Titanium alloy metal forging process parameter regulation method based on digital twinning

Abstract

The present application relates to the field of titanium alloy forging regulation technology, in particular to a titanium alloy metal forging process parameter regulation method based on digital twinning, comprising: deploying isothermal forging presses, multi-section intelligent temperature control systems, vacuum or protective atmosphere environments and multi-source online monitoring sensor networks in the physical space to collect multi-source state data during the forging process. A virtual space constructs a digital twin of a constitutive model library based on Gleeble thermal simulation experiments, a finite element real-time simulation engine and process decision algorithm, realizes real-time data transmission between the virtual and real spaces through a data interaction link, drives the simulation to generate a virtual mirror, compares key field variables with the threshold values of the constitutive model library in real time to trigger parameter optimization, and returns the optimized parameters to realize the coordinated closed-loop adjustment of the press speed, temperature control power and atmosphere control logic. The present method can accurately characterize the damage behavior of titanium alloy materials and maintain the dynamic stability of the forging process state.

Description

Technical Field

[0001] This invention relates to the field of titanium alloy forging control technology, and in particular to a method for controlling process parameters in titanium alloy metal forging based on digital twins. Background Technology

[0002] Existing isothermal forging processes for titanium alloys mostly employ an offline, pre-set process parameter implementation mode, using conventional zoned temperature control equipment and isothermal forging presses to complete the forging operation. Basic data such as temperature and displacement during the forging process are collected by scattered sensors. Simulation analysis primarily utilizes offline finite element methods, relying on constitutive models that are mostly general-purpose metal material models, without incorporating Gleeble thermal simulation experimental data to construct a dedicated model. Parameter control during the forging process often employs single-parameter open-loop adjustment methods, with the vacuum or protective atmosphere environment, temperature control system, and forging press operation control operating independently, lacking a unified control and coordination mechanism.

[0003] General-purpose material constitutive models cannot accurately characterize the material damage behavior of titanium alloys under the coupled effects of different temperatures and strain rates. Offline simulations cannot synchronously map the real-time state of physical forging, and parameter optimization operations cannot be triggered in a timely manner when key field variables deviate from the safe range. The mode of independent control of single parameters cannot adapt to the coupled influence of multiple parameters in the forging process. The control actions of different execution terminals are disconnected from each other, which can easily lead to instability in the process state of forging.

[0004] It is necessary to rely on real-time simulation to form a virtual image of the forging process, establish a constitutive model library that fits the damage characteristics of titanium alloy materials, and complete the real-time comparison of field variables to trigger parameter optimization. At the same time, it is necessary to realize multi-parameter coordinated closed-loop regulation of press running speed, temperature control system output power, and atmosphere control logic. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and propose a method for controlling process parameters in titanium alloy metal forging based on digital twins.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a method for controlling process parameters in titanium alloy metal forging based on digital twins, comprising: Deploy isothermal forging presses, multi-segment intelligent temperature control systems, fully enclosed vacuum or protective atmosphere environments, and online monitoring sensor networks consisting of high-temperature industrial cameras, laser displacement sensors, and pressure sensors in physical space to collect multi-source state data during the forging process. A digital twin containing a constitutive model library, a finite element real-time simulation engine, and a process decision algorithm is constructed in virtual space. The constitutive model library is based on Gleeble thermal simulation experimental data and is used to characterize the material damage behavior of titanium alloys at different temperatures and strain rates. The multi-source state data collected in the physical space is transmitted to the virtual space in real time through the data interaction link, driving the finite element real-time simulation engine to perform transient field calculations on the current forging process and generate a virtual image. The key field variables in the virtual image are compared with the constitutive model library in real time. When the key field variables are detected to deviate from the safety threshold defined by the constitutive model library, the process decision algorithm is triggered to optimize the parameters. The process parameters optimized by the process decision algorithm are transmitted back to the physical space through a data interaction link to perform closed-loop regulation of the operating speed of the isothermal forging press, the output power of the multi-segment intelligent temperature control system, and the control logic of the fully enclosed vacuum or protective atmosphere environment.

[0007] As a further aspect of the present invention, the deployment of an isothermal forging press, a multi-segment intelligent temperature control system, a fully enclosed vacuum or protective atmosphere environment, and an online monitoring sensor network composed of a high-temperature industrial camera, a laser displacement sensor, and a pressure sensor in physical space to collect multi-source state data during the forging process includes: The real-time running speed and position of the slider, as well as the real-time load data on the titanium alloy billet, are obtained through the servo control system built into the isothermal forging press. The surface temperature and temperature rise rate data of each key section of the mold are collected by using the independent thermocouple feedback of each zone of the multi-segment intelligent temperature control system. The pressure or vacuum level data of the gas inside the furnace is collected by a pressure gauge in the fully enclosed vacuum or protective atmosphere environment, and the flow rate data of the protective gas is recorded simultaneously. The high-temperature industrial camera is used to acquire images of the outline of the titanium alloy billet in the mold cavity, and to extract the shape changes and flash bridge height data of the billet during the filling process. The laser displacement sensor is used to scan the parting surface of the mold and collect data on the elastic opening of the mold during the forging process. All collected load data, temperature data, gas data, shape change data, and elastic opening data are timestamped and normalized, and then packaged to generate the multi-source state data.

[0008] As a further aspect of the present invention, the construction of a digital twin in virtual space, comprising a constitutive model library, a finite element real-time simulation engine, and a process decision algorithm, includes: Import microstructure images of titanium alloys under different heat treatment states and corresponding mechanical property test results, and construct the constitutive model library using a neural network fitting algorithm. The constitutive model library includes a flow stress prediction sub-model, a damage accumulation criterion sub-model, and a dynamic recrystallization kinetics sub-model. Based on the geometric dimensions of physical space, a three-dimensional geometric model including titanium alloy billet, mold and press is established, and the three-dimensional geometric model is imported into the finite element real-time simulation engine to divide into fine unstructured meshes; In the finite element real-time simulation engine, material property mapping rules are set, the flow stress prediction sub-model in the constitutive model library is embedded into the material card of the mesh element, and a thermo-mechanical coupling solver is configured. In the process decision algorithm, an optimization objective vector is set, which includes evaluation indicators in three dimensions: minimization of maximum forming load, uniformity of grain size, and constraint of mold elastic deformation. The finite element real-time simulation engine is interface-bound with the process decision algorithm, so that the simulation calculation results are directly used as the fitness function value input of the process decision algorithm.

[0009] As a further aspect of the present invention, the step of transmitting multi-source state data collected in the physical space to the virtual space in real time via a data interaction link, driving the finite element real-time simulation engine to perform transient field calculations on the current forging process, and generating a virtual image, includes: The virtual image includes the internal stress field, strain field, temperature field, and grain evolution state of the forging; The current forging stroke position is parsed from the multi-source state data, and the forging stroke position is used as the starting point of the time step of the finite element real-time simulation engine. The billet contour data acquired by the high-temperature industrial camera is compared with the geometric shape of the simulation model. The deviation value between the billet contour data and the simulation model is calculated. The deviation value is used to correct the grid node coordinates of the billet in the simulation model and eliminate simulation drift error. The mold elastic opening data collected by the laser displacement sensor is used as the input boundary condition and applied to the contact surface of the mold in the simulation model to simulate the real mold constraint state. The load data collected by the pressure sensor is verified in real time with the contact force calculated by the simulation. If the difference between the load data collected by the pressure sensor and the contact force calculated by the simulation exceeds the tolerance, the friction coefficient in the simulation model is dynamically adjusted. The explicit integration calculation of the finite element real-time simulation engine is started, and a virtual image of the internal stress field, strain field, temperature field and grain evolution state of the forging at the current moment is output.

[0010] As a further aspect of the present invention, the key field variables in the virtual image are compared with the constitutive model library in real time. When the key field variables are detected to deviate from the safety threshold defined in the constitutive model library, the process decision algorithm is triggered to perform parameter optimization, including: The equivalent strain rate and temperature value of the core of the forging are extracted from the virtual image and input into the damage accumulation criterion sub-model in the constitutive model library to calculate the current damage variable value. Temperature gradient data of the forging surface is extracted from the virtual image, its absolute value of thermal gradient is calculated, and it is compared with the dynamic recrystallization critical value in the constitutive model library; When the value of the damage variable is greater than the damage tolerance set by the constitutive model library, or when the absolute value of the thermal gradient exceeds the dynamic recrystallization critical value, it is determined that there is a risk of process instability. In response to the determination of the risk of process instability, the process decision algorithm is activated, and the current combination of process parameters and the corresponding field variable data are used as the initial population and input into the process decision algorithm. The process decision algorithm traverses new combinations of process parameters in the solution space and calls the finite element real-time simulation engine to quickly evaluate each new set of parameters until a parameter solution set that makes the key field variables return to within the safety threshold is found.

