Aluminum profile pre-aging stability treatment method

Abstract

The application provides an aluminum profile pre-aging stability treatment method and belongs to the technical field of aluminum profile production. A data set is established through cross-section thickness scanning, a multi-scale asymptotic expansion homogenization algorithm and thermodynamic path integral Monte Carlo simulation are used to predict a temperature field and precipitate phase evolution, a double-layer game model is used to optimize partition heating power distribution, temperature history synchronous control is realized through dynamic adjustment of a model prediction controller, an artificial intelligence organization uniformity evaluation model and a learning rate dynamic adjustment mechanism are combined to realize online optimization, and finally, process parameters are adjusted through mechanical property detection feedback, so that the technical problem of large differences in precipitate phase organization states caused by uneven temperature distribution in different cross-section thickness regions during the pre-aging process of the aluminum profile is solved.

Description

Technical Field

[0001] This invention belongs to the field of aluminum profile production technology, and specifically relates to a method for pre-aging stability treatment of aluminum profiles. Background Technology

[0002] Pre-aging treatment of aluminum profiles is a crucial process that enhances material stability by promoting the diffusion and precipitation of solute atoms in the alloy through heating. In existing technologies, traditional pre-aging treatments typically employ a uniform heating method, applying the same heating power to the entire aluminum profile and relying on furnace temperature control for aging. However, due to the complex cross-sectional shape of aluminum profiles and the significant difference in heat capacity between thick-walled and thin-walled regions, there is a marked asynchrony in the heating rates and temperature responses of different parts. In current uniform heating processes, the lag in heating in thick-walled regions and the excessively rapid heating in thin-walled regions result in significant differences in the temperature history of precipitate nucleation and growth across different parts. This ultimately leads to uneven precipitate size distribution, excessive deviations in precipitate volume fraction, and poor consistency in the microstructure across different parts of the aluminum profile. In other words, existing technologies suffer from the technical problem of uneven temperature distribution across different cross-sectional thicknesses during aluminum profile pre-aging, resulting in significant differences in the microstructure of the precipitates. Summary of the Invention

[0003] In view of this, the present invention provides a method for pre-aging stability treatment of aluminum profiles, which can solve the technical problem in the prior art that uneven temperature distribution in different cross-sectional thickness regions during the pre-aging process of aluminum profiles leads to large differences in the microstructure of precipitated phases.

[0004] This invention is implemented as follows: A method for pre-aging stability treatment of aluminum profiles includes the following steps: After the aluminum profile is naturally placed, it is fed into an aging furnace. A cross-sectional thickness scanning device is used to scan and measure the thickness at various locations on the cross-section of the aluminum profile, establishing a cross-sectional thickness distribution dataset. Based on the cross-sectional thickness distribution dataset, a multi-scale asymptotic expansion homogenization algorithm is used to calculate the predicted temperature field evolution of the aluminum profile during the pre-aging process. A thermodynamic path integral Monte Carlo simulation algorithm is used to calculate the equilibrium state of precipitates under different temperature regimes. The predicted temperature field evolution and the equilibrium state of precipitates are input into an aging path co-optimizer to obtain a zoned heating power allocation scheme. The aging furnace applies differentiated heating power to different cross-sectional thickness regions of the aluminum profile according to the zoned heating power allocation scheme. Simultaneously, a multi-point temperature sensor array collects temperature data from various parts of the aluminum profile in real time. The temperature data is dynamically adjusted by a model prediction controller. The model prediction controller adjusts the temperature data accordingly. The current temperature deviation and the target temperature arrival time deviation are calculated. When the absolute value of the current temperature deviation is greater than the set temperature threshold or the absolute value of the target temperature arrival time deviation is greater than the set time threshold, the heating power correction is calculated using the furnace thermal inertia compensation function and output to each heating zone. After the aluminum profile is held in the aging furnace for a preset time, the distribution of precipitated phases in each part of the aluminum profile is predicted and evaluated using the microstructure uniformity evaluation model to obtain the microstructure uniformity index. When the microstructure uniformity index is less than the first threshold, the heating power is adjusted again. When the microstructure uniformity index is greater than or equal to the first threshold, the next step is initiated. After the aluminum profile is taken out of the furnace, mechanical properties are tested to obtain the yield strength values ​​of different parts. The yield strength dispersion coefficient is calculated. When the yield strength dispersion coefficient is less than or equal to the second threshold, the pre-aging treatment is completed. When the yield strength dispersion coefficient is greater than the second threshold, the process parameters for this batch are recorded and the heating power allocation scheme for the next batch is adjusted.

[0005] The cross-sectional thickness scanning device uses the laser triangulation principle to perform non-contact scanning of the aluminum profile cross-section. The scanning accuracy is 0.01 mm and the scanning interval is 5 mm. The resulting cross-sectional thickness distribution dataset contains thickness information of each cross-sectional position within the entire length of the aluminum profile.

[0006] Among them, the multi-scale asymptotic expansion homogenization algorithm decouples the macroscopic temperature field and the microscopic organization evolution through the asymptotic expansion method. It solves the equivalent heat conduction equation at the macroscopic scale and calculates the local phase transition dynamics in the representative volume element at the microscopic scale. The two scales are coupled through the homogenization operator.

[0007] Among them, the multi-scale asymptotic expansion homogenization algorithm establishes a macroscopic finite element mesh for the aluminum profile cross section, and establishes a microscopic representative volume element in each macroscopic element. The macroscopic temperature field is solved using the Fourier heat conduction equation, the microscopic phase transition dynamics are described by the phase field equation, and the homogenization operator feeds back the changes in microscopic phase transition latent heat and diffusion coefficient to the macroscopic temperature field equation.

[0008] Among them, the thermodynamic path integral Monte Carlo simulation algorithm is based on Feynman path integral theory, which represents the atomic diffusion and phase transition behavior in the aging process as the path evolution of the system in the configuration space, and calculates the partition function through Monte Carlo sampling.

[0009] The thermodynamic path integral Monte Carlo simulation algorithm discretizes the time axis into virtual time slices, each virtual time slice corresponding to a system configuration. It samples all system configuration paths using the Monte Carlo method, calculates the effect of each system configuration path, and obtains the precipitate equilibrium state with the minimum free energy.

[0010] Among them, the time-dependent path co-optimizer adopts a two-level game model for optimization, including an upper-level model with temperature uniformity as the objective and a lower-level model with organizational consistency as the objective. The two objective functions influence each other through temperature history coupling terms.

[0011] The upper-level model objective function is used to minimize the weighted combination of the variance of the temperature field at each part of the aluminum profile and the variance of the time to reach the target temperature, while the lower-level model objective function is used to minimize the product of the standard deviation of the precipitate size distribution and the deviation of the precipitate volume fraction.

[0012] The two-layer game model is solved using an alternating iterative method. First, the parameters of the lower-layer model are fixed to optimize the upper-layer model to obtain the optimal heating power allocation. Then, the parameters of the upper-layer model are fixed to optimize the lower-layer model to obtain the optimal aging temperature curve. The iteration is repeated until the change in the objective function of both layers is less than the set convergence threshold.

