A low-pressure casting aluminum alloy wheel microstructure-mechanical property prediction method and related device

Abstract

The application discloses a low-pressure casting aluminum alloy wheel microstructure-mechanical property prediction method and a related device, relates to the aluminum alloy wheel metal forming and industrial artificial intelligence technical field, and the method comprises the following steps: firstly, obtaining the mold temperature sequence of at least one sampling area on an aluminum alloy wheel in a low-pressure casting process, and performing pretreatment; then, respectively inputting the pretreated mold temperature sequence of each sampling area into a pre-trained microstructure prediction model to obtain the secondary dendrite arm spacing prediction value of each sampling area; then, respectively inputting the pretreated mold temperature sequence of each sampling area and the secondary dendrite arm spacing prediction value into a pre-trained corresponding mechanical property prediction model to obtain at least one mechanical property prediction value of each sampling area. Compared with the prior art, the application has the remarkable advantages of non-destructiveness, real-time performance and high efficiency.

Description

Technical Field

[0001] This application relates to the fields of aluminum alloy wheel metal forming and industrial artificial intelligence technology, and in particular to a method and related device for predicting the microstructure and mechanical properties of low-pressure cast aluminum alloy wheels. Background Technology

[0002] Aluminum alloys are widely used in automobile wheel manufacturing due to their low density, good casting performance, strong corrosion resistance, and excellent mechanical properties. Currently, the assessment of the microstructure and mechanical properties of aluminum alloy wheels typically relies on traditional metallographic analysis and tensile testing. These methods are time-consuming, costly, and susceptible to human error during testing and interpretation, constituting destructive testing that fails to meet the real-time quality control and rapid feedback requirements of low-pressure casting production lines.

[0003] In addition, some methods have attempted to establish the relationship between process parameters and microstructure properties using empirical formulas, statistical regression, or numerical simulation, but they generally suffer from the following problems: 1) They rely on data obtained under laboratory conditions, making it difficult to characterize the fluctuation characteristics in the actual production environment; 2) They do not adequately mine mold temperature data with multiple measurement points and obvious time series, making it difficult to fully express the complex relationship between temperature field evolution and microstructure formation; 3) When using a unified modeling strategy for different mechanical property indicators, it is difficult to simultaneously take into account the prediction accuracy of different indicators. Summary of the Invention

[0004] The purpose of this application is to provide a method and related apparatus for predicting the microstructure and mechanical properties of low-pressure cast aluminum alloy wheels, which can improve the detection efficiency and accuracy of the microstructure and mechanical properties of aluminum alloy wheels.

[0005] To achieve the above objectives, this application provides the following solution: Firstly, this application provides a method for predicting the microstructure and mechanical properties of low-pressure cast aluminum alloy wheels, including: Obtain the mold temperature sequence of at least one sampling area on an aluminum alloy wheel during the low-pressure casting process; Each mold temperature sequence is preprocessed to obtain the preprocessed mold temperature sequence for each sampling area; The pre-processed mold temperature sequence of each sampling area is input into the pre-trained microstructure prediction model to obtain the predicted value of the secondary dendrite arm spacing of each sampling area. The pre-processed mold temperature sequence and the predicted value of the secondary dendrite arm spacing of each sampling area are input into the pre-trained corresponding mechanical property prediction model to obtain at least one mechanical property prediction value for each sampling area; wherein each mechanical property prediction value corresponds to an independent mechanical property prediction model.

[0006] In a second aspect, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the microstructure-mechanical property prediction method for low-pressure cast aluminum alloy wheels described in the first aspect.

[0007] Thirdly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the microstructure-mechanical property prediction method for low-pressure cast aluminum alloy wheels described in the first aspect.

[0008] Fourthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the microstructure-mechanical property prediction method for low-pressure cast aluminum alloy wheels described in the first aspect.

