Aluminum alloy automobile wheel hub low pressure casting process parameter adaptive control system

Abstract

The application discloses an aluminum alloy automobile hub low-pressure casting process parameter self-adaptive control system and relates to the technical field of intelligent manufacturing. The system comprises a state observer operation module, an adaptive optimal decision control module and an error suppression execution module. The state observer operation module takes multi-source heterogeneous working condition data as the input of a composite digital twin and outputs real-time estimated values. The adaptive optimal decision control module compares the real-time estimated values with preset trigger thresholds, determines corresponding pressure maintaining trigger signals, takes the real-time estimated values as the input of a data-driven model during pressure maintaining processing, and determines a corresponding optimal pressure maintaining pressure curve. The error suppression execution module corrects the optimal pressure maintaining pressure curve, determines corresponding target pressure set values at each moment, and determines corresponding driving instructions according to the pressure error between actual pressure values and the target pressure set values. The application can fundamentally solve the problems of invisible internal state and experience-dependent control.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing technology, specifically to an adaptive control system for low-pressure casting process parameters of aluminum alloy automotive wheel hubs. Background Technology

[0002] Driven by the trend towards lightweighting and energy conservation in the automotive industry, aluminum alloy wheels have gradually replaced traditional steel wheels, becoming the mainstream choice in the market due to their significant advantages such as light weight, excellent heat dissipation, aesthetics, and durability. Related research indicates that reducing wheel weight by 10% can improve overall vehicle fuel efficiency by 6%, which is of great significance for energy conservation, emission reduction, and improved vehicle performance. Simultaneously, aluminum alloy's excellent thermal conductivity allows for rapid heat dissipation, maintaining the normal operating temperature of the braking system and enhancing driving safety; its diverse exterior designs also greatly enhance the vehicle's aesthetics and brand image.

[0003] As the automotive market continues to demand higher performance, lighter weight, and better aesthetics, the market size for aluminum alloy wheels is constantly expanding, with widespread applications in passenger cars, sports cars, and some commercial vehicles. Against this backdrop, how to efficiently and reliably produce high-quality aluminum alloy wheels has become a critical issue that the industry urgently needs to address.

[0004] Low-pressure casting, as the mainstream technology for aluminum alloy wheel production, possesses numerous unique advantages. It is an anti-gravity casting process, where molten metal fills the mold smoothly from bottom to top under low pressure, effectively avoiding turbulence and reducing defects such as air entrapment and oxide film inclusions, significantly improving casting quality. After filling, further pressurization allows the casting to solidify and crystallize under pressure, resulting in a denser microstructure and significantly enhanced mechanical properties.

[0005] Chinese invention patent application CN121956845A discloses a method and system for real-time prediction and control of deformation state of superplastic forming parts based on digital twins. This method aims to solve the technical challenges of invisible deformation state of parts during superplastic forming, reliance on experience for control, and open-loop control leading to unstable quality. The core of this method lies in: offline, fusing parametric finite element simulation and physical experimental data to construct a multi-task learning artificial intelligence prediction model with intake volume time-series data as input and mold cavity pressure and the overall deformation state of the part as output; online, using this model, based on real-time collected intake volume data, dynamically predicting the deformation state of the part's internal strain field, thickness field, etc., achieving process transparency; and further combining this with a model predictive control algorithm, comparing the predicted state with the ideal path, and real-time reverse optimization and adjustment of the intake pressure to form a closed-loop intelligent control system. This invention achieves transparent and proactive precise control of the internal state of the "black box" forming process, which is beneficial for improving part quality consistency and yield.

[0006] However, the above and similar technical solutions still have the following shortcomings: In the low-pressure casting process of aluminum alloy automobile wheels, the core internal state that determines the quality of the casting (such as the position of the flow front of the molten metal in the cavity, the real-time pressure state of the solidification shrinkage point, and the evolution process of the microstructure) cannot be directly and accurately measured online, resulting in a significant blind spot in the control process. Therefore, the current control system can only rely on indirect parameters such as the gating pressure and the mold surface temperature for estimation, which makes the control of the holding pressure timing and pressure highly empirical and lagging. This not only causes the holding time to be excessively extended in order to suppress shrinkage defects, resulting in energy waste and reduced production efficiency, but also makes it difficult to fundamentally guarantee the stability and consistency of the internal quality of the casting. Summary of the Invention

[0007] The purpose of this invention is to provide an adaptive control system for low-pressure casting process parameters of aluminum alloy automotive wheels to solve the problems mentioned in the background art.

[0008] To achieve the above objectives, the present invention provides the following technical solution: an adaptive control system for low-pressure casting process parameters of aluminum alloy automotive wheel hubs, comprising: State observer operation module: It collects multi-source heterogeneous operating condition data through the set sensor network, and uses the multi-source heterogeneous operating condition data as the input of the constructed composite digital twin, and outputs the corresponding real-time estimated value. Adaptive optimal decision control module: compares the real-time estimated value with the preset trigger threshold, and determines the corresponding pressure holding trigger signal based on the comparison result. At the same time, during the pressure holding process, the real-time estimated value is used as the input of the data-driven model set in the composite digital twin, and the output obtains the comprehensive cost value corresponding to each candidate path. Based on the comprehensive cost value, the corresponding optimal pressure holding curve is determined. Error suppression execution module: corrects the optimal pressure holding curve, obtains the corresponding corrected pressure holding curve, and determines the target pressure setting value at each moment. At the same time, when the pressure holding trigger signal is triggered, it collects the actual pressure value in the gas path and determines the corresponding drive command based on the pressure error between the actual pressure value and the target pressure setting value.

