Mold temperature control method, electronic equipment and storage medium

Abstract

The invention discloses a die temperature control method, electronic equipment and a storage medium, and the method comprises the steps: obtaining an infrared image of a die-casting die and the real-time temperature data of an internal pre-buried thermocouple, fusing the infrared image and the thermocouple data, generating a three-dimensional correction temperature field containing the surface and internal temperature of the die, and storing the three-dimensional correction temperature field in the die-casting die; when the three-dimensional correction temperature field is abnormal, an adjustment value of a target adjustment parameter is output through a parameter prediction model, thermodynamic equation residual constraint is introduced into a loss function when the parameter prediction model is trained, the adjustment value is sent to a mold temperature control system to be executed, a new temperature field is collected in the next production cycle, and if the deviation between the new temperature field and the target temperature field exceeds a threshold value, the mold temperature control system is started. And if so, incrementally adjusting the parameters through a PID algorithm. By means of the die temperature control method, the die temperature of the die-casting model in the die-casting process can be controlled, and the speed and precision of die temperature control can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of die casting, in particular to a mold temperature control method, an electronic device and a storage medium. BACKGROUND

[0002] Due to the requirements of vehicle body strength and weight reduction, the application of vehicle body integrated die casting process is becoming more and more widely used. Among them, accurate mold temperature control plays a key role in reducing die casting defects such as porosity, cold shut, etc. The large die casting mold used in the vehicle body integrated die casting process often has hundreds of water and oil circuits. How to accurately control the mold temperature to reduce casting defects and improve the economy and product quality of the vehicle body integrated die casting process has become a technical problem to be solved in the field. SUMMARY

[0003] In a first aspect, the embodiments of the present application provide a mold temperature control method, wherein the mold temperature control method comprises:

[0004] acquiring an infrared image of a die casting mold and real-time temperature data of an internally embedded thermocouple;

[0005] fusing the infrared image and the thermocouple data to generate a three-dimensional corrected temperature field containing the surface and internal temperature of the mold;

[0006] when the three-dimensional corrected temperature field is abnormal, outputting an adjustment value of a target adjustment parameter through a parameter prediction model, wherein the parameter prediction model introduces a thermodynamic equation residual constraint in the loss function during training;

[0007] sending the adjustment value to a mold temperature control system for execution;

[0008] collecting a new temperature field in the next production cycle, and if the deviation from the target temperature field is greater than a threshold value, incrementally adjusting the parameter through a PID algorithm.

[0009] In this way, compared with the related art, the mold temperature control method provided by the embodiments of the present application realizes accurate judgment of the mold temperature by fusing the three-dimensional corrected temperature field generated by the internal infrared image of the mold and the temperature data of the internal thermocouple point sensor, and then outputs the adjustment value of the adjustment parameter through the parameter prediction model. The parameter prediction model introduces a thermodynamic equation residual constraint during training to avoid the parameter prediction model outputting solutions that violate physical laws and improve the accuracy of the parameter prediction model. At the same time, the deviation value between the temperature field adjusted by the adjustment parameter output by the parameter prediction model and the predicted temperature field is judged, and the PID algorithm is used for fine tuning to realize closed-loop control. Therefore, by the above method steps and their combination, the accuracy of mold temperature control is improved, the decision-making efficiency is improved, and the reliability of the entire system is enhanced.

[0010] In a second aspect, an electronic device is provided, comprising a processor, wherein

[0011] The processor implements the mold temperature control method of the first aspect when executing a program.

[0012] The electronic device is connected with a temperature compensator, and the temperature compensator is used to dynamically calibrate the thermocouple reading according to the ambient temperature.

[0013] In a third aspect, a computer readable storage medium is provided, and the computer readable storage medium stores a program, wherein the program is executed by a processor to control the electronic device of the third aspect to run.