[0009] According to the specific embodiments provided in this application, this application has the following technical effects: This application provides a method and related apparatus for predicting the microstructure and mechanical properties of low-pressure cast aluminum alloy wheels. First, the mold temperature sequence of at least one sampling area on the aluminum alloy wheel during the low-pressure casting process is obtained and preprocessed. Then, the preprocessed mold temperature sequence of each sampling area is input into a pre-trained microstructure prediction model to obtain the predicted value of the secondary dendrite arm spacing of each sampling area. Next, the preprocessed mold temperature sequence and the predicted value of the secondary dendrite arm spacing of each sampling area are input into the corresponding pre-trained mechanical property prediction model to obtain at least one predicted value of mechanical property for each sampling area. Each predicted value of mechanical property corresponds to an independent mechanical property prediction model. Compared to existing technologies, which suffer from low detection efficiency, destructive nature, and difficulty in characterizing fluctuations in actual production environments, the detection method based on mold temperature and machine learning described in this application utilizes online mold temperature data acquired during actual production to achieve rapid prediction of the microstructure and mechanical properties of wheels. This method offers significant advantages such as non-destructiveness, real-time performance, high efficiency, and high accuracy. Furthermore, each detection indicator uses an independent prediction model, which can further improve the accuracy of the detection. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1This is a diagram illustrating the application environment of the microstructure-mechanical property prediction method for low-pressure cast aluminum alloy wheels in one embodiment of this application. Figure 2 A flowchart illustrating the microstructure and mechanical properties prediction method for low-pressure cast aluminum alloy wheels provided in an embodiment of this application. Figure 3 A flowchart of the overall method for predicting the microstructure and mechanical properties of low-pressure cast A356 aluminum alloy wheels provided in another embodiment of this application; Figure 4 A schematic diagram of the overall framework of the microstructure-mechanical property prediction method for low-pressure cast A356 aluminum alloy wheels provided in another embodiment of this application; Figure 5 A schematic diagram showing the specific sampling location of a wheel provided in an embodiment of this application; Figure 6 A schematic diagram comparing the prediction effects of different models on the secondary dendrite arm spacing provided in an embodiment of this application; Figure 7 This is a schematic diagram of the yield strength random forest prediction results provided in an embodiment of this application; Figure 8 This is a schematic diagram of the tensile strength support vector regression prediction results provided in an embodiment of this application; Figure 9 A schematic diagram of the elongation support vector regression prediction results provided in an embodiment of this application; Figure 10 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0012] Current technologies are insufficient for mining multi-point, time-series mold temperature data, making it difficult to fully express the complex relationship between temperature field evolution and microstructure formation. The industry urgently needs a method that can effectively predict the secondary dendrite arm spacing and mechanical properties of wheel products based solely on real-time mold temperature data from actual production.

[0013] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0014] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0015] The microstructure-mechanical property prediction method for low-pressure cast aluminum alloy wheels provided in this application embodiment can be applied to, for example... Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be set up independently, integrated into server 104, or placed in the cloud or on another server. Terminal 102 can send the mold temperature sequence of at least one sampling area on the aluminum alloy wheel during the low-pressure casting process to server 104. After receiving the data, server 104 preprocesses each mold temperature sequence to obtain preprocessed mold temperature sequences for each sampling area; it then inputs the preprocessed mold temperature sequences for each sampling area into a pre-trained microstructure prediction model to obtain predicted values ​​for the secondary dendrite arm spacing of each sampling area; finally, it inputs the preprocessed mold temperature sequences and their predicted secondary dendrite arm spacing values ​​for each sampling area into a pre-trained corresponding mechanical property prediction model to obtain at least one predicted mechanical property value for each sampling area. Server 104 can then feed back the obtained predicted values ​​for the secondary dendrite arm spacing and at least one predicted mechanical property value for each sampling area to terminal 102. In addition, in some embodiments, the microstructure-mechanical property prediction method for low-pressure cast aluminum alloy wheels can also be implemented by the server 104 or the terminal 102 separately. For example, the terminal 102 can directly process the mold temperature sequence of at least one sampling area on the aluminum alloy wheel during the low-pressure casting process, or the server 104 can obtain the mold temperature sequence of at least one sampling area on the aluminum alloy wheel during the low-pressure casting process from the data storage system and process it.

[0016] The terminal 102 can be, but is not limited to, various desktop computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. The server 104 can be implemented using a standalone server or a server cluster consisting of multiple servers, or it can be a cloud server.

[0017] In one exemplary embodiment, such as Figure 2 As shown, a method for predicting the microstructure and mechanical properties of low-pressure cast aluminum alloy wheels is provided. This method is executed by computer equipment, specifically by a terminal or server alone, or by both a terminal and a server. In this embodiment, the method is applied to... Figure 1 Taking server 104 as an example, the explanation includes the following steps 201 to 204. Wherein: Step 201: Obtain the mold temperature sequence of at least one sampling area on the aluminum alloy wheel during the low-pressure casting process.

[0018] Step 202: Preprocess each mold temperature sequence to obtain the preprocessed mold temperature sequence for each sampling area.