[0009] Furthermore, the output obtains the corresponding real-time estimates, including: SA1: Data Acquisition: Using a non-contact infrared thermal imager, thermocouples, pressure transmitters, and gas mass flow meters, the corresponding temperature matrix, measured thermocouple temperature values, pouring pressure, and pressurized gas are acquired to construct the corresponding multi-source heterogeneous operating condition data. SA2: Initial construction of the digital twin: Using multiphysics simulation software, a corresponding physical mechanism model is constructed. Using the multi-source heterogeneous working condition data and deep learning model, a corresponding data-driven model is constructed. The physical mechanism model and the data-driven model are combined to construct the corresponding initial composite digital twin. SA3: Twin Update: Through the data-driven model, the corresponding predicted data value is obtained, and the model parameters for updating the initial composite digital twin are updated according to the residual between the predicted data value and the multi-source heterogeneous operating condition data. SA4: Model Correction: Based on the updated model parameters, update the initial composite digital twin corresponding to the current moment to obtain the corresponding final composite digital twin. Use the multi-source heterogeneous operating condition data as the input to the final composite digital twin and output the real-time estimated value corresponding to the current moment, including the position and shape of the molten metal flow front, the real-time predicted solid fraction of typical hot spot positions, the real-time predicted local pressure, and the real-time predicted local pressure change rate.

[0010] Furthermore, the corresponding initial composite digital twin is constructed, including: SA2.1: Digital Reconstruction: Based on the design drawings of the casting mold, a three-dimensional geometric model is created, and the component material properties, initial operating conditions and boundary operating conditions of the three-dimensional geometric model are set through multiphysics simulation software to construct the corresponding physical mechanism model; SA2.2: Data Fusion: By using time-series process data, process setting data, and result label data from historical process data, a corresponding training dataset is constructed. At the same time, a corresponding data-driven model is set using a convolutional neural network model and a long short-term memory network model. The data-driven model and the physical mechanism model are combined to construct a corresponding initial composite digital twin. The initial composite digital twin is then trained and validated using the training dataset.

[0011] Furthermore, the time-series process data in the training dataset is used as the input to the data-driven model, the process setting data in the training dataset is used as the input to the physical mechanism model, and the result label data in the training dataset is used as the output of the initial composite digital twin, thereby training and validating the initial composite digital twin.

[0012] Furthermore, the model parameters for updating the initial composite digital twin are updated, including: SA3.1: Data Prediction: Based on the set of optimal estimated values ​​and the posterior estimation error covariance matrix obtained from the output of the initial composite digital twin, the set of optimal estimated values ​​and the posterior estimation error covariance matrix obtained from the output of the initial composite digital twin are combined with the control command obtained from the output of the current time to obtain the predicted state vector and the prior estimation error covariance matrix obtained from the output of the current time. SA3.2: Data extraction: Compare the multi-source heterogeneous operating condition data corresponding to the current time with the predicted state vector corresponding to the current time to obtain the residual vector between the multi-source heterogeneous operating condition data and the predicted state vector. At the same time, combine the multi-source heterogeneous operating condition data with the prior estimation error covariance matrix corresponding to the current time to obtain the Kalman gain matrix corresponding to the current time. SA3.3: Parameter Update: Combine the Kalman gain matrix, residual vector and predicted state vector at the current time to determine the optimal set of estimates at the current time. At the same time, based on the Kalman gain matrix, identity matrix and prior estimation error covariance matrix at the current time, determine the posterior estimation error covariance matrix at the current time.

[0013] Furthermore, the corresponding optimal holding pressure curve is determined, including: SB1: Switching determination: Based on the real-time estimated value, determine the real-time predicted solid fraction and real-time predicted local pressure change rate corresponding to each typical hot spot location, and compare the real-time predicted solid fraction and real-time predicted local pressure change rate with a preset trigger threshold to determine the corresponding pressure holding trigger signal; SB2: Path Determination: Based on the pressurized gas pressure setpoint corresponding to each discrete time within a preset time period, the pressurized gas pressure setpoint and the set of optimal estimates corresponding to the current time are both used as inputs to the data-driven model. The output obtains the predicted total volume of the isolated liquid phase region and the sum of squares of the accumulated pressure sequence corresponding to each candidate path. At the same time, the predicted total volume of the isolated liquid phase region and the sum of squares of the accumulated pressure sequence are combined to obtain the comprehensive cost value corresponding to each candidate path. The minimum comprehensive cost value is determined from all comprehensive cost values. The candidate path corresponding to the minimum comprehensive cost value is the optimal pressure holding curve.