[0013] The zoned heating power distribution scheme divides the heating area of ​​the aging furnace into several independent control zones. Each independent control zone corresponds to different cross-sectional thickness ranges of aluminum profiles. The power density of the thick-walled area is different from that of the thin-walled area, and the power density of the transition area is determined by linear interpolation based on the cross-sectional thickness.

[0014] Among them, the model predictive controller establishes a dynamic model of furnace thermal inertia, adjusts the heating power in advance through multi-step prediction and rolling optimization to offset the hysteresis effect, and uses the prediction time domain to predict the future temperature evolution trajectory based on the current temperature state and furnace thermal inertia parameters.

[0015] The furnace thermal inertia compensation function calculates the heating power correction based on the current temperature deviation and the target temperature arrival time deviation. The inputs include the current temperature deviation, the target temperature arrival time deviation, the furnace heat capacity, and the current power value of the heating zone.

[0016] Among them, the uniformity assessment model utilizes a scale-invariant learning mechanism based on multi-resolution training to enhance its adaptability to different scales through joint training with different input resolutions. The input layer receives the predicted results of temperature field evolution, alloy composition, and aging time.

[0017] The model prediction controller is designed with a dynamic learning rate adjustment function. The dynamic learning rate adjustment function calculates the learning rate adjustment factor based on the current batch prediction error, the moving average of historical prediction errors, and the model output confidence. The learning rate parameter of the tissue uniformity evaluation model is dynamically adjusted according to the learning rate adjustment factor.

[0018] The yield strength dispersion coefficient is the standard deviation of the yield strength values ​​at different parts of the aluminum profile divided by the average yield strength value, and is used to evaluate the consistency of mechanical properties.

[0019] The next batch of zoned heating power allocation schemes uses a feedback adjustment method to fine-tune the power density of thick-walled and thin-walled regions based on the correlation between the process parameters of this batch and the yield strength dispersion coefficient.

[0020] This invention establishes a cross-sectional thickness distribution dataset and combines it with a multi-scale asymptotic expansion homogenization algorithm to predict temperature field evolution. It employs a two-layer game model-based time-aging path co-optimizer to generate a zoned heating power allocation scheme, applying differentiated heating power to thick-walled and thin-walled regions to ensure consistent heating curves across different cross-sectional thicknesses. The model predictive controller adjusts the heating power in advance through multi-step prediction and rolling optimization to offset the temperature response lag effect caused by furnace thermal inertia, achieving precise synchronous control of the temperature history in each part. Since the nucleation and growth process of precipitates is highly sensitive to temperature history, ensuring that each part experiences a similar temperature evolution path significantly reduces the standard deviation of precipitate size distribution and effectively controls the deviation of precipitate volume fraction, thereby achieving a high degree of consistency in the microstructure of precipitates across different parts of the aluminum profile. In summary, this invention solves the technical problem mentioned in the background art where uneven temperature distribution in different cross-sectional thickness regions during the pre-aging process of aluminum profiles leads to large differences in the microstructure of precipitates. Attached Figure Description

[0021] Figure 1 This is a flowchart of the method of the present invention.

[0022] Figure 2 This is a scan result of the thickness distribution of the aluminum profile section.

[0023] Figure 3 A comparison of temperature field evolution curves for regions with different thicknesses.

[0024] Figure 4 This is a curve showing the dynamic adjustment of heating power during the aging process.

[0025] Figure 5 Statistical chart of the size distribution and volume fraction of precipitated phases in various parts. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below.

[0027] like Figure 1 The diagram shown is a flowchart of a pre-aging stability treatment method for aluminum profiles provided by the present invention. This method includes the following steps: S01. After the aluminum profile is left to stand naturally for more than 72 hours, it is sent into the aging furnace. The thickness of the aluminum profile at each position of the cross section is scanned and measured using a cross section thickness scanning device to establish a cross section thickness distribution dataset. S02. Based on the cross-sectional thickness distribution dataset, the temperature field evolution prediction results of aluminum profiles during the pre-aging process are calculated using a multi-scale asymptotic expansion homogenization algorithm. The thermodynamic path integral Monte Carlo simulation algorithm is used to calculate the equilibrium state of precipitates under different temperature regimes. The temperature field evolution prediction results and the equilibrium state of precipitates are input into the aging path co-optimizer to obtain the partitioned heating power allocation scheme. S03. The aging furnace applies differentiated heating power to different cross-sectional thickness areas of the aluminum profile according to the zoned heating power distribution scheme. At the same time, a multi-point temperature sensor array collects temperature data of various parts of the aluminum profile in real time, and the temperature data is dynamically adjusted by the model prediction controller. S04. The model prediction controller calculates the deviation between the current temperature and the time to reach the target temperature based on the temperature data. When the absolute value of the current temperature deviation is greater than 2℃ or the absolute value of the time to reach the target temperature is greater than 3 minutes, the heating power correction is calculated using the furnace thermal inertia compensation function and output to each heating zone. S05. After the aluminum profile is held in the aging furnace for a preset time, the distribution of precipitated phases in each part of the aluminum profile is predicted and evaluated using the microstructure uniformity evaluation model to obtain the microstructure uniformity index. When the microstructure uniformity index is less than 0.85, return to S03 to continue adjusting the heating power. When the microstructure uniformity index is greater than or equal to 0.85, proceed to the next step. S06. After the aluminum profiles are taken out of the furnace, mechanical properties are tested to obtain the yield strength values ​​of different parts. The yield strength dispersion coefficient is calculated. When the yield strength dispersion coefficient is less than or equal to 0.06, the pre-aging treatment is completed. When the yield strength dispersion coefficient is greater than 0.06, the process parameters of this batch are recorded and the zonal heating power distribution scheme for the next batch is adjusted.

[0028] The cross-sectional thickness scanning device uses the laser triangulation principle to perform non-contact scanning of the aluminum profile cross-section. The scanning accuracy is 0.01 mm and the scanning interval is 5 mm. The resulting cross-sectional thickness distribution dataset contains thickness information of each cross-sectional position within the entire length of the aluminum profile.

[0029] The multi-scale asymptotic expansion homogenization algorithm decouples the macroscopic temperature field and microstructure evolution through asymptotic expansion. It solves the equivalent heat conduction equation at the macroscopic scale and calculates the local phase transition dynamics within representative volume elements at the microscopic scale. These two scales are coupled through a homogenization operator, enabling simulations spanning six orders of magnitude, from millimeter-scale cross-sections to nanometer-scale precipitates. The algorithm first establishes a macroscopic finite element mesh for the aluminum profile cross-section with a mesh size of 2 mm. Then, it establishes microscopic representative volume elements with a size of 100 nm within each macroscopic element. The macroscopic temperature field is solved using the Fourier heat conduction equation, while the microscopic phase transition dynamics are described using the phase field equation. The homogenization operator feeds back the changes in latent heat of microscopic phase transitions and diffusion coefficients to the macroscopic temperature field equation. The predicted temperature field evolution results include temperature-time curves at various locations on the aluminum profile throughout the pre-aging process.