[0011] As a further aspect of the present invention, the process decision algorithm traverses new combinations of process parameters in the solution space and calls the finite element real-time simulation engine to quickly evaluate each new set of parameters until a parameter solution set that causes the key field variables to regress to within a safe threshold is found, including: The decision variables of the process decision algorithm are set, including the downward speed of the isothermal forging press, the preheating holding time in the multi-segment intelligent temperature control system, and the holding pressure during the deformation process; A set of constraints is constructed within the process decision algorithm. The set of constraints includes the ultimate vacuum degree of the fully enclosed vacuum or protective atmosphere environment, the allowable yield strength of the mold material, and the phase transformation temperature range of the titanium alloy billet. The decision variables are iteratively evolved using a non-dominated sorting genetic algorithm. In each generation of evolution, dominant individuals on the Pareto front are selected as candidate process parameters. For each of the candidate process parameters, the finite element real-time simulation engine is invoked to perform a simplified and rapid simulation, calculating only the interaction results of the stress field and temperature field. Based on the results of the simplified simulation, the fitness ranking of individuals in the process decision algorithm is updated, and the final parameter solution set is output after reaching the maximum number of iterations or convergence.

[0012] As a further aspect of the present invention, the process parameters optimized by the process decision algorithm are transmitted back to the physical space via a data interaction link to perform closed-loop regulation of the operating speed of the isothermal forging press, the output power of the multi-segment intelligent temperature control system, and the control logic of the fully enclosed vacuum or protective atmosphere environment, including: The parameter solution set output by the process decision algorithm is analyzed to extract the slider descent speed command and acceleration curve for the next forging stage; The slider downward speed command is sent to the PLC controller of the isothermal forging press to replace the original speed curve and realize dynamic intervention of the loading rate. Extract the mold preheating temperature setpoint from the parameter solution set, calculate its deviation from the current measured temperature, and adjust the power output of each heating unit in the multi-segment intelligent temperature control system through a PID control algorithm; Based on the requirements for forging atmosphere stability in the parameter solution set, adjust the pumping speed of the vacuum pump or the opening of the inlet valve in the fully enclosed vacuum or protective atmosphere environment to maintain a constant pressure inside the furnace. After the adjustment command is issued, the feedback data from the physical space is continuously monitored to confirm that the process parameters have been received and executed by the physical equipment.

[0013] As a further aspect of the present invention, the downward speed command of the slider is sent to the PLC controller of the isothermal forging press to replace the original speed curve, including segmentation processing logic for the speed curve: The slider downward speed command is decomposed into several consecutive stages, each stage corresponding to a specific interval in the forging stroke; In the early stages of forging, the downward speed of the slider is reduced based on the optimization results in order to reduce the peak impact load of the pressure sensor in the multi-source state data. During the filling stage, the downward speed of the slider is dynamically adjusted according to the flow state of the blank in the mold cavity to ensure that the flash height monitored by the high-temperature industrial camera does not exceed the design threshold. During the final forging stage, the downward speed of the slider is locked, and the mold temperature is increased in conjunction with the multi-segment intelligent temperature control system to promote the recrystallization process of the grains through heat conduction. During the pressure holding stage, the slider speed is set to zero, and the clamping force of the isothermal forging press is maintained until the finite element real-time simulation engine determines that the internal stress field tends to stabilize.

[0014] As a further aspect of the present invention, an online self-updating step for the constitutive model library is also included: After each batch of forging tasks is completed, the multi-source status data and the corresponding virtual image data accumulated during the current batch of forging are retrieved. Extract the actual formed key dimensions of the forging from the multi-source state data, extract the simulation-predicted key dimensions of the forging from the virtual mirror data, and calculate the dimensional deviation between the actual formed key dimensions of the forging and the simulation-predicted key dimensions of the forging. When the dimensional deviation exceeds the allowable range in a statistical sense, the average strain rate, peak temperature and deformation amount in the current batch forging process are extracted. The average strain rate, peak temperature, and deformation are used as inputs, and the actual microstructure inspection results are used as labels to incrementally train the flow stress prediction sub-model and dynamic recrystallization kinetics sub-model in the constitutive model library. The updated trained model parameters are written into the constitutive model library to replace the original outdated parameters, thereby enabling the digital twin model to evolve on its own.

[0015] As a further aspect of the present invention, the billet contour data acquired by the high-temperature industrial camera is compared with the geometric shape of the simulation model to calculate the deviation value between the billet contour data and the simulation model. The deviation value is then used to correct the grid node coordinates of the billet in the simulation model, eliminating simulation drift errors. This includes: During a preset time interval in the forging process, a real-time image of the billet inside the mold cavity is captured by the high-temperature industrial camera, and the real-time image is binarized and edge detected to extract the real-time outer contour pixel coordinates of the billet. The real-time outer contour pixel coordinates and the outer surface of the three-dimensional solid model of the titanium alloy billet in the finite element real-time simulation engine are projected onto the camera imaging plane, and the projection result is converted into the contour pixel coordinates of the simulation model. Calculate the contour matching error between the real-time outer contour pixel coordinates and the simulation model contour pixel coordinates in the image plane. The contour matching error is the deviation value between the blank contour data and the simulation model. Determine whether the contour matching error exceeds a preset pixel threshold. If it does, start the grid node coordinate correction process. In the mesh node coordinate correction process, the contour matching error is converted into a position offset vector in three-dimensional space through inverse projection transformation; The position offset vector is decomposed into components along the normal and tangential directions of the outer surface of the billet. Based on a preset weighting factor, the position offset vector is applied to the mesh nodes corresponding to the three-dimensional solid model of the titanium alloy billet in the simulation model, and the coordinates of the mesh nodes are updated to eliminate simulation drift error.

[0016] Compared with the prior art, the advantages and positive effects of the present invention are as follows: Based on Gleeble thermal simulation experimental data, a constitutive model library characterizing the damage behavior of titanium alloys at different temperatures and strain rates was established. Key field variables in the virtual image were compared in real-time with safety thresholds defined in this constitutive model library, triggering a process decision algorithm to perform parameter optimization. The construction logic of the constitutive model aligns with the material damage evolution characteristics of titanium alloy forging. The comparison benchmark for field variables closely matches the material's own mechanical response characteristics. The triggering judgment for parameter optimization is synchronously correlated with the real-time material state during the forging process. The accuracy of the judgment results matches the actual critical damage state of the material. The initiation logic of parameter optimization no longer relies on the fuzzy judgment criteria of general models, and the comparison process of field variables forms a continuous correspondence with the transient changes in the forging process.

[0017] The optimized process parameters, determined by the process decision algorithm, are transmitted back to the physical space via a data interaction link. This enables closed-loop regulation of the isothermal forging press's operating speed, the output power of the multi-segment intelligent temperature control system, and the control logic of the fully enclosed vacuum or protective atmosphere environment. The regulation actions of these three types of process parameters form a synchronous and coordinated execution relationship. The operating parameter adjustments of different physical execution terminals are mutually adapted, and the process execution status in the physical space and the optimized output results in the virtual space form a continuous dynamic correspondence. The regulatory effect covers the core execution units of the forging process, and the scope of parameter adjustment is no longer limited to a single execution device. The control logic adjustments of each execution terminal form a unified adaptation system, and the parameter interaction between the virtual and physical spaces forms a continuous closed-loop response path. Attached Figure Description