[0014] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, but not limiting the technical solutions of the present application. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 A flowchart of a mold temperature control method provided by the embodiment of the present application;

[0016] Figure 2 A flowchart of a mold temperature control method provided by the embodiment of the present application, which includes a trained parameter prediction model and lightens the parameter prediction model;

[0017] Figure 3 A flowchart of introducing a thermodynamic exothermic residual constraint in a loss function when training a parameter prediction model provided by the embodiment of the present application;

[0018] Figure 4 A flowchart of PID incremental adjustment provided by the embodiment of the present application;

[0019] Figure 5 A flowchart of generating a three-dimensional correction temperature field containing mold surface and internal temperature provided by the embodiment of the present application;

[0020] Figure 6 A flowchart of a mold temperature control method provided by the embodiment of the present application, which includes an artificial correction instruction;

[0021] Figure 7 A flowchart of updating the parameter prediction model based on the artificial correction instruction for reinforcement learning provided by the embodiment of the present application. DETAILED DESCRIPTION

[0022] In order to be able to understand the features and technical contents of the embodiments of the present application more clearly, the implementation of the embodiments of the present application will be described in detail below with reference to the accompanying drawings, and the accompanying drawings are only used for reference and do not limit the embodiments of the present application.

[0023] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to be limiting.

[0024] In the following description, reference is made to the "some embodiments" which describe a subset of all possible embodiments, but it is to be understood that "some embodiments" can be the same subset or a different subset of all possible embodiments and can be combined with each other, without conflict.

[0025] It should also be noted that the terms "first", "second", "third" and the like in the description and in the claims, are used for distinguishing between similar elements and not necessarily for describing a sequential or chronological order. It is to be understood that the terms so used are interchangeable under appropriate circumstances such that the embodiments of the application described herein are, for example, capable of use in either order.

[0026] Also, the use of "example", "comprise" or "comprising", "include" or "including", "have" or "having", "contain" or "containing", "consist of" or "consist of" herein, are used in the sense of "including" rather than "consisting of".

[0027] The embodiments of the present application provide a mold temperature control method, as shown in Figure 1 The mold temperature control method comprises the following steps:

[0028] 100: Obtain the infrared image of the die casting mold and the real-time temperature data of the internally embedded thermocouple;

[0029] 200: Fuse the infrared image and the thermocouple data to generate a three-dimensional corrected temperature field containing the surface and internal temperature of the mold;

[0030] 300: When the three-dimensional corrected temperature field is abnormal, output the adjustment value of the target adjustment parameter through the parameter prediction model, wherein the parameter prediction model introduces the thermodynamic equation residual constraint in the loss function during training;

[0031] 400: Send the adjustment value to the mold temperature control system for execution;

[0032] 500: Collect a new temperature field in the next production cycle, and if the deviation from the target temperature field is greater than the threshold, incrementally adjust the parameter through the PID algorithm.

[0033] For the mold temperature control method provided by the embodiment of the application, the three-dimensional correction temperature field fused by the internal infrared image of the mold and the temperature data of the internal thermocouple point position sensing is obtained, the temperature of the mold is accurately judged, then the adjustment value of the adjustment parameter output by the parameter prediction model is output, and the thermodynamic equation residual constraint is introduced into the parameter prediction model during training, so as to avoid the solution output by the parameter prediction model from violating the physical law, improve the accuracy of the parameter prediction model, and at the same time, the deviation value between the temperature field adjusted by the adjustment parameter output by the parameter prediction model and the predicted temperature field is judged, the PID algorithm is used for fine tuning, and closed-loop control is realized. Therefore, by the above method steps and their combination, the accuracy of the mold temperature control is improved, the decision efficiency is improved, and the reliability of the entire system is enhanced.

[0034] In some embodiments, in order to more accurately obtain more valuable data through the pre-embedded thermocouple in the mold, the thermocouple can be embedded in the deep cavity, the reverse structure and the point cooling pipe outlet of the die casting mold, and the hidden area with a distance of ≤5 mm from the cooling pipe, the wall thickness mutation within ±3 mm range and the corresponding mold cavity position of the casting with a historical defect rate > 10%. Therefore, the temperature value of the point position where the temperature anomaly is easy to occur in the mold, the internal area judged by the mold surface infrared image, or the area where the die casting defect often occurs can be obtained by the thermocouple, thereby improving the accuracy of the three-dimensional correction temperature field reflecting the actual temperature of the mold.

