A machine learning-based hub low-pressure casting mold temperature field self-adaptive adjustment method

Abstract

The application discloses a kind of based on machine learning's wheel hub low-pressure casting mould temperature field self-adapting adjustment method, corresponding relationship of temperature variation of each wind water cooling and mould different position is obtained by machine learning, and when the measured temperature curve is corrected to qualified temperature curve, the adjustment that wind water cooling parameter needs to make is obtained by the method of calculation, then the wind water cooling parameter of die casting machine is regulated by PLC control program.The application can also be applied to the production of other castings low-pressure casting.The application lets low-pressure casting die casting machine-computer-thermocouple complete the interaction of data, obtains the wind cooling water cooling parameter adjustment scheme needed under different die casting machine working conditions by the method of machine learning, adjusts parameter by computer control die casting machine, realizes the self-adapting adjustment of mould temperature field, and simultaneously improves the yield of low-pressure casting.

Description

Technical Field

[0001] This invention relates to an adaptive temperature field adjustment method for a low-pressure casting mold for wheel hubs, and particularly to a machine learning-based adaptive temperature field adjustment method for a low-pressure casting mold for wheel hubs. Background Technology

[0002] Low-pressure casting is a common metal casting process used to manufacture high-quality metal parts with complex shapes. It is a relatively new casting method characterized by smooth mold filling, low cost, and good casting quality, and is widely used in the production of automotive parts such as wheel hubs.

[0003] The production process of low-pressure casting in enterprises is as follows: Figure 3 For example, in step 302, after obtaining a new wheel hub model, the simulation department will use simulation to obtain a set of process parameters (including pressure parameters and air / water cooling parameters) that make the simulation model qualified. In step 303, the simulation results do not match the actual results; the process parameters that are qualified in simulation cannot produce qualified castings in actual production. Therefore, after obtaining the simulation parameters, they need to be applied to the die-casting machines in the pilot production workshop. The process parameters are adjusted during the pilot production process in the pilot workshop until the castings are qualified, thus obtaining a qualified actual casting process. In step 304, when mass production of this wheel type is required, the qualified pilot production process is applied to a large number of die-casting machines in the production workshop. Then, the process parameters are adjusted during the die-casting process until each die-casting machine can produce qualified wheel hubs. In step 306, during the production process, the operating conditions of each die-casting machine may change due to prolonged working time (before low-pressure casting, the inner surface of the mold needs to be sprayed with heat-insulating coating; prolonged mold use will cause the coating to decrease, affecting the mold temperature field). In this case, it is necessary to return to step 305 and readjust the process parameters. When the paint consumption is high, it is necessary to stop the machine to re-spray paint. At this time, the temperature field of the mold will change due to the machine stoppage, and it is also necessary to return to process 305 and readjust the process parameters.

[0004] The Influence of Mold Temperature Field on the Quality of Low-Pressure Casting Aluminum Alloy Wheels: In the low-pressure casting process, the mold temperature field directly affects the solidification process of the casting, thereby influencing the generation of porosity and inclusion defects, as well as the grain size and mechanical properties of the casting. In low-pressure casting production, a first-version qualified process for each wheel type needs to be obtained and then applied to different die-casting machines for mass production. However, the operating conditions of each die-casting machine are different, resulting in different mold temperature fields during low-pressure casting under the same process parameters. The temperature field largely reflects whether there are defects in the casting obtained from this casting process; deviations in the temperature field will lead to defects in the cast wheel. For controlling the mold temperature field, if the mold temperature field is lower than that corresponding to the qualified process, the holding time needs to be reduced, and molten aluminum should be used to heat the mold. If the mold temperature field is higher than that corresponding to the qualified process, the air-cooling and water-cooling parameters of each die-casting machine need to be adjusted differently based on the first-version qualified process. Furthermore, after adjusting the parameters and entering stable production, and after stopping the machine to re-spray paint, changes in the mold working conditions can affect the temperature field and cause defects. At this time, workers need to readjust the air-cooling and water-cooling process parameters.

