Integrated industrial control system and method of vacuum die-casting machine

Abstract

The invention discloses an integrated industrial control system and method for a vacuum die-casting machine, and belongs to the technical field of industrial control, and the method comprises the steps that a multi-dimensional data collection module obtains multi-source state data and upstream batch data in the die-casting process; the flow field reconstruction and preprocessing module synchronously filters the data to obtain first feature data, and maps the first feature data into flow field reconstruction data through a reduced-order flow field observation model; the adaptive strategy generation module determines a dynamic impedance matching control strategy through multi-physical field coupling deduction in combination with mold thermal deformation data, downstream quality feedback and flow field data; and the collaborative execution and interaction module generates a driving instruction according to the strategy, sends the driving instruction to an execution mechanism, packages an execution result and returns the execution result to the manufacturing execution system. According to the method, real-time reconstruction of the three-dimensional flow field is achieved through the reduced-order model, prospective gas entrapment prediction and continuous impedance matching are combined, the problem of computing power and data synchronization is solved, cross-process closed-loop quality control is achieved, and the forming quality and production stability of vacuum die casting are improved.

Description

Technical Field

[0001] This invention relates to the field of industrial control technology, and more particularly to an integrated industrial control system and method for a vacuum die-casting machine. Background Technology

[0002] In the field of modern new energy vehicles and high-end motor manufacturing, the motor housing, as a core support and sealing component, directly determines the power density and operational reliability of the motor due to its casting quality. Such housings typically possess geometric characteristics such as thin walls, deep cavities, complex internal cooling channels, and high airtightness requirements. Traditional high-pressure die casting processes are prone to defects such as air entrapment, porosity, and cold shuts during the filling process. To address this issue, existing technologies generally employ a strategy of vacuum-assisted die casting combined with closed-loop feedback control. Specifically, the conventional approach involves installing vacuum sensors at key locations in the mold to monitor the absolute pressure within the cavity in real time. The control system compares the measured vacuum level with the target curve based on a preset PID algorithm. When the system detects that the vacuum level deviates from the set value due to residual gas, it dynamically adjusts the speed of the injection punch, for example, reducing the speed to wait for venting when the vacuum level is insufficient, or accelerating to ensure smooth filling when the vacuum level is good. This control logic based on single-point pressure feedback achieves basic automated execution, ensures the standardization of the production process, and improves the internal quality of the castings to a certain extent.

[0003] However, the aforementioned conventional technical solutions reach bottlenecks in control accuracy and system integration when dealing with complex motor housings made of ultra-high-strength aluminum alloys or ultra-thin walls. Existing solutions suffer from the bias of representing the whole with a localized view, incorrectly equating the value of a single-point vacuum sensor with the exhaust status of the entire complex cavity. The internal flow channels of the motor housing mold are tortuous, and the vacuum sensor can only reflect the local pressure near its installation location, failing to detect the residual gas state in dead zones far from the sensor. When the sensor indicates that the vacuum is within acceptable limits, the acceleration of the punch often causes molten metal to prematurely block the exhaust channels in unmonitored areas, resulting in hidden gas entrapment defects. Furthermore, existing die-casting equipment is often in an automated island state, with its control system typically employing closed logic and lacking standardized data interfaces with manufacturing execution systems, enterprise resource planning systems, and warehouse management systems. The equipment cannot adaptively adjust based on batch rheological characteristics differences in upstream materials, factory-level energy dispatch instructions, or real-time quality feedback from downstream machining and airtightness testing processes. This results in the control benchmark for each injection cycle being isolated from the overall production data, severely restricting the overall manufacturing yield and production collaboration efficiency of complex housing castings. Summary of the Invention

[0004] This invention overcomes the shortcomings of the prior art and provides an integrated industrial control system and method for a vacuum die-casting machine.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is: an integrated industrial control system for a vacuum die-casting machine, comprising:

[0006] The multi-dimensional data acquisition module is used to acquire multi-source state data during the die-casting process and upstream batch data from the factory's manufacturing execution system.

[0007] The flow field reconstruction and preprocessing module is connected to the multi-dimensional data acquisition module. It is used to perform spatiotemporal synchronization and filtering on the multi-source state data to obtain first feature data, and to perform mapping calculation on the first feature data based on the preset reduced-order flow field observation model to obtain flow field reconstruction data characterizing the dynamics of the three-dimensional fluid front inside the cavity.

[0008] An adaptive strategy generation module, connected to the flow field reconstruction and preprocessing module, is used to acquire mold thermal deformation reference data and downstream process quality feedback data, and determine the dynamic impedance matching control strategy of the target exhaust node through multi-physics coupling deduction based on the flow field reconstruction data, the mold thermal deformation reference data, and the downstream process quality feedback data.

[0009] The collaborative execution and interaction module, connected to the adaptive strategy generation module, is used to generate underlying drive instructions based on the dynamic impedance matching control strategy, send the underlying drive instructions to the external die-casting actuator, and encapsulate the execution result into a standard protocol data packet and send it back to the factory manufacturing execution system.

[0010] In a preferred embodiment of the present invention, the flow field reconstruction and preprocessing module includes a wavelet denoising unit, a feature mode extraction unit, and a transient mapping unit;

[0011] The wavelet noise reduction unit uses a precise time protocol based on a hardware clock to perform spatiotemporal synchronization of the multi-source state data; and uses a combination of discrete wavelet transform and soft threshold function to filter out high-frequency impulse noise and output the first feature data.

[0012] The feature mode extraction unit is used to extract the flow field spatial mode matrix offline;

[0013] The transient mapping unit constructs the first feature data into a transient state vector; by solving a set of low-dimensional ordinary differential equations with embedded thermo-fluid-structure interaction viscosity penalty terms, it obtains reduced-order characteristic coefficients; and then linearly combines and projects the reduced-order characteristic coefficients with the flow field spatial mode matrix to reconstruct the flow field reconstruction data.

[0014] In a preferred embodiment of the present invention, the adaptive strategy generation module includes a boundary correction unit, a swirling air prediction unit, and an impedance matching unit;

[0015] The boundary correction unit solves for the dynamic geometric deformation offset of the mold based on the three-dimensional thermoelastic partial differential equations and the mold thermal deformation reference data, so as to update the physical boundary of the reduced-order flow field observation model in real time.

[0016] The air-entrainment prediction unit predicts the motion topology of the molten metal front within a preset look-ahead time window based on the dynamic evolution equation of the fluid volume function interface, and determines the probability of the target exhaust node generating a closed air bag.

[0017] When the probability exceeds the safety threshold, the impedance matching unit solves the continuous opening change sequence according to the acoustic impedance matching principle. This sequence balances the flow resistance characteristics of the gas flow in the exhaust pipe with the dynamic driving torque generated by the molten metal propelling and compressing the gas. This sequence is then used as the dynamic impedance matching control strategy.

[0018] In a preferred embodiment of the present invention, the multidimensional data acquisition module includes a high-frequency sensing unit, a thermal field sensing unit, and an industrial interconnection interface unit.

[0019] The industrial interconnection interface unit extracts the spectral analysis parameters of the current aluminum alloy material by parsing the received data packets, and uses them as the upstream batch data;

[0020] The system is also equipped with a preset composition-viscosity mapping relationship library; the multidimensional data acquisition module calculates the dynamic viscosity coefficient of the current aluminum liquid according to the spectral analysis parameters and the composition-viscosity mapping relationship library; and uses the dynamic viscosity coefficient as the initial physical parameter to overwrite the configuration file of the reduced-order flow field observation model to realize adaptive preloading of process parameters based on the characteristics of incoming materials.

[0021] In a preferred embodiment of the present invention, the adaptive strategy generation module uses a unified architecture protocol to pull the visual inspection report of the shrinkage hole on the cutting surface of the downstream computer numerical control machining center and the micro-leakage test log of the airtightness testing equipment as the downstream process quality feedback data.

[0022] The adaptive strategy generation module uses a preset homogeneous coordinate transformation matrix to reverse-map the three-dimensional spatial coordinates of historical defects presented in the downstream process quality feedback data to the corresponding grid nodes of the reduced-order flow field observation model; and corrects the propulsion velocity field of the fluid front by adjusting the convective heat transfer coefficient or local flow resistance coefficient of the corresponding grid nodes.