[0018] Figure 1 This is a flowchart of the process parameter control method for titanium alloy metal forging based on digital twins as described in this invention; Figure 2 A flowchart for building a digital twin in virtual space; Figure 3 Diagram for assessing the instability risk of TC4 titanium alloy forging process; Figure 4 Temperature response curve of titanium alloy forging die; Figure 5 This is a comparison chart of prediction errors before and after the constitutive model library update. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0020] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0021] See Figure 1 This method establishes a real-time interactive and closed-loop control architecture between physical and virtual spaces. In the physical space, an isothermal forging press, a multi-segment intelligent temperature control system, a fully enclosed vacuum or protective atmosphere environment, and an online monitoring sensor network consisting of high-temperature industrial cameras, laser displacement sensors, and pressure sensors are deployed to collect multi-source state data during the forging process. In the virtual space, a digital twin containing a constitutive model library, a finite element real-time simulation engine, and a process decision algorithm is constructed. The constitutive model library, based on Gleeble thermal simulation experimental data, characterizes the material damage behavior of titanium alloys at different temperatures and strain rates. Multi-source state data collected in the physical space is transmitted to the virtual space in real time via a data interaction link, driving the finite element real-time simulation engine to perform transient field calculations on the current forging process, generating a virtual image containing information such as stress, strain, and temperature fields. Key field variables in the virtual image are compared with those in the constitutive model library in real time. When a key field variable deviates from the safety threshold defined in the constitutive model library, the process decision algorithm is triggered to optimize parameters. The process parameters optimized by the process decision algorithm are transmitted back to the physical space through the data interaction link. The operating speed of the isothermal forging press, the output power of the multi-segment intelligent temperature control system, and the control logic of the fully enclosed vacuum or protective atmosphere environment are adjusted in a closed loop, thereby realizing intelligent and adaptive control of the forging process.

[0022] In one embodiment of the invention, the real-time running speed and position of the slider, as well as the real-time load data applied to the titanium alloy billet, are acquired through the servo control system built into the isothermal forging press. Utilizing independent thermocouple feedback in each zone of the multi-segment intelligent temperature control system, surface temperature and temperature rise rate data of key sections of the mold are collected. Pressure gauges within a fully enclosed vacuum or protective atmosphere environment are used to collect furnace gas pressure or vacuum level data, and the flow rate data of the protective gas is recorded simultaneously. A high-temperature industrial camera is used to acquire images of the contour of the titanium alloy billet within the mold cavity, extracting shape changes and flash bridge height data during the filling process. A laser displacement sensor is used to scan the mold parting surface, collecting data on the elastic opening of the mold during forging. All collected load data, temperature data, gas data, shape change data, and elastic opening data are timestamped, normalized, and packaged to generate multi-source state data.

[0023] The construction of a digital twin in virtual space is accomplished in the following ways, see below. Figure 2 Microstructure photographs of titanium alloys under different heat treatment states and corresponding mechanical property test results were imported. A constitutive model library was constructed using a neural network fitting algorithm. This library includes a flow stress prediction sub-model, a damage accumulation criterion sub-model, and a dynamic recrystallization kinetics sub-model. Based on the physical space geometry, a three-dimensional geometric model including the titanium alloy billet, mold, and press was established and imported into a finite element real-time simulation engine, where a fine unstructured mesh was generated. Material property mapping rules were set in the finite element real-time simulation engine, and the flow stress prediction sub-model from the constitutive model library was embedded into the material cards of the mesh elements. A thermo-mechanical coupling solver was configured. An optimization objective vector was set in the process decision algorithm, which included evaluation indicators in three dimensions: minimizing the maximum forming load, grain size uniformity, and mold elastic deformation constraint. The finite element real-time simulation engine and the process decision algorithm were interface-bound, allowing the simulation results to be directly used as the fitness function value input for the process decision algorithm.

[0024] In practical implementation, the physical space multi-source state data acquisition and virtual space digital twin construction of the titanium alloy metal forging process parameter control method based on digital twins can be illustrated through the isothermal forging process of a TC11 titanium alloy disc for aero-engines. In physical space, a 100MN isothermal forging press, a multi-segment intelligent temperature control system comprising three independent temperature control zones (upper die, lower die, and die sleeve), a fully enclosed protective atmosphere environment protected by high-purity argon, and an online monitoring sensor network consisting of high-temperature industrial cameras, laser displacement sensors, and pressure sensors are deployed. The servo control system built into the isothermal forging press acquires real-time data on the slider's running speed and position, as well as the real-time load data applied to the titanium alloy billet at a frequency of 1000Hz. For example, when the forging stroke is 50mm, the recorded real-time load is 42.5MN. The independent thermocouples in each zone of the multi-segment intelligent temperature control system provide feedback on the surface temperature and temperature rise rate of key sections of the die at a sampling frequency of 10Hz, such as maintaining the temperature in the center area of ​​the upper die at 950°C ± 5°C. A pressure gauge continuously collects argon pressure data within a fully enclosed vacuum or protective atmosphere environment, while a synchronous flow meter records the argon inlet flow rate, maintaining the furnace pressure at 0.5 standard atmospheres and a flow rate of 20 L / min. A high-temperature industrial camera acquires images of the contour of the TC11 titanium alloy billet within the mold cavity at a rate of 5 frames per second. Image processing algorithms extract real-time data on the shape changes and flash bridge height of the billet during the filling process, for example, monitoring the flash height increase from 0 mm to 3.5 mm. A laser displacement sensor moves along the mold parting surface at a scanning frequency of 2000 Hz, collecting data on the elastic opening of the mold during forging, recording a maximum elastic opening of 0.12 mm. A central data acquisition unit timestamps and aligns the load data, temperature data, gas data, shape change data, and elastic opening data acquired by all sensors, normalizing data of different dimensions to the [0,1] interval, and finally packaging them into a structured multi-source state data package.

[0025] In virtual space, the construction of a digital twin begins with the establishment of a materials database. Metallographic images of TC11 titanium alloy in annealed, solution-treated, and double-heat-treated states, along with corresponding mechanical property test results such as room temperature tensile strength and high-temperature creep, are imported. A constitutive model library is constructed using a three-layer backpropagation neural network fitting algorithm. This library includes a flow stress prediction sub-model, a damage accumulation criterion sub-model based on the Ayada criterion, and a dynamic recrystallization kinetics sub-model based on the Avrami equation. Based on the three-dimensional geometric dimensions of the 100MN isothermal forging press, die, and TC11 titanium alloy disc blank in physical space, corresponding three-dimensional geometric models are established in CAD software. These models are then imported into the finite element real-time simulation engine component of ANSYS or similar software. For the severely deformed blank regions, tetrahedral unstructured meshes with an average size of 1 mm are generated. In the finite element real-time simulation engine, material property mapping rules are set up. The flow stress prediction sub-model from the constitutive model library is embedded into the material card of each mesh element through a user subroutine interface, and an implicit-explicit hybrid solver capable of solving thermo-mechanical coupling problems is configured. A three-dimensional optimization objective vector F is defined in the process decision algorithm. The three dimensions of the objective vector F correspond to minimizing the maximum forming load, maximizing the uniformity of the grain size in the forging cross-section, and minimizing the maximum elastic deformation of the die, respectively. The calculation output port of the finite element real-time simulation engine is interface-bound with the input port of the process decision algorithm, so that the maximum forming force, standard deviation of grain size distribution, and die deformation obtained from each simulation calculation are directly used as the fitness function value input of the process decision algorithm to evaluate the parameter quality.

[0026] In some embodiments, the data acquisition in the physical space can be further refined. For example, pressure sensors not only measure the total load but also measure the local forces at different locations on the mold through a distributed arrangement. These local force data, after normalization, are used as part of the load data in the multi-source state data to more precisely assess the stress equilibrium state of the mold. Before edge detection, the image data acquired by the high-temperature industrial camera undergoes filtering, noise reduction, and perspective correction to eliminate high-temperature radiation noise and image distortion caused by the shooting angle, ensuring that the extracted blank contour pixel coordinates accurately reflect the true shape in three-dimensional space. The laser displacement sensor arranges multiple scanning lines along the mold parting surface to acquire a two-dimensional elastic opening distribution surface data, not just data from a single line. This distribution surface data, after gridding interpolation, serves as a complete representation of the elastic opening data in the multi-source state data. Optionally, the construction of the virtual space constitutive model library can employ different machine learning architectures. In addition to backpropagation neural networks, convolutional neural networks can also be used to automatically extract features from microstructure photographs and, together with thermodynamic parameters, as input to construct a flow stress prediction sub-model. In a real-time finite element simulation engine, the mesh generation strategy can be dynamically adjusted based on the real-time load of simulation computing resources. While ensuring computational accuracy, a coarser mesh is used for non-critical areas, while a finer mesh is maintained for critical areas such as the contact boundary between the billet and the die, and the core of the billet, to balance computational speed and accuracy. The weights of the optimization objective vector F in the process decision algorithm can be manually configured or adaptively learned based on the final performance requirements of the product.