[0019] Step 203: Input the pre-processed mold temperature sequence of each sampling area into the pre-trained microstructure prediction model to obtain the predicted value of the secondary dendrite arm spacing of each sampling area.

[0020] Step 204: Input the pre-processed mold temperature sequence and the predicted value of the secondary dendrite arm spacing of each sampling area into the pre-trained corresponding mechanical property prediction model to obtain at least one mechanical property prediction value for each sampling area; wherein each mechanical property prediction value corresponds to an independent mechanical property prediction model.

[0021] As an optional implementation, the microstructure prediction model is the TransformerBiGRU model.

[0022] As an optional implementation, when the predicted mechanical properties include the predicted yield strength, the mechanical property prediction model corresponding to the predicted yield strength is a random forest model.

[0023] As an optional implementation, when the predicted mechanical properties include the predicted tensile strength, the mechanical property prediction model corresponding to the predicted tensile strength is a support vector regression model based on the radial basis function kernel.

[0024] As an optional implementation, when the predicted mechanical properties include the predicted elongation, the mechanical property prediction model corresponding to the predicted elongation is a support vector regression model based on the radial basis function kernel.

[0025] As an optional implementation, the preprocessing includes one or more of the following operations: missing value imputation, outlier removal, data standardization, or data normalization.

[0026] As an optional implementation, the process of constructing the dataset for training the microstructure prediction model and the mechanical property prediction model includes: 1-1) Obtain the original dataset, specifically including: collecting the mold temperature sequence of each sampling area on several aluminum alloy wheels during the low-pressure casting process; for each sampling area of ​​each aluminum alloy wheel, obtain the corresponding secondary dendrite arm spacing measurement data and mechanical property measurement data through metallographic inspection and tensile test.

[0027] 1-2) The original dataset is preprocessed to obtain a preprocessed dataset; the preprocessed dataset is used for training the microstructure prediction model and the mechanical property prediction model.

[0028] To aid understanding by those skilled in the art, the following embodiments further illustrate the points.

[0029] This embodiment focuses on low-pressure cast A356 aluminum alloy wheels (also referred to as A356 aluminum wheels in this article).

[0030] Low-pressure casting is an important forming process for A356 aluminum alloy wheels. The thermal history during its solidification process significantly affects the microstructure evolution of the casting and the final mechanical properties.

[0031] In the low-pressure casting process, mold temperature is an important parameter reflecting the heat transfer state, cooling conditions and local solidification behavior. Moreover, compared with microstructure testing and mechanical testing, mold temperature is easier to obtain continuously, online and in real time on the production site.

[0032] Secondary dendrite arm spacing is a crucial parameter for evaluating the refinement of the solidification microstructure in A356 aluminum alloy. Its magnitude is closely related to the local cooling rate and further affects mechanical properties such as yield strength, tensile strength, and elongation. If the secondary dendrite arm spacing and mechanical properties of critical areas can be predicted based on online mold temperature data acquired during production, then rapid assessment of casting quality will be of great significance.

[0033] To address the problem of difficulty in quickly and accurately predicting the quality of cast products during actual production in existing technologies, this embodiment provides a prediction method based on machine learning, which predicts "mold temperature-microstructure-mechanical properties". This method is applicable to various A356 aluminum alloy wheel types that include multiple temperature measurement points and can effectively achieve rapid assessment of casting quality.

[0034] refer to Figure 3 and Figure 4 The technical solution of this embodiment includes: S1, collecting low-pressure casting (i.e.) Figure 3 The mold temperature data during the actual production process of A356 aluminum wheels (die casting) was collected. Samples were selected near the temperature measurement point (i.e., the sampling area), and their microstructure characterization parameters, namely the secondary dendrite arm spacing data (i.e.,... Figure 3 The dataset consists of SDAS (Simplified Chinese SDAS), and mechanical property data, including yield strength, tensile strength and elongation. The correspondence between the three types of data is established to construct the original dataset.

[0035] S2. Preprocess the original dataset, including missing value imputation, outlier removal, data matching, and data standardization or normalization, to obtain the modeling dataset.

[0036] S3. Based on the modeling dataset, establish a prediction model for the secondary dendrite arm spacing. In the selected optimal prediction model, use the mold temperature as input and the secondary dendrite arm spacing as output for prediction.