[0023] In a preferred embodiment of the present invention, the collaborative execution and interaction module includes an instruction compilation unit and an energy efficiency coordination unit;

[0024] The instruction compilation unit converts the dynamic impedance matching control strategy into a high-frequency pulse width modulation signal to drive the actuator at the end of the vacuum exhaust block; and based on the fluid resistance change trend predicted by the flow field reconstruction data, it calculates the dynamic driving torque required for the injection system to overcome inertia and viscous friction, and generates a torque feedforward control signal for the main injection servo motor.

[0025] The energy efficiency coordination unit subscribes to the real-time power limit data of the energy management system and dynamically issues frequency conversion step-down commands during the non-peak stage of the die-casting cycle to optimize the power flow.

[0026] This invention provides an integrated industrial control method for a vacuum die-casting machine, comprising:

[0027] Acquire multi-source state data during the die-casting process and upstream batch data from the factory's manufacturing execution system;

[0028] The multi-source state data is spatiotemporally synchronized and filtered to obtain first feature data; and the first feature data is mapped and calculated based on a preset reduced-order flow field observation model to obtain flow field reconstruction data characterizing the dynamics of the three-dimensional fluid front inside the cavity.

[0029] Acquire mold thermal deformation reference data and downstream process quality feedback data; and based on the flow field reconstruction data, the mold thermal deformation reference data, and the downstream process quality feedback data, determine the dynamic impedance matching control strategy of the target exhaust node through multi-physics field coupling deduction;

[0030] The underlying drive instructions are generated according to the dynamic impedance matching control strategy and sent to the external die-casting actuator; at the same time, the execution result is encapsulated into a standard protocol data packet and sent back to the factory manufacturing execution system.

[0031] In a preferred embodiment of the present invention, the step of performing spatiotemporal synchronization and filtering processing on the multi-source state data, and performing mapping calculation based on a reduced-order flow field observation model, includes:

[0032] Nanosecond-level spatiotemporal alignment of the multi-source state data is performed using a precision time protocol based on a hardware clock.

[0033] The aligned data is decomposed into multiple scales using the Dobessi wavelet basis; and the high-frequency detail coefficients are shrunk using a soft thresholding function before the first feature data is reconstructed.

[0034] The first feature data is constructed into a transient state vector and input into the reduced-order flow field observation model constructed by the intrinsic orthogonal decomposition algorithm and the Galerkin projection mechanism;

[0035] By solving a set of low-dimensional ordinary differential equations with embedded viscosity penalty terms for thermal-fluid-structure interaction, reduced-order characteristic coefficients are obtained; then, the reduced-order characteristic coefficients are linearly combined and projected with the pre-extracted flow field spatial mode matrix to reconstruct the flow field reconstruction data.

[0036] In a preferred embodiment of the present invention, the step of determining the dynamic impedance matching control strategy through multiphysics coupling deduction based on flow field reconstruction data, mold thermal deformation reference data, and downstream process quality feedback data includes:

[0037] Based on the three-dimensional thermoelastic partial differential equations and the mold thermal deformation reference data, the dynamic geometric deformation offset of the mold is calculated to update the physical boundary.

[0038] By utilizing the dynamic evolution equation of the fluid volume function interface, time integration is performed on the reconstructed flow field data within a preset look-ahead time window to predict the motion topology of the molten metal front.

[0039] When the risk of closed gas bag formation is predicted at the target exhaust node, a flow resistance network model is established based on the one-dimensional unsteady flow energy equation and mass conservation equation of compressible gas.

[0040] Solving for the objective function generates a sequence of servo proportional throttle valve opening changes with continuous time domain resolution, which serves as the dynamic impedance matching control strategy. The objective function is used to balance the flow resistance characteristics of the gas flow in the exhaust pipe with the dynamic driving torque generated by the molten metal propelling and compressing the gas.

[0041] In a preferred embodiment of the present invention, after acquiring the upstream batch data, the method further includes: extracting the spectral analysis parameters of the current aluminum alloy material from the upstream batch data; calculating the dynamic viscosity coefficient of the current aluminum liquid by combining it with a preset composition-viscosity mapping relationship library; and overwriting the dynamic viscosity coefficient as an initial physical parameter into the reduced-order flow field observation model.

[0042] In the process of determining the dynamic impedance matching control strategy through multiphysics coupling deduction, the following steps are also included: analyzing the three-dimensional spatial coordinates of historical defects in the downstream process quality feedback data; mapping the three-dimensional spatial coordinates of historical defects to the corresponding grid nodes of the reduced-order flow field observation model through affine transformation of spatial coordinates; and adjusting the convective heat transfer coefficient or local flow resistance coefficient of the corresponding grid nodes to correct the propulsion velocity field of the fluid front.

[0043] This invention addresses the shortcomings of the prior art and has the following beneficial effects:

[0044] (1) This invention transforms high-dimensional fluid computation problems into online solutions of low-dimensional ordinary differential equations through a reduced-order flow field observation model. By combining discrete wavelet transform and precise time protocol, real-time reconstruction of the three-dimensional flow field of complex motor housing cavity is achieved within the die-casting cycle. This method solves the contradiction between the inability of traditional single-point vacuum sensors to characterize the global flow state and the difficulty of adapting complex calculations to the transient filling process. It transforms the evolution characteristics of the fluid front from an invisible black box state to a perceptible control variable, thereby significantly improving the perception accuracy of complex thin-walled parts under high-speed filling.

[0045] (2) This invention calculates the geometric offset caused by mold thermal deformation in real time through a boundary correction unit, and uses a homogeneous coordinate transformation matrix to reverse-map the three-dimensional coordinates of quality defects fed back from downstream machining to the flow field mesh, thus realizing cross-process data closure and dynamic self-adaptation of physical boundaries. This overcomes the problems of exhaust control failure caused by mold thermal deviation and information silos between die casting and subsequent quality inspection processes in conventional solutions, compensates for cross-process cumulative errors from the whole process dimension, and greatly enhances the robustness of the production system to process fluctuations.

[0046] (3) This invention uses an air entrapment prediction unit to perform forward-looking deduction based on the interface dynamic evolution equation and generates a continuous proportional throttle valve opening sequence based on the acoustic impedance matching principle, thus realizing the control transition from passive pressure feedback to active flow guidance. This method solves the problems of shock wave splashing and exhaust plate blockage induced by traditional two-position switch control, ensuring that the metal liquid front advances in a stable laminar flow state and effectively eliminating the hidden air entrapment defects in complex deep cavity parts such as motor housings.

[0047] (4) This invention converts the impedance matching strategy into a high-frequency pulse width modulation signal through an instruction compilation unit, drives the high-speed actuator, and synchronously calculates the torque feedforward signal of the injection motor, realizing microsecond-level electromechanical coordination between the pumping resistance adjustment and the injection driving torque. With the adaptive preloading of the viscosity of upstream material batches, the system can achieve millisecond-level response to the characteristics of aluminum liquid and the fluctuation of filling resistance in different furnaces. While ensuring ultra-high filling quality, it greatly optimizes energy utilization efficiency, thereby promoting the transformation of vacuum die casting process from single equipment control to integrated intelligent manufacturing. Attached Figure Description

[0048] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0049] Figure 1This is a preferred embodiment of the system architecture and industrial internet topology diagram of the present invention;

[0050] Figure 2 This is a flowchart of the steps of a preferred embodiment of the present invention;

[0051] Figure 3 This is a flowchart of the physical field coupling deduction process generated by the adaptive strategy in a preferred embodiment of the present invention. Detailed Implementation

[0052] 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.