[0027] It is understandable that the packaging and generation of multi-source state data is a standardized step in the migration of physical information to digital space. Timestamp alignment ensures that data from the isothermal forging press servo control system, multi-segment intelligent temperature control system thermocouples, pressure gauges, high-temperature industrial cameras, and laser displacement sensors have a strictly unified time reference, enabling precise synchronization between virtual space simulation calculations and physical processes. Normalization eliminates the differences caused by different physical dimensions and numerical magnitudes of force, temperature, displacement, and image pixels, converting all data to the same dimensionless numerical range, providing standardized input for subsequent virtual space data analysis and model driving. The construction of the dynamic recrystallization kinetics sub-model in the constitutive model library relies on thermal simulation experimental data, and its general form can be expressed as:

[0028] in: This refers to the volume fraction of dynamic recrystallization. For true response, The critical strain. The strain is the strain when the recrystallization volume fraction reaches 50%. and The model parameters, which are related to the material, deformation temperature, and strain rate, are learned from Gleeble thermal simulation experimental data using a neural network fitting algorithm. The three-dimensional geometric model is established strictly according to the physical entity dimensions, and the configuration of the thermo-mechanical coupling solver in the finite element real-time simulation engine includes the definition and solution settings for various physical fields such as heat conduction, thermal radiation, plastic work-generated heat, and contact heat conduction.

[0029] In one embodiment of the present invention, multi-source state data collected in physical space is transmitted in real time to virtual space via a data interaction link, driving a finite element real-time simulation engine to perform transient field calculations on the current forging process. The generated virtual image includes the internal stress field, strain field, temperature field, and grain evolution state of the forging. The current forging stroke position is parsed from the multi-source state data and used as the starting point of the time step of the finite element real-time simulation engine. The mold elastic opening data collected by the laser displacement sensor is used as the input boundary condition and applied to the contact surface of the mold in the simulation model to simulate the real mold constraint state. The load data collected by the pressure sensor is verified in real time with the simulated contact force. If the difference between the load data collected by the pressure sensor and the simulated contact force exceeds the tolerance, the friction coefficient in the simulation model is dynamically adjusted. The explicit integration calculation of the finite element real-time simulation engine is started, and a virtual image of the internal stress field, strain field, temperature field, and grain evolution state of the forging at the current moment is output. During the forging process, real-time images of the billet inside the mold cavity are captured using a high-temperature industrial camera at preset time intervals. These real-time images are then binarized and edge-detected to extract the real-time outer contour pixel coordinates of the billet. These real-time outer contour pixel coordinates are projected onto the camera's imaging plane along with the outer surface of the 3D solid model of the titanium alloy billet in the finite element real-time simulation engine. The projection result is then converted into the contour pixel coordinates of the simulation model. The contour matching error between the real-time outer contour pixel coordinates and the simulation model contour pixel coordinates in the image plane is calculated; this contour matching error represents the deviation between the billet contour data and the simulation model. If the contour matching error exceeds a preset pixel threshold, a mesh node coordinate correction process is initiated. In this process, the contour matching error is converted into a position offset vector in 3D space through inverse projection transformation. This position offset vector is decomposed into components along the normal and tangential directions of the billet's outer surface. Based on a preset weighting factor, the position offset vector is applied to the corresponding mesh nodes of the 3D solid model of the titanium alloy billet in the simulation model, updating the coordinates of these mesh nodes to eliminate simulation drift errors.

[0030] In practical implementation, the data interaction, virtual image generation, and simulation drift error elimination of the process parameter control method for titanium alloy metal forging based on digital twins can be illustrated through the isothermal forging process of an aero-engine turbine disk. Multi-source state data collected in physical space is transmitted in real time to virtual space via a data interaction link, driving the finite element real-time simulation engine to perform transient field calculations on the current forging process. The generated virtual image includes the internal stress field, strain field, temperature field, and grain evolution state of the forging. The current forging stroke position is parsed from the multi-source state data; for example, when the slider position sensor reading is 120 mm, the forging stroke position of 120 mm is used as the starting point of the time step of the finite element real-time simulation engine. The mold elastic opening data collected by the laser displacement sensor is used as input boundary conditions applied to the contact surface of the mold in the simulation model to simulate the real mold constraint state; for example, if the laser displacement sensor measures an elastic opening of 0.15 mm at the center point of the mold parting surface, a displacement boundary condition of 0.15 mm is applied to the contact constraint conditions of the corresponding node in the simulation model. The load data collected by the pressure sensor is verified in real time against the simulated contact force. If the difference between the load data collected by the pressure sensor and the simulated contact force exceeds the tolerance (e.g., the measured load is 50 MN while the simulated contact force is 52 MN, and the difference exceeds the preset tolerance of 1 MN), the friction coefficient in the simulation model is dynamically adjusted from 0.3 to 0.28. Explicit integration calculation of the finite element real-time simulation engine is initiated, outputting a virtual image of the stress field, strain field, temperature field, and grain evolution state inside the forging at the current moment. At preset time intervals during the forging process, such as every 0.5 seconds, real-time images of the billet inside the mold cavity are captured by a high-temperature industrial camera. The real-time images are binarized and edge-detected to extract the real-time outer contour pixel coordinates of the billet. The real-time outer contour pixel coordinates are projected onto the camera imaging plane along with the outer surface of the 3D solid model of the titanium alloy billet in the finite element real-time simulation engine, and the projection result is converted into the contour pixel coordinates of the simulation model. Calculate the contour matching error between the real-time outer contour pixel coordinates and the simulation model contour pixel coordinates in the image plane. The contour matching error is the deviation between the blank contour data and the simulation model. It can be calculated using the following formula:

[0031] in: Indicates contour matching error. Indicates the number of contour point samples. Indicates the first Real-time outer contour pixel coordinates Indicates the first The simulation model outlines pixel coordinates. It determines whether the outline matching error exceeds a preset pixel threshold. If it does, a mesh node coordinate correction process is initiated. In this process, the outline matching error is converted into a position offset vector in three-dimensional space through inverse projection transformation. The position offset vector is decomposed into components along the normal and tangential directions of the billet's outer surface. Based on a preset weighting factor, the position offset vector is applied to the mesh nodes corresponding to the three-dimensional solid model of the titanium alloy billet in the simulation model, updating the mesh node coordinates to eliminate simulation drift errors. In some embodiments, the generation of the virtual image can include more field variables. For example, the grain evolution state field includes not only the average grain size but also microstructure variables such as grain size distribution and recrystallization fraction. These variables are calculated by coupling the finite element real-time simulation engine with the cellular automata model. The starting point of the time step of the finite element real-time simulation engine can be dynamically adjusted based on the timestamps in the multi-source state data to ensure strict synchronization between the virtual image and the physical process.

[0032] In some embodiments, the calculation of contour matching error can employ more complex metrics. For example, in addition to Euclidean distance, shape context descriptors or Hausdorff distance can be introduced to comprehensively evaluate contour similarity, thereby more accurately characterizing the geometric deviation between the simulation model and the physical entity. Optionally, the specific implementation of the inverse projection transformation depends on the camera calibration parameters. By pre-calibrating the high-temperature industrial camera, the camera's intrinsic parameter matrix and distortion coefficients, as well as the camera's pose relative to the mold, are obtained, thereby establishing a precise mapping relationship between image pixel coordinates and three-dimensional world coordinates, which is used to convert the contour matching error into a three-dimensional position offset vector. It can be understood that applying the mold elastic opening data collected by the laser displacement sensor as a boundary condition is an important way to combine physical measurement data with the simulation model. The elastic deformation of the mold directly affects the geometric dimensions of the cavity and the stress state of the blank. Introducing this measured data into the simulation can significantly improve the accuracy of stress field and contact force calculations in the virtual image and reduce model errors caused by assuming a rigid mold.