[0037] S4. Based on the modeling dataset, establish a mechanical property prediction model. In the selected optimal prediction model, use the mold temperature data and the secondary dendrite arm spacing prediction results obtained in step S3 as inputs to predict the mechanical properties of the sample.

[0038] S5. Obtain A356 aluminum wheels (i.e., ...) during actual production. Figure 3 The temperature profile of the casting mold for the unknown sample (i.e.) Figure 3 After obtaining the temperature data, input it into the model described in steps S3 and S4, and output the predicted values ​​of SDAS, yield strength, tensile strength and elongation near the temperature measurement point.

[0039] Compared with existing technologies, the improvements of this embodiment are as follows: 1) Modeling based on real data from the actual production process better reflects the fluctuations in working conditions at the low-pressure casting A356 aluminum alloy wheel production site, improving the model's engineering applicability. 2) Predicting SDAS and mechanical properties solely based on online mold temperature data eliminates the need for offline metallographic testing and tensile testing for each sample, offering non-destructive, real-time, and high-efficiency characteristics. 3) Employing a cascade prediction path of "mold temperature—secondary dendrite arm spacing—mechanical properties" aligns with the evolutionary logic of "process—structure—property" during casting, demonstrating good physical rationality. 4) Differentiated model configurations are used for different prediction targets. The microstructure prediction model uses TransformerBiGRU, the yield strength prediction uses a random forest model, and the tensile strength and elongation prediction uses a support vector regression model, which helps improve the prediction accuracy and generalization ability of each indicator. 5) It can provide data support for quality early warning, process parameter optimization, and production process control in low-pressure casting production lines, showing promising application prospects.

[0040] The following sections will introduce the data acquisition and dataset establishment, data preprocessing, the "mold temperature - secondary dendrite arm spacing" prediction model, the mechanical property prediction model, online prediction and application, and system deployment stages in this embodiment.

[0041] (1) Data collection and dataset establishment.

[0042] In this embodiment, the research object is a low-pressure cast A356 aluminum alloy wheel. The chemical composition of A356 aluminum alloy, by mass percentage, includes: Si 6.5%–7.5%, Mg 0.25%–0.45%, Fe≤0.20%, Ti 0.08%–0.20%, Sr 0.01%–0.03%, with the balance being Al and unavoidable impurities.

[0043] The data collection targets were A356 aluminum alloy wheel castings produced under continuous and stable production conditions on the factory production line. Six sampling areas were set up for each wheel, such as... Figure 5 As shown, each region corresponds to one set of mold temperature curve data, one secondary dendrite arm spacing value, and one set of mechanical property values. The above three types of data together constitute the original dataset.

[0044] Explanation of the setting of 6 sampling areas in this embodiment: Because the overall structure, mold cavity, and cooling and temperature measurement arrangements of the A356 aluminum alloy wheel have a certain degree of symmetry in the circumferential direction, under continuous and stable production conditions, the temperature variation patterns, solidification characteristics, and mechanical properties of the mold at symmetrical positions on the wheel are similar. Therefore, as... Figure 5 As shown, selecting six typical areas covering half of the wheel can represent the changes in microstructure and properties of the main structural parts of the wheel under typical cooling conditions. Furthermore, this sampling method can reduce the number of repeated samples and destructive tests, thus saving on testing costs.

[0045] The mold temperature data acquisition equipment consists of K-type thermocouples calibrated according to national standards. These thermocouples are installed in each sampling area, with a sampling frequency of 10 Hz, and the sampling time covers the entire low-pressure casting cycle. To ensure the comparability of data from different batches, process data such as casting number, production time, holding pressure, and holding time are recorded simultaneously for each production run.

[0046] Metallographic samples from the sampling area were obtained by wire cutting, and after mounting, polishing, and etching, the α-Al dendrite morphology was observed under an optical microscope. The secondary dendrite arm spacing was measured using image analysis software. At least five micrographs were taken from each sampling area, and at least 10 valid SDAS values ​​were measured for each micrograph. The average value was taken as the secondary dendrite arm spacing characterization value for that area, in micrometers.

[0047] Mechanical property test specimens were taken from the region corresponding to the metallographic sampling location. Tensile specimens of appropriate size were prepared by wire cutting according to the national standard GB / T 228.1 "Metallic materials, tensile testing—Part 1: Test at room temperature". After grinding and polishing, tensile testing was conducted using a tensile testing machine to obtain the yield strength, tensile strength, and elongation data at that location. The yield strength is in MPa, the tensile strength is in MPa, and the elongation is in %.