[0053] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein. Therefore, the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0054] Application Overview:

[0055] This invention addresses the vacuum die-casting of complex three-dimensional components for new energy vehicles, such as motor housings. These parts possess geometric features including ultra-thin walls, deep cavities, and complex internal cooling channels, demanding extremely high airtightness and mechanical properties. To achieve comprehensive factory control, data integration is needed between die-casting equipment and manufacturing execution systems, quality inspection systems, and warehouse management systems, forming a collaborative manufacturing process from materials to finished products. The limitation of existing technologies in this scenario lies in their control logic's reliance on a single-point vacuum sensor installed in a localized area of ​​the mold. This sensor detects pressure changes at that point to indirectly determine the venting status of the entire cavity. However, the mold cavity for complex three-dimensional components is a three-dimensional structure, and the flow front of the molten metal during filling is spatially highly uneven. Single-point data cannot characterize the actual gas residue in areas far from the sensor. Therefore, when the sensor indicates that the vacuum level meets the standard, a large amount of gas may still exist in the dead zones of the cavity. If the control system accelerates injection based on this false signal, the gas may be encapsulated by the molten metal, creating a hidden gas entrapment defect. This hidden gas entrapment defect cannot be corrected by conventional closed-loop feedback.

[0056] The underlying reason why existing technologies cannot solve the above problems lies in the fact that conventional vacuum die-casting control schemes are based on an implicit simplification: that controlling the filling speed over time automatically achieves high-quality filling of the entire cavity space. This assumption is flawed when applied to parts with complex three-dimensional geometries. During the filling process, the rheological behavior of molten metal is strongly influenced by various random or time-varying factors, such as batch composition fluctuations of the molten aluminum, thermal expansion and deformation of the mold, and uniformity of the release agent spraying. Even with more complex PID controllers or an increase in the number of sensors, conventional technologies are essentially passive responses to results, rather than active perception of the process space. When a sensor detects an abnormal pressure, the flow field morphology leading to the defect has already solidified; adjustments at this point are merely post-hoc compensations and cannot correct the ongoing air entrapment behavior.

[0057] In addition, conventional control systems use closed logic, and their control reference is derived from a preset ideal curve. They cannot receive batch difference information from upstream materials, nor can they adaptively correct based on the type of defects reported by downstream airtightness testing. As a result, the system lacks any compensation means when facing cumulative errors across processes.

[0058] This invention breaks away from the conventional architecture of single-point feedback control at the equipment level. It introduces the reduced-order model from computational fluid dynamics and the horizontal integration technology of the factory information system into the bottom layer of die casting control. By fusing material data from a multi-source sensor array with the upstream MES system, it drives a lightweight digital twin model embedded in the controller. It predicts the evolution trend of the fluid front and, combined with mold thermal deformation data and downstream CNC machining quality feedback, dynamically generates an impedance matching strategy for the vacuum valve and the injection system.

[0059] Exemplary system:

[0060] like Figure 1 As shown, an integrated industrial control system for a vacuum die-casting machine includes:

[0061] The multi-dimensional data acquisition module is used to acquire multi-source state data during the die-casting process and upstream batch data from the factory's manufacturing execution system.

[0062] The flow field reconstruction and preprocessing module is connected to the multi-dimensional data acquisition module. It is used to perform spatiotemporal synchronization and filtering on multi-source state data to obtain the first feature data. Based on the preset reduced-order flow field observation model, the first feature data is mapped and calculated to obtain flow field reconstruction data that characterizes the dynamics of the three-dimensional fluid front inside the cavity.

[0063] The adaptive strategy generation module, connected to the flow field reconstruction and preprocessing module, is used to acquire mold thermal deformation reference data and downstream process quality feedback data. Based on the flow field reconstruction data, mold thermal deformation reference data, and downstream process quality feedback data, the dynamic impedance matching control strategy of the target exhaust node is determined through multi-physics coupling deduction.

[0064] The collaborative execution and interaction module, connected to the adaptive strategy generation module, is used to generate underlying drive instructions based on the dynamic impedance matching control strategy and send the underlying drive instructions to the external die-casting actuator. At the same time, it encapsulates the execution results into standard protocol data packets and sends them back to the factory manufacturing execution system.

[0065] During system integration, introducing high-dimensional fluid dynamics calculations into the underlying industrial control system presents challenges related to computing power bottlenecks and data synchronization. Traditional computational fluid dynamics (CFD) simulations typically take several hours to complete, while actual industrial processes last only 10 to 1000 milliseconds. Due to the limited computing resources of industrial controllers and edge gateways, conventional solvers cannot be directly deployed. Therefore, it is essential to control model complexity while ensuring accuracy, guaranteeing that the output latency is below the critical time constant. This constitutes the core challenge of real-time reconstruction of 3D flow fields.

[0066] Establishing data links between the equipment layer, workshop-level management layer, and downstream testing equipment presents a significant technical challenge: parsing and semantic mapping of heterogeneous network protocols. Die-casting machines typically use proprietary industrial fieldbus protocols; Manufacturing Execution Systems (MES) and Enterprise Resource Planning (ERP) systems are mostly based on Ethernet and standard data exchange formats; while downstream airtightness testing equipment and machining equipment each follow different communication standards. To achieve cross-system data interaction, it is essential to parse these heterogeneous protocols in real time and map the data models of different devices into a unified semantic expression.

[0067] like Figure 2 As shown, based on the above system architecture, the workflow of this system within a single die-casting cycle is as follows:

[0068] The system extracts batch parameters of upstream materials and real-time physical status parameters during equipment operation through a multi-dimensional data acquisition module;

[0069] The flow field reconstruction and preprocessing module performs spatiotemporal alignment and order reduction mapping operations on the above heterogeneous data to reconstruct the three-dimensional dynamic distribution model of the metal fluid inside the mold cavity.

[0070] The adaptive strategy generation module integrates the microscopic thermal deformation parameters of the mold with downstream quality inspection feedback, infers potential gas retention risks based on flow field reconstruction data, and generates a dynamic impedance matching strategy for the exhaust assembly accordingly.

[0071] The collaborative execution and interaction module converts the matching strategy into underlying electrical signals to drive the relevant hardware to execute, and simultaneously uploads the full process operation parameters of this cycle to the workshop-level network system to build a complete cross-process data closed loop.

[0072] The specific execution process involves the following modules:

[0073] The multidimensional data acquisition module is used to build a communication and interaction mechanism between the underlying physical state perception of the system and the upper-level information network. Its function is not limited to the acquisition of physical quantities of a single device, but extends to the factory-level information network. Specifically, the multidimensional data acquisition module includes a high-frequency sensing unit, a thermal field sensing unit, and an industrial interconnection interface unit. The three units work together and independently acquire data in their respective dimensions. After preliminary integration, the acquired data is output to the downstream module. The high-frequency sensing unit and the thermal field sensing unit are responsible for acquiring real-time physical state data during the die-casting process, while the industrial interconnection interface unit is responsible for acquiring upstream batch data.

[0074] Furthermore, when executing the method logic for acquiring multi-source state data, the high-frequency sensing unit uses the EtherCAT industrial real-time Ethernet bus to simultaneously read the absolute encoder displacement data of the injection servo motor, the high-frequency dynamic pressure data of the booster cylinder, and the miniature piezoresistive vacuum data distributed in key nodes such as the mold exhaust channel and overflow groove, with an ultra-high sampling period of 100μs. The real-time physical quantities provide accurate time reference and spatial distribution information for subsequent flow field reconstruction, ensuring that the dynamic monitoring of the molten metal filling process has sufficient spatiotemporal resolution.

[0075] Meanwhile, the thermal field sensing unit uses fiber optic grating sensors (FBGs) embedded in the deep cavity structural ribs and areas prone to heat generation in the mold to collect temperature gradient data and thermal stress distribution data inside the mold. By detecting the drift of the center wavelength of the reflected light, the fiber optic grating sensors can be immune to electromagnetic interference on site and achieve micron-level thermal deformation sensing. Thermodynamic data provides key input for mold thermal deformation compensation, ensuring that the flow field reconstruction model can accurately reflect the influence of the actual mold state on the filling process.

[0076] In the data flow of acquiring upstream batch data, the industrial interconnection interface unit proactively sends a physical parameter request to the factory manufacturing execution system before the target die casting batch operation. The interface unit extracts the spectral analysis parameters of the mass fraction of silicon, magnesium and iron elements in the aluminum alloy material of the current furnace by parsing the received JSON format data packet, as well as the hydrogen content test index after the refining and degassing process.