[0030] The thermodynamic path integral Monte Carlo simulation algorithm, based on Feynman path integral theory, represents atomic diffusion and phase transition behavior during aging as path evolution of the system in configuration space. The algorithm calculates the partition function through Monte Carlo sampling, predicts the precipitate equilibrium state and transition kinetics under different temperature regimes, and introduces importance sampling and parallel tempering techniques to explore energy topography within a reasonable computation time. Specifically, the calculation process involves discretizing the time axis into 200 virtual time slices, each corresponding to a system configuration. All system configuration paths are sampled using the Monte Carlo method, and the effect of each path is calculated. System configuration paths with smaller effects contribute more to the partition function, ultimately yielding the precipitate equilibrium state with the minimum free energy. This precipitate equilibrium state includes information on the precipitate equilibrium temperature, precipitate size distribution, and precipitate volume fraction.

[0031] The aging path co-optimizer employs a two-level game model for optimization, comprising an upper-level model targeting temperature uniformity and a lower-level model targeting microstructure consistency. The upper-level model's objective function minimizes the weighted combination of the variance of the temperature field at different locations on the aluminum profile and the variance of the time it takes to reach the target temperature. Inputs include the power values ​​of each heating zone, the cross-sectional thickness distribution dataset, and material thermophysical parameters; the output is the temperature field uniformity evaluation value. The lower-level model's objective function minimizes the product of the standard deviation of the precipitate size distribution and the deviation of the precipitate volume fraction. Inputs include the predicted temperature field evolution, alloy composition, and aging time; the output is the microstructure consistency evaluation value. The two objective functions influence each other through a temperature history coupling term. The temperature field distribution of the upper-level model directly affects the precipitate evolution of the lower-level model, while the latent heat release of the phase transformation in the lower-level model, in turn, affects the temperature field calculation of the upper-level model. The upper-level model's constraints are that the sum of the power values ​​of each heating zone is less than or equal to the total power of the aging furnace, the power value of each heating zone is greater than or equal to 0, and the difference between the maximum and minimum temperature field values ​​is less than or equal to 15℃. The lower-level model is constrained by an average precipitate size between 8 nm and 25 nm and a precipitate volume fraction between 1.2% and 3.5%. The game theory model is solved using an alternating iterative method. First, the lower-level model parameters are fixed to optimize the upper-level model and obtain the optimal heating power allocation. Then, the upper-level model parameters are fixed to optimize the lower-level model and obtain the optimal aging temperature curve. This iteration is repeated until the change in the objective function of both models is less than 0.01. The aging time refers to the holding time of the aluminum profile in the aging furnace. The material thermophysical parameters include the thermal conductivity, specific heat capacity, and density of the aluminum profile.

[0032] The zonal heating power distribution scheme divides the heating area of ​​the aging furnace into several independent control zones. Each independent control zone corresponds to different cross-sectional thickness ranges of the aluminum profile, with a power density of 0.8 in the thick-walled zone. Up to 1.5 The power density in the thin-walled region is 0.3. Up to 0.7 The power density in the transition region is determined by linear interpolation based on the cross-sectional thickness. The zoned heating power allocation scheme includes the heating power setpoint for each independent control zone.

[0033] The multi-point temperature sensor array includes multiple temperature sensors arranged at different locations on the aluminum profile. The temperature sensors acquire data at a frequency of 1 second, with a temperature measurement accuracy of 0.5℃. The temperature data includes real-time temperature values ​​at various locations on the aluminum profile. The model predictive controller establishes a dynamic model of the furnace's thermal inertia and adjusts the heating power in advance to offset the hysteresis effect through multi-step prediction and rolling optimization. The model predictive controller uses a 5-step prediction time domain, with each prediction step having a 1-minute time interval. Based on the current temperature state and the furnace's thermal inertia parameters, it predicts the temperature evolution trajectory within the next 5 minutes, calculates the heating power control sequence that minimizes the deviation between the predicted trajectory and the target trajectory, and takes the first value of the heating power control sequence as the control output at the current moment. The above process is repeated at the next moment to form rolling optimization. The furnace's thermal inertia parameters include the furnace's heat capacity, the furnace wall thermal conductivity, and the heating element response time constant.

[0034] The current temperature deviation is the maximum absolute value of the difference between the real-time temperature value of each part in the temperature data and the target temperature value corresponding to the zone heating power allocation scheme. The target temperature arrival time deviation is the maximum absolute value of the difference between the time when the temperature data shows that each part reaches the target temperature value and the predetermined arrival time of the zone heating power allocation scheme.

[0035] The furnace thermal inertia compensation function is used to calculate the heating power correction based on the current temperature deviation and the target temperature arrival time deviation. The inputs include the current temperature deviation, the target temperature arrival time deviation, the furnace heat capacity, and the current power value of the heating zone. The output is the heating power correction. The function calculation process is as follows: divide the current temperature deviation by 2℃ to obtain the normalized temperature deviation value; divide the target temperature arrival time deviation by 3 minutes to obtain the normalized time deviation value; divide the furnace heat capacity by the standard furnace heat capacity of 5000kJ / ℃ to obtain the normalized heat capacity value; and divide the current power value of the heating zone by the rated power of 50kJ / ℃. The power normalization value is obtained. The heating power correction is equal to the sum of the temperature deviation normalization value and the time deviation normalization value, multiplied by the heat capacity normalization value, divided by the power normalization value, and then multiplied by the correction factor 8. The current power value of the heating zone is the actual heating power currently output by each heating zone. The heating power correction amount is added to the current power value of the heating zone to form a new heating power setting value.

[0036] The preset time is determined based on the alloy composition and cross-sectional dimensions of the aluminum profile. For 6-series aluminum alloys, the preset time is 2 to 4 hours, and for 7-series aluminum alloys, the preset time is 3 to 5 hours.

[0037] The tissue uniformity assessment model utilizes a scale-invariant learning mechanism based on multi-resolution training. Through joint training with different input resolutions, it enhances adaptability to different scales and improves performance stability in multi-resolution deployment scenarios. The structure of the tissue uniformity assessment model is as follows: the input layer receives the predicted temperature field evolution, alloy composition, and aging time. The input data is converted into a 128-dimensional feature vector by a feature extraction module. This 128-dimensional feature vector is then fed into three fully connected layers for nonlinear mapping. The number of neurons in each fully connected layer is 256, 128, and 64, respectively, and the activation function is a modified linear unit. The output layer is a single-neuron regression layer, outputting a tissue uniformity index. The tissue uniformity index ranges from 0 to 1; a higher index value indicates a more uniform tissue. The steps for establishing the training dataset for the microstructure uniformity assessment model include: collecting pre-aging experimental data of 1000 batches of aluminum profiles with different alloy compositions and process parameters. Each batch includes temperature field evolution curves, aging time, and microscopic images of precipitates in various parts after furnace exit. Image analysis is performed on the microscopic images of precipitates to obtain precipitate size distribution, precipitate volume fraction, and precipitate spatial distribution uniformity. The standard deviation of the characteristic parameters of precipitates in each part is calculated, and the standard deviation of the characteristic parameters of precipitates is normalized to the range of 0 to 1 as the microstructure uniformity index label to form training sample pairs. The steps for training the microstructure uniformity assessment model include: dividing the 1000 training sample pairs into a training set and a validation set in an 8:2 ratio. The training set is used for model parameter optimization, and the validation set is used for model performance evaluation. The mean squared error is used as the loss function, and the optimization algorithm adopts adaptive moment estimation. The initial learning rate is set to 0.001, and the model is trained for 300 rounds. After each round of training, the model performance is evaluated on the validation set. Training stops when the validation set loss no longer decreases for 20 consecutive rounds, and the model parameters with the minimum validation set loss are saved.