[0035] In some embodiments, as shown in Figure 2 The mold temperature control method further comprises:

[0036] 010: training a parameter mapping model based on a preset training set, and establishing a mapping relationship between the pipeline parameters and the mold temperature;

[0037] 020: training the parameter mapping model to obtain a parameter prediction model, and performing lightweight processing on the parameter prediction model;

[0038] Specifically, in step 010, the preset training set can be composed of CAE (computer aided engineering) simulation data, the parameter mapping model is pre-trained through the CAE simulation data, and the mapping relationship between the pipeline parameters and the mold temperature is established. In step 020, for the parameter mapping model obtained by training, the parameter prediction model is obtained by performing reinforcement learning on the training set composed of the results output by the parameter mapping model and the real-time temperature field data.

[0039] The parameter prediction model training in step 020 further comprises:

[0040] 021: pruning the channels of the neural network of the parameter prediction model, and removing the nodes with a weight absolute value < 0.001;

[0041] 022: model compression of the parameter prediction model;

[0042] 023: deploying the parameter prediction model on an edge computing device.

[0043] Specifically, in step 021, the L1 norm of the neurons of the parameter prediction model can be calculated (L1_morm), and the importance of the neurons is determined according to the result, so as to screen out the nodes with a weight absolute value <0.001. According to experience, this part accounts for 12% to 18% of the total parameters of the model, so as to reduce the parameters to reduce the response delay of the parameter prediction model; in step 022, the FP32 parameters of the parameter prediction model can be converted to INT8, and through compression, the volume of the parameter prediction model can be reduced to less than 30% of the original volume; in step 023, the parameter prediction model is deployed on an edge computing device, and parameter optimization is performed for the edge computing device, so as to reduce the end-to-end response delay.

[0044] In some embodiments, as shown in FIG. 3, a thermodynamic equation residual constraint is introduced in the loss function during training of the parameter prediction model in step 300, specifically: Figure 3

[0045] 310: adding a heat conduction equation residual term in the loss function:

[0046] ;

[0047] wherein = 0.7 to 1.2;

[0048] 320: limiting the search space based on the historical qualified parameter set, and the constraint conditions include that the point cooling time and the flow meet , is a pipe diameter coefficient.

[0049] Specifically, in step 310, and are the temperature field matrix predicted by the parameter prediction model and the real simulation temperature field matrix, respectively. The predicted temperature field matrix corresponds to the number of three-dimensional grid nodes of the mold, and represents the predicted temperature distribution corresponding to the parameter combination calculated by the parameter prediction model according to the input temperature field. The real simulation temperature field matrix is generated by high-precision thermodynamic simulation of the training parameters through CAE, so that it can be used as label data for supervised learning to guide the optimization direction of the model. is a weight coefficient for balancing the prediction accuracy and the compliance with the physical law, and the value range is 0.7 to 0.12. According to experiments, when is 0.9, the comprehensive effect in the mold of the automobile structure part is optimal. ​It is used to characterize the second-order rate of change of the temperature field in space and reflect the diffusion intensity of heat transfer; α is the thermal diffusivity of the mold material; The temperature-time derivative reflects the rate of temperature change per unit time and the mold's heating / cooling rate. This data is calculated by differentially calculating the time series data from the infrared camera and the thermocouple. By adding the residual term of the heat conduction equation to the loss function during parameter prediction model training in step 310, the parameter prediction model can autonomously avoid nonphysical solutions, preventing it from outputting data that does not conform to real-world physical laws.

[0050] In step 320, Minimum volume flow rate to maintain turbulent state (m³ / s), The maximum safe flow rate to prevent water hammer effect (m³ / s), where:

[0051] ;

[0052] ;

[0053] =4000, is the kinematic viscosity of cooling water, is the pressure limit of the pipeline, A is the cross-sectional area of ​​the pipeline; k is the correction factor that characterizes the relationship between pipeline diameter and flow-time, which is related to the size of the pipe diameter. is the on-off time of the cooling point, Through step 320, the parameters output by the parameter prediction model can be kept within a preset range, thereby avoiding insufficient flow or overpressure in the mold cooling pipeline.