[0077] The multi-dimensional data acquisition module initially encapsulates the synchronously acquired multi-source state data and upstream batch data, and transmits them along the data bus to the flow field reconstruction and preprocessing module, using them as the basic input conditions for subsequent fluid rheological characteristic deduction and calculation, and establishing a data transmission link between cross-level manufacturing nodes.

[0078] The flow field reconstruction and preprocessing module, acting as an intermediary between the underlying physical devices and the upper-level virtual digital twin space, has the core function of converting the low-dimensional sensing signals transmitted by the multi-dimensional data acquisition module into a high-dimensional three-dimensional flow field spatial representation. Internally, the module mainly includes a wavelet denoising unit, a feature mode extraction unit, and a transient mapping unit. In the system data stream, the raw heterogeneous data output from the multi-dimensional data acquisition module is first cleaned and aligned by the wavelet denoising unit, then the purified primary feature data is output to the transient mapping unit. The transient mapping unit then calls the flow field spatial mode matrix pre-frozen in the system memory by the feature mode extraction unit, and through online solving of the reduced-order flow field observation model, finally outputs the reconstructed flow field data to the adaptive strategy generation module.

[0079] Furthermore, the wavelet denoising unit undertakes the tasks of spatiotemporal synchronization and filtering of multi-source state data. In the operating environment of the vacuum die-casting machine, the multi-source state data input to this unit includes heterogeneous signals collected by sensing elements from different physical locations and electrical characteristics, such as displacement, pressure, and temperature parameters. Spatiotemporal synchronization aims to accurately align these asynchronous signals from different hardware channels in a unified absolute time coordinate system; filtering aims to extract effective characteristic waveforms that truly represent the physical state of the metal fluid from the original signals subjected to strong electromagnetic and mechanical interference. Given the extreme physical characteristics of the die-casting process—ultra-high speed, high pressure, and high temperature—the filling speed of the molten metal in the mold is extremely high. Any tiny time synchronization error or noise contamination will be significantly amplified during the subsequent three-dimensional flow field reconstruction process of the transient mapping unit, leading to the failure of the digital twin model simulation.

[0080] Preferably, in spatiotemporally synchronized data flow, existing industrial automation modules typically rely on network time protocols to achieve time alignment between nodes. The principle of network time protocols is that a time server within the network periodically broadcasts timestamps, and each slave node receives them, calculates the network latency, and adjusts its local clock accordingly.

[0081] However, the aforementioned network time protocol has limitations in the thin-walled die-casting scenario of new energy vehicle motor housings. The synchronization accuracy of the network time protocol is typically at the millisecond level, and its synchronization error can produce unpredictable jitter due to fluctuations in the industrial Ethernet network load. When the die-casting machine injects molten metal at a high speed of tens of meters per second, a 1ms clock deviation means that the metal fluid front has advanced by several centimeters. The aforementioned spatial misalignment will reduce the spatial accuracy of subsequent flow field reconstruction. Therefore, the wavelet denoising unit of this invention abandons the conventional network time protocol and instead adopts a precision time protocol based on a hardware clock.

[0082] Specifically, the principle of the precision time protocol is to use the hardware-level media access control layer to mark timestamps, thereby eliminating the uncertainties and delays caused by operating system kernel scheduling and network protocol stack processing. In the synchronization logic of the wavelet denoising unit, the system configures the master programmable logic controller as the global time master node and the intelligent sensor nodes distributed in the injection mechanism and mold cavity as slave nodes.

[0083] Before initiating the die-casting cycle, the master node sends a synchronization message with a precise hardware timestamp via the bus. The slave nodes receive this message, record its arrival time, and send a delay request message back to the master node. Through multiple rounds of hardware-level message interaction, the wavelet denoising unit calculates the asymmetric delay of the transmission link and the local clock offset. In specific process parameter applications, this unit strictly controls the time synchronization error of the entire sensor network to within 100ns. This high-precision synchronization ensures that when the leading edge of the molten metal touches a specific cavity node, the pressure jump signal captured by the sensor at that node can achieve a precise physical-temporal correspondence with the displacement signal recorded by the injection encoder.

[0084] After completing spatiotemporal synchronization, the wavelet denoising unit performs filtering on the data stream. Industrial die-casting sites experience complex noise interference, primarily including hydrodynamic noise caused by high-frequency pulsation of the hydraulic pump station, low-frequency mechanical vibration noise generated by the operation of the heavy-duty mold-locking mechanism, and high-frequency electromagnetic interference noise radiated by the high-power servo motor driver on site.

[0085] The built-in filtering algorithms in conventional modules are typically infinite impulse response low-pass filters, moving average filters, or Kalman filters. Moving average and low-pass filters smooth the signal by averaging or cutting off high-frequency components, but while filtering out noise, they also dull the true edges of the physical signal. When high-speed molten metal instantaneously impacts a vacuum pressure probe installed inside the mold, it physically generates a steep step signal. Conventional low-pass filters flatten this step signal, causing a lag in the system's judgment of fluid arrival time. Furthermore, Kalman filtering theoretically assumes that system noise follows a Gaussian white noise distribution, while fluid impact noise in the die-casting process is often typical non-Gaussian impulse noise. This can lead to abnormal updates in the Kalman filter's gain matrix and cause signal distortion.

[0086] To overcome the shortcomings of existing filtering mechanisms, the wavelet denoising unit of this invention employs a denoising logic combining discrete wavelet transform and a soft threshold function. The principle of wavelet transform is to map a one-dimensional time-domain signal onto a two-dimensional time-frequency phase plane by introducing scaling and translation operations. It possesses excellent localization analysis capabilities, providing high time resolution in the high-frequency band and high frequency resolution in the low-frequency band. This multi-resolution analysis characteristic is extremely suitable for processing die-casting injection signals containing abrupt changes.

[0087] Specifically, this unit selects a Dobesi wavelet basis sequence with compact support and high regularity. Since the molten metal impact step signal captured by the die-casting sensor is mathematically highly similar to the low-order waveform of the Dobesi wavelet, this morphological matching can concentrate the energy of the effective signal on a few wavelet coefficients to the maximum extent. The wavelet denoising unit performs multi-scale decomposition on the synchronized original state data sequence, setting the decomposition level to 3 to 5 levels. Too few decomposition levels cannot effectively isolate low-frequency mechanical vibration noise, while too many levels will introduce excessive boundary distortion and consume too much edge controller computing power. After decomposition, the original signal is transformed into low-frequency approximation coefficients reflecting the signal trend and high-frequency detail coefficients containing noise abrupt changes.

[0088] Furthermore, the wavelet denoising unit employs a soft thresholding function to nonlinearly shrink the high-frequency detail coefficients. The core logic of soft thresholding is to retain coefficients with absolute values ​​greater than a set threshold and attenuate their values, while setting coefficients with absolute values ​​less than the set threshold to zero. A representative soft thresholding formula configured internally in this invention is as follows: ,in, This represents the new high-frequency detail coefficients after threshold shrinkage processing. Represents the high-frequency detail coefficients at the original j-th scale and k-th translation position after wavelet decomposition. The sign function is used to extract the positive and negative polarities of the original coefficients. The function for finding the maximum value is used to ensure that the result is non-negative. This is a globally preset threshold calculated based on adaptive signal-noise variance. Regarding the application range of process parameters, the unit performs statistical analysis on the background noise of the sensor under no-load conditions, multiplying the standard deviation of the noise by a specific constant to set... The threshold value is typically set between 80% and 120% of the original noise amplitude. After shrinkage, the unit performs an inverse discrete wavelet transform to reconstruct the filtered first feature data and outputs it to the transient mapping unit. This module architecture filters out high-frequency pulsations and electromagnetic interference while preserving the steep step physical characteristics formed instantaneously by the metal fluid impact sensor, providing a high-quality input source for the transient mapping unit.