[0014] Furthermore, the preset trigger threshold includes a preset solid fraction trigger threshold and a preset negative local pressure change rate threshold. The preset solid fraction trigger threshold is compared with the real-time predicted solid fraction, and the preset negative local pressure change rate threshold is compared with the real-time predicted local pressure change rate. Based on the comparison results, the corresponding pressure holding trigger signal is determined, specifically as follows: When the real-time predicted solid fraction is not less than the preset solid fraction trigger threshold and the real-time predicted local pressure change rate corresponding to the same typical hot spot location is less than the preset negative local pressure change rate threshold, a pressure holding trigger signal is triggered, and the pressure holding process begins; otherwise, the pressure holding trigger signal is not triggered, and the current operating state continues.

[0015] Furthermore, the corresponding driver instructions are determined, including: SC1: Curve Correction: Based on the optimal pressure path, obtain the target pressure setpoint corresponding to each moment in the optimal pressure path, and correct the target pressure setpoint at each moment through the set feedforward controller to obtain the target pressure correction value corresponding to each moment. Connect the target pressure correction values ​​in chronological order to construct the corrected holding pressure curve. SC2: Feedback Correction: Acquire the actual pressure value at each moment in the gas path, and determine the target pressure correction value corresponding to the same time point in the corrected pressure holding curve based on the timestamp corresponding to the actual pressure value. Compare the actual pressure value and the target pressure correction value at the same time point to obtain the corresponding pressure error and the total pressure error accumulation. At the same time, determine the corresponding drive command through the set PID controller based on the pressure error corresponding to each moment and the total pressure error accumulation.

[0016] Compared with the prior art, the beneficial effects of the present invention are: Firstly, this invention constructs a composite digital twin that integrates a physical mechanism model and a data-driven model through a state observer operation module. It takes multi-source heterogeneous working condition data collected by a sensor network as input and outputs real-time estimates of key internal state quantities such as the position and shape of the molten metal flow front, the real-time predicted solid fraction of typical hot spots, and the real-time predicted local pressure and rate of change. This can fundamentally solve the problems of invisible internal states and control relying on experience. Secondly, this invention is based on the real-time estimated value output by the composite digital twin and sets up a dual threshold triggering mechanism to accurately determine the time of switching the pressure holding stage. Through a data-driven model, it performs rapid simulation and prediction of different pressure control paths in the future, evaluates the predicted total volume of the isolated liquid phase region and the sum of squares of the accumulated pressure sequence for each candidate path, and calculates the corresponding comprehensive cost value. The path with the minimum comprehensive cost value is selected as the optimal pressure holding curve. Thus, while ensuring the internal quality of the casting, it can automatically find the pressure holding strategy with the best energy consumption and time, achieving a balance between quality, efficiency and energy consumption. Thirdly, this invention pre-corrects the optimal pressure holding curve based on the known dynamic characteristics of the controlled object, generating a corrected pressure holding curve to actively compensate for errors caused by the inherent dynamic characteristics of the system. After the pressure holding is triggered, the actual pressure value of the gas path is collected in real time and compared with the target pressure correction value in the correction curve. The PID feedback controller generates a drive command based on the pressure error and its integral and derivative to dynamically adjust the control valve, thereby suppressing unknown disturbances and ensuring that the actual pressure can accurately track the set curve, thus improving the accuracy, stability and anti-interference capability of pressure control. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the adaptive control system in this invention; Figure 2 This is a flowchart illustrating the state observer operation method in this invention; Figure 3 This is a schematic diagram of the update process of the initial composite digital twin in this invention; Figure 4 This is a flowchart illustrating the adaptive optimal decision control method of the present invention; Figure 5 This is a flowchart illustrating the error suppression execution method of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] refer to Figure 1This embodiment provides an adaptive control system for low-pressure casting process parameters of aluminum alloy automotive wheel hubs. This adaptive control system specifically includes a state observer operation module, an adaptive optimal decision control module, and an error suppression execution module. The state observer operation module collects corresponding multi-source heterogeneous operating condition data based on the sensor network deployed on the low-pressure casting machine and mold. Simultaneously, it fuses the established physical mechanism model with the data-driven model to construct a corresponding composite digital twin. The collected multi-source heterogeneous operating condition data is used as the input to the composite digital twin to output corresponding real-time estimated values, including the position and morphology of the molten metal flow front, the real-time predicted solid fraction of typical hot spots, the real-time predicted local pressure, and the real-time predicted rate of change of local pressure. It is worth noting that during the operation of the composite digital twin, based on the infrared thermal images and thermocouple data from the acquired multi-source heterogeneous operating conditions data, the corresponding surface temperature prediction field and temperature prediction curve are obtained. The surface temperature prediction field and temperature prediction curve are then compared with the real-time acquired infrared thermal images and thermocouple data to obtain the corresponding residual magnitude. At the same time, based on the obtained residual magnitude, the model parameters of the composite digital twin are corrected in real time.

[0020] Furthermore, the adaptive optimal decision control module monitors the real-time estimated values ​​obtained by the state observer operation module through a preset trigger threshold. Based on the comparison between the real-time estimated values ​​and the preset trigger threshold, it determines the corresponding pressure holding trigger signal. Simultaneously, during the pressure holding process based on the pressure holding trigger signal, the real-time estimated values ​​obtained by the state observer operation module are used as input to the data-driven model set in the composite digital twin. This model outputs the predicted total volume of the isolated liquid phase region and the sum of squares of the accumulated pressure sequence for each candidate path. Based on the predicted total volume of the isolated liquid phase region and the sum of squares of the accumulated pressure sequence, the comprehensive cost value corresponding to each candidate path is determined. Furthermore, by comparing the comprehensive cost values ​​corresponding to each candidate path, the corresponding optimal pressure holding curve is determined.