[0038] The model prediction controller is designed with a dynamic learning rate adjustment function, which is used to adjust the learning rate parameter of the tissue evenness assessment model during online learning. The dynamic learning rate adjustment function calculates the learning rate adjustment factor based on the current batch prediction error, the historical prediction error moving average, and the model output confidence score. The current batch prediction error is divided by the historical prediction error moving average to obtain the error ratio, and the model output confidence score is divided by the standard confidence score of 0.9 to obtain the confidence score normalization value. The learning rate adjustment factor is equal to the error ratio multiplied by the confidence score normalization value. When the learning rate adjustment factor is less than 0.5, the learning rate parameter is set to 0.1 times the initial learning rate to reduce the learning step size and avoid overfitting. When the learning rate adjustment factor is within the range [0.5, 1.5), the learning rate parameter remains at the initial learning rate to maintain a normal learning speed. When the learning rate adjustment factor is greater than or equal to 1.5, the learning rate parameter is set to 3 times the initial learning rate to accelerate the learning speed and adapt to new conditions. The current batch prediction error is the absolute value of the difference between the tissue evenness index output by the tissue evenness assessment model and the actual measured tissue evenness index. The historical prediction error moving average is the arithmetic mean of the prediction errors of the current batch over the past 10 batches. The model output confidence score is calculated from the probability distribution entropy value output by the softmax layer inside the tissue homogeneity assessment model; the smaller the entropy value, the higher the model output confidence score.

[0039] The precipitate distribution includes precipitate size distribution, precipitate volume fraction, and precipitate spatial distribution uniformity. The microstructure uniformity index is calculated by analyzing the standard deviation of precipitate size distribution, precipitate volume fraction deviation, and precipitate spatial distribution uniformity in various parts of the aluminum profile. The microstructure uniformity index ranges from 0 to 1. When the microstructure uniformity index is greater than or equal to 0.85, it indicates that the difference in microstructure in various parts of the aluminum profile is within the allowable range and meets the requirements of subsequent bending processes.

[0040] The mechanical property testing involves sampling different parts of the aluminum profile for tensile testing to measure the yield strength value. The yield strength value is the stress value when the material undergoes plastic deformation. The yield strength dispersion coefficient is the standard deviation of the yield strength values ​​at different parts of the aluminum profile divided by the average yield strength value, used to evaluate the consistency of mechanical properties. When the yield strength dispersion coefficient is less than or equal to 0.06, it indicates that the mechanical property fluctuation is within ±10MPa, meeting the product quality requirements.

[0041] The process parameters for this batch include the zoned heating power allocation scheme, preset time, and actual temperature curves for each heating zone. The zoned heating power allocation scheme for the next batch, based on the correlation between the process parameters of this batch and the yield strength dispersion coefficient, employs a feedback adjustment method to fine-tune the power density in the thick-walled and thin-walled regions, with an adjustment range of ±0.1. .

[0042] The specific implementation methods of the above steps are described in detail below.

[0043] The specific implementation of step S01 involves placing the aluminum profile, after being unloaded from the extrusion production line, in a constant temperature and humidity environment for at least 72 hours to allow the residual stress inside the aluminum profile to be fully released, thus preventing deformation caused by the release of residual stress during subsequent heat treatment. After natural placement, the operator hoists the aluminum profile onto the worktable of the cross-sectional thickness scanning device. The cross-sectional thickness scanning device uses the principle of laser triangulation, emitting a laser beam onto the surface of the aluminum profile through a laser emitter. After the laser beam hits the surface of the aluminum profile, diffuse reflection occurs. The receiver receives the reflected light spot and calculates the distance from the laser emitter to the surface of the aluminum profile using trigonometric relationships. The scanning device continuously scans along the length of the aluminum profile at 5mm intervals, while simultaneously performing a 360-degree rotation scan of the cross-sectional profile at each scanning position to obtain the thickness information of each cross-sectional position along the entire length of the aluminum profile. The scanning accuracy reaches 0.01mm, and all scanned data is stored as a cross-sectional thickness distribution dataset, which includes the spatial coordinates and corresponding thickness value of each scanning point.

[0044] The specific implementation of step S02 involves inputting the cross-sectional thickness distribution dataset into a computer workstation. The workstation first calls the multi-scale asymptotic expansion homogenization algorithm module. Based on asymptotic expansion theory, the algorithm decomposes the heat transfer problem of the aluminum profile into two levels: macroscopic and microscopic. At the macroscopic scale, a three-dimensional finite element model of the aluminum profile is established based on the cross-sectional thickness distribution dataset, with a mesh size set to 2 mm. The macroscopic temperature field distribution is solved using the Fourier heat conduction equation. At the microscopic scale, a microscopic representative volume element with a size of 100 nm is established for each macroscopic mesh unit. The phase field equation is used to describe the evolution process of the precipitated phase within the microscopic representative volume element. The homogenization operator transmits information such as the latent heat of phase change and the change in diffusion coefficient at the microscopic scale to the macroscopic scale, realizing bidirectional coupling calculation between the two scales. The temperature change curve of each location of the aluminum profile during the entire pre-aging process is obtained as the temperature field evolution prediction result. The computer workstation then invokes the thermodynamic path integral Monte Carlo simulation algorithm module. Based on Feynman path integral theory, the algorithm represents the atomic diffusion and phase transformation behavior of aluminum alloys during aging as the path integral of a quantum system on a virtual time axis. By discretizing the time axis into 200 virtual time slices, each corresponding to an atomic configuration state, the importance of all possible configuration evolution paths is sampled using the Monte Carlo method. The partition function and free energy of the system are calculated to predict the equilibrium temperature, size distribution, and volume fraction of the precipitated phase under different temperature conditions, thus obtaining the precipitated phase equilibrium state. The computer workstation inputs the temperature field evolution prediction results and the precipitated phase equilibrium state into the aging path co-optimizer. The optimizer uses a two-level game model for optimization. The upper-level model aims to minimize the weighted sum of the variance of the temperature field at each part of the aluminum profile and the variance of the time to reach the target temperature. Constraints include that the sum of the power values ​​of each heating zone does not exceed 50% of the total power of the aging furnace. The power values ​​in each heating zone are non-negative, and the difference between the maximum and minimum temperature field values ​​does not exceed 15℃. The lower-level model aims to minimize the product of the standard deviation of the precipitate size distribution and the deviation of the precipitate volume fraction. Constraints include an average precipitate size between 8nm and 25nm and a precipitate volume fraction between 1.2% and 3.5%. The two models are linked through a temperature history coupling term and solved using an alternating iterative algorithm. In each iteration, the variables of the lower-level model are first fixed to optimize the upper-level model to obtain the heating power allocation, and then the variables of the upper-level model are fixed to optimize the lower-level model to obtain the aging temperature curve. The iteration converges when the change in the objective function of both models is less than 0.01, and the partitioned heating power allocation scheme is output. In the scheme, the heating area of ​​the aging furnace is divided into several independent control zones corresponding to different cross-sectional thickness ranges, and the power density of the thick-walled region is set to 0.8. Up to 1.5 The power density in the thin-walled region is set to 0.3. Up to 0.7 The power density in the transition region is determined by linear interpolation based on the cross-sectional thickness.