[0089] Specifically, the transient mapping unit receives the first feature data and performs mapping calculations on it based on a preset reduced-order flow field observation model to obtain flow field reconstruction data. Flow field reconstruction data refers to a three-dimensional numerical matrix characterizing the velocity vector, absolute pressure scalar, and temperature distribution of the molten metal at millions of spatial nodes within the complex three-dimensional cavity of the die-casting mold. However, when dealing with ultra-thin-walled and deep-cavity structures such as motor housings, completing the numerical simulation of a single filling process requires a significant amount of time. Since the actual physical filling time of a vacuum die-casting machine is only between 10ms and 1000ms, full-order calculation methods suffer from severe lag. Furthermore, while purely data-driven deep learning models have fast inference speeds, they are prone to outputting abnormal results that violate fundamental principles of fluid mechanics due to the lack of strict physical conservation laws.

[0090] Therefore, the feature mode extraction unit and transient mapping unit of this invention work together to adopt a reduced-order flow field observation model based on the intrinsic orthogonal decomposition algorithm combined with the Galerkin projection mechanism. The principle of the intrinsic orthogonal decomposition algorithm is to extract the optimal set of orthogonal basis functions, i.e., the flow field spatial mode matrix, from the high-dimensional flow field snapshot dataset by utilizing the correlation between the system state in time and space.

[0091] By projecting high-dimensional partial differential equations onto these low-dimensional modal bases, nonlinear systems containing millions of variables can be simplified to a system of ordinary differential equations with only dozens of variables. The reduced-order flow field observation model inherits the conservation properties of the physical equations while significantly reducing computational complexity. To adapt the model to the unique physical environment of thin-walled die-casting of motor housings, the characteristic mode extraction unit underwent adaptive improvements, introducing a thermo-fluid-structure interaction viscosity penalty term into the Galerkin projection equations. This penalty term maps the temperature gradient data collected by local thermodynamic state sensors to a nonlinear adjustment factor for local viscosity, enabling the model to accurately capture the characteristics of molten metal flow rate attenuation and solidification front advancement caused by rapid cooling in thin-walled regions.

[0092] During the off-line training phase executed by the feature mode extraction unit, the system uses full-order fluid dynamics software to set a wide range of input boundaries for the target mold to conduct simulation experiments.

[0093] The specific process parameters cover the following ranges: the injection punch speed varies continuously from 0.5 m / s to 8 m / s; the initial pouring temperature of the aluminum alloy melt is set between 630℃ and 710℃; and the initial cavity vacuum degree is distributed between 20 mbar and 200 mbar. A high-dimensional flow field snapshot sample data matrix is ​​constructed by extracting the flow velocity and pressure parameters of each grid node at the discrete time step through extensive simulations.

[0094] Subsequently, the unit performs singular value decomposition, sets the cumulative energy contribution rate threshold to 99.5%, retains only the core feature vectors of the top 10 to 50 orders, constructs the flow field spatial mode matrix, and stores it in the memory of the control system.

[0095] In the online mapping calculation process of die casting production, the transient mapping unit receives first characteristic data containing instantaneous punch displacement, high-frequency pressure, and local mold temperature, and reconstructs it into the physical transient state vector at the current moment. Subsequently, the unit uses the fourth-order Runge-Kutta numerical integration algorithm to solve the low-dimensional ordinary differential equation system with embedded thermo-fluid-structure interaction viscosity penalty terms. Within an extremely short computation cycle of 2 ms, the edge controller quickly calculates the set of scalar coefficients reflecting the dynamic weights of the current flow field, i.e., the reduced-order characteristic coefficients. Finally, the transient mapping unit calls the high-dimensional flow field spatial mode matrix in memory to perform the representative mathematical reconstruction operation of this module. ,in, The output flow field reconstruction data includes three-dimensional velocity and pressure fields. The time-averaged background flow field ground state distribution in the cavity space, representing the extraction of the decoupling, is shown. Represents the total number of spatial modes retained. The dynamic weights represent the i-th order reduced-order characteristic coefficients obtained through online real-time solution. This represents the i-th order flow field spatial mode matrix stored in memory. The transient mapping unit instantly reconstructs the three-dimensional dynamic morphology of the flow field covering the entire mold cavity through simple linear multiply-accumulate projection operations.

[0096] like Figure 3As shown, the adaptive strategy generation module serves as the decision-making hub in the integrated industrial control system. Its main function is to leverage the 3D dynamic flow field data output by the flow field reconstruction and preprocessing module, integrating multi-physics boundary conditions and cross-process manufacturing feedback to achieve forward-looking process simulation and control command generation. In the system's data flow architecture, this module plays a crucial role, aggregating not only the underlying transient physical mapping data but also integrating feedback data from the horizontal information system, and then outputting dynamic control strategies to the collaborative execution and interaction module. Specifically, the internal logic architecture of the adaptive strategy generation module includes a boundary correction unit, an air entrainment prediction unit, and an impedance matching unit. These units sequentially execute the calculation processes of physical boundary calibration, flow field trend extrapolation, and control parameter solving according to a time sequence, jointly constructing a complete closed loop for multi-physics coupled simulation.

[0097] Specifically, the boundary correction unit is used to acquire reference data on mold thermal deformation and downstream process quality feedback data. In the continuous vacuum die-casting production of complex three-dimensional parts, large mold structures inevitably accumulate significant heat after hundreds of filling and spraying cooling cycles. In actual manufacturing environments, the motor housing mold absorbs heat from the high-temperature molten aluminum alloy, resulting in complex nonlinear thermal expansion and thermal stress-induced warping deformation. This geometric deformation causes the mold parting surface gap and the venting groove size of the micro-venting plate to shrink or expand. If the control system ignores the time-varying characteristics of these physical boundaries, the fluid venting resistance calculated based on the initial static dimensions will deviate significantly, causing the venting strategy to fail.

[0098] To eliminate the mismatch error between the static physical model and the hot mold entity, the boundary correction unit first receives the mold temperature gradient and internal stress data collected by the thermal field sensing unit. Then, based on the three-dimensional thermoelastic partial differential equations, this unit performs real-time numerical solutions for the dynamic geometric deformation offset of the mold. The boundary correction unit utilizes the preloaded thermal expansion coefficient tensor and elastic modulus parameters of the mold material, and maps the discrete temperature measurement point data into a continuous global thermal field distribution through a radial basis function interpolation algorithm, thereby calculating the normal displacement vector of each mesh node on the inner wall of the cavity.

[0099] After acquiring geometric deformation data, the boundary correction unit simultaneously receives and parses feedback data from downstream quality inspection processes. Conventional die-casting equipment often exists in an information silo, lacking awareness of internal defects revealed in castings during subsequent machining. The boundary correction unit involved in this invention directly interconnects with a workshop-level industrial Ethernet via the OPC UA unified architecture protocol, actively pulling visual inspection reports of shrinkage cavities on cutting surfaces from downstream CNC machining centers, as well as micro-leakage test logs from helium mass spectrometry gas tightness testing equipment.

[0100] In a preferred embodiment, the boundary correction unit performs an affine transformation of spatial coordinates when processing quality feedback data from downstream processes. Since the coordinate system of the downstream processing equipment is established based on the machine tool spindle or fixture positioning reference, while the coordinate system of the die-casting mold flow field model has the mold center as its origin, this unit uses a preset homogeneous coordinate transformation matrix to accurately reverse-map the three-dimensional spatial coordinates of shrinkage defects or leakage channels present in historical scrap to the corresponding mesh nodes of the die-casting mold fluid model. The mesh nodes weighted with historical defects and the thermally calibrated physical boundary topology are then updated in real-time to the reduced-order flow field observation model as entirely new dynamic constraints. This mechanism enables the control system to achieve self-learning and self-correction capabilities across processes.

[0101] After physical boundary calibration, the gas entrainment prediction unit performs multiphysics coupling simulation. This unit directly receives the current flow field reconstruction data after boundary correction and is responsible for predicting the evolution trend of molten metal flow within a very short time window. Existing closed-loop control technology is essentially a passive response mechanism, meaning the controller only initiates adjustment of the injection speed after receiving an abnormal electrical signal from the local pressure sensor. Within this lag response period, the leading edge of the aluminum alloy melt, advancing at a speed of several meters per second, has usually already converged and encapsulated residual gas. To overcome the inherent technical obstacle of irreversible physical defects in post-compensation, the gas entrainment prediction unit adopts a forward-looking state extrapolation algorithm, whose core operating logic is based on the dynamic evolution equation of the fluid volume function interface.