[0021] Furthermore, the error suppression execution module is used to correct the optimal pressure holding curve determined by the adaptive optimal decision control module based on known disturbance sources, thereby obtaining the corresponding corrected pressure holding curve. Simultaneously, based on the pressure holding trigger signal determined by the adaptive optimal decision control module, when the pressure holding trigger signal is triggered, the actual pressure value in the gas path is acquired, and the pressure error between the actual pressure value and the target pressure setpoint in the corrected pressure holding curve is obtained. Based on the acquired pressure error, the corresponding drive command is determined.

[0022] Specifically, when designing the production of the A356 aluminum alloy wheel hub according to the adaptive control system in this embodiment, the core data comparison table is as follows compared with the traditional technical solution: Table 1: Comparison of Core Data

[0023] In other words, as shown in the core data comparison table in Table 1 above, traditional technical solutions rely on indirect parameters such as gating pressure and mold surface temperature to determine the timing of pressure holding, resulting in a false trigger rate of 18%. However, the adaptive control system in this embodiment uses a composite digital twin to observe the invisible internal states such as the molten metal flow front and the solidity of the hot spot in real time, and accurately triggers pressure holding based on dual thresholds, reducing the false trigger rate to 2%. This fundamentally solves the problem of "black box production relying on guesswork."

[0024] Meanwhile, traditional technical solutions, in order to avoid the risk of shrinkage porosity, are forced to uniformly set a fixed holding time of 180 seconds, resulting in a large amount of ineffective energy consumption. However, the adaptive control system in this embodiment can dynamically match the real-time solidification requirements of the casting, which shortens the average holding time by 37.8%, reduces energy consumption per piece by 33.3%, and increases output per shift by 47.9%. Furthermore, it can also avoid mold wear and production cycle waste caused by excessive holding pressure.

[0025] Meanwhile, traditional technical solutions, unable to detect internal shrinkage conditions, suffer from a thermal shrinkage rate of up to 12.4%, and exhibit significant performance fluctuations within the same batch. In contrast, the adaptive control system in this embodiment, by predicting the evolution of isolated liquid phase regions and dynamically optimizing the holding pressure curve, can increase the internal quality pass rate from 86.2% to 98.7%, reducing performance fluctuations by more than 60%. This fundamentally ensures the mechanical performance stability of high-end wheel hubs and significantly reduces the cost of rework.

[0026] In this embodiment, multi-source heterogeneous operating condition data acquired is used as input to the constructed composite digital twin, with the output being the corresponding real-time estimated value. (Reference) Figure 2 and Figure 3 This embodiment provides a method for operating a state observer, which specifically includes the following steps: Step SA1: Data Acquisition. This involves deploying multiple sensors on the low-pressure casting machine and the mold to construct a corresponding three-dimensional sensing network and acquire corresponding multi-source heterogeneous operating condition data. Specifically, a non-contact infrared thermal imager (e.g., resolution ≥ 640×480, frame rate ≥ 30Hz, thermal sensitivity < 0.05℃, spectral range 7.5μm ≤ 14μm) is fixedly installed on the gantry or independent support above the casting machine. The distance between the non-contact infrared thermal imager and the mold surface is 1.5-2.5 meters to ensure that the field of view of the non-contact infrared thermal imager can cover the entire opening and closing area of ​​the mold. In other words, the non-contact infrared thermal imager acquires a continuous two-dimensional temperature distribution map of the mold surface, i.e., the corresponding temperature matrix.

[0027] Furthermore, during the mold processing stage, at least one thermocouple (such as a type K or N armored thermocouple with a temperature resistance of ≥800℃) is installed in areas where shrinkage defects are prone to occur in the casting, i.e., typical hot spots (such as the center of the spoke, the connection between the spoke and the rim (R angle), the rim flange, and near the gate) to collect the measured temperature value of the thermocouple corresponding to each typical hot spot.

[0028] Furthermore, a pressure transmitter (e.g., 0-1MPa range, 0.5%FS accuracy) and a gas mass flow meter (e.g., thermal mass flow meter) are installed on the pressurization pipeline of the insulation furnace. The pressure transmitter is used to collect the corresponding pouring pressure, and the gas mass flow meter is used to collect the corresponding pressurized gas.

[0029] Specifically, through the central controller, information is exchanged with non-contact infrared thermal imagers, thermocouples, pressure transmitters, and gas mass flow meters to collect the corresponding temperature matrix, measured temperature values ​​of thermocouples, pouring pressure, and pressurized gas. At the same time, the collected temperature matrix, measured temperature values ​​of thermocouples, pouring pressure, and pressurized gas are spliced ​​and combined to construct the corresponding multi-source heterogeneous operating condition data.