[0033] In one embodiment of the present invention, key field variables in the virtual image are compared with the constitutive model library in real time. When a key field variable is detected to deviate from the safety threshold defined in the constitutive model library, a process decision algorithm is triggered to optimize parameters. The equivalent strain rate and temperature value of the core of the forging are extracted from the virtual image and input into the damage accumulation criterion sub-model in the constitutive model library to calculate the current damage variable value. Temperature gradient data of the forging surface are extracted from the virtual image, its absolute value of thermal gradient is calculated, and compared with the dynamic recrystallization critical value in the constitutive model library. When the damage variable value is greater than the damage tolerance set by the constitutive model library, or the absolute value of thermal gradient exceeds the dynamic recrystallization critical value, it is determined that there is a risk of process instability. In response to the determination of the risk of process instability, the process decision algorithm is activated, and the current combination of process parameters and the corresponding field variable data are used as the initial population and input into the process decision algorithm. The process decision algorithm traverses new combinations of process parameters in the solution space and calls the finite element real-time simulation engine to quickly evaluate each new set of parameters until a parameter solution set that makes the key field variables return to the safety threshold is found. The decision variables for the process decision algorithm are defined, including the descent speed of the isothermal forging press, the preheating holding time in the multi-segment intelligent temperature control system, and the holding pressure during deformation. A set of constraints is constructed within the process decision algorithm, including the ultimate vacuum level of the fully enclosed vacuum or protective atmosphere environment, the allowable yield strength of the die material, and the phase transformation temperature range of the titanium alloy billet. A non-dominated sorting genetic algorithm is used to iteratively evolve the decision variables. In each generation, dominant individuals on the Pareto front are selected as candidate process parameters. For each candidate process parameter, a simplified fast simulation is performed using a finite element real-time simulation engine, calculating only the interaction results of the stress and temperature fields. Based on the simplified simulation output, the fitness ranking of individuals in the process decision algorithm is updated, and the final parameter solution set is output after reaching the maximum number of iterations or convergence.

[0034] In practical implementation, the process instability risk assessment and parameter optimization process of the digital twin-based titanium alloy metal forging process parameter control method can be illustrated through the forging process of a TC4 titanium alloy structural component for aerospace applications. Key field variables in the virtual image are compared with a constitutive model library in real time. When a key field variable deviates from the safety threshold defined in the constitutive model library, a process decision algorithm is triggered to optimize parameters. When the forging stroke reaches 65 mm, the equivalent strain rate and temperature value at a specific integration point in the core of the forging are extracted from the virtual image; for example, the equivalent strain rate is 0.8 s². -1The temperature is 925°C. The equivalent strain rate and temperature value are input into the damage accumulation criterion sub-model in the constitutive model library, and the current damage variable value is calculated to be 0.12. Temperature data of two points 5 mm apart on the surface of the forging are extracted from the virtual image, and the absolute value of its thermal gradient is calculated to be 15°C / mm. This is compared with the dynamic recrystallization critical value of 8°C / mm set for TC4 titanium alloy in the constitutive model library. When the damage variable value of 0.12 is greater than the damage tolerance of 0.1 set in the constitutive model library, or the absolute value of the thermal gradient of 15°C / mm exceeds the dynamic recrystallization critical value of 8°C / mm, it is determined that there is a risk of process instability. In response to the determination of the risk of process instability, the process decision algorithm is activated, and the current combination of process parameters and the corresponding field variable data are used as the initial population and input into the process decision algorithm. The process decision algorithm traverses the new combination of process parameters in the solution space and calls the finite element real-time simulation engine to quickly evaluate each new set of parameters until a parameter solution set that makes the key field variables return to the safety threshold is found. The decision variables for the process decision algorithm are defined, including the downward speed of the isothermal forging press, the preheating holding time in the multi-segment intelligent temperature control system, and the holding pressure during deformation. A set of constraints is constructed within the process decision algorithm, including the ultimate vacuum level of the fully enclosed vacuum or protective atmosphere environment, the allowable yield strength of the H13 steel die material, and the phase transformation temperature range of the titanium alloy billet. A non-dominated sorting genetic algorithm is used to iteratively evolve the decision variables, with a population size of 50 and a maximum number of iterations of 100. In each generation, dominant individuals on the Pareto front are selected as candidate process parameters. For each candidate process parameter, a simplified fast simulation is performed using a finite element real-time simulation engine, calculating only the interaction results of the stress and temperature fields without solving the complete grain evolution field, thus shortening the evaluation time. Based on the results of the simplified simulation, the fitness ranking of individuals in the process decision algorithm is updated, and the final parameter solution set is output after reaching the maximum number of iterations or convergence. For example, the optimized parameter combination is press speed 1.5mm / s, mold preheating temperature 920°C, and holding pressure 75MPa.

[0035] In some embodiments, the calculation of damage variable values ​​can be based on different damage models. For example, in addition to classical models based on strain and stress, the damage accumulation criterion sub-model can integrate a continuous damage mechanics model that considers void evolution, making the physical meaning of the damage variable values ​​closer to the evolution process of microscopic defects inside the material. The calculation of the absolute value of the thermal gradient can be based on denser temperature field data. The temperature of a region of the grid nodes on the surface of the forging is extracted from the virtual image, and the temperature gradient field of that region is calculated using the spatial difference method. The maximum value of the gradient field is taken as the absolute value of the thermal gradient, which is used to compare with the dynamic recrystallization critical value, thereby more sensitively capturing the risk of local overheating. Optionally, when constructing the initial population, the process decision algorithm can use the Latin hypercube sampling method instead of completely random generation to ensure that the decision variable space is covered by more uniform and representative sample points, thereby improving the search efficiency and solution quality of the non-dominated sorting genetic algorithm in the early stages of iteration. For each candidate process parameter, a simplified and rapid simulation is performed using a real-time finite element simulation engine. The mesh can employ a coarser global mesh or activate local sub-models, significantly improving the computational speed of a single simulation by sacrificing some spatial resolution. This allows the process decision algorithm to complete the evaluation of a generation of samples within seconds. It can be understood that extracting the equivalent strain rate and temperature values ​​of the forging core from the virtual image is fundamental to assessing the risk of internal damage. The core region is typically under complex triaxial compressive stress, but its equivalent strain and temperature level directly affect the nucleation and growth of voids within the material. Inputting these two field variables into the damage accumulation criterion sub-model is a crucial step in predicting whether internal cracks will occur during the forging process. The mathematical expression of the damage accumulation criterion sub-model can be:

[0036] in: For damage variable values, For equivalent change, For stress triaxiality and temperature The relevant fracture strain, a functional relationship, is obtained through mapping using a neural network sub-model in the constitutive model library. It can be understood that temperature gradient data extracted from the forging surface from the virtual image is used to determine dynamic recrystallization behavior. Rapid cooling and large temperature gradients in the surface region indicate a significant temperature difference between the surface and subsurface layers, potentially leading to uncoordinated dynamic recrystallization processes and resulting in inhomogeneous microstructures. Comparing the calculated absolute value of the thermal gradient with the dynamic recrystallization critical value stored in the constitutive model library serves as a feedforward judgment mechanism to prevent microstructural abnormalities and control grain size uniformity.

[0037] See Figure 3This is a chart for assessing the instability risk of TC4 titanium alloy forging. It visually demonstrates the correlation between forging stroke and key process risk indicators, serving as a core visualization basis for determining process instability during digital twin forging. The damage variable value represents the degree of accumulation of microscopic damage within the material, with a safety threshold of 0.1. The absolute value of the thermal gradient reflects the degree of temperature non-uniformity on the forging surface, with a dynamic recrystallization critical value of 8°C / mm. When the forging stroke reaches 65mm, both indicators exceed the safety threshold, triggering process intervention. The damage variable value between 0 and 65mm remains in the range of 0.02–0.07, far below the safety threshold. The absolute value of the thermal gradient fluctuates between 3 and 8°C / mm, not exceeding the dynamic recrystallization critical value, indicating a stable process. The damage variable value between 65 and 100mm rapidly climbs to 0.09–0.14, repeatedly exceeding the damage tolerance of 0.1, indicating a risk of internal cracking. The absolute value of the thermal gradient suddenly rises to 8–16°C / mm, far exceeding the critical value of 8°C / mm, which can easily lead to uneven microstructure between the surface and the core.