[0054] In some embodiments, as Figure 4 As shown, the PID increment adjustment in step 500 is specifically as follows:

[0055] 510: Calculate temperature field deviation matrix ;

[0056] 520: Generate adjustment amount by region:

[0057] ;

[0058] in =0.8, =0.05, =0.15;

[0059] The generated adjustment amount is limited to a single adjustment amplitude not exceeding ±15% of the parameter range.

[0060] Specifically, the temperature field deviation matrix in step 510 is is derived from the difference between the real temperature field of the mold and the target temperature field; in step 520, the proportional coefficient = 0.8, is the area average temperature deviation, reflecting the mean value of the abnormal area , the integral coefficient = 0.05, is the sum of historical deviation accumulation, the differential coefficient = 0.15, is the current deviation matrix, reflecting the latest value, is the last cycle deviation matrix, reflecting the previous cycle value; for the generated adjustment amount, the adjustment amplitude does not exceed ± 15% of the parameter range, which is the safety limit considering the pressure bearing of the pipeline, and by setting the upper limit of the adjustment amplitude, the water hammer effect caused by sudden increase of flow in the pipeline is prevented. Through steps 510, 520 and the limitation of the adjustment amount, the adjustment amount is generated in real time according to the deviation of the three-dimensional modified temperature field, as a supplementary fine adjustment amount to the target adjustment parameter output by the parameter prediction model, and by setting the values of the proportional coefficient, the integral coefficient and the differential coefficient respectively, the output of the adjustment amount is optimized, which can adjust the temperature of the mold faster, reduce the power consumption of the die casting system and reduce the scrap rate.

[0061] And in step 520, when the adjustment amount output by the parameter prediction model for the first time is greater than or equal to 15% for the first time, the coarse adjustment operation is performed, and the parameters of the mold temperature control system are set to the adjustment value output by the parameter prediction model. In the next production cycle after performing coarse adjustment, new temperature field is collected and the difference between the new temperature field and the target temperature field is calculated, i.e. the temperature field deviation matrix When the following conditions are met , the PID fine adjustment layer is activated:

[0062] ;

[0063] Wherein, is the fine output of PID, is the local temperature compensation gain, is the historical cumulative error compensation, is the average temperature difference of the key area, is the temperature deviation time integral, preferably, = 0.3, = 0.02, and when , the emergency shutdown protocol is triggered to stop production. Specifically, in the coarse adjustment stage, the adjustment value output by the parameter prediction model is directly used to adjust the mold temperature control system in full amount to minimize the temperature field deviation matrix greater than 10℃, and reduce the risk of PID overshoot, while when the temperature field deviation matrix between 5℃ and 10℃, it can be considered that through the PID fine-tuning layer, the mold temperature can be ensured within the expected range, ensuring the stable production, while when the temperature field deviation matrix is greater than 10℃, it can be considered that the temperature in the mold has reached the critical temperature of the aluminum alloy melt solidification defect at this time, and the probability of casting defects is close to 100%, and the die casting production should be stopped immediately to protect the mold and equipment. Thus, by layering the coarse and fine tuning of the PID, the adjustment time can be saved, the overshoot amount can be reduced, and the steady-state error can be reduced, thereby improving the production efficiency and reducing the waste rate, and reducing the damage risk of the die casting equipment, the mold temperature control system and the mold.

[0064] In some embodiments, as shown in FIG. 2A, the generating of the three-dimensional corrected temperature field containing the mold surface and internal temperature in step 200 specifically includes: Figure 5

[0065] 210: mapping the thermocouple data to the mold internal body grid temperature nodes;

[0066] 220: fusing the surface infrared temperature and internal node data by Laplace smoothing algorithm;

[0067] 230: using radial basis function interpolation to fill in the blind area in the undercut area.