[0030] Step SA2: Initial Construction of the Digital Twin. This involves constructing a corresponding physical mechanism model using multiphysics simulation software. Using the multi-source heterogeneous operating condition data obtained in Step SA1 and the established deep learning model (e.g., a convolutional neural network-long short-term memory network model), a corresponding data-driven model is constructed. Simultaneously, the constructed physical mechanism model and the data-driven model are combined to create the initial composite digital twin. Details are as follows: Step SA2.1: Digital Reconstruction. This involves creating a 3D geometric model based on the casting mold design drawings using 3D CAD software (such as UGNX, CATIA). The model includes the mold itself (e.g., upper mold, lower mold, side mold), cavity, gating system, cooling channels, and venting channels. Simultaneously, the constructed 3D geometric model is meshed, with the mesh density of the spoke area being greater than that of the rim area.

[0031] Furthermore, by using the material library of multiphysics simulation software (such as ProCAST, FLOW-3D CAST, or AnyCasting), corresponding material properties are set for each component in the created 3D geometric model. For example, using A356 aluminum alloy, the material properties corresponding to the casting in the 3D geometric model are set, along with physical property parameters such as specific heat capacity, thermal conductivity, density, viscosity, solid / liquid phase temperature, latent heat of crystallization, and shrinkage rate, which vary with temperature. Using H13 mold steel, the material properties corresponding to the mold in the 3D geometric model are set, along with the corresponding thermophysical parameters. Simultaneously, based on the initial preheating temperature of the mold (e.g., 350℃) and the initial pouring temperature of the molten aluminum (e.g., 720℃), the initial operating conditions of the 3D geometric model are set. The system is based on the established pressure boundary (i.e., defining a time-varying pressure curve at the riser or gate, the initial setting of which can be specifically set according to historical process cards)), heat transfer boundary (i.e., defining the natural convection and radiation heat transfer coefficients with the environment on the outer surface of the mold), and zonal settings for the interface heat transfer coefficient (which can assign different initial values ​​to different areas of the mold and casting contact surface according to actual needs, such as setting 800W / (m²) in areas with thicker coatings or expected poor contact (e.g., deep cavities, corners). 2 The initial value of the interfacial heat transfer coefficient (·K) is set at 2000 W / (m²) in a flat, well-contact region. 2 ·K) initial value of the interface heat transfer coefficient. ), set the boundary operating conditions of the three-dimensional geometric model.

[0032] In other words, the set component material properties, initial operating conditions, and boundary operating conditions are combined with the created three-dimensional geometric model to construct the corresponding physical mechanism model.

[0033] Step SA2.2: Data Fusion. This involves collecting historical process data for the low-pressure casting process of aluminum alloy automotive wheels, including corresponding time-series process data (i.e., the acquired temperature matrix, measured thermocouple temperature values, and time-series casting pressure curves), process setting data (i.e., the casting temperature, pressure settings at each stage, and cooling water start-up time for this batch of production), and result label data (i.e., by acquiring industrial CT scan images of the castings, marking the location, size, and morphology of internal defects such as shrinkage porosity and gas pores, and simultaneously determining the microstructure information such as the secondary dendrite arm spacing and solidification time at preset locations through metallographic samples). Simultaneously, based on the timestamps corresponding to the time-series process data, process setting data, and result label data, time synchronization and alignment processing is performed to match time-series process data, process setting data, and result label data with the same timestamp, constructing a corresponding training dataset.

[0034] Furthermore, a corresponding data-driven model is constructed using a convolutional neural network model and a long short-term memory network model. This data-driven model is then combined with the physical mechanism model constructed in step SA2.1 to create an initial composite digital twin. Specifically, the constructed training dataset is used as the input to the initial composite digital twin. The time-series process data in the training dataset serves as the input to the data-driven model, the process setting data as the input to the physical mechanism model, and the result label data as the output of the initial composite digital twin. This training dataset is then used to train and validate the initial composite digital twin.

[0035] Step SA3: Twin Update. This involves using the data-driven model set in step SA2.2 to output predicted data values ​​corresponding to historical process data (including predicted temperature values ​​for all pixels on the mold surface and predicted thermocouple temperature values ​​for each typical hot spot location; and constructing a corresponding predicted temperature matrix based on the predicted temperature value for each pixel). The output predicted data values ​​(i.e., the predicted temperature matrix and thermocouple predicted temperature values) are compared with the temperature matrix and measured thermocouple temperature values ​​(i.e., actual measured data values) from the multi-source heterogeneous operating condition data in step SA1 to obtain the residuals between the predicted and actual measured data values ​​(i.e., the residuals between the temperature matrix and the predicted temperature matrix, and the residuals between the measured thermocouple temperature values ​​and the predicted thermocouple temperature values). In other words, the initial composite digital twin is updated using a state estimation algorithm (such as the extended Kalman filter algorithm) and the obtained residuals to obtain the corresponding final composite digital twin. Specifically: Step SA3.1: Data Prediction. This involves combining the set of optimal estimates from the previous time step (including the position and morphology of the molten metal flow front, the real-time predicted solid fraction at typical hot spots, the real-time predicted local pressure, and the real-time predicted local pressure change rate) and the posterior estimation error covariance matrix obtained from the initial composite digital twin output, with the control command for the current time step (i.e., the target pressure setpoint for the pressure servo system), to obtain the predicted state vector and the prior estimation error covariance matrix for the current time step. Specifically: ; in: Let k be the predicted state vector at time k. This is the set of optimal estimates corresponding to time k-1. This is the control command corresponding to time k. Let be the prior estimation error covariance matrix corresponding to time k. For Jacobian matrices, Let be the transpose of the Jacobian matrix. Let be the posterior estimation error covariance matrix corresponding to time k-1. The process noise covariance matrix is... This is the state transition function.