[0102] Specifically, the fluid volume function method is a computational fluid dynamics Eulerian grid technique used to trace phase interfaces. The gas entrainment prediction unit performs time-integration calculations on the multiphase flow volume profile composed of molten metal and cavity gas in the low-dimensional state space of the reduced-order model. ,in, The value represents the volume fraction of the liquid phase within a computational grid cell. Its value is strictly defined between 0 and 1. 0 indicates that the grid is completely occupied by gas, 1 indicates that the grid is completely filled with liquid metal, and values ​​between 0 and 1 represent the precise location of the gas-liquid interface. Represents variables that evolve over physical time; This represents the reconstructed three-dimensional fluid velocity vector field. By solving this partial differential equation, the gas entrainment prediction unit can perform explicit integration forward along the time axis based on the current velocity distribution at time t, thereby accurately depicting the motion topology of the metallic fluid front.

[0103] When performing extrapolation prediction, the time step and prediction horizon parameters of the air entrainment prediction unit are strictly constrained, with the prediction time horizon set to the range of 10 ms to 20 ms in the future. If the prediction time is less than 10 ms, sufficient physical response time cannot be reserved for the downstream mechanical and hydraulic actuators; if the prediction time exceeds 20 ms, the prediction distortion of the fluid front morphology will increase exponentially due to the cumulative effect of the reduced-order model truncation error and numerical dissipation. Within this look-ahead time window, the system monitors the flow evolution of target exhaust nodes such as the mold exhaust block inlet and the overflow channel end in a high-frequency manner. If the prediction calculation shows that a certain liquid metal jet has prematurely jumped over and blocked a key exhaust port due to the nonlinear acceleration effect caused by the sudden change in wall thickness, and the model shows that there are still gaseous grid groups in the deep cavity region that have not been filled by liquid metal, the air entrainment prediction unit determines that the probability of generating a closed air bag in this area exceeds the safety threshold and immediately triggers the subsequent intervention and adjustment mechanism.

[0104] Upon receiving a high-risk gas entrapment warning, the impedance matching unit intervenes to execute the physical calculation logic for generating a dynamic control strategy. Current vacuum die-casting exhaust control primarily employs a logic switch control mode, where the system sends a fully open or fully closed signal to a two-position, two-way solenoid valve. When the vacuum valve is fully open for evacuation, the negative pressure generated in the exhaust channel may trigger a shock wave phenomenon at the metal molten front, causing aluminum molten metal to splash and block the micro-wave grooves on the exhaust plate. When the exhaust capacity is insufficient, the reverse back pressure generated by the compressed gas inside the cavity increases, forcing a decrease in metal flow velocity and potentially leading to cold shut defects. To overcome the limitations of the aforementioned switch-based control, the impedance matching unit introduces the concept of impedance matching from the acoustic system and the microfluidic network.

[0105] Specifically, the physical meaning of fluid impedance matching refers to dynamically adjusting the throttling cross-sectional area at the end of the exhaust system so that the flow resistance characteristics of the gas flow in the exhaust pipe can be balanced with the dynamic driving torque generated by the molten metal pushing and extruding gas inside the mold cavity. The impedance matching unit, combined with the three-dimensional spatial coordinate characteristics of the downstream easily shrinkable or leaky areas input by the boundary correction unit, calculates the optimal local exhaust rate required for that specific area during the filling process. Subsequently, based on the one-dimensional unsteady flow energy equation and mass conservation equation of compressible gas, this unit establishes a flow resistance network model of the vacuum servo proportional valve.

[0106] Furthermore, the impedance matching unit solves for a continuous-time sequence of servo proportional throttle valve opening changes. This opening sequence refines the original 0 or 100 solenoid valve control state into a continuously adjustable analog control curve with a resolution of 0.1, and the time update frequency of each opening command in this sequence reaches 1 ms. By outputting the above dynamic impedance matching control strategy, this system can ensure that residual gas inside the cavity is extracted by the vacuum pump in a laminar flow state, avoiding splashing and blockage induced by sudden changes in vacuum pressure, while also offsetting the exhaust area attenuation effect caused by mold thermal deformation. This impedance matching calculation integrates multi-dimensional physical field boundary conditions, cross-process quality feedback, and forward-looking flow field topology evolution information, enabling the exhaust system control command to transform from passive following to active guidance adapted to flow field dynamics.

[0107] The collaborative execution and interaction module transforms high-dimensional abstract strategies into underlying physical actions and completes a data loop throughout the entire lifecycle. Specifically, this module translates the high-dimensional dynamic control strategies output by the adaptive strategy generation module into electrical and physical signals recognizable by the underlying servo drives and actuators, while simultaneously aggregating all characteristic parameters of the production process to the host computer system. Internally, the collaborative execution and interaction module includes an instruction compilation unit and an energy efficiency coordination unit. In the system's data and control flow, the instruction compilation unit receives dynamic impedance matching control strategies to drive external mechanical and hydraulic mechanisms, while the energy efficiency coordination unit monitors power consumption during execution. After execution, the module extracts the actual response curve of the underlying hardware, aligns and packages it with the initial control commands, and finally transmits it back to the factory manufacturing execution system via a standard industrial network protocol, thus establishing a complete link from virtual simulation to physical manufacturing and data archiving.

[0108] Furthermore, the instruction compilation unit parses the dynamic impedance matching control strategy into low-level driving instructions. In the die-casting process of complex three-dimensional parts, the control of the exhaust system affects the displacement efficiency of the gas-liquid two-phase flow inside the cavity. Traditional discrete switching control has significant technical shortcomings when dealing with products such as new energy vehicle motor housings, which have extremely high requirements for exhaust flow. When the solenoid valve is fully opened, the sudden change in the cross-sectional area of ​​the flow channel can cause cavitation and fluid shock waves, resulting in the extracted gas carrying metal droplets splashing, which can easily physically block the micron-level exhaust channels. At the same time, the mechanical spring reset and coil self-inductance physical characteristics of conventional solenoid valves result in an opening and closing response delay on the order of tens of milliseconds, which cannot meet the millisecond-level dynamic impedance matching requirements of the reduced-order flow field observation model.

[0109] Preferably, to overcome the limitations of hysteresis and nonlinear abrupt changes in the execution of underlying hardware, the instruction compilation unit abandons the traditional discrete voltage-driven mode and instead adopts a high-frequency pulse width modulation signal-driven architecture. This unit receives the continuous-time domain valve opening sequence with a resolution of 0.1 from the adaptive strategy generation module and maps it into a high-frequency pulse width modulation signal with a frequency range of 5 kHz to 10 kHz.

[0110] By adjusting the duty cycle parameter within a single pulse period, this high-frequency signal directly drives a voice coil motor or a high-speed servo proportional throttle valve located at the end of the vacuum exhaust block. The voice coil motor operates based on the Ampere force principle, eliminating the complex armature and return spring mechanical transmission structure of traditional solenoid valves. Its mover is extremely lightweight, reducing the system's mechanical displacement tracking response time to within 2 ms. Transforming the software-level continuous impedance strategy into a hardware-level high-frequency pulse adjustment mapping step allows the vacuum exhaust valve's opening area to smoothly and linearly expand and contract with the transient pressure changes generated by the advancing molten metal front, ensuring that the mixed gas within the cavity is always extracted by the vacuum system in a shock-free laminar flow state.

[0111] When performing coordinated control of evacuation and injection, the instruction compilation unit not only controls the vacuum valve independently but also simultaneously generates torque feedforward control signals for the main injection servo motor. Feedback control is essentially error-driven; that is, the system only increases the output current to enhance the driving torque when the actual speed of the punch falls below the set target due to resistance from molten metal filling. In the high-speed filling stage of ultra-thin-wall die casting, hysteresis compensation relying on accumulated errors often leads to periodic oscillations in the molten metal front velocity. To achieve precise temporal and spatial coordination between evacuation and injection, the instruction compilation unit employs a control topology that combines feedforward and feedback.