[0045] The specific implementation of step S03 is that the operator sets the heating power setting value of each independent control zone in the aging furnace control system according to the zone heating power distribution scheme. The aging furnace control system adjusts the output power of the resistance heating element of each heating zone according to the power setting value, and applies differentiated heating to the aluminum profile with different cross-sectional thicknesses. The thick-walled area has a higher heating power because of its large heat capacity and slow temperature rise, while the thin-walled area has a lower heating power because of its small heat capacity and fast temperature rise, thus ensuring that the temperature rise rate of each part of the aluminum profile is consistent. The aging furnace is equipped with a multi-point temperature sensor array. The temperature sensors are thermocouples and are distributed at key locations along the length and cross-section of the aluminum profile. The sampling frequency is 1 second, and the temperature measurement accuracy is 0.5℃. The temperature sensors collect the temperature values ​​of various parts of the aluminum profile in real time and transmit them to the model predictive controller through the data acquisition system. The model predictive controller has a furnace thermal inertia dynamic model built inside. The model describes the influence of parameters such as furnace heat capacity, furnace wall thermal conductivity, and heating element response time constant on the temperature control response speed. The controller adopts a rolling time domain optimization strategy, with the prediction time domain set to 5 steps and each step time interval being 1 minute. Based on the current temperature data and the furnace thermal inertia dynamic model, the controller predicts the temperature evolution trajectory of various parts of the aluminum profile in the next 5 minutes. Through optimization algorithms, it calculates the heating power control sequence that minimizes the deviation between the predicted trajectory and the target temperature curve. The first value of the control sequence is taken as the control output for each heating zone at the current moment, realizing the dynamic adjustment of heating power.

[0046] The specific implementation of step S04 involves the model predictive controller extracting real-time temperature values ​​for each component from the temperature data, comparing them with the corresponding target temperature values ​​in the zoned heating power allocation scheme, calculating the absolute value of the temperature difference for each component, and taking the maximum value as the current temperature deviation value. Simultaneously, it calculates the actual time for each component to reach the target temperature value, compares it with the predetermined arrival time in the zoned heating power allocation scheme, calculates the absolute value of the time difference, and takes the maximum value as the target temperature arrival time deviation value. The controller determines whether the absolute value of the current temperature deviation value is greater than a threshold of 2°C or whether the absolute value of the target temperature arrival time deviation value is greater than a threshold of 3 minutes. If either condition is met, it indicates a significant deviation in temperature control, requiring the activation of the furnace thermal inertia compensation function for power correction. The furnace thermal inertia compensation function first normalizes the input parameters: dividing the current temperature deviation by the reference value of 2℃ to obtain the normalized temperature deviation value; dividing the target temperature arrival time deviation by the reference value of 3 minutes to obtain the normalized time deviation value; dividing the furnace heat capacity by the standard furnace heat capacity of 5000kJ / ℃ to obtain the normalized heat capacity value; and dividing the current actual output power value of each heating zone by the rated power of 50kJ / ℃. To obtain the power normalization value, the function calculation process involves adding the temperature deviation normalization value and the time deviation normalization value, multiplying by the heat capacity normalization value, dividing by the power normalization value, and finally multiplying by the correction factor 8. The heating power correction is obtained. The physical meaning of the correction coefficient is to convert the dimensionless deviation signal into a correction quantity with power dimensions. The model predictive controller outputs the calculated heating power correction to the aging furnace control system. The control system adds the heating power correction to the current power value of each heating zone to form a new power setpoint, thereby achieving rapid compensation for temperature deviation.

[0047] The specific implementation of step S05 is that the aluminum profile is heat-treated in an aging furnace for a set preset time. The preset time is set to 2 to 4 hours for 6 series aluminum alloys and 3 to 5 hours for 7 series aluminum alloys. After the heat treatment time is completed, the tissue uniformity evaluation model starts to work. The microstructure uniformity assessment model adopts a deep learning architecture. The model input layer receives three types of data: temperature field evolution prediction results, alloy composition, and aging time. The input data first enters the feature extraction module, which extracts the spatial distribution features and time series features of the temperature field through convolution and pooling operations, compressing the high-dimensional input data into a 128-dimensional feature vector. The 128-dimensional feature vector then enters three fully connected layers for nonlinear mapping. The first fully connected layer contains 256 neurons, the second fully connected layer contains 128 neurons, and the third fully connected layer contains 64 neurons. Each fully connected layer is followed by a modified linear unit activation function to introduce nonlinear transformation capability. The output layer is a single-neuron regression layer that outputs the microstructure uniformity index. The microstructure uniformity index ranges from 0 to 1. The larger the value, the smaller the differences in the size distribution, volume fraction, and spatial distribution uniformity of precipitates in different parts of the aluminum profile. The model predictive controller determines whether the microstructure uniformity index is less than the threshold of 0.85. When the microstructure uniformity index is less than 0.85, it means that the difference in microstructure state of various parts of the aluminum profile exceeds the allowable range. The controller issues an instruction to return to step S03 to readjust the heating power of each heating zone. The learning rate parameter of the microstructure uniformity evaluation model is adjusted by the learning rate dynamic adjustment function to adapt to the process characteristics of the current batch. When the microstructure uniformity index is greater than or equal to 0.85, it means that the microstructure uniformity meets the requirements and proceeds to the next step.

[0048] The specific implementation of step S06 is as follows: After the aluminum profile is taken out of the aging furnace and naturally cooled to room temperature, the operator cuts tensile specimens from different parts of the aluminum profile according to the standard sampling method. The sampling locations include thick-walled areas, thin-walled areas, and transition areas. Three parallel specimens are taken from each location. The specimens are clamped on a universal testing machine for tensile testing. The test is carried out according to the standard for tensile testing of metallic materials. The yield strength value of each specimen is recorded. The arithmetic mean of the yield strength values ​​of all specimens is calculated as the average yield strength value. The standard deviation of the yield strength values ​​of all specimens is calculated. The standard deviation is divided by the average yield strength value to obtain the yield strength dispersion coefficient. The yield strength dispersion coefficient reflects the degree of consistency of mechanical properties in different parts of the aluminum profile. Quality inspectors determine whether the yield strength dispersion coefficient is less than or equal to the threshold of 0.06. When the yield strength dispersion coefficient is less than or equal to 0.06, it indicates that the mechanical property fluctuation is within ±10MPa, meeting product quality requirements, and the pre-aging treatment of this batch of aluminum profiles is complete. When the yield strength dispersion coefficient is greater than 0.06, it indicates that the consistency of mechanical properties does not meet the requirements. Process personnel record the process parameters for this batch, such as the zoned heating power distribution scheme, preset time, and actual temperature curves of each heating zone. They analyze the correlation between the yield strength dispersion coefficient and the process parameters and use a feedback adjustment method to optimize the zoned heating power distribution scheme for the next batch. For areas with low yield strength, they increase the power density in the corresponding thick-walled or thin-walled areas; for areas with high yield strength, they decrease the power density in the corresponding areas. The fine-tuning range is set to ±0.1. This enables continuous optimization and improvement of process parameters.