[0036] Step SA3.2: Data Extraction. This involves extracting the temperature matrix and measured thermocouple temperature values ​​corresponding to the current time based on the multi-source heterogeneous operating condition data. The temperature matrix and measured thermocouple temperature values ​​are then compared with the predicted state vector obtained in Step SA3.1 to obtain the residual between the predicted and measured data values. Specifically: ; in: Let K be the residual vector at time k. This represents the actual measured data value at time k. Let k be the predicted state vector at time k. For observation functions.

[0037] Furthermore, by extracting the corresponding linearized matrix from the multi-source heterogeneous operating condition data at the current time, and combining it with the prior estimation error covariance matrix obtained in step SA3.1 at the current time, the Kalman gain matrix at the current time is obtained, specifically: in: Let K be the Kalman gain matrix at time k. Let be the prior estimation error covariance matrix corresponding to time k. For Jacobian matrices, Let be the transpose of the Jacobian matrix. To observe the noise covariance matrix.

[0038] Step SA3.3: Parameter Update. This involves combining the residual vector and Kalman gain matrix obtained in Step SA3.2 with the predicted state vector at the current time step to determine the optimal set of estimates for the current time step. Specifically: ; in: This represents the set of optimal estimates corresponding to time k. Let k be the predicted state vector at time k. Let K be the Kalman gain matrix at time k. Let be the residual vector at time k.

[0039] Furthermore, based on the Kalman gain matrix obtained in step SA3.2 for the current time step, the difference between the Kalman gain matrix and the identity matrix is ​​obtained. This difference is then combined with the prior estimation error covariance matrix for the current time step to determine the posterior estimation error covariance matrix for the current time step. Specifically,... ; in: Let be the posterior estimation error covariance matrix corresponding to time k. Let K be the Kalman gain matrix at time k. Let be the prior estimation error covariance matrix corresponding to time k. For Jacobian matrices, It is an identity matrix.

[0040] Step SA4: Model Correction. Based on the optimal set of estimates and the posterior estimation error covariance matrix determined in Step SA3.3, the model parameters of the initial composite digital twin for the current moment are updated to obtain the final composite digital twin for the current moment. Simultaneously, the multi-source heterogeneous operating condition data obtained in Step SA1 is used as input to the final composite digital twin, outputting real-time estimates for the current moment, including the position and morphology of the molten metal flow front, the real-time predicted solid fraction at typical hot spot locations, the real-time predicted local pressure, and the real-time predicted rate of change of local pressure.

[0041] In this embodiment, during the pressure holding process, based on the real-time estimated value obtained in step SA4 and the data-driven model set in the final composite digital twin, the comprehensive cost value corresponding to each candidate path is determined. Then, based on the determined comprehensive cost value, the optimal pressure path, i.e., the corresponding optimal pressure holding curve, is selected from all candidate paths. (Reference) Figure 4 This embodiment provides an adaptive optimal decision control method, which specifically includes the following steps: Step SB1: Switching Determination. Based on the real-time estimated values ​​obtained in Step SA4, the real-time predicted solid fraction and real-time predicted local pressure change rate corresponding to each typical hot spot location are determined. Simultaneously, the determined real-time predicted solid fraction and real-time predicted local pressure change rate are compared with corresponding preset trigger thresholds (which can be specifically set according to actual needs, so this embodiment does not elaborate on them; each typical hot spot location has a corresponding preset solid fraction trigger threshold and preset negative local pressure change rate threshold). Based on the comparison results, the corresponding pressure holding trigger signal is determined. Specifically: When the real-time predicted solid fraction at the same typical hot spot location is not less than the preset solid fraction trigger threshold, and the corresponding real-time predicted local pressure change rate is less than the preset negative local pressure change rate threshold, a pressure holding trigger signal is triggered, and the pressure holding process begins. Otherwise, the pressure holding trigger signal is not triggered, and the current operating state continues.

[0042] Step SB2: Path Determination. Based on the pressurized gas pressure setpoint corresponding to the preset time period (which can be specifically set according to actual needs, so it is not specifically described in this embodiment), the pressurized gas pressure setpoint corresponding to each discrete time within the preset time period is determined. The pressurized gas pressure setpoint corresponding to each discrete time and the set of optimal estimates corresponding to the current time determined in step SA3.3 are both used as inputs to the data-driven model set in the final composite digital twin obtained in step SA4, so as to output the predicted total volume of the isolated liquid phase region and the sum of squares of the accumulated pressure sequence corresponding to each candidate path.

[0043] Furthermore, based on the predicted total volume of the isolated liquid phase region and the sum of squares of the accumulated pressure sequence corresponding to each candidate path, the predicted total volume of the isolated liquid phase region and the sum of squares of the accumulated pressure sequence are combined to obtain the comprehensive cost value corresponding to each candidate path (it is worth noting that only the numerical values ​​are added here, without any unit merging). Simultaneously, the comprehensive cost values ​​of each candidate path are compared to determine the corresponding minimum comprehensive cost value, and the candidate path corresponding to the minimum comprehensive cost value is set as the corresponding optimal pressure path, i.e., the corresponding optimal holding pressure curve.