[0068] Specifically, in step 210, by analyzing the thermocouple data and spatial coordinates, the position of the thermocouple and the spatial coordinates are matched to the three-dimensional body network, thereby generating the temperature nodes inside the mold, wherein the spatial coordinates can be obtained through the CAD model of the mold, and when mapping the thermocouple and the spatial coordinates, a direct attenuation factor is introduced, where d is the depth, thereby solving the problem of heat transfer distortion in the deep cavity area of the mold; in step 220, the Laplace regularization fusion algorithm is used to fuse the mold surface infrared temperature collected by the infrared camera with the mold internal node data collected by the thermocouple, through the following formula:

[0069] ;

[0070] wherein in the discrete solution formula, the matrix represents the surface infrared temperature field of the mold, is the smoothing weight, according to the experimental results, for a general mold, the optimal value is =0.8, is the Laplace matrix, ​I is an identity matrix, and the infrared accuracy of the mold surface and the smoothness of the internal thermocouple data are balanced by the preferred smoothing weight; in step 230, based on the known temperature points, i.e., the surface infrared points of the mold and the thermocouple points inside the mold, the temperature of the inverted blind area inside the mold is predicted by radial basis function (RBF) interpolation to improve the accuracy of the three-dimensional correction temperature field.

[0071] In some embodiments, as shown in Figure 6 the mold temperature control method further comprises:

[0072] 600: Calculate the confidence of the parameter prediction model;

[0073] 700: If the confidence of the parameter prediction model is less than 90%, accept the manual correction instruction and store it in the historical experience database.

[0074] Specifically, in step 600, the confidence is calculated by

[0075] ;

[0076] Calculate the confidence conf, where is the temperature field predicted by the parameter prediction model, is the target temperature field, is the maximum allowed deviation threshold, which can be set to 10°C; in step 700, when the confidence of the parameter prediction model is less than 90%, the operator can be prompted that the parameter output by the parameter prediction model has a deviation by superimposing a deviation heat map on the CAD model, or by describing the deviation value and deviation coordinates in text, etc. The operator can then adjust the temperature of the mold accordingly, and the operator's correction record is automatically clustered and stored in the historical experience database. Through the confidence quantification mechanism, the reliability of the parameter prediction model is objectively measured, and the confidence is used to determine whether the parameter prediction model is fully automatic or needs human intervention, and the data of manual correction is collected, which facilitates the improvement of the model in the future.

[0077] In some embodiments, as shown in Figure 7 the mold temperature control method further comprises:

[0078] 800: When the cumulative number of manual corrections exceeds a threshold, update the parameter prediction model weight based on reinforcement learning.

[0079] Specifically, in step 800, when the number of manual corrections exceeds a threshold, e.g., 5 or 10, etc., an experience pool is constructed based on the recent correction records, reinforcement learning is performed on the parameter prediction model, and a reward function can be constructed, e.g.:

[0080] ;

[0081] wherein, After artificial correction of the parameters, the deviation matrix between the actually measured temperature field and the target temperature field in a new round of die casting production cycle is obtained, so that the parameter prediction model has a continuous learning ability, the experience of artificial correction is integrated into the parameter prediction model, which can not only reduce the demand for artificial intervention, but also improve the confidence of the parameter prediction model, and then improve the accuracy of the mold temperature control.

[0082] The electronic device includes a processor, wherein the processor executes the method provided in any of the above embodiments when executing a program; and the electronic device is connected with a temperature compensator, and the temperature compensator is used for dynamically calibrating the thermocouple reading according to the ambient temperature.

[0083] The electronic device includes a processor, wherein the processor executes the method provided in any of the above embodiments when executing a program; and the electronic device is connected with a temperature compensator, and the temperature compensator is used for dynamically calibrating the thermocouple reading according to the ambient temperature.

[0084] It should be noted that in the present application, the terms "comprising", "including", or any other variant thereof are intended to cover non-exclusive inclusion, so that processes, methods, products or devices including a series of elements not only include those elements, but also include other elements not explicitly listed, or further include elements inherent to such processes, methods, products or devices. Without more limitations, the element defined by the statement "including a" does not exclude the presence of additional identical elements in the process, method, product or device including the element.

[0085] In several embodiments provided by the present application, it should be understood that the disclosed system, device and method can be implemented by other ways. The above-described device embodiments are only schematic, for example, the division of units is only a logical function division, and there can be another division way in actual implementation, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the various components shown or discussed can be through some interface, indirect coupling or communication connection between devices or units, which can be electrical, mechanical or other forms.

