casting is calculated, and the engineering feature vector of the casting is extracted. By combining the three-dimensional wall thickness field and the engineering feature vector, the three-dimensional geometric model is divided into feature regions, and the defect detection data is reconstructed based on the feature region division results. Based on the data structure reconstruction results, a dual-channel adaptive fusion analysis is performed to obtain the critical defect feature values ​​of multiple feature regions. Combined with the prior alloy high-temperature mechanical parameters, the global critical temperature of the casting is determined. The feasibility of the process is determined based on the global critical temperature, and the upper limit of the heat preservation time for solution treatment is determined. A heat transfer proxy model is constructed, and the heating process parameters are determined by performing Bayesian optimization search through the heat transfer proxy model, with the global critical temperature and the upper limit of the holding time as constraints. Output the solution heating curve corresponding to the heating process parameters to perform heat treatment on the casting.

2. The heat treatment strengthening process for die-cast automotive aluminum alloy parts as described in claim 1, characterized in that, Acquire the three-dimensional geometric model of the casting and defect detection data of historical castings; calculate the three-dimensional wall thickness field of the casting based on the three-dimensional geometric model; and extract the engineering feature vector of the casting, including: Obtain the three-dimensional geometric model of the casting; By using the historical inspection database of the target scenario and combining it with a preset traceability period, the defect inspection data is extracted. The defect inspection data includes at least the associated spatial location of the defect, dimensional features, heat treatment information of the corresponding casting, and inspection result labels. The three-dimensional geometric model is subjected to adaptive voxelization, and the wall thickness is calculated by traversing multiple voxels based on the voxelization result. The wall thickness calculation is implemented based on either the ray intersection method or the median transformation method. Based on the voxel mesh corresponding to the voxelization processing results, the wall thickness calculation results are stored in a structured manner to obtain the three-dimensional wall thickness field. The three-dimensional wall thickness field, the three-dimensional geometric model, and the defect detection data are jointly analyzed to extract the engineering feature vector. The engineering feature vector includes at least the global wall thickness feature, the deep cavity structure feature, the rib structure feature, the heat transfer efficiency feature, the thermal bottleneck feature, and the defect location feature.

3. The heat treatment strengthening process for die-cast automotive aluminum alloy parts as described in claim 1, characterized in that, Combining the three-dimensional wall thickness field and the engineering feature vector, the three-dimensional geometric model is divided into feature regions, and the defect detection data is reconstructed based on the feature region division results, including: Based on the wall thickness range and structural feature type, the three-dimensional geometric model is divided into multiple feature regions, wherein the structural feature types include deep cavity structure region, rib structure region and thin wall transition region; Each defect in the defect detection data is assigned to its corresponding feature region according to its spatial coordinates; The castings were divided into qualified and unqualified groups based on the heat treatment results. For each region in the qualified and unqualified groups, a parameterized feature probability distribution of historically detected defects within the region is fitted, wherein the parameterized feature probability distribution includes at least the distribution of equivalent radius, surface distance, radius-distance ratio, and defect density; The parameterized feature probability distribution is associated with the qualified and unqualified groups and output to obtain the data structure reconstruction result.

4. The heat treatment strengthening process for die-cast automotive aluminum alloy parts as described in claim 1, characterized in that, Based on the data structure reconstruction results, a dual-channel adaptive fusion analysis is performed to obtain the critical defect characteristic values ​​of multiple feature regions. Combined with prior alloy high-temperature mechanical parameters, the global critical temperature of the casting is determined, including: For each feature region, extract the qualified group and the unqualified group from the data structure reconstruction result respectively; Based on the defect data of the qualified group and the unqualified group, a dual-channel parallel analysis is performed by combining the ROC curve method and the logistic regression method, and the analysis results are fused through an adaptive fusion strategy to obtain the critical defect feature value of each feature region. Based on the critical defect characteristic values ​​of all feature regions, and combined with the high-temperature mechanical property parameters of the alloy, the global critical temperature constraint is determined.