[0049] It should be noted that one of the key technical ideas of this invention is to use a two-layer game model to collaboratively optimize temperature field uniformity and microstructure consistency. By establishing an upper-layer model with temperature uniformity as the objective and a lower-layer model with microstructure consistency as the objective, the two-layer models are mutually influenced and constrained by the temperature history coupling term. Compared with the traditional single-objective optimization method that only focuses on temperature control and ignores the microstructure evolution law, the two-layer game model can take into account the uniform precipitation of precipitates while ensuring a reasonable temperature field distribution. This avoids the phenomenon of temperature reaching the target but microstructure being uneven. It fundamentally solves the problem of inconsistent mechanical properties caused by uneven thickness of large-size aluminum profiles. Compared with the traditional empirical parameter tuning method that requires a lot of trial and error experiments, the two-layer game model directly obtains the optimal process parameters through theoretical calculation, which significantly shortens the process development cycle.

[0050] The second key technical approach is to introduce a model predictive controller to establish a dynamic model of the furnace body's thermal inertia and to use a rolling time-domain optimization strategy to achieve feedforward compensation of heating power. Traditional temperature control systems use feedback control principles, which only adjust the power after a temperature deviation occurs. Due to the large thermal inertia of the aging furnace, the response lag can reach 5 to 10 minutes. During the lag period, the aluminum profile may undergo irreversible microstructural transformation. The model predictive controller effectively counteracts the lag effect caused by the thermal inertia of the furnace body by predicting the future temperature evolution trajectory and adjusting the heating power in advance. This improves the temperature control accuracy from ±5℃ to ±2℃ using traditional methods, shortens the response time to less than 2 minutes, and significantly improves the microstructure uniformity of the precipitated phase.

[0051] The third key technological approach is to use a uniformity assessment model to directly predict the microstructure state from temperature field data. Traditional methods can only obtain organizational information through destructive testing and cannot assess the organizational state in real time during the aging process. The uniformity assessment model establishes a nonlinear mapping relationship between temperature history and organizational state through deep learning. It can output the uniformity index in real time based on the temperature field evolution prediction results during the aging process, realizing online monitoring of organizational state, providing a basis for judgment for closed-loop control, and avoiding the generation of defective products.

[0052] The synergistic effect of the three key technical approaches mentioned above forms a complete closed-loop system from process parameter design to process control and quality assessment. The two-layer game model provides the theoretically optimal partitioned heating power allocation scheme as the control target. The model predictive controller ensures that the temperature field evolves according to the optimal scheme through feedforward compensation and dynamic adjustment. The uniformity assessment model monitors the microstructure in real time and triggers feedback adjustment when it deviates from the target. The synergistic work of the three enables the pre-aging process to transform from traditional open-loop experience control to closed-loop intelligent control. Compared with traditional methods that rely on operator experience and have large batch-to-batch fluctuations, this invention achieves a significant improvement in process stability and product consistency through multi-level synergistic optimization, providing uniform and controllable billets for aluminum profile bending forming.

[0053] It should be noted that this invention also solves the following technical problem: the significant lag effect in temperature control caused by furnace thermal inertia during traditional pre-aging processes. This invention establishes a dynamic model of furnace thermal inertia using a model predictive controller. It employs a five-step predictive time domain to predict the temperature evolution trajectory within the next five minutes based on the current temperature state and furnace thermal inertia parameters. It calculates the heating power control sequence that minimizes the deviation between the predicted trajectory and the target trajectory. The first value of the control sequence is taken as the control output at the current moment, and this process is repeated at the next moment to form rolling optimization. When the temperature deviation or time deviation exceeds a threshold, the furnace thermal inertia compensation function calculates the heating power correction by multiplying the sum of the normalized temperature deviation and the normalized time deviation by the normalized heat capacity value and then dividing by the normalized power value. This advance adjustment of the heating power offsets the lag effect, achieving rapid and accurate temperature response control, thereby solving the temperature control lag problem caused by furnace thermal inertia.

[0054] Specifically, the principle of this invention is as follows: This invention employs a multi-scale asymptotic expansion homogenization algorithm to decouple the macroscopic temperature field solution from the microscopic precipitate evolution. An equivalent heat conduction equation is established at the millimeter-level cross-sectional scale using an asymptotic expansion method, while local phase transition dynamics are calculated at the nanometer-level representative volume element scale. The two scales are coupled through a homogenization operator to achieve multi-physics collaborative simulation across six orders of magnitude. The two-layer game model of the aging path co-optimizer associates the temperature uniformity objective with the tissue consistency objective through a temperature history coupling term. The upper-layer model optimizes the regional heating power allocation to minimize the temperature field variance in each part, while the lower-layer model optimizes the aging temperature curve based on the temperature field evolution prediction results to minimize the standard deviation and volume fraction deviation of the precipitate size distribution. The model predictive controller establishes a dynamic model of the furnace's thermal inertia, employing a five-step prediction time domain to pre-calculate the future temperature evolution trajectory, and generates a heating power control sequence through rolling optimization to compensate for the system's lag response. The aforementioned multi-scale simulation and bi-layer optimization mechanism ensures the synchronization of temperature history between thick-walled and thin-walled regions. Since the nucleation driving force and growth rate of precipitated phases are both determined by temperature history, consistent temperature history leads to convergence of precipitated phase evolution processes, ultimately achieving a high degree of uniformity in the tissue state of each part.

[0055] The following provides a specific embodiment 1 of the present invention, and the specific implementation of each step in this embodiment 1 is described in detail below.

[0056] The specific implementation of step S01 is as follows: After the aluminum profile is left to stand naturally for more than 72 hours, it is sent into an aging furnace. The thickness of the aluminum profile at various locations on the cross-section is scanned and measured using a cross-sectional thickness scanning device to establish a cross-sectional thickness distribution dataset. The cross-sectional thickness scanning device uses the laser triangulation principle to perform non-contact scanning of the aluminum profile cross-section, with a scanning accuracy of 0.01 mm and a scanning interval of 5 mm. The resulting cross-sectional thickness distribution dataset contains thickness information at various cross-sectional locations along the entire length of the aluminum profile.

[0057] The specific implementation of step S02 is as follows: Based on the cross-sectional thickness distribution dataset, the temperature field evolution prediction results of the aluminum profile during the pre-aging process are calculated using a multi-scale asymptotic expansion homogenization algorithm. The thermodynamic path integral Monte Carlo simulation algorithm is then used to calculate the equilibrium state of the precipitated phases under different temperature regimes. The temperature field evolution prediction results and the precipitated phase equilibrium states are input into the aging path co-optimizer to obtain a partitioned heating power allocation scheme. The multi-scale asymptotic expansion homogenization algorithm decouples the macroscopic temperature field and microstructure evolution through an asymptotic expansion method. It solves the equivalent heat conduction equation at the macroscopic scale and calculates the local phase transformation dynamics within a representative volume element at the microscopic scale. The two scales are coupled through a homogenization operator, achieving simulation across six orders of magnitude from millimeter-scale cross-sections to nanometer-scale precipitated phases. The algorithm first establishes a macroscopic finite element mesh for the aluminum profile cross-section with a mesh size of 2 mm. Then, a microscopic representative volume element with a size of 100 nm is established within each macroscopic element. The macroscopic temperature field is solved using the Fourier heat conduction equation as follows: .