[0038] In one embodiment of the present invention, the process parameters optimized by the process decision algorithm are transmitted back to the physical space via a data interaction link to perform closed-loop regulation of the operating speed of the isothermal forging press, the output power of the multi-segment intelligent temperature control system, and the control logic of the fully enclosed vacuum or protective atmosphere environment. The parameter solution set output by the process decision algorithm is analyzed to extract the slider descent speed command and acceleration curve for the next forging stage. The slider descent speed command is sent to the PLC controller of the isothermal forging press, replacing the original speed curve and achieving dynamic intervention in the loading rate. The mold preheating temperature setpoint in the parameter solution set is extracted, and its deviation from the current measured temperature is calculated. The power output of each heating unit in the multi-segment intelligent temperature control system is adjusted using a PID control algorithm. Based on the requirements for forging atmosphere stability in the parameter solution set, the pumping speed of the vacuum pump or the opening of the inlet valve in the fully enclosed vacuum or protective atmosphere environment is adjusted to maintain a constant furnace pressure. After the adjustment command is issued, the feedback data from the physical space is continuously monitored to confirm that the process parameters have been received and executed by the physical equipment. The slide descent speed command is sent to the PLC controller of the isothermal forging press, replacing the original speed curve. This process includes segmented processing logic for the speed curve. The slide descent speed command is decomposed into several consecutive stages, each corresponding to a specific interval in the forging stroke. In the initial stage of forging, the slide descent speed is reduced based on optimization results to decrease the peak impact load of the pressure sensor in the multi-source state data. In the filling stage, the slide descent speed is dynamically adjusted according to the flow state of the billet in the die cavity to ensure that the flash height monitored by the high-temperature industrial camera does not exceed the design threshold. In the final forging stage, the slide descent speed is locked, and the die temperature is increased in conjunction with the multi-segment intelligent temperature control system to promote the recrystallization process of the grains through heat conduction. In the holding pressure stage, the slide speed is set to zero, and the clamping force of the isothermal forging press is maintained until the finite element real-time simulation engine determines that the internal stress field tends to stabilize.

[0039] In practical implementation, the optimized closed-loop adjustment of process parameters and segmented processing of speed curves in the titanium alloy metal forging process based on digital twins can be illustrated through the isothermal forging process of an aero-engine compressor disk forging. The process parameters optimized by the process decision algorithm are transmitted back to the physical space via a data interaction link to perform closed-loop adjustment of the operating speed of the isothermal forging press, the output power of the multi-segment intelligent temperature control system, and the control logic of the fully enclosed vacuum or protective atmosphere environment. The parameter solution set output by the process decision algorithm is analyzed to extract the slider descent speed command and acceleration curve for the next forging stage. For example, the parameter solution set output is {"Speed ​​command": [1.2,1.8,0.5,0.0]mm / s, "Acceleration curve": [0.05,0.02,-0.1,0.0]m / s²}. The slider descent speed command is sent to the PLC controller of the isothermal forging press, replacing the original speed curve, thus achieving dynamic intervention in the loading rate. Extract the mold preheating temperature setpoint from the parameter solution set, for example, the upper mold is set to 940°C and the lower mold to 930°C. Calculate the deviation between the setpoint and the current measured temperature, and adjust the power output of each heating unit in the multi-segment intelligent temperature control system using a PID control algorithm. Based on the requirements for forging atmosphere stability in the parameter solution set, such as maintaining the argon pressure in the furnace at 0.3 atmospheres, adjust the pumping speed of the vacuum pump or the opening of the inlet valve in the fully enclosed vacuum or protective atmosphere environment to maintain a constant furnace pressure. After the adjustment command is issued, continuously monitor the feedback data from the physical space to confirm that the process parameters have been received and executed by the physical equipment, for example, monitoring that the actual slider speed is stable at 1.2 mm / s, the mold temperature is stable at 938°C, and the furnace pressure is stable at 0.305 atmospheres.

[0040] The slider descent speed command is sent to the PLC controller of the isothermal forging press, replacing the original speed curve. This process includes segmented processing logic for the speed curve. The slider descent speed command is decomposed into several consecutive stages, each corresponding to a specific interval in the forging stroke, as shown in Table 1. In the early stage of forging, the slider descent speed is reduced based on optimization results to decrease the peak impact load of the pressure sensor in the multi-source state data. For example, in the stroke range of 0-20mm, the speed is reduced from the originally planned 2.0mm / s to 1.2mm / s. In the filling stage, the slider descent speed is dynamically adjusted according to the flow state of the billet in the die cavity to ensure that the flash height monitored by the high-temperature industrial camera does not exceed the design threshold. For example, in the stroke range of 20-80mm, when the flash height growth rate is detected to be accelerating, the speed is dynamically reduced from 1.8mm / s to 1.5mm / s. During the final forging stage, the downward speed of the slide block is locked, and the die temperature is increased in conjunction with a multi-segment intelligent temperature control system. Heat conduction is used to promote the recrystallization process of the grains. For example, in the stroke range of 80-95mm, the speed is locked at 0.5mm / s, while the die temperature is increased by 10°C. During the holding pressure stage, the slide block speed is set to zero, and the clamping force of the isothermal forging press is maintained until the finite element real-time simulation engine determines that the internal stress field tends to stabilize. For example, after the stroke is greater than 95mm, the speed is set to zero and the maximum clamping force is maintained for 120 seconds (see Table 1).

[0041] Table 1: Optimized slider speed curve segmentation instruction table

[0042] In some embodiments, the specific implementation of the PID control algorithm to adjust the power output of a multi-segment intelligent temperature control system can be more refined. Power adjustment amount The calculation can be based on the integral separation PID algorithm, and the formula is: ,in Let be the power adjustment amount at time t. The deviation between the temperature setpoint and the measured value at time t. , , For proportional, integral, and differential coefficients, This is an integral switching function; the integral term is turned off when the absolute value of the deviation exceeds a threshold to prevent integral saturation. Each temperature control zone operates independently with a set of adjustable PID controllers.

[0043] See Figure 4This is a temperature response curve of a titanium alloy forging die, showing the actual temperature changes of the upper and lower dies over time under a multi-segment intelligent temperature control system, and comparing them with the set temperature. From 0–40s, the upper die rapidly heats up from 900°C, reaching a plateau around 915°C after about 30s, consistently below the set value of 940°C. The lower die gradually heats up from 890°C, reaching a plateau around 905°C after about 40s, consistently below the set value of 930°C. Both show a stepped increase with slight fluctuations, reflecting the PID control characteristics of the temperature control system. From 40–120s, the upper die stabilizes in the 914–917°C range, with a temperature difference of approximately 23–26°C from the set value of 940°C, failing to reach the target temperature. The lower die stabilizes in the 904–907°C range, also with a temperature difference of approximately 23–26°C from the set value of 930°C, similarly failing to reach the target temperature. Both showed small, high-frequency fluctuations, indicating that the temperature control system was continuously fine-tuning, but had not yet converged to the set value.

[0044] In one embodiment of the present invention, the online self-updating of the constitutive model library is achieved through the following steps: After each batch of forging tasks is completed, the multi-source state data and corresponding virtual mirror data accumulated during the current batch of forging are retrieved. The key dimensions of the actually formed forging are extracted from the multi-source state data, and the key dimensions of the forging predicted by simulation are extracted from the virtual mirror data. The dimensional deviation between the key dimensions of the actually formed forging and the key dimensions of the forging predicted by simulation are calculated. When the dimensional deviation exceeds the allowable range in a statistical sense, the average strain rate, peak temperature, and deformation during the current batch of forging are extracted. Using the average strain rate, peak temperature, and deformation as inputs, and the actual microstructure inspection results as labels, incremental training is performed on the flow stress prediction sub-model and the dynamic recrystallization kinetics sub-model in the constitutive model library. The updated model parameters are written into the constitutive model library, replacing the original outdated parameters, thus realizing the self-evolution of the digital twin model.