5. The heat treatment strengthening process for die-cast automotive aluminum alloy parts as described in claim 1, characterized in that, Based on the global critical temperature, process feasibility is assessed, and the upper limit of the holding time for solution treatment is determined, including: The global critical temperature is compared with the preset effective lower limit of the solution temperature; If the global critical temperature is less than the lower limit of the effective temperature of the solution treatment, it is determined that the casting cannot be safely solution treated under the current defect conditions, and process feasibility warning information is output and the process optimization process is terminated. If the global critical temperature is greater than or equal to the lower limit of the effective solution temperature, the casting is determined to be suitable for solution treatment. Based on the temperature difference between the global critical temperature and the lower limit of the effective solution temperature, combined with the prior alloy high-temperature creep parameters and the preset target creep strain limit, the upper limit of the holding time for solution treatment is determined.

6. The heat treatment strengthening process for die-cast automotive aluminum alloy parts as described in claim 1, characterized in that, Constructing a heat transfer proxy model includes: Construct and train an autoencoder-based feature compression layer, which is used to compress and reconstruct the three-dimensional wall thickness field; The heat transfer proxy model is constructed by concatenating the compressed reconstruction output of the feature compression layer with the engineering feature vector as the first input, the heating process parameters as the second input, and the key temperature index of the casting as the output. The heat transfer proxy model is iteratively trained by combining the heat treatment information and the detection result labels in the defect detection data.

7. The heat treatment strengthening process for die-cast automotive aluminum alloy parts as described in claim 1, characterized in that, Using the global critical temperature and the upper limit of the holding time as constraints, the heating process parameters are determined through Bayesian optimization search using the heat transfer surrogate model, including: The number of heating steps is determined based on the engineering feature vector. Based on the heating step number, an optimization problem is defined, wherein the optimization problem uses the holding time and heating rate corresponding to the heating step number as optimization variables, and minimizes the final maximum temperature difference output by the heat transfer proxy model as the optimization objective; By combining the global critical temperature, the upper limit of the heat preservation time, and the heating control characteristics of the target heating equipment, constraints are determined, and a Bayesian optimization search is performed on the optimization problem to obtain the heating process parameters.

8. The heat treatment strengthening process for die-cast automotive aluminum alloy parts as described in claim 1, characterized in that, Output the solution heating curve corresponding to the heating process parameters, and perform heat treatment on the casting, including: Based on the heating process parameters and combined with the heating control characteristics of the target heating equipment, a preset sinusoidal micro-oscillation is superimposed during the heat preservation stage; The heat transfer proxy model calculates the final temperature field with and without sinusoidal micro-oscillations, and calculates the maximum improvement in the final temperature difference accordingly. If the improvement in the final maximum temperature difference is greater than or equal to the preset improvement threshold, the micro-oscillation is determined to be effective. The amplitude and period of the sinusoidal micro-oscillation are used as optimization variables. Based on the Bayesian optimization method, the optimal micro-oscillation parameters are searched for in the heat preservation stage, and the solution heating curve of the step-micro-oscillation is output. If the improvement in the final maximum temperature difference is less than the preset improvement threshold, the micro-oscillation is deemed invalid, and the solution heating curve of the stepped heating is directly output based on the heating process parameters.

9. The heat treatment strengthening process for die-cast automotive aluminum alloy parts as described in claim 8, characterized in that, The search for optimal micro-oscillation parameters during the heat preservation stage based on the Bayesian optimization method also includes: Set a set of constraints including upper limit of amplitude, lower limit of amplitude, upper limit of period, and lower limit of period.

10. The heat treatment strengthening process for die-cast automotive aluminum alloy parts as described in claim 1, characterized in that, Also includes: Based on the actual heat treatment test results of the casting after performing the solution heating curve, update the heat treatment information and test result labels in the defect detection data; Based on preset data update trigger conditions, update trigger judgment is performed on the updated defect detection data; If the data update triggering condition is met, the heating process parameters are updated based on the updated defect detection data; If the data update triggering condition is not met, the current heating process parameters will remain unchanged.

Images