[0058] In the formula, This refers to the macroscopic temperature field, with units of °C. For reference temperature, the value is taken as 100℃; Time, in seconds; For reference time, the value is 1 second; Equivalent thermal conductivity, in units of ; For reference thermal conductivity, the value is taken as 200. ; The gradient operator is used. Microscopic phase transition dynamics are described by the phase field equation, and the homogenization operator feeds back the changes in latent heat of microscopic phase transition and diffusion coefficient to the macroscopic temperature field equation. The predicted temperature field evolution results include temperature-time curves at various locations of the aluminum profile throughout the pre-aging process. The thermodynamic path integral Monte Carlo simulation algorithm, based on Feynman path integral theory, represents atomic diffusion and phase transition behavior during aging as path evolution of the system in configuration space. The algorithm calculates the partition function through Monte Carlo sampling, predicts the precipitate equilibrium state and transition dynamics under different temperature regimes, and introduces importance sampling and parallel tempering techniques to explore energy topography within a reasonable computation time. Specifically, the calculation process involves discretizing the time axis into 200 virtual time slices, each corresponding to a system configuration. All system configuration paths are sampled using the Monte Carlo method, and the action of each system configuration path is calculated. System configuration paths with smaller actions contribute more to the partition function, ultimately yielding the precipitate equilibrium state with the minimum free energy. The precipitate equilibrium state includes information on the precipitate equilibrium temperature, precipitate size distribution, and precipitate volume fraction. The aging path co-optimizer employs a two-level game model, comprising an upper-level model targeting temperature uniformity and a lower-level model targeting tissue consistency. The objective function of the upper-level model is expressed as follows: .

[0059] In the formula, This is the evaluation value for temperature field uniformity; The variance of the temperature field in various parts of the aluminum profile is given in units of 1. ; The reference temperature field variance is set to 25. ; This is the weighting coefficient, with a default value of 0.6; The variance of the time to reach the target temperature, in units of ; The reference time variance is set to 9. The objective function of the lower-level model is expressed as follows: .

[0060] In the formula, This is the organizational consistency evaluation value; The standard deviation of the precipitated phase size distribution is given in nm; the standard deviation of the reference size is given in 5 nm. The deviation in volume fraction of precipitated phase is expressed in % (%). The reference volume fraction deviation is set to 0.5%. The two objective functions influence each other through a temperature history coupling term. The temperature field distribution of the upper model directly affects the evolution of precipitates in the lower model, while the latent heat release of phase change in the lower model, in turn, affects the temperature field calculation of the upper model. The constraints of the upper model are that the sum of the power values ​​of each heating zone is less than or equal to the total power of the aging furnace, the power value of each heating zone is greater than or equal to 0, and the difference between the maximum and minimum temperature field values ​​is less than or equal to 15℃. The constraints of the lower model are that the average size of the precipitates is between 8nm and 25nm, and the volume fraction of the precipitates is between 1.2% and 3.5%. The game theory model is solved using an alternating iterative method. First, the parameters of the lower model are fixed to optimize the upper model to obtain the optimal heating power allocation. Then, the parameters of the upper model are fixed to optimize the lower model to obtain the optimal aging temperature curve. This iteration is repeated until the change in the objective functions of both models is less than 0.01. The zonal heating power distribution scheme divides the heating area of ​​the aging furnace into several independent control zones. Each independent control zone corresponds to different cross-sectional thickness ranges of the aluminum profile, with a power density of 0.8 in the thick-walled zone. Up to 1.5 The power density in the thin-walled region is 0.3. Up to 0.7 The power density in the transition region is determined by linear interpolation based on the cross-sectional thickness. The formula for calculating the power density in the transition region is as follows: .

[0061] In the formula, represents the power density in the transition region, with units of . ; Power density in thin-walled regions, in units of The value range is 0.3. Up to 0.7 ; Power density in thick-walled regions, in units of The value range is 0.8 Up to 1.5 ; This represents the thickness of the cross-section at the current location, in mm. This represents the maximum thickness threshold for the thin-walled region, in mm. This represents the minimum thickness threshold for thick-walled regions, expressed in mm.

[0062] The specific implementation method of step S03 is the same as described above, and will not be repeated in detail here.

[0063] The specific implementation of step S04 is as follows: The model predictive controller calculates the current temperature deviation value and the target temperature arrival time deviation value based on the temperature data. When the absolute value of the current temperature deviation value is greater than 2℃ or the absolute value of the target temperature arrival time deviation value is greater than 3 minutes, the heating power correction amount is calculated using the furnace thermal inertia compensation function and output to each heating zone. The current temperature deviation value is the maximum absolute value of the difference between the real-time temperature value of each part in the temperature data and the target temperature value corresponding to the zone heating power allocation scheme. The calculation formula is expressed as follows: .

[0064] In the formula, This is the current temperature deviation value, in °C. This represents the number of temperature measurement points. For the first Real-time temperature values ​​at each measuring point, in °C; For the first The target temperature value at each measuring point is expressed in °C. The target temperature arrival time deviation is the maximum absolute value of the difference between the time it takes for each part to reach the target temperature value as displayed by the temperature data and the predetermined arrival time of the zoned heating power distribution scheme. The calculation formula is as follows: .

[0065] In the formula, This represents the deviation in time from the target temperature, in minutes. For the first The actual time for each measuring point to reach the target temperature value, in minutes; For the first The predetermined time for each measuring point to reach the target temperature value, in minutes. The furnace thermal inertia compensation function is expressed as follows: .

[0066] In the formula, This is a heating power correction amount, in units of ; This is the current temperature deviation value, in °C. This represents the deviation in time from the target temperature, in minutes. This refers to the furnace body's heat capacity, expressed in kJ / ℃, typically ranging from 3000 kJ / ℃ to 8000 kJ / ℃. This is the current power value of the heating zone, in units of... .