[0045] In practical implementation, the constitutive model library for the constitutive model library of the method for controlling process parameters in titanium alloy metal forging based on digital twins can be illustrated through a batch management scenario of the forging production process of a TC17 titanium alloy integral bladed disk for an aero-engine. After each batch of forging tasks is completed, for example, after the production of 10 bladed disk forgings with the same drawing number, the multi-source state data and corresponding virtual mirror data accumulated during the current batch of forging are retrieved from the database. The key dimensions of the actually formed forgings are extracted from the multi-source state data, such as the blade profile contour error and tenon slot spacing obtained through a coordinate measuring machine; the key dimensions of the forgings predicted by simulation are extracted from the virtual mirror data, such as the node coordinates and shapes at corresponding positions read from the simulation result file of the final time step. The dimensional deviation between the key dimensions of the actually formed forgings and the key dimensions of the simulated forgings is calculated. For example, if the simulated value of the contour error of a certain characteristic section of the blade is 0.08 mm, while the average measured value of this error for 10 forgings is 0.12 mm, then the dimensional deviation is 0.04 mm. When dimensional deviations exceed the statistically permissible range—for example, a t-test of dimensional deviations in 10 forgings yields a p-value less than 0.05—indicating a significant difference between simulation and actual measurements, the model update process is triggered. The average strain rate, peak temperature, and deformation during the current batch of forging are extracted. Using these as inputs and the actual microstructure inspection results as labels, incremental training is performed on the flow stress prediction sub-model and dynamic recrystallization kinetics sub-model in the constitutive model library. The microstructure inspection results include the average grain size and recrystallization fraction obtained from metallographic images sampled from the same location in each forging. The updated model parameters are then written into the constitutive model library, replacing the original outdated parameters, thus achieving the self-evolution of the digital twin model.

[0046] In practical implementation, the criterion for determining whether dimensional deviations exceed the statistically permissible range can be based on a combination of preset thresholds and statistical tests. An upper limit for the permissible dimensional deviation is set, for example, the profile error deviation must not exceed 0.05 mm. Simultaneously, statistical analysis is performed on the same dimensional deviation sequence of multiple forgings within a batch, calculating its mean and standard deviation. If the mean exceeds the permissible upper limit, or if hypothesis testing indicates a significant difference between the batch's mean deviation and zero, the dimensional deviation is determined to exceed the statistically permissible range, triggering subsequent model parameter extraction and training processes. When extracting process parameters, the average strain rate is obtained by averaging the strain rate field data over time throughout the forging process; the peak temperature is taken as the maximum value of the temperature field data during the entire forging process; and the deformation is obtained by calculating the engineering strain of the initial and final height of the billet. Optionally, incremental training of sub-models in the constitutive model library can employ an online sequence learning algorithm. New training data (mean strain rate, peak temperature, and deformation as inputs, and measured microstructure results as labels) are input in batches as streaming data. The model is rapidly iteratively updated based on the original parameters. A regularization term for remembering old data is added to the loss function to prevent catastrophic forgetting. The loss function for model parameter updates... It can be designed as:

[0047] in: For the loss on new data, For the current parameters of the model, To update the model parameters, This is a memory strength coefficient used to balance the learning of new knowledge with the retention of old knowledge. Optionally, updating the trained model parameters and writing them into the constitutive model library can employ a versioning management strategy. Each update generates a new version of the model parameters, which coexists with the old version in the model library. New forging tasks use the latest version of the model by default, but for specific materials or processes, historical versions of the model can be specified for simulation to facilitate comparison of the predictive performance of different versions and backtesting analysis. It is understandable that the average strain rate, peak temperature, and deformation are extracted as inputs because these parameters are the core driving factors of the material's constitutive behavior. The average strain rate reflects the rate of deformation, the peak temperature represents the level of thermal activation energy, and the deformation describes the degree of work hardening; together, they determine the flow stress and dynamic recrystallization behavior. Correlating these macroscopically measurable process parameters with microstructural results is an effective way to correct and improve the predictive accuracy of the constitutive model. Using actual microstructural inspection results as training labels allows the model's optimization objective to directly target the final microstructural properties, achieving reverse model calibration from process to performance.

[0048] See Figure 5This is a comparison chart of prediction errors before and after the constitutive model library update, visually demonstrating the improvement in the accuracy of forging dimension prediction brought about by the online self-updating of the digital twin constitutive model library. The chart compares the changes in prediction errors before and after updates for five forging batches. In all batches, the prediction error after the update is consistently significantly lower than before, indicating that the model self-updating strategy effectively improves prediction accuracy. Both curves show a continuous decrease with increasing batches, reflecting the convergence trend of the model's continuous optimization under the accumulation of multiple batches of data. The error after the update gradually decreased from 0.08mm to 0.04mm, while the error before the update gradually decreased from 0.12mm to 0.08mm, both converging towards a smaller error range. The stable reduction in error between batches indicates that the incremental training strategy has good persistence and robustness. By accumulating actual forming data and virtual mirror data for each batch, the model significantly reduced the dimension prediction deviation after incremental training, proving the feasibility of the digital twin model's self-evolution mechanism.

[0049] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for controlling process parameters in titanium alloy metal forging based on digital twins, characterized in that, The method includes: Deploy isothermal forging presses, multi-segment intelligent temperature control systems, fully enclosed vacuum or protective atmosphere environments, and online monitoring sensor networks consisting of high-temperature industrial cameras, laser displacement sensors, and pressure sensors in physical space to collect multi-source state data during the forging process. A digital twin containing a constitutive model library, a finite element real-time simulation engine, and a process decision algorithm is constructed in virtual space. The constitutive model library is based on Gleeble thermal simulation experimental data and is used to characterize the material damage behavior of titanium alloys at different temperatures and strain rates. The multi-source state data collected in the physical space is transmitted to the virtual space in real time through the data interaction link, driving the finite element real-time simulation engine to perform transient field calculations on the current forging process and generate a virtual image. The key field variables in the virtual image are compared with the constitutive model library in real time. When the key field variables are detected to deviate from the safety threshold defined by the constitutive model library, the process decision algorithm is triggered to optimize the parameters. The process parameters optimized by the process decision algorithm are transmitted back to the physical space through a data interaction link to perform closed-loop regulation of the operating speed of the isothermal forging press, the output power of the multi-segment intelligent temperature control system, and the control logic of the fully enclosed vacuum or protective atmosphere environment.

2. The method for controlling process parameters in titanium alloy metal forging based on digital twins according to claim 1, characterized in that, The physical deployment of an isothermal forging press, a multi-segment intelligent temperature control system, a fully enclosed vacuum or protective atmosphere environment, and an online monitoring sensor network composed of high-temperature industrial cameras, laser displacement sensors, and pressure sensors is used to collect multi-source status data during the forging process, including: The real-time running speed and position of the slider, as well as the real-time load data on the titanium alloy billet, are obtained through the servo control system built into the isothermal forging press. The surface temperature and temperature rise rate data of each key section of the mold are collected by using the independent thermocouple feedback of each zone of the multi-segment intelligent temperature control system. The pressure or vacuum level data of the gas inside the furnace is collected by a pressure gauge in the fully enclosed vacuum or protective atmosphere environment, and the flow rate data of the protective gas is recorded simultaneously. The high-temperature industrial camera is used to acquire images of the outline of the titanium alloy billet in the mold cavity, and to extract the shape changes and flash bridge height data of the billet during the filling process. The laser displacement sensor is used to scan the parting surface of the mold and collect data on the elastic opening of the mold during the forging process. All collected load data, temperature data, gas data, shape change data, and elastic opening data are timestamped and normalized, and then packaged to generate the multi-source state data.

3. The method for controlling process parameters in titanium alloy metal forging based on digital twins according to claim 2, characterized in that, The construction of a digital twin in virtual space, including a constitutive model library, a real-time finite element simulation engine, and process decision algorithms, includes: Import microstructure images of titanium alloys under different heat treatment states and corresponding mechanical property test results, and construct the constitutive model library using a neural network fitting algorithm. The constitutive model library includes a flow stress prediction sub-model, a damage accumulation criterion sub-model, and a dynamic recrystallization kinetics sub-model. Based on the geometric dimensions of physical space, a three-dimensional geometric model including titanium alloy billet, mold and press is established, and the three-dimensional geometric model is imported into the finite element real-time simulation engine to divide into fine unstructured meshes; In the finite element real-time simulation engine, material property mapping rules are set, the flow stress prediction sub-model in the constitutive model library is embedded into the material card of the mesh element, and a thermo-mechanical coupling solver is configured. In the process decision algorithm, an optimization objective vector is set, which includes evaluation indicators in three dimensions: minimization of maximum forming load, uniformity of grain size, and constraint of mold elastic deformation. The finite element real-time simulation engine is interface-bound with the process decision algorithm, so that the simulation calculation results are directly used as the fitness function value input of the process decision algorithm.