[0048] After clarifying the correspondences of the data obtained above, construct the original dataset according to the common dataset format for machine learning.

[0049] (2) Data preprocessing.

[0050] After importing the raw data into the database, duplicate records and invalid records caused by sensor malfunctions are first deleted. For individual abnormal sampling points, linear interpolation or the mean of adjacent time points is used to complete the data. Samples with an abnormality rate exceeding 10% are discarded.

[0051] The mold temperature time series is denoised, preferably by using Savitzky-Golay filtering or moving average filtering to reduce the impact of on-site electromagnetic interference and measurement jitter on model training.

[0052] Anomaly sample identification employs a combined discrimination strategy: first, the 3σ method is used to remove obviously abnormal temperature records; second, the box plot method is used to identify local abnormal distributions; and finally, the isolated forest algorithm is used to identify anomaly samples that deviate from the overall process rules.

[0053] In terms of modeling datasets, Z-score standardization was applied to the input of the TransformerBiGRU model, while Z-score standardization or max-min normalization was applied to the input of the random forest and support vector regression models to improve the stability of kernel function fitting.

[0054] (3) Prediction model of “mold temperature - secondary dendrite arm spacing”.

[0055] After data preprocessing, a prediction model for "mold temperature - secondary dendrite arm spacing" is established. The training set, validation set, and test set are divided in an 8:1:1 ratio. That is, the training set and validation set data are used for modeling, pre-training, and parameter optimization during the model building phase, and the test set is used for accuracy evaluation after the model is built.

[0056] To obtain the optimal model for this task, support vector regression, random forest, Transformer, BiGRU, and TransformerBiGRU models were compared, and the coefficient of determination R was used as the evaluation metric. 2 Root mean square error (RMSE) and mean absolute error (MAE).

[0057] like Figure 6 As shown, the model training results indicate that the TransformerBiGRU model performs best in predicting the secondary dendrite arm spacing on the test set, with a determination coefficient R0. 2The R² value was 0.932, the MAE was 1.87 μm, and the RMSE was 2.46 μm. In comparison, the BiGRU model had a lower R² value. 2 The R-value of the Transformer model is 0.905. 2 The R value for the random forest model is 0.911. 2 The R-value is 0.884, indicating a support vector regression model. 2 The value is 0.873. Therefore, TransformerBiGRU is determined to be the optimal prediction model for "mold temperature - secondary dendrite arm spacing".

[0058] The TransformerBiGRU model comprises: an input layer, a linear mapping layer, a positional encoding layer, a 2-layer Transformer encoder, a 2-layer BiGRU layer, a Dropout layer, and a fully connected regression output layer. The Transformer encoder has 4 attention heads, the Transformer has a hidden dimension of 128, the BiGRU has a unidirectional hidden dimension of 64, the Dropout coefficient is 0.2, the optimizer is Adam, the initial learning rate is 0.001, the batch size is 32, the maximum number of training epochs is 200, and early stopping is used to prevent overfitting.

[0059] The specific link relationships between the layers of this model are as follows: The output of the input layer is connected to the input of the linear mapping layer, the output of the linear mapping layer is connected to the input of the position encoding layer, the output of the position encoding layer is connected to the input of the first Transformer encoder layer, the output of the first Transformer encoder layer is connected to the input of the second Transformer encoder layer, the output of the second Transformer encoder layer is connected to the input of the first BiGRU layer, the output of the first BiGRU layer is connected to the input of the second BiGRU layer, the output of the second BiGRU layer is connected to the input of the Dropout layer, and the output of the Dropout layer is connected to the input of the fully connected regression output layer. The fully connected regression output layer outputs the predicted value of the secondary dendrite arm spacing.

[0060] Through the above connection method, the model first uses the Transformer encoder layer to extract the global correlation features between multiple measurement points and multiple time points in the mold temperature sequence, then uses the BiGRU layer to further extract the bidirectional dynamic features of the temperature sequence in the time dimension, and finally uses the Dropout layer for regularization processing, and the fully connected regression output layer completes the prediction of the secondary dendrite arm spacing.

[0061] The Transformer encoder layer and BiGRU layer are explained in detail below: Each Transformer encoder layer is used to extract the global dependencies between different measurement points and different time points in the mold temperature sequence. Its attention mechanism can be represented as: ; In the formula: Represents the query matrix; Represents the key matrix; Represents a value matrix; The dimension of the bond vector is represented; the superscript T indicates transpose. This structure can learn the correlation between temperature evolution in different mold regions from a global perspective, making it suitable for handling coupled information of mold temperature fields at multiple measurement points.