[0086] The above are only preferred embodiments of the present application, and are not used to limit the protection scope of the present application, and any modification, equivalent replacement and improvement made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A mold temperature control method, characterized in that: include: Obtain infrared images of die-casting molds and real-time temperature data of embedded thermocouples inside; Fusion of infrared images and thermocouple data to generate a three-dimensional corrected temperature field including mold surface and internal temperatures; When the three-dimensional corrected temperature field is abnormal, the adjustment value of the target adjustment parameter is output through the parameter prediction model. When the parameter prediction model is trained, the residual constraint of the thermodynamic equation is introduced into the loss function. Send the adjustment value to the mold temperature control system for execution; In the next production cycle, a new temperature field is collected. If the deviation from the target temperature field exceeds the threshold, the parameters are incrementally adjusted using the PID algorithm.

2. The mold temperature control method according to claim 1, characterized in that: Obtaining infrared images of die-casting molds and real-time temperature data of pre-buried thermocouples inside also includes burying thermocouples in the deep cavity, undercut structure and spot cooling pipe outlet of the die-casting mold, in hidden areas ≤5mm away from the cooling pipe, within the range of ±3mm at the sudden change in wall thickness, and in the corresponding mold cavity positions of castings with a historical defect rate of >10%.

3. The mold temperature control method according to claim 1, wherein: The method further comprises: Based on the preset training set, the parameter mapping model is trained to establish the mapping relationship between pipeline parameters and mold temperature; Based on the parameter mapping model, training is performed to obtain a parameter prediction model, and the parameter prediction model is lightweight processed; Among them, parameter prediction model training also includes: Perform channel pruning on the neural network of the parameter prediction model and remove nodes with absolute weight values ​​less than 0.001; Perform model compression on parameter prediction models; Deploy the parameter prediction model on edge computing devices.

4. The mold temperature control method according to claim 1, wherein: When training the parameter prediction model, the residual constraint of the thermodynamic equation is introduced into the loss function, specifically: Add the heat conduction equation residual term to the loss function: ,in =0.7-1.2, is the temperature field matrix predicted by the parameter prediction model, is the real simulation temperature field matrix; is the weight coefficient used to balance the prediction accuracy and compliance with physical laws; The search space is limited based on the historical qualified parameter set, and the constraints include: the cooling time and flow rate meet the requirements. , is the pipe diameter coefficient, To maintain the minimum volume flow rate of turbulent state, The maximum safe flow rate to prevent water hammer effect.

5. The mold temperature control method according to claim 1, wherein: The PID algorithm incremental adjustment parameters are specifically: Calculate the temperature field deviation matrix ΔT; Generate adjustments by region: ,in =0.8, =0.05, =0.15, is the regional average temperature deviation, is the cumulative sum of historical deviations, is the current deviation matrix, is the deviation matrix of the previous cycle; The generated adjustment amount is limited to a single adjustment amplitude not exceeding ±15% of the parameter range.

6. The mold temperature control method according to claim 1, wherein: Generating a three-dimensional corrected temperature field including the mold surface and internal temperatures includes: Map the thermocouple data to the temperature nodes of the mold internal body mesh; The surface infrared temperature and internal node data are fused by Laplace smoothing algorithm; Radial basis function interpolation is used to fill the blind spots in the undercut area.

7. The mold temperature control method according to claim 1, wherein: The method further comprises: Calculate the confidence level of the parameter prediction model; If the confidence level of the parameter prediction model is less than 90%, manual correction instructions are accepted and stored in the historical experience database.

8. The mold temperature control method according to claim 7, characterized in that: The method further comprises: When the cumulative number of manual corrections exceeds a threshold, the parameter prediction model weights are updated based on reinforcement learning.

9. An electronic device comprising a processor, characterized in that: When the processor executes the program, the mold temperature control method according to any one of claims 1 to 6 is implemented; The electronic device is connected to a temperature compensator, which is used to dynamically calibrate the thermocouple readings based on the ambient temperature.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, it controls the operation of the electronic device according to claim 9.

Images