[0005] In the mass production stage of low-pressure casting, the air-cooling and water-cooling process parameters need to be adjusted when each die-casting machine first casts a new wheel shape and when the mold temperature field changes during the die-casting process. Currently, these adjustments are all done manually in enterprises. Manual adjustments are based on worker experience, and each person's experience and adjustment methods differ, resulting in varying outcomes. Some workers adjust quickly, while others adjust slowly. Generally, it takes at least two hours to try, wasting a significant amount of time. Scrap castings generated during the process also need to be remelted into molten aluminum, consuming a large amount of energy. Furthermore, manual adjustments require at least one worker to monitor each die-casting machine in real time, adjusting the air-cooling and water-cooling process parameters based on the size and location of defects in each casting, wasting substantial human resources, significantly increasing production costs, and reducing enterprise profits.

[0006] The existing invention CN109242192A discloses a method for processing and optimizing low-pressure casting production data. Based on the large amount of data generated during production, it selects filling temperature, mold temperature, filling time, and production cycle number, and employs a process combining data collection, processing, analysis, and training with machine learning methods to analyze the relationship between these parameters and the product yield, identifying the main influencing factors and optimizing the process to some extent. However, this method is not applicable to low-pressure casting of aluminum alloy wheels. Furthermore, it does not reflect the mold temperature field, does not involve extracting air-cooling process parameters during low-pressure casting, and cannot discuss the correspondence between each air-cooling point and the temperature changes at different locations in the mold.

[0007] In summary, the existing low-pressure cast aluminum alloy wheel hub solutions have the following drawbacks:

[0008] (1) The adjustment time is too long during mass production, which wastes a lot of time and resources.

[0009] (2) Manual adjustment is required. Each die-casting machine requires an employee to monitor and adjust it, resulting in excessively high labor costs.

[0010] (3) There are certain requirements for the experience of employees. For new employees, it takes more time to adjust the process parameters to make the wheel hub quality qualified.

[0011] (4) Each employee has different experience and different methods of adjusting parameters. Although the final wheel hub meets the quality requirements, there are still quality differences, and it is impossible to make each wheel hub reach the best quality. Summary of the Invention

[0012] The temperature field of a low-pressure casting mold largely reflects the quality of the cast part after forming. However, manually adjusting the air-cooling and water-cooling process parameters is based on personal experience, and each person's adjustment method is different, requiring varying amounts of time. This invention aims to solve the problem of needing to spend a lot of manpower, resources, and time to adjust air-cooling and water-cooling parameters.

[0013] This invention provides a method for adjusting the temperature field of a low-pressure cast aluminum alloy wheel hub mold by combining machine learning. This method uses the temperature-time curves of thermocouples arranged at different positions on the mold to reflect the temperature field of the mold.

[0014] For example, in case 401, after obtaining the simulation process of the new wheel shape, a trial production is needed to obtain the first qualified process. According to the method of this invention, the computer is connected to the thermocouple signal output terminal and the die-casting machine. Thermocouples are used to collect temperature-time curves at different locations on the mold, reflecting the mold temperature field. After data collection, the data is transmitted to the computer via a data cable and saved. The die-casting machine transmits the process parameter settings for each step to the computer via the data cable, saving them along with the corresponding thermocouple temperature curves. Simultaneously, the computer controls the adjustment of the die-casting machine's process parameters through a PLC program. When the artificial neural network model outputs the air-water cooling process parameters, these parameters are applied to the die-casting machine. The number and distribution of thermocouples are based on the hub structure and only need to be able to characterize the mold temperature field. During the low-pressure casting process, the adjusted air-water cooling process parameters (the air-water cooling flow rate is generally not changed during production; in this paper, air-water cooling parameters refer to the on / off time of each air-water cooling point) and the temperature change curve of each thermocouple under the corresponding process are extracted.