[0112] Based on the predicted fluid resistance trend from the reconstructed flow field data, the dynamic driving torque required for the injection system to overcome inertia and viscous friction is calculated in advance. The mathematical formula for torque feedforward is: In this feedforward control formula, The target feedforward torque command that the instruction compilation unit outputs to the servo driver at time t; The equivalent moment of inertia parameter representing the transmission chain of the injection mechanism referred to the motor shaft end; This represents the optimal reference angular velocity trajectory of the punch derived from the flow field reconstruction model; The equivalent viscous friction coefficient representing the moving parts of the system; This represents the dynamic reaction load torque of the fluid inside the mold cavity, predicted by a reduced-order flow field observation model, evolving over time. Through this feedforward channel, the system directly converts the predicted physical load changes into a forward drive current gain, enabling the injection punch to be completely immune to velocity disturbances caused by sudden changes in fluid resistance. Fine-tuning of the punch acceleration and the high-frequency extension and retraction of the vacuum proportional valve are interlocked at the microsecond level on the time axis, constructing a highly coordinated electromechanical-hydraulic integrated execution network.

[0113] The energy efficiency coordination unit simultaneously optimizes the power flow during the non-peak phase of the die-casting cycle. Within a single die-casting cycle, apart from the injection and pressurization phases that require instantaneous peak power for tens of milliseconds, the auxiliary action phases such as mold closing, spraying, cooling, and mold opening and part removal, which last for tens of seconds, have lower requirements for hydraulic system flow and pressure. The energy efficiency coordination unit analyzes the global process status timestamps output by the adaptive strategy generation module and dynamically issues frequency conversion pressure reduction commands to the main drive servo pump station during the non-peak process phase, smoothly reducing the standby holding pressure from the conventional 140 bar to the maintenance range of 20 bar to 30 bar.

[0114] After completing the underlying physical drive, this module executes and sends the multi-dimensional execution results back to the factory network. The existing result-oriented data processing method cuts off the clues for quality traceability. When defects such as internal porosity are found in batches in downstream processes, process engineers cannot reconstruct the molding boundary conditions at the moment the defects occurred.

[0115] After the injection cycle ends, the system immediately extracts the high-frequency physical parameters actually collected during this cycle, including the actual displacement response curve of the servo proportional vacuum valve, the dynamic pressure build-up trajectory of the injection cylinder, the evolution sequence of the local temperature gradient of the mold, and the pressure history of key nodes characterizing the air bag distribution in the flow field reconstruction data. The system performs strict time-series alignment operations on the above multidimensional heterogeneous data to ensure that the microscopic features at the same time section are accurately correlated on multiple physical curves.

[0116] This invention employs a unified architecture information model protocol based on Ethernet or a manufacturing message specification protocol based on object-oriented modeling. The collaborative execution and interaction module has a built-in information model mapping gateway that encapsulates the aligned multidimensional physical curves and discrete process parameters into standardized object-oriented data packets with hierarchical attributes, timestamp identifiers, and unit metadata.

[0117] Once the factory's manufacturing execution system receives the standardized data packet, the platform uses the industrial internet to strongly bind this set of complete microscopic process data with the unique traceability code assigned to the physical casting on-site via a laser marking machine or RFID component.

[0118] Example 2:

[0119] This embodiment focuses on illustrating how the collaborative execution and interaction module can be deeply integrated with the factory-level manufacturing execution system, energy management system, and downstream testing equipment;

[0120] To address the shortcomings of existing die-casting equipment in detecting upstream material variations, this invention proposes an adaptive pre-loading of process parameters based on incoming material characteristics. After acquiring upstream batch data from the factory manufacturing execution system, the multi-dimensional data acquisition module analyzes the upstream batch data and extracts the alloy composition ratio of the current aluminum melt, such as the specific mass percentages of silicon, copper, and magnesium. Subsequently, the system calls a preset composition-viscosity mapping relationship library. This library, built based on an improved Carlo-Yasuda rheological model, can calculate the dynamic viscosity coefficient and yield stress of the current aluminum melt based on the alloy composition and the current furnace temperature. The system uses the calculated dynamic viscosity coefficient as the initial physical parameter and automatically overwrites it into the configuration file of the reduced-order flow field observation model. This feedforward parameter correction mechanism allows the control system to perform model calibration for the specific rheological properties of the current batch of material before the die-casting cycle begins, fundamentally eliminating flow field prediction deviations caused by material batch fluctuations.

[0121] Furthermore, addressing the disconnect between die casting and subsequent processing and inspection in existing technologies, a closed-loop quality compensation mechanism based on feedback from downstream processes is constructed. During multi-physics coupled deduction, the adaptive strategy generation module monitors data topics published by downstream CNC machining centers and airtightness testing equipment in real time via industrial Ethernet. When downstream process quality feedback data is parsed, the actual dimensional deviation coordinates and leakage point coordinates are extracted. For example, if there is a micron-level insufficient machining allowance at a certain coordinate on the downstream feedback motor housing stator mounting surface, the system will combine mold thermal deformation reference data to reverse-determine that the area has an abnormal shrinkage rate due to local overheating during die casting. In this case, the system uses the actual dimensional deviation coordinates and leakage point coordinates as penalty terms, introducing them into the evolution equation of the flow field reconstruction data.

[0122] Specifically, by adjusting the convective heat transfer coefficient or local flow resistance coefficient of the corresponding grid nodes in the region, the propulsion velocity field of the fluid front is corrected; in the next die-casting cycle, the system will automatically open the local cooling water valve in the region in advance or fine-tune the injection pressurization curve based on the corrected flow field prediction results; the cross-process data closed loop enables the die-casting machine to have the ability to learn and evolve on its own, realizing true full-process quality control.

[0123] Preferably, this embodiment also involves collaborative optimization with the factory's energy management system. Under the dual-carbon background, frequent start-stop and long-term high-load operation of vacuum pump units will lead to huge energy consumption. The energy efficiency collaboration unit in the collaborative execution and interaction module subscribes to the real-time power limit data and time-of-use electricity price data of the energy management system through the OPC UA protocol. In the process of generating dynamic impedance matching control strategy, the energy efficiency collaboration unit introduces a multi-objective optimization algorithm to construct a comprehensive fitness function that includes casting quality penalty term and energy consumption cost term. When the factory is in the peak electricity consumption period or the total power is close to the limit, the system automatically relaxes the vacuum degree constraint conditions of non-critical areas on the premise that the flow field morphology of the key stress area and sealing area of ​​the motor housing does not deteriorate. Through peak-shaving and valley-filling start-stop scheduling, such as using pulsed gas extraction instead of continuous gas extraction in the non-critical exhaust stage, the instantaneous peak power of the system is effectively reduced. The collaborative mechanism breaks the limitation of traditional control systems that only pursue a single quality target and realizes the deep integration of manufacturing process control and enterprise resource management.

[0124] Based on the preferred embodiments of the present invention described above, those skilled in the art can make various changes and modifications without departing from the inventive concept. The technical scope of this invention is not limited to the contents of the specification, but must be determined according to the scope of the claims.

Claims

1. An integrated industrial control system for a vacuum die-casting machine, characterized in that, include: The multi-dimensional data acquisition module is used to acquire multi-source state data during the die-casting process and upstream batch data from the factory's manufacturing execution system. The flow field reconstruction and preprocessing module is connected to the multi-dimensional data acquisition module. It is used to perform spatiotemporal synchronization and filtering on the multi-source state data to obtain first feature data, and to perform mapping calculation on the first feature data based on the preset reduced-order flow field observation model to obtain flow field reconstruction data characterizing the dynamics of the three-dimensional fluid front inside the cavity. An adaptive strategy generation module, connected to the flow field reconstruction and preprocessing module, is used to acquire mold thermal deformation reference data and downstream process quality feedback data, and determine the dynamic impedance matching control strategy of the target exhaust node through multi-physics coupling deduction based on the flow field reconstruction data, the mold thermal deformation reference data, and the downstream process quality feedback data. The collaborative execution and interaction module, connected to the adaptive strategy generation module, is used to generate underlying drive instructions based on the dynamic impedance matching control strategy, send the underlying drive instructions to the external die-casting actuator, and encapsulate the execution result into a standard protocol data packet and send it back to the factory manufacturing execution system.