[0062] Each BiGRU layer is used to extract bidirectional temporal features of the temperature sequence, and its forward and reverse hidden states are as follows: ; in, and Representing time respectively and time The hidden state; Indicates time Input features; symbols Indicates a positive hidden state; symbol Indicates the reverse hidden state; This represents the state update function of the gated loop unit.

[0063] This structure can simultaneously utilize information from before and after the mold temperature sequence, thereby improving the ability to express the entire solidification process of aluminum alloys.

[0064] By stacking two layers, the level of abstraction of features and the range of receptive fields can be improved.

[0065] After feature extraction via two Transformer layers and two BiGRU layers, the predicted secondary dendrite arm spacing is output by a fully connected regression layer: ; In the formula: Indicates the first The deep feature vector obtained by integrating the output of the last layer BiGRU of each sample; b represents the regression layer weights; b represents the bias term. Indicates the first Predicted secondary dendrite arm spacing for each sample.

[0066] The TransformerBiGRU model first uses the Transformer layer to extract the global correlation between temperature features at multiple measurement points, then uses the BiGRU layer to extract the bidirectional dependence of the temperature sequence in the time dimension, and finally outputs the predicted value of the secondary dendrite arm spacing through the regression layer.

[0067] (4) Mechanical performance prediction model.

[0068] After obtaining the prediction model for the secondary dendrite arm spacing, further prediction models for yield strength, tensile strength, and elongation are established. The input features of the mechanical property models include: the mold temperature curve and the predicted value of the secondary dendrite arm spacing obtained from the previous stage.

[0069] For the yield strength prediction task, models including linear regression, ridge regression, lasso regression, random forest, gradient boosting, and support vector regression were tested. The training results show that the yield strength prediction model achieves R0 on the test set. 2 The yield strength prediction model is selected based on the random forest model, which has a yield strength of 0.918, a MAE of 5.6 MPa, and an RMSE of 7.8 MPa, outperforming other models.

[0070] Random forest models learn the nonlinear mapping between input features and yield strength by constructing multiple decision trees and integrating their outputs. Let a random forest consist of M decision trees, then the... The predicted yield strength values ​​for each sample are: ; In the formula, This represents the input feature vector of the model; Indicates the first The output of the regression tree; This indicates the total number of regression trees; This represents the predicted yield strength.

[0071] Random forests reduce the overfitting risk of a single decision tree through bootstrap sampling and random feature selection mechanisms, making them suitable for fitting complex nonlinear relationships between yield strength and modulus temperature and tissue characteristics. In this random forest model, the number of decision trees is set to 300, the maximum depth to 12, and the minimum number of leaf node samples to 3. Five-fold cross-validation is used for parameter optimization. Figure 7 The image shows the prediction results of the model.

[0072] For the task of predicting tensile strength and elongation, linear regression, ridge regression, lasso regression, random forest, gradient boosting, and support vector regression were tested respectively. The results show that the support vector regression model using the radial basis function kernel function has the best prediction performance for both. The R-value of the tensile strength prediction model on the test set is [not specified]. 2The R-value is 0.904, MAE is 6.9 MPa, and RMSE is 9.1 MPa. The elongation prediction model achieves R-value on the test set. 2 The value was 0.881, MAE was 0.58%, and RMSE was 0.74%. Figures 8 to 9 As shown, the model prediction results are as follows. Since the elongation is more significantly affected by factors such as shrinkage porosity, local segregation and micro-defects, its prediction accuracy is slightly lower than that of the strength index, but it still meets the requirements of online quality assessment.

[0073] Support vector regression models map input features to a high-dimensional feature space using kernel functions, achieving regression fitting while keeping model complexity under control. Its regression function... It can be represented as: ; In the formula, The sample to be predicted; For the first One training sample; Indicates the total number of training samples; Indicates the first Each training sample corresponds to a Lagrange multiplier with an upper boundary constraint on an insensitive loss function; Indicates the first Each training sample corresponds to a Lagrange multiplier with a lower boundary constraint that is insensitive to the loss function; For bias terms; This is the kernel function.

[0074] Support vector regression models are well-suited for handling nonlinear fitting problems with small sample sizes, making them appropriate for tensile strength and elongation prediction tasks. Penalty parameters for tensile strength prediction models. Set to 80, kernel function parameter Set to 0.06, loss tolerance coefficient Set to 0.08. For the elongation prediction task, the parameters are set to... =60, =0.08, =0.05.