[0015] For example, in case 402, the air-water cooling process parameters for each low-pressure casting are used as data along with the temperature-time curves collected by all thermocouples during the casting process. The temperature curves of all thermocouples corresponding to each process parameter are used as input to an artificial neural network model. The air-water cooling process parameters are represented in interval form. For example, if the water cooling start time is 20 seconds after the start of die casting and the stop time is 100 seconds after the start of die casting, the water cooling at this point is represented as (20, 100). The air-water cooling parameters at the qualified temperature are also represented using this method. The air-water cooling adjustment scheme at that location is characterized by subtracting the air-water cooling parameters at the corresponding positions under other processes from the qualified air-water cooling parameters to obtain the qualified temperature curve. Example: In a certain process, the water-cooling process parameter at a certain point is (20, 100). When the temperature curve is acceptable, the water-cooling process parameter at this point is (10, 110). The difference is: [(10, 110) - (20, 100)] = (-10, 10). Therefore, to obtain a acceptable temperature curve, the water-cooling start-up time should be 10 seconds earlier, and the shut-off time should be 10 seconds later. All air-cooling and water-cooling parameters under a process are processed in this way to form a dataset, which is the output.

[0016] For example, in case 403, all data is divided into two groups: one group (1 / 4) serves as the test group, and the other group (3 / 4) serves as the training group. The data in the training group is used to train the artificial neural network model, while the data in the test group is used to verify the accuracy of the artificial neural network model and to adjust the parameters of the artificial neural network model to achieve the highest prediction accuracy.

[0017] For example, in case 404, an artificial neural network (ANN) model is trained using the input and output of the training group data. After training, the input of the test group is input, and the prediction result is output. Simultaneously, the parameters of the ANN model are adjusted by comparing the prediction results of the test group with the actual process parameters of the test group to maximize its prediction accuracy (e.g., ...). Figure 5 These parameters that need to be adjusted include the number of neurons, initial weights, and allowable error. By using different parameters to predict on the test set and comparing the errors between the prediction results under different parameters and the actual process parameters, a more ideal set of artificial neural network model parameters is obtained.

[0018] For example, after obtaining the first qualified process, 405 will be applied to other die-casting machines. During the production process of other die-casting machines using this process, the computer will automatically collect the temperature curve of each thermocouple and transmit these temperature curves to the computer for judgment.

[0019] For example, in step 406, the computer determines whether the temperature of the mold at the start of casting has reached the initial temperature requirement. If it is lower than the initial temperature requirement, then proceed to step 407; if it has reached the initial temperature requirement, then proceed to step 408.

[0020] For example, in 407, if the temperature of the mold at the start of casting is lower than the initial temperature requirement, the computer will reduce the holding time of the die-casting machine, use the temperature of the molten aluminum to heat the mold, and then adjust the holding time back to the original state and proceed to process 408 after the mold temperature rises to the required initial temperature.

[0021] For example, in 408, if the mold reaches the initial temperature requirement at the start of casting, the computer will collect the temperature curve during casting as input to the artificial neural network, and calculate the air-cooling and water-cooling parameter adjustment scheme through the artificial neural network model.

[0022] For example, in model 409, the artificial neural network model calculates the air-cooling and water-cooling parameter adjustment scheme that can achieve the qualified temperature curve on this die-casting machine based on the input measured temperature curve, and realizes automatic adjustment through PLC program control of the die-casting machine (e.g., Figure 6 It can also be manually adjusted at any time.

[0023] For example, in process 410, prolonged production can affect the mold's operating conditions, leading to different mold temperature fields under the same air-cooling and water-cooling process parameters. Therefore, during stable production, the computer continuously collects temperature curve data. When the temperature curve fluctuates significantly (the maximum temperature difference from the acceptable temperature curve reaches 5℃), the process for 406 and below is repeated. This cycle allows the die-casting machine to adaptively adjust, ensuring the mold temperature field remains within the acceptable range. Simulations show that temperature fluctuations within 5℃ ensure that the casting quality and defects do not change significantly.

[0024] The essence of this invention is to obtain the correspondence between each air-water cooling point and the temperature changes at different locations in the mold through machine learning, and then calculate the necessary adjustments to the air-water cooling parameters to correct the measured temperature curve to a qualified temperature curve. Finally, the air-water cooling parameters of the die-casting machine are controlled by a PLC program. This invention can also be applied to the production of other castings in low-pressure casting.