2. The integrated industrial control system for a vacuum die-casting machine according to claim 1, characterized in that: The flow field reconstruction and preprocessing module includes a wavelet denoising unit, a feature mode extraction unit, and a transient mapping unit; The wavelet noise reduction unit uses a precise time protocol based on a hardware clock to perform spatiotemporal synchronization of the multi-source state data; and uses a combination of discrete wavelet transform and soft threshold function to filter out high-frequency impulse noise and output the first feature data. The feature mode extraction unit is used to extract the flow field spatial mode matrix offline; The transient mapping unit constructs the first feature data into a transient state vector; by solving a set of low-dimensional ordinary differential equations with embedded thermo-fluid-structure interaction viscosity penalty terms, it obtains reduced-order characteristic coefficients; and then linearly combines and projects the reduced-order characteristic coefficients with the flow field spatial mode matrix to reconstruct the flow field reconstruction data.

3. The integrated industrial control system for a vacuum die-casting machine according to claim 1, characterized in that: The adaptive strategy generation module includes a boundary correction unit, an air-entrainment prediction unit, and an impedance matching unit. The boundary correction unit solves for the dynamic geometric deformation offset of the mold based on the three-dimensional thermoelastic partial differential equations and the mold thermal deformation reference data, so as to update the physical boundary of the reduced-order flow field observation model in real time. The air-entrainment prediction unit predicts the motion topology of the molten metal front within a preset look-ahead time window based on the dynamic evolution equation of the fluid volume function interface, and determines the probability of the target exhaust node generating a closed air bag. When the probability exceeds the safety threshold, the impedance matching unit solves the continuous opening change sequence according to the acoustic impedance matching principle. This sequence balances the flow resistance characteristics of the gas flow in the exhaust pipe with the dynamic driving torque generated by the molten metal propelling and compressing the gas. This sequence is then used as the dynamic impedance matching control strategy.

4. The integrated industrial control system for a vacuum die-casting machine according to claim 1, characterized in that: The multidimensional data acquisition module includes a high-frequency sensing unit, a thermal field sensing unit, and an industrial interconnection interface unit. The industrial interconnection interface unit extracts the spectral analysis parameters of the current aluminum alloy material by parsing the received data packets, and uses them as the upstream batch data; The system is also equipped with a preset component-viscosity mapping relationship library; The multidimensional data acquisition module calculates the dynamic viscosity coefficient of the current aluminum liquid based on the spectral analysis parameters and the composition-viscosity mapping relationship library; and uses the dynamic viscosity coefficient as the initial physical parameter to overwrite the configuration file of the reduced-order flow field observation model, thereby realizing adaptive preloading of process parameters based on the characteristics of incoming materials.

5. The integrated industrial control system for a vacuum die-casting machine according to claim 1, characterized in that: The adaptive strategy generation module uses a unified architecture protocol to pull the visual inspection report of the shrinkage hole on the cutting surface of the downstream computer numerical control machining center and the micro-leakage test log of the airtightness testing equipment as the quality feedback data of the downstream process. The adaptive strategy generation module uses a preset homogeneous coordinate transformation matrix to reverse-map the three-dimensional spatial coordinates of historical defects presented in the downstream process quality feedback data to the corresponding grid nodes of the reduced-order flow field observation model; and corrects the propulsion velocity field of the fluid front by adjusting the convective heat transfer coefficient or local flow resistance coefficient of the corresponding grid nodes.

6. The integrated industrial control system for a vacuum die-casting machine according to claim 1, characterized in that: The collaborative execution and interaction module includes an instruction compilation unit and an energy efficiency coordination unit; The instruction compilation unit converts the dynamic impedance matching control strategy into a high-frequency pulse width modulation signal to drive the actuator at the end of the vacuum exhaust block; and based on the fluid resistance change trend predicted by the flow field reconstruction data, it calculates the dynamic driving torque required for the injection system to overcome inertia and viscous friction, and generates a torque feedforward control signal for the main injection servo motor. The energy efficiency coordination unit subscribes to the real-time power limit data of the energy management system and dynamically issues frequency conversion step-down commands during the non-peak stage of the die-casting cycle to optimize the power flow.

7. An integrated industrial control method for a vacuum die-casting machine, comprising an integrated industrial control system for a vacuum die-casting machine according to any one of claims 1-6, characterized in that, include: Acquire multi-source state data during the die-casting process and upstream batch data from the factory's manufacturing execution system; The multi-source state data is subjected to spatiotemporal synchronization and filtering to obtain the first feature data; The first feature data is mapped and calculated based on a preset reduced-order flow field observation model to obtain flow field reconstruction data that characterizes the dynamics of the three-dimensional fluid front inside the cavity. Obtain reference data on mold thermal deformation and quality feedback data from downstream processes; Based on the flow field reconstruction data, the mold thermal deformation reference data, and the downstream process quality feedback data, the dynamic impedance matching control strategy of the target exhaust node is determined through multi-physics coupling deduction. The underlying drive instructions are generated according to the dynamic impedance matching control strategy and sent to the external die-casting actuator; at the same time, the execution result is encapsulated into a standard protocol data packet and sent back to the factory manufacturing execution system.

8. The integrated industrial control method for a vacuum die-casting machine according to claim 7, characterized in that: The process of performing spatiotemporal synchronization and filtering on the multi-source state data, and performing mapping calculations based on a reduced-order flow field observation model, includes: Nanosecond-level spatiotemporal alignment of the multi-source state data is performed using a precision time protocol based on a hardware clock. The aligned data is decomposed into multiple scales using the Dobessi wavelet basis; and the high-frequency detail coefficients are shrunk using a soft thresholding function before the first feature data is reconstructed. The first feature data is constructed into a transient state vector and input into the reduced-order flow field observation model constructed by the intrinsic orthogonal decomposition algorithm and the Galerkin projection mechanism; By solving a set of low-dimensional ordinary differential equations with embedded viscosity penalty terms for thermal-fluid-structure interaction, reduced-order characteristic coefficients are obtained; then, the reduced-order characteristic coefficients are linearly combined and projected with the pre-extracted flow field spatial mode matrix to reconstruct the flow field reconstruction data.

9. The integrated industrial control method for a vacuum die-casting machine according to claim 7, characterized in that: The dynamic impedance matching control strategy is determined through multi-physics coupling deduction based on flow field reconstruction data, mold thermal deformation reference data, and downstream process quality feedback data, including: Based on the three-dimensional thermoelastic partial differential equations and the mold thermal deformation reference data, the dynamic geometric deformation offset of the mold is calculated to update the physical boundary. By utilizing the dynamic evolution equation of the fluid volume function interface, time integration is performed on the reconstructed flow field data within a preset look-ahead time window to predict the motion topology of the molten metal front. When the risk of closed gas bag formation is predicted at the target exhaust node, a flow resistance network model is established based on the one-dimensional unsteady flow energy equation and mass conservation equation of compressible gas. Solving for the objective function generates a sequence of servo proportional throttle valve opening changes with continuous time domain resolution, which serves as the dynamic impedance matching control strategy. The objective function is used to balance the flow resistance characteristics of the gas flow in the exhaust pipe with the dynamic driving torque generated by the molten metal propelling and compressing the gas.

10. An integrated industrial control method for a vacuum die-casting machine according to claim 7, characterized in that: After acquiring the upstream batch data, the process also includes: extracting the spectral analysis parameters of the current aluminum alloy material from the upstream batch data; calculating the dynamic viscosity coefficient of the current molten aluminum by combining it with a preset composition-viscosity mapping relationship library; and overwriting the dynamic viscosity coefficient as an initial physical parameter into the reduced-order flow field observation model. In the process of determining the dynamic impedance matching control strategy through multiphysics coupling deduction, the following steps are also included: analyzing the three-dimensional spatial coordinates of historical defects in the downstream process quality feedback data; mapping the three-dimensional spatial coordinates of historical defects to the corresponding grid nodes of the reduced-order flow field observation model through affine transformation of spatial coordinates; and adjusting the convective heat transfer coefficient or local flow resistance coefficient of the corresponding grid nodes to correct the propulsion velocity field of the fluid front.

Images