4. The method for controlling process parameters in titanium alloy metal forging based on digital twins according to claim 3, characterized in that, The process of transmitting multi-source state data collected in the physical space to the virtual space in real time via a data interaction link, driving the finite element real-time simulation engine to perform transient field calculations on the current forging process, and generating a virtual image includes: The virtual image includes the internal stress field, strain field, temperature field, and grain evolution state of the forging; The current forging stroke position is parsed from the multi-source state data, and the forging stroke position is used as the starting point of the time step of the finite element real-time simulation engine. The billet contour data acquired by the high-temperature industrial camera is compared with the geometric shape of the simulation model. The deviation value between the billet contour data and the simulation model is calculated. The deviation value is used to correct the grid node coordinates of the billet in the simulation model and eliminate simulation drift error. The mold elastic opening data collected by the laser displacement sensor is used as the input boundary condition and applied to the contact surface of the mold in the simulation model to simulate the real mold constraint state. The load data collected by the pressure sensor is verified in real time with the contact force calculated by the simulation. If the difference between the load data collected by the pressure sensor and the contact force calculated by the simulation exceeds the tolerance, the friction coefficient in the simulation model is dynamically adjusted. The explicit integration calculation of the finite element real-time simulation engine is started, and a virtual image of the internal stress field, strain field, temperature field and grain evolution state of the forging at the current moment is output.

5. The method for controlling process parameters in titanium alloy metal forging based on digital twins according to claim 4, characterized in that, The key field variables in the virtual image are compared with the constitutive model library in real time. When the key field variables are detected to deviate from the safety threshold defined in the constitutive model library, the process decision algorithm is triggered to perform parameter optimization, including: The equivalent strain rate and temperature value of the core of the forging are extracted from the virtual image and input into the damage accumulation criterion sub-model in the constitutive model library to calculate the current damage variable value. Temperature gradient data of the forging surface is extracted from the virtual image, its absolute value of thermal gradient is calculated, and it is compared with the dynamic recrystallization critical value in the constitutive model library; When the value of the damage variable is greater than the damage tolerance set by the constitutive model library, or when the absolute value of the thermal gradient exceeds the dynamic recrystallization critical value, it is determined that there is a risk of process instability. In response to the determination of the risk of process instability, the process decision algorithm is activated, and the current combination of process parameters and the corresponding field variable data are used as the initial population and input into the process decision algorithm. The process decision algorithm traverses new combinations of process parameters in the solution space and calls the finite element real-time simulation engine to quickly evaluate each new set of parameters until a parameter solution set that makes the key field variables return to within the safety threshold is found.

6. The method for controlling process parameters in titanium alloy metal forging based on digital twins according to claim 5, characterized in that, The process decision algorithm traverses new combinations of process parameters in the solution space and calls the finite element real-time simulation engine to quickly evaluate each new set of parameters until a parameter solution set that causes the key field variables to regress to within a safe threshold is found, including: The decision variables of the process decision algorithm are set, including the downward speed of the isothermal forging press, the preheating holding time in the multi-segment intelligent temperature control system, and the holding pressure during the deformation process; A set of constraints is constructed within the process decision algorithm. The set of constraints includes the ultimate vacuum degree of the fully enclosed vacuum or protective atmosphere environment, the allowable yield strength of the mold material, and the phase transformation temperature range of the titanium alloy billet. The decision variables are iteratively evolved using a non-dominated sorting genetic algorithm. In each generation of evolution, dominant individuals on the Pareto front are selected as candidate process parameters. For each of the candidate process parameters, the finite element real-time simulation engine is invoked to perform a simplified and rapid simulation, calculating only the interaction results of the stress field and temperature field. Based on the results of the simplified simulation, the fitness ranking of individuals in the process decision algorithm is updated, and the final parameter solution set is output after reaching the maximum number of iterations or convergence.

7. The method for controlling process parameters in titanium alloy metal forging based on digital twins according to claim 6, characterized in that, The process parameters optimized by the aforementioned process decision algorithm are transmitted back to the physical space via a data interaction link. This enables closed-loop regulation of the operating speed of the isothermal forging press, the output power of the multi-segment intelligent temperature control system, and the control logic of the fully enclosed vacuum or protective atmosphere environment, including: The parameter solution set output by the process decision algorithm is analyzed to extract the slider descent speed command and acceleration curve for the next forging stage; The slider downward speed command is sent to the PLC controller of the isothermal forging press to replace the original speed curve and realize dynamic intervention of the loading rate. Extract the mold preheating temperature setpoint from the parameter solution set, calculate its deviation from the current measured temperature, and adjust the power output of each heating unit in the multi-segment intelligent temperature control system through a PID control algorithm; Based on the requirements for forging atmosphere stability in the parameter solution set, adjust the pumping speed of the vacuum pump or the opening of the inlet valve in the fully enclosed vacuum or protective atmosphere environment to maintain a constant pressure inside the furnace. After the adjustment command is issued, the feedback data from the physical space is continuously monitored to confirm that the process parameters have been received and executed by the physical equipment.

8. The method for controlling process parameters in titanium alloy metal forging based on digital twins according to claim 7, characterized in that, The downward speed command of the slider is sent to the PLC controller of the isothermal forging press to replace the original speed curve, including the segmentation processing logic of the speed curve: The slider downward speed command is decomposed into several consecutive stages, each stage corresponding to a specific interval in the forging stroke; In the early stages of forging, the downward speed of the slider is reduced based on the optimization results in order to reduce the peak impact load of the pressure sensor in the multi-source state data. During the filling stage, the downward speed of the slider is dynamically adjusted according to the flow state of the blank in the mold cavity to ensure that the flash height monitored by the high-temperature industrial camera does not exceed the design threshold. During the final forging stage, the downward speed of the slider is locked, and the mold temperature is increased in conjunction with the multi-segment intelligent temperature control system to promote the recrystallization process of the grains through heat conduction. During the pressure holding stage, the slider speed is set to zero, and the clamping force of the isothermal forging press is maintained until the finite element real-time simulation engine determines that the internal stress field tends to stabilize.

9. The method for controlling process parameters in titanium alloy metal forging based on digital twins according to claim 8, characterized in that, It also includes an online self-updating step for the constitutive model library: After each batch of forging tasks is completed, the multi-source status data and the corresponding virtual image data accumulated during the current batch of forging are retrieved. Extract the actual formed key dimensions of the forging from the multi-source state data, extract the simulation-predicted key dimensions of the forging from the virtual mirror data, and calculate the dimensional deviation between the actual formed key dimensions of the forging and the simulation-predicted key dimensions of the forging. When the dimensional deviation exceeds the allowable range in a statistical sense, the average strain rate, peak temperature and deformation amount in the current batch forging process are extracted. The average strain rate, peak temperature, and deformation are used as inputs, and the actual microstructure inspection results are used as labels to incrementally train the flow stress prediction sub-model and dynamic recrystallization kinetics sub-model in the constitutive model library. The updated trained model parameters are written into the constitutive model library to replace the original outdated parameters, thereby enabling the digital twin model to evolve on its own.

10. The method for controlling process parameters in titanium alloy metal forging based on digital twins according to claim 9, characterized in that, The billet contour data acquired by the high-temperature industrial camera is compared with the geometric shape of the simulation model to calculate the deviation value between the billet contour data and the simulation model. This deviation value is then used to correct the mesh node coordinates of the billet in the simulation model, eliminating simulation drift errors. This includes: During a preset time interval in the forging process, a real-time image of the billet inside the mold cavity is captured by the high-temperature industrial camera, and the real-time image is binarized and edge detected to extract the real-time outer contour pixel coordinates of the billet. The real-time outer contour pixel coordinates and the outer surface of the three-dimensional solid model of the titanium alloy billet in the finite element real-time simulation engine are projected onto the camera imaging plane, and the projection result is converted into the contour pixel coordinates of the simulation model. Calculate the contour matching error between the real-time outer contour pixel coordinates and the simulation model contour pixel coordinates in the image plane. The contour matching error is the deviation value between the blank contour data and the simulation model. Determine whether the contour matching error exceeds a preset pixel threshold. If it does, start the grid node coordinate correction process. In the mesh node coordinate correction process, the contour matching error is converted into a position offset vector in three-dimensional space through inverse projection transformation; The position offset vector is decomposed into components along the normal and tangential directions of the outer surface of the billet. Based on a preset weighting factor, the position offset vector is applied to the mesh nodes corresponding to the three-dimensional solid model of the titanium alloy billet in the simulation model, and the coordinates of the mesh nodes are updated to eliminate simulation drift error.

Images