[0075] (5) Online prediction and application.

[0076] In actual production, when an unknown product only has online collected mold temperature data, the standardization process is first completed according to the method of “(2) data preprocessing”.

[0077] The processed mold temperature data is input into the TransformerBiGRU model to obtain the predicted value of the secondary dendrite arm spacing of the corresponding part (sampling area); then the mold temperature characteristics and the predicted value of the secondary dendrite arm spacing are input into the yield strength random forest model, the tensile strength support vector regression model and the elongation support vector regression model to obtain the predicted results of yield strength, tensile strength and elongation of the corresponding part.

[0078] In a set of online validation samples, after a set of mold temperature sequences were input into the system, the model output the secondary dendrite arm spacing, yield strength, tensile strength, and elongation of the region. Subsequent comparison with measured results showed that the error could be stably controlled within 10%, indicating that this embodiment can achieve relatively accurate online prediction.

[0079] To enable production early warning, minimum or maximum thresholds can be set for each indicator. When any predicted value is not at the set quality threshold, the system can issue an early warning, indicating that there is a risk of the wheel quality failing to meet the standard, and reminding the operators to make process adjustments.

[0080] (6) System deployment.

[0081] This method can be deployed on industrial computers, edge computing servers, or manufacturing execution systems in low-pressure casting production lines, and can be linked with on-site temperature acquisition modules, database modules, and quality early warning modules.

[0082] The data acquisition module is responsible for reading thermocouple signals from each mold and simultaneously recording the production batch number; the data processing module is responsible for data cleaning, time series alignment, standardization, and normalization; the prediction module calls the TransformerBiGRU model to output the predicted value of the secondary dendrite arm spacing; the mechanical property prediction module calls the random forest and support vector regression models respectively to output the predicted values ​​of yield strength, tensile strength, and elongation; the result display module displays the predicted values ​​and alarm status on the host computer interface and writes them to the database. Under edge deployment conditions, the complete prediction time for a single wheel is about 3 seconds, which can meet the online application requirements of low-pressure casting production cycle.

[0083] This solution is particularly suitable for online quality assessment during the production process of low-pressure cast A356 aluminum alloy wheels, and can also be extended to quality prediction scenarios for similar cast aluminum alloy wheels such as A357 and AlSi7Mg, or similar thin-walled aluminum alloy castings.

[0084] Compared to traditional methods that require metallographic examination and tensile testing for each sample, this approach can predict the SDAS and mechanical properties of a product solely based on online mold temperature data. It is also the first time that mold temperature has been used as an input feature to predict microstructure properties.

[0085] This scheme adopts a series prediction path of "mold temperature - secondary dendrite arm spacing - mechanical properties", which conforms to the evolution logic of "process - microstructure - properties" in the casting process and has good physical rationality.

[0086] This approach employs differentiated model configurations for different prediction objectives. Microstructure prediction utilizes the TransformerBiGRU machine learning model, yield strength uses a random forest model, and tensile strength and elongation use support vector regression models. Selecting different models for different tasks ensures that each prediction task is matched to its respective data characteristics, effectively improving overall prediction performance and avoiding the uneven adaptation problems caused by a single modeling strategy.

[0087] Technical advantages of this solution: This method accurately reflects the operational fluctuations in the actual production of low-pressure cast A356 aluminum alloy wheels, improving the model's applicability in production line environments. This effectiveness primarily stems from the technical aspect of "building a predictive model based on actual production data." Many existing prediction methods are often based on laboratory conditions or idealized data, resulting in a limited sample environment that fails to cover the complex factors of equipment fluctuations, temperature changes, and batch variations in real production lines. This method directly models based on data collected from the production site, making it easier to maintain stable output when used for predicting unknown products, thus exhibiting stronger engineering adaptability and wider applicability.

[0088] This invention achieves a chain-like online prediction from process to microstructure to performance, improving the physical rationality and reliability of the prediction results. This effect mainly stems from the cascaded technical approach of "first predicting the microstructure through mold temperature, and then combining the two to predict mechanical properties." Existing technologies only utilize the correspondence between microstructure and mechanical properties for prediction. In actual production, obtaining the microstructure still requires manual inspection, which is time-consuming, costly, and destructive. In this application, the inclusion of mold temperature makes the microstructure an intermediate variable in performance prediction, conforming to the physical logic of "process-microstructure-performance," and the method of obtaining the mold temperature is simpler, significantly enhancing the interpretability of the prediction route and the stability of the prediction results—a technical effect that is difficult to achieve with existing technologies.