[0025] This invention enables data interaction between the low-pressure casting die-casting machine, computer, and thermocouple. It utilizes machine learning to obtain air-cooling and water-cooling parameter adjustment schemes required under different die-casting machine operating conditions. The computer controls the die-casting machine to adjust the parameters, achieving adaptive adjustment of the mold temperature field and improving the yield of low-pressure casting.

[0026] Terminology Explanation

[0027] Low-pressure cast aluminum alloy wheels: such as Figure 1Before casting begins, molten aluminum at 700℃ is added to a holding furnace below the mold. At the start of casting, the mold is closed and sealed, and dry compressed air is pumped into the container through the air inlet valve. Under the air pressure above the molten aluminum, it moves upwards along the riser pipe, filling the mold cavity and solidifying under pressure. When the molten aluminum reaches the gate, the furnace is depressurized, and the unsolidified molten aluminum flows back into the holding furnace along the riser pipe, preparing for the next filling. After depressurization, the mold is opened, and a mechanical gripper places the wheel hub into a cooling water tank, awaiting manual surface defect inspection.

[0028] Machine learning: Machine learning refers to the process by which machines learn from large amounts of historical data using statistical algorithms, and then use the generated experience models to guide business operations. It is a multidisciplinary field that specifically studies how computers can simulate or implement human learning behavior to acquire new knowledge or skills, and reorganize existing knowledge structures to continuously improve their performance.

[0029] Artificial Neural Networks (ANNs) are network structures that process external information. Their design prototype is the neural network of the human brain in biology. Within the neural network structure, a large number of artificial neurons are arranged as nodes. These neurons are highly complex and interconnected, and by adding functions, they achieve a degree of non-linear mapping output. Because ANNs possess adaptive, self-learning, and self-organizing capabilities, the network learns to analyze and reason about input vectors through training, thereby processing massive amounts of information. Due to the large number of complex and interconnected neurons in the neural network, artificial neural networks can efficiently process the large amounts of complex and redundant information continuously generated externally, and learn the correlations between data information for use in new data. Figure 2 Its structure consists of an input layer, hidden layers, and an output layer. Here, m = (1,2,3...m) represents the input value of a neuron, i = (1,2,3...i) represents the input neurons in the hidden layer (the more neurons there are, the more complex the structure), and j = (1,2,3...j) represents the output of a neuron.

[0030] Advantages of artificial neural networks:

[0031] (1) Parallel distributed processing: The arrangement of neurons in artificial neural networks is not random, but often hierarchical or in a regular sequence. Signals can reach the input of a batch of neurons at the same time. This structure is very suitable for parallel computing. At the same time, if each neuron is regarded as a small processing unit, the whole system can be a distributed computing system, avoiding matching conflicts and achieving fast computing speed.

[0032] (2) Learnability: A relatively small artificial neural network can store a large amount of knowledge and can continuously learn automatically and improve its knowledge reserves by learning algorithms, or by using sample guidance systems to simulate real environments, or by adaptively learning from inputs.

[0033] (3) Generalization ability: Artificial neural networks are a type of large-scale nonlinear system that can fully approximate complex nonlinear relationships. When the input changes slightly, its output can maintain a fairly small difference from the output generated by the original input.

[0034] (4) Universality: It has a unified internal knowledge representation form. Any knowledge rule can be stored in the connection weights of the same neural network through learning from examples, which facilitates the organization and management of the knowledge base.

[0035] Therefore, artificial neural network machine learning methods have a great advantage for data with complex internal relationships and large quantities in this invention. Attached Figure Description

[0036] Figure 1 A schematic diagram of a low-pressure cast aluminum alloy wheel hub structure;

[0037] Figure 2 This is a schematic diagram of a neural network structure;

[0038] Figure 3 This is a flowchart of the low-pressure casting process;

[0039] Figure 4 Flowchart of a method for adjusting the temperature field of low-pressure cast aluminum alloy wheel hub molds using machine learning;

[0040] Figure 5 This is a schematic diagram of the construction of an artificial neural network model;