[0044] In this embodiment, the optimal pressure path determined in step SB2 is corrected using a known disturbance source to obtain the corresponding corrected pressure holding curve. Based on the corrected pressure holding curve and the actual pressure values ​​collected in the gas path, the corresponding drive command is determined. (Reference) Figure 5 This embodiment provides an error suppression execution method, which specifically includes the following steps: Step SC1: Curve Correction. Based on the optimal pressure path determined in Step SB2, the target pressure setpoint for each moment along the optimal pressure path is obtained. Then, using a set feedforward controller, the target pressure setpoint for each moment is corrected to obtain the corrected target pressure value for that moment. Specifically: ; in: The target pressure correction value, The steady-state gain of the controlled object (i.e., from valve command to furnace pressure). The filter time constant is For the Laplace operator, Set a value for the target pressure. The dominant time constant of the controlled object. The pure time delay of the controlled object. It is a natural constant.

[0045] In other words, based on the target pressure correction value corresponding to each moment, all target pressure correction values ​​are connected sequentially according to time order to construct the corrected holding pressure curve and the target pressure correction value corresponding to each moment.

[0046] Step SC2: Feedback Correction. This involves using a pressure transmitter installed in the gas path to acquire the actual pressure value at each moment in real time. Simultaneously, the timestamp corresponding to the acquired actual pressure value is used to determine the target pressure correction value for the same timestamp from the corrected holding pressure curve constructed in Step SC1. The actual pressure value and the target pressure correction value for the same timestamp are then compared to obtain the pressure difference between them, i.e., the corresponding pressure error.

[0047] Furthermore, based on the pressure error at each moment in the corrected holding pressure curve, the cumulative total pressure error corresponding to the corrected holding pressure curve is obtained. Then, based on the pressure error at each moment and the obtained cumulative total pressure error, the corresponding drive command is determined through the configured PID controller, specifically: ; in: This is the driving instruction corresponding to time k. The proportional gain of the PID controller. This represents the pressure error at time k. The integral gain of the PID controller. This is the cumulative sum of total pressure errors. The derivative gain of the PID controller, This represents the pressure error at time k-1. The sampling period.

[0048] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended embodiments and their equivalents.

Claims

1. An adaptive control system for low-pressure casting process parameters of aluminum alloy automotive wheel hubs, characterized in that, Including: State observer operation module: It collects multi-source heterogeneous operating condition data through the set sensor network, and uses the multi-source heterogeneous operating condition data as the input of the constructed composite digital twin, and outputs the corresponding real-time estimated value. Adaptive optimal decision control module: compares the real-time estimated value with the preset trigger threshold, and determines the corresponding pressure holding trigger signal based on the comparison result. At the same time, during the pressure holding process, the real-time estimated value is used as the input of the data-driven model set in the composite digital twin, and the output obtains the comprehensive cost value corresponding to each candidate path. Based on the comprehensive cost value, the corresponding optimal pressure holding curve is determined. Error suppression execution module: corrects the optimal pressure holding curve, obtains the corresponding corrected pressure holding curve, and determines the target pressure setting value at each moment. At the same time, when the pressure holding trigger signal is triggered, it collects the actual pressure value in the gas path and determines the corresponding drive command based on the pressure error between the actual pressure value and the target pressure setting value.

2. The adaptive control system for low-pressure casting process parameters of aluminum alloy automotive wheel hubs according to claim 1, characterized in that, The output retrieves the corresponding real-time estimates, including: SA1: Data Acquisition: Using a non-contact infrared thermal imager, thermocouples, pressure transmitters, and gas mass flow meters, the corresponding temperature matrix, measured thermocouple temperature values, pouring pressure, and pressurized gas are acquired to construct the corresponding multi-source heterogeneous operating condition data. SA2: Initial construction of the digital twin: Using multiphysics simulation software, a corresponding physical mechanism model is constructed. Using the multi-source heterogeneous working condition data and deep learning model, a corresponding data-driven model is constructed. The physical mechanism model and the data-driven model are combined to construct the corresponding initial composite digital twin. SA3: Twin Update: Through the data-driven model, the corresponding predicted data value is obtained, and the model parameters for updating the initial composite digital twin are updated according to the residual between the predicted data value and the multi-source heterogeneous operating condition data. SA4: Model Correction: Based on the updated model parameters, update the initial composite digital twin corresponding to the current moment to obtain the corresponding final composite digital twin. Use the multi-source heterogeneous operating condition data as the input to the final composite digital twin and output the real-time estimated value corresponding to the current moment, including the position and shape of the molten metal flow front, the real-time predicted solid fraction of typical hot spot positions, the real-time predicted local pressure, and the real-time predicted local pressure change rate.