[0089] This approach improves the prediction accuracy and generalization ability of different quality indicators, making it more suitable for engineering applications. This effect primarily stems from the technique of "adopting differentiated model configurations for different prediction targets." SDAS, yield strength, tensile strength, and elongation differ in data distribution characteristics, nonlinearity, and the weight of their influence on input variables. Existing technologies use a uniform model to process all output indicators, which easily leads to uneven prediction results with significant variations. This solution selects different models for different tasks (indicators), matching each prediction task with its respective data characteristics, thus more effectively improving overall prediction performance and avoiding the uneven adaptation problem caused by a single modeling strategy.

[0090] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 10 As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores the mold temperature sequence of various sampling areas on the aluminum alloy wheel during the low-pressure casting process. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a method for predicting the microstructure and mechanical properties of low-pressure cast aluminum alloy wheels.

[0091] Those skilled in the art will understand that Figure 10 The structures shown are merely block diagrams of some structures related to the present application and do not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than shown in the figures, or combine certain components, or have different component arrangements. In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0092] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0093] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0094] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0095] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0096] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0097] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for predicting the microstructure and mechanical properties of low-pressure cast aluminum alloy wheels, characterized in that, include: Obtain the mold temperature sequence of at least one sampling area on an aluminum alloy wheel during the low-pressure casting process; Each mold temperature sequence is preprocessed to obtain the preprocessed mold temperature sequence for each sampling area; The pre-processed mold temperature sequence of each sampling area is input into the pre-trained microstructure prediction model to obtain the predicted value of the secondary dendrite arm spacing of each sampling area. The pre-processed mold temperature sequence and the predicted value of the secondary dendrite arm spacing of each sampling area are input into the pre-trained corresponding mechanical property prediction model to obtain at least one mechanical property prediction value for each sampling area; wherein each mechanical property prediction value corresponds to an independent mechanical property prediction model.

2. The method for predicting the microstructure and mechanical properties of low-pressure cast aluminum alloy wheels according to claim 1, characterized in that, The microstructure prediction model is the TransformerBiGRU model.

3. The method for predicting the microstructure and mechanical properties of low-pressure cast aluminum alloy wheels according to claim 1, characterized in that, When the predicted mechanical properties include the predicted yield strength, the mechanical property prediction model corresponding to the predicted yield strength is a random forest model.

4. The method for predicting the microstructure and mechanical properties of low-pressure cast aluminum alloy wheels according to claim 1, characterized in that, When the predicted mechanical properties include the predicted tensile strength, the mechanical property prediction model corresponding to the predicted tensile strength is a support vector regression model based on the radial basis kernel function.

5. The method for predicting the microstructure and mechanical properties of low-pressure cast aluminum alloy wheels according to claim 1, characterized in that, When the predicted mechanical properties include the predicted elongation, the mechanical property prediction model corresponding to the predicted elongation is a support vector regression model based on the radial basis kernel function.

6. The method for predicting the microstructure and mechanical properties of low-pressure cast aluminum alloy wheels according to claim 1, characterized in that, The preprocessing includes one or more of the following operations: missing value imputation, outlier removal, data standardization, or data normalization.

7. The method for predicting the microstructure and mechanical properties of low-pressure cast aluminum alloy wheels according to claim 1, characterized in that, The process of constructing the datasets used to train the microstructure prediction model and the mechanical property prediction model includes: The process of obtaining the original dataset includes: collecting the mold temperature sequence of each sampling area on several aluminum alloy wheels during the low-pressure casting process; and obtaining the corresponding secondary dendrite arm spacing measurement data and mechanical property measurement data for each sampling area of ​​each aluminum alloy wheel through metallographic inspection and tensile test. The original dataset is preprocessed to obtain a preprocessed dataset; the preprocessed dataset is used for training the microstructure prediction model and the mechanical property prediction model.

8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that the processor executes the computer program to implement the microstructure-mechanical property prediction method for low-pressure cast aluminum alloy wheels according to any one of claims 1-7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the method for predicting the microstructure and mechanical properties of low-pressure cast aluminum alloy wheels as described in any one of claims 1-7.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the method for predicting the microstructure and mechanical properties of low-pressure cast aluminum alloy wheels as described in any one of claims 1-7.

Images