[0041] Figure 6 A schematic diagram illustrating the adaptive adjustment of air-cooling and water-cooling process parameters using machine learning;

[0042] Figure 7 A comparison diagram of the three temperature curves at thermocouple 1;

[0043] Figure 8 An X-ray image of the first cast wheel hub;

[0044] Figure 9 An X-ray image of the cast wheel hub after the air-cooling parameters have been adjusted;

[0045] Figure 10 The tensile result diagram of the wheel hub obtained from the first die casting;

[0046] Figure 11 The tensile result diagram of the wheel hub obtained from the second die casting; Detailed Implementation

[0047] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to embodiments. The content mentioned in the embodiments is not intended to limit the present invention.

[0048] A method for adjusting the temperature field of low-pressure cast aluminum alloy wheel hub molds by combining machine learning is illustrated in the flowchart below. Figure 4 This method uses the temperature-time curves of thermocouples placed at different locations on the mold to reflect the temperature field of the mold.

[0049] For example, in case 401, after obtaining the simulation process of the new wheel shape, a trial production is needed to obtain the first qualified process. According to the method of this invention, the computer is connected to the thermocouple signal output terminal and the die-casting machine. Thermocouples are used to collect temperature-time curves at different locations on the mold, reflecting the mold temperature field. After data collection, the data is transmitted to the computer via a data cable and saved. The die-casting machine transmits the process parameter settings for each step to the computer via the data cable, saving them along with the corresponding thermocouple temperature curves. Simultaneously, the computer controls the adjustment of the die-casting machine's process parameters through a PLC program. When the artificial neural network model outputs the air-water cooling process parameters, these parameters are applied to the die-casting machine. The number and distribution of thermocouples are based on the hub structure and only need to be able to characterize the mold temperature field. During the low-pressure casting process, the adjusted air-water cooling process parameters (the air-water cooling flow rate is generally not changed during production; in this paper, air-water cooling parameters refer to the on / off time of each air-water cooling point) and the temperature change curve of each thermocouple under the corresponding process are extracted.

[0050] For example, in case 402, the air-water cooling process parameters for each low-pressure casting are used as data along with the temperature-time curves collected by all thermocouples during the casting process. The temperature curves of all thermocouples corresponding to each process parameter are used as input to an artificial neural network model. The air-water cooling process parameters are represented in interval form. For example, if the water cooling start time is 20 seconds after the start of die casting and the stop time is 100 seconds after the start of die casting, the water cooling at this point is represented as (20, 100). The air-water cooling parameters at the qualified temperature are also represented using this method. The air-water cooling adjustment scheme at that location is characterized by subtracting the air-water cooling parameters at the corresponding positions under other processes from the qualified air-water cooling parameters to obtain the qualified temperature curve. Example: In a certain process, the water-cooling process parameter at a certain point is (20, 100). When the temperature curve is acceptable, the water-cooling process parameter at this point is (10, 110). The difference is: [(10, 110) - (20, 100)] = (-10, 10). Therefore, to obtain a acceptable temperature curve, the water-cooling start-up time should be 10 seconds earlier, and the shut-off time should be 10 seconds later. All air-cooling and water-cooling parameters under a process are processed in this way to form a dataset, which is the output.

[0051] For example, in case 403, all data is divided into two groups: one group (1 / 4) serves as the test group, and the other group (3 / 4) serves as the training group. The data in the training group is used to train the artificial neural network model, while the data in the test group is used to verify the accuracy of the artificial neural network model and to adjust the parameters of the artificial neural network model to achieve the highest prediction accuracy.

[0052] For example, in case 404, an artificial neural network (ANN) model is trained using the input and output of the training group data. After training, the input of the test group is input, and the prediction result is output. Simultaneously, the parameters of the ANN model are adjusted by comparing the prediction results of the test group with the actual process parameters of the test group to maximize its prediction accuracy (e.g., ...). Figure 5 These parameters that need to be adjusted include the number of neurons, initial weights, and allowable error. By using different parameters to predict on the test set and comparing the errors between the prediction results under different parameters and the actual process parameters, a more ideal set of artificial neural network model parameters is obtained.