3. The adaptive control system for low-pressure casting process parameters of aluminum alloy automotive wheel hubs according to claim 2, characterized in that, The corresponding initial composite digital twin is constructed, including: SA2.1: Digital Reconstruction: Based on the design drawings of the casting mold, a three-dimensional geometric model is created, and the component material properties, initial operating conditions and boundary operating conditions of the three-dimensional geometric model are set through multiphysics simulation software to construct the corresponding physical mechanism model; SA2.2: Data Fusion: By using time-series process data, process setting data, and result label data from historical process data, a corresponding training dataset is constructed. At the same time, a corresponding data-driven model is set using a convolutional neural network model and a long short-term memory network model. The data-driven model and the physical mechanism model are combined to construct a corresponding initial composite digital twin. The initial composite digital twin is then trained and validated using the training dataset.

4. The adaptive control system for low-pressure casting process parameters of aluminum alloy automotive wheel hubs according to claim 3, characterized in that, The time-series process data in the training dataset is used as the input to the data-driven model, the process setting data in the training dataset is used as the input to the physical mechanism model, and the result label data in the training dataset is used as the output of the initial composite digital twin. The initial composite digital twin is then trained and validated.

5. The adaptive control system for low-pressure casting process parameters of aluminum alloy automotive wheel hubs according to claim 2, characterized in that, The model parameters for updating the initial composite digital twin are updated, including: SA3.1: Data Prediction: Based on the set of optimal estimated values ​​and the posterior estimation error covariance matrix obtained from the output of the initial composite digital twin, the set of optimal estimated values ​​and the posterior estimation error covariance matrix obtained from the output of the initial composite digital twin are combined with the control command obtained from the output of the current time to obtain the predicted state vector and the prior estimation error covariance matrix obtained from the output of the current time. SA3.2: Data extraction: Compare the multi-source heterogeneous operating condition data corresponding to the current time with the predicted state vector corresponding to the current time to obtain the residual vector between the multi-source heterogeneous operating condition data and the predicted state vector. At the same time, combine the multi-source heterogeneous operating condition data with the prior estimation error covariance matrix corresponding to the current time to obtain the Kalman gain matrix corresponding to the current time. SA3.3: Parameter Update: Combine the Kalman gain matrix, residual vector and predicted state vector at the current time to determine the optimal set of estimates at the current time. At the same time, based on the Kalman gain matrix, identity matrix and prior estimation error covariance matrix at the current time, determine the posterior estimation error covariance matrix at the current time.

6. The adaptive control system for low-pressure casting process parameters of aluminum alloy automotive wheel hubs according to claim 1, characterized in that, The corresponding optimal holding pressure curve is determined, including: SB1: Switching determination: Based on the real-time estimated value, determine the real-time predicted solid fraction and real-time predicted local pressure change rate corresponding to each typical hot spot location, and compare the real-time predicted solid fraction and real-time predicted local pressure change rate with a preset trigger threshold to determine the corresponding pressure holding trigger signal; SB2: Path Determination: Based on the pressurized gas pressure setpoint corresponding to each discrete time within a preset time period, the pressurized gas pressure setpoint and the set of optimal estimates corresponding to the current time are both used as inputs to the data-driven model. The output obtains the predicted total volume of the isolated liquid phase region and the sum of squares of the accumulated pressure sequence corresponding to each candidate path. At the same time, the predicted total volume of the isolated liquid phase region and the sum of squares of the accumulated pressure sequence are combined to obtain the comprehensive cost value corresponding to each candidate path. The minimum comprehensive cost value is determined from all comprehensive cost values. The candidate path corresponding to the minimum comprehensive cost value is the optimal pressure holding curve.

7. The adaptive control system for low-pressure casting process parameters of aluminum alloy automotive wheel hubs according to claim 6, characterized in that, The preset trigger thresholds include a preset solid fraction trigger threshold and a preset negative local pressure change rate threshold. The preset solid fraction trigger threshold is compared with the real-time predicted solid fraction, and the preset negative local pressure change rate threshold is compared with the real-time predicted local pressure change rate. Based on the comparison results, the corresponding pressure holding trigger signal is determined, specifically: When the real-time predicted solid fraction is not less than the preset solid fraction trigger threshold and the real-time predicted local pressure change rate corresponding to the same typical hot spot location is less than the preset negative local pressure change rate threshold, a pressure holding trigger signal is triggered, and the pressure holding process begins; otherwise, the pressure holding trigger signal is not triggered, and the current operating state continues.

8. The adaptive control system for low-pressure casting process parameters of aluminum alloy automotive wheel hubs according to claim 1, characterized in that, The corresponding driver instructions were determined, including: SC1: Curve Correction: Based on the optimal pressure path, obtain the target pressure setpoint corresponding to each moment in the optimal pressure path, and correct the target pressure setpoint at each moment through the set feedforward controller to obtain the target pressure correction value corresponding to each moment. Then, connect the target pressure correction values ​​in chronological order to construct the corrected holding pressure curve. SC2: Feedback Correction: Acquire the actual pressure value at each moment in the gas path, and determine the target pressure correction value corresponding to the same time point in the corrected pressure holding curve based on the timestamp corresponding to the actual pressure value. Compare the actual pressure value and the target pressure correction value at the same time point to obtain the corresponding pressure error and the total pressure error accumulation. At the same time, determine the corresponding drive command through the set PID controller based on the pressure error corresponding to each moment and the total pressure error accumulation.

Images