[0053] For example, after obtaining the first qualified process, 405 will be applied to other die-casting machines. During the production process of other die-casting machines using this process, the computer will automatically collect the temperature curve of each thermocouple and transmit these temperature curves to the computer for judgment.

[0054] For example, in step 406, the computer determines whether the temperature of the mold at the start of casting has reached the initial temperature requirement. If it is lower than the initial temperature requirement, then proceed to step 407; if it has reached the initial temperature requirement, then proceed to step 408.

[0055] For example, in 407, if the temperature of the mold at the start of casting is lower than the initial temperature requirement, the computer will reduce the holding time of the die-casting machine, use the temperature of the molten aluminum to heat the mold, until the mold temperature rises to the required initial temperature, then adjust the holding time back to the original state and proceed to process 408.

[0056] For example, in 408, if the mold reaches the initial temperature requirement at the start of casting, the computer will collect the temperature curve during casting as input to the artificial neural network, and calculate the air-cooling and water-cooling parameter adjustment scheme through the artificial neural network model.

[0057] For example, in model 409, the artificial neural network model calculates the air-cooling and water-cooling parameter adjustment scheme that can achieve the qualified temperature curve on this die-casting machine based on the input measured temperature curve, and realizes automatic adjustment through PLC program control of the die-casting machine (e.g., Figure 6 It can also be manually adjusted at any time.

[0058] For example, in process 410, prolonged production can affect the mold's operating conditions, leading to different mold temperature fields under the same air-cooling and water-cooling process parameters. Therefore, during stable production, the computer continuously collects temperature curves. When the temperature curve fluctuates significantly (the maximum temperature difference from the acceptable temperature curve reaches 5℃), the process for 406 and below is repeated. Through this cycle, the die-casting machine achieves adaptive adjustment, ensuring that the mold temperature field remains within the acceptable range. Simulations show that temperature fluctuations within 5℃ ensure that there are no significant changes in casting quality and defects.

[0059] Experimental conclusion: such as Figure 7 When using the first version of the qualified process on other die-casting machines (casting had stabilized), the temperature curve at thermocouple 1 differed from the qualified temperature curve by a maximum of approximately 10℃. After acquiring the measured temperature curves via computer, calculating and modifying the air-cooling and water-cooling process parameters of the die-casting machine, the difference between the new temperature curve and the qualified temperature curve at all points during the next casting was less than 3℃. Manual inspection showed no defects on the surface of the wheel hubs from both castings, but X-ray inspection revealed significant shrinkage porosity at the spokes of the first casting (e.g.,...). Figure 8 The modification scheme for air-cooling and water-cooling parameters was calculated using artificial neural networks, and the resulting cast wheel hub successfully eliminated defects (such as...). Figure 9 According to GB / T228-2002, tensile tests were conducted on the wheel hub's center, spokes, and rim obtained from two die-casting processes. The tensile results of the wheel hub obtained from the first die-casting are as follows: Figure 10 The tensile strengths of the samples at the wheel center, spokes, and rim were 249 MPa, 200 MPa, and 214 MPa, respectively. The tensile results of the wheel hub obtained from the second die casting are as follows... Figure 11 The tensile strengths at the wheel center, spokes, and rim are 267 MPa, 230 MPa, and 230 MPa, respectively. Compared to before the adjustment, the tensile strength at all three locations has improved, especially at the spokes, where the tensile strength has significantly increased due to the elimination of defects. Manually adjusting the air-cooling system to achieve this effect would take at least two hours and result in the scrapping of dozens of wheel hubs. For companies that require a large number of low-pressure die-casting machines to operate simultaneously, this invention can save significant resource and labor costs.

[0060] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0061] It should be understood that the above detailed description of the technical solutions of the present invention with reference to preferred embodiments is illustrative and not restrictive. Those skilled in the art can modify the technical solutions described in the embodiments or make equivalent substitutions for some of the technical features based on reading this specification; however, these modifications or substitutions do not cause the essence of the corresponding technical solutions to depart from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for adjusting the temperature field of a low-pressure cast aluminum alloy wheel hub mold by combining machine learning, characterized in that the steps include... include: (1) Connect the computer to the thermocouple signal output terminal and the die-casting machine; the thermocouple is used to collect temperature-time curves at different positions of the mold to reflect the temperature field of the mold. After the data is collected, the data is transmitted to the computer and saved; the die-casting machine transmits the process parameter settings for each time to the computer and saves them together with the corresponding thermocouple temperature curves; at the same time, the computer controls the adjustment of the process parameters of the die-casting machine through the PLC program. When the artificial neural network model outputs the air-cooling process parameters, the air-cooling process parameters are applied to the die-casting machine; the number and distribution of thermocouples are based on the structure of the hub to characterize the temperature field of the mold. Extract the air-water cooling process parameters after each adjustment and the temperature change curve of each thermocouple under the corresponding process during the low-pressure casting process. (2) The air-water cooling process parameters of each low-pressure casting and the temperature-time change curves collected by all thermocouples during the casting process are used as data; the temperature curves of all thermocouples corresponding to each process parameter are used as inputs to the artificial neural network model; the air-water cooling process parameters are represented in interval form, and the air-water cooling parameters when qualified are also represented in interval form; the air-water cooling parameters when qualified are subtracted from the air-water cooling parameters at the corresponding positions under other processes to characterize the air-water cooling adjustment scheme at that position when a qualified temperature curve is to be obtained. All air-cooling and water-cooling parameters under a certain process are processed and combined into a dataset, which is a single output. (3) Divide all data into two groups: one as the test group and the other as the training group; The training group's data is used to train the artificial neural network model, while the test group's data is used to verify the accuracy of the artificial neural network model and adjust the parameters of the artificial neural network model to achieve the highest prediction accuracy. (4) Train the artificial neural network model using the input and output of the training group data. After training, input the test group data and output the prediction results. At the same time, adjust the parameters of the artificial neural network model by comparing the prediction results of the artificial neural network model after the test group input with the actual test group process parameters to make its prediction accuracy reach the highest level. By using different parameters to predict the test set, compare the error between the prediction results under different parameters and the actual process parameters to obtain the parameters of the artificial neural network model. (5) Once the first qualified process is obtained, it will be applied to other die casting machines. During the production process of other die casting machines using this process, the computer will automatically collect the temperature curve of each thermocouple and transmit these temperature curves to the computer for judgment. (6) The computer determines whether the temperature of the mold at the start of casting has reached the initial temperature requirement. If it is lower than the initial temperature requirement, proceed to step (7); if it has reached the initial temperature requirement, proceed to step (8). (7) If the temperature of the mold at the start of casting is lower than the initial temperature requirement, the computer will reduce the holding time of the die casting machine, use the temperature of the aluminum liquid to heat the mold, and then adjust the holding time back to the original state and proceed to step (8) after the mold temperature rises to the required initial temperature. (8) If the temperature of the mold reaches the initial temperature requirement at the start of casting, the computer will collect the temperature curve during casting as the input of the artificial neural network, and calculate the air-cooling parameter adjustment scheme through the artificial neural network model. (9) The artificial neural network model calculates the air-cooling parameter adjustment scheme that can achieve the qualified temperature curve on the die-casting machine by inputting the measured temperature curve, and realizes automatic adjustment by controlling the PLC program of the die-casting machine, while also making manual intervention adjustments at any time. (10) During stable production, the computer will continue to collect the temperature curve. When the temperature curve fluctuates significantly, repeat step (6) and the following steps. The largest fluctuation in step (10) is a temperature difference of 5°C from the qualified temperature curve.

2. The method according to claim 1, characterized in that, In step (1), the flow rate of the air-water cooling system is not changed during the production process. The air-water cooling parameters are the on-time and off-time of each air-water cooling system.

3. The method according to claim 1, characterized in that, In step (3), all data are divided into two groups: one group accounts for 1 / 4 and serves as the test group; the other group accounts for 3 / 4 and serves as the training group.

4. The method according to claim 1, characterized in that, The parameters that need to be adjusted in step (4) include the number of neurons, the initial weights, and the allowable error.

Images