[0067] The specific implementation of step S05 is as follows: After the aluminum profile is held at a temperature in an aging furnace for a preset time, the distribution of precipitated phases in various parts of the aluminum profile is predicted and evaluated using a microstructure uniformity evaluation model to obtain a microstructure uniformity index. When the microstructure uniformity index is less than 0.85, the process returns to S03 to continue adjusting the heating power; when the microstructure uniformity index is greater than or equal to 0.85, the process proceeds to the next step. The preset time is determined based on the alloy composition and cross-sectional dimensions of the aluminum profile. For 6-series aluminum alloys, the preset time is 2 to 4 hours, and for 7-series aluminum alloys, the preset time is 3 to 5 hours. The microstructure uniformity evaluation model utilizes a scale-invariant learning mechanism based on multi-resolution training. Through joint training with different input resolutions, it enhances adaptability to different scales and improves performance stability in multi-resolution deployment scenarios. The structure of the tissue uniformity assessment model is as follows: The input layer receives the predicted temperature field evolution results, alloy composition, and aging time. The input data is converted into a 128-dimensional feature vector by a feature extraction module. The 128-dimensional feature vector enters three fully connected layers for nonlinear mapping. The number of neurons in each fully connected layer is 256, 128, and 64, respectively, and the activation function is a modified linear unit. The output layer is a single-neuron regression layer, which outputs a tissue uniformity index. The tissue uniformity index ranges from 0 to 1, with a larger index indicating more uniform tissue. The prediction error for the current batch is the absolute value of the difference between the tissue uniformity index output by the tissue uniformity assessment model and the actual measured tissue uniformity index, calculated using the following formula: .

[0068] In the formula, This represents the prediction error for the current batch. The tissue homogeneity index is the output of the tissue homogeneity assessment model, with a value ranging from 0 to 1. The actual measurement of tissue uniformity index, with a value ranging from 0 to 1. The historical prediction error moving average is the arithmetic mean of the prediction errors of the current batch over the past 10 batches, and the calculation formula is as follows: .

[0069] In the formula, This is the moving average of historical forecast errors; For the first The prediction error of the current batch in each historical batch. The learning rate dynamic adjustment function is expressed as follows: .

[0070] In the formula, The adjusted learning rate parameter; The initial learning rate is 0.001. The learning rate adjustment factor is described in detail below: .

[0071] In the formula, This represents the prediction error for the current batch. This is the moving average of historical forecast errors; Output confidence scores for the model. When hour, ;when hour, ;when hour, .

[0072] The specific implementation of step S06 is as follows: After the aluminum profile is removed from the furnace, its mechanical properties are tested to obtain the yield strength values ​​of different parts. The yield strength dispersion coefficient is calculated. When the yield strength dispersion coefficient is less than or equal to 0.06, the pre-aging treatment is completed. When the yield strength dispersion coefficient is greater than 0.06, the process parameters for this batch are recorded and the zonal heating power allocation scheme for the next batch is adjusted. The yield strength dispersion coefficient is the standard deviation of the yield strength values ​​of different parts of the aluminum profile divided by the average yield strength value, and is expressed as follows: .

[0073] In the formula, The coefficient of variation of yield strength; This represents the standard deviation of the yield strength values ​​at different locations of the aluminum profile, in MPa. This represents the average yield strength, expressed in MPa. The formula for calculating the standard deviation of the yield strength is as follows: .

[0074] In the formula, This represents the standard deviation of the yield strength values ​​at different locations of the aluminum profile, in MPa. This represents the number of sampling test points; For the first The yield strength value at each test point, in MPa; The average yield strength is expressed in MPa, and the calculation formula is as follows: .

[0075] In the formula, This is the average yield strength value, in MPa. This represents the number of sampling test points; For the first The yield strength value at each test point is expressed in MPa.

[0076] To better understand and implement this invention, a specific application scenario is provided in Example 2: A technical team undertook a pre-aging treatment task for a batch of 7-series aluminum alloy profiles. These profiles are used in the manufacture of aerospace structural components, with complex cross-sections and significant thickness variations, ranging from a minimum of 2.5 mm to a maximum of 18 mm. Traditional uniform heating methods resulted in over-aging of thin-walled areas and insufficient aging of thick-walled areas, causing inconsistent springback during subsequent bending forming. The technical team decided to use the pre-aging stability treatment method described in this invention to solve this problem.

[0077] After quenching, the aluminum profiles are left to stand naturally for 76 hours, and then sent to an aging furnace equipped with a zoned heating system. For example... Figure 2 As shown, the cross-sectional thickness scanning device uses the laser triangulation principle to scan the entire length of the profile with a scanning accuracy of 0.01 mm and a scanning interval of 5 mm, collecting thickness data at 4800 cross-sectional locations. The scanning results show that there are three typical thickness regions along the length of the profile: the first region has an average thickness of 3.2 mm, the second region has an average thickness of 9.6 mm, and the third region has an average thickness of 16.8 mm. The thickness distribution dataset contains detailed geometric information for each cross-section.

[0078] Based on the cross-sectional thickness distribution dataset, a multi-scale asymptotic expansion homogenization algorithm was used to calculate the temperature field evolution. The algorithm established a finite element mesh at the macroscopic scale with a mesh size of 2 mm, generating 12,000 macroscopic elements. Within each macroscopic element, a 100 nm microscopic representative volume element was established, and the Fourier heat conduction equation and the phase-field equation were coupled and solved using an asymptotic expansion method. Figure 3 As shown, the temperature field evolution prediction shows that under the same heating power, the temperature rise rate of the 2.5mm thin-walled region is 1.8℃ / min, while the temperature rise rate of the 18mm thick-walled region is only 0.4℃ / min, and the time difference between the two reaching the target temperature of 175℃ is as high as 52 minutes.

[0079] The thermodynamic path integral Monte Carlo simulation algorithm was used to calculate the precipitation behavior of 7-series aluminum alloys under different temperature regimes. The time axis was discretized into 200 imaginary time slices, and Monte Carlo sampling was used for calculation. System configuration paths are calculated. The results are shown in Table 1.

[0080] Table 1. Equilibrium parameters of precipitated phases under different temperature regimes

[0081] The time-dependent path co-optimizer employs a two-layer game model for partitioned heating power allocation. The upper-layer model aims at achieving temperature uniformity, and its inputs include the furnace's total power of 480kW, cross-sectional thickness distribution dataset, thermal conductivity of 7-series aluminum alloy of 118W / (m·K), specific heat capacity of 896J / (kg·K), and density of 2810. Material thermophysical parameters were used. The lower-level model aimed for microstructure consistency, inputting the predicted temperature field evolution and an aging time of 4 hours. After 37 alternating iterations, the two-level model converged, with the change in the objective function of the upper-level model decreasing to 0.008 and the change in the objective function of the lower-level model decreasing to 0.007. The optimization results divided the aging furnace into 6 independent control zones, corresponding to different cross-sectional thickness ranges. As shown in Table 2, the power density in the thin-walled zone was set to 0.45. The power density in the medium-thick wall region is 0.88. The power density in the thick-walled region is 1.32. The power density in the transition region is determined by linear interpolation.

[0082] Table 2 Zoned Heating Power Allocation Scheme

[0083] The aging furnace starts according to the zoned heating power distribution scheme, and a multi-point temperature sensor array collects the temperature of various parts of the profile in real time. The sensor array contains 48 measuring points, distributed in different cross-sectional thickness areas of the profile, with a sampling frequency of 1 second and a temperature measurement accuracy of 0.5℃. The model predictive controller establishes a dynamic model of the furnace body's thermal inertia. The measured value of the furnace body's heat capacity is 4850 kJ / ℃, the thermal conductivity of the furnace wall is 1.8 W / (m·K), and the response time constant of the heating element is 42 seconds. The controller uses a 5-step prediction time domain, with each prediction step occurring at a time interval of 1 minute, to predict the temperature evolution trajectory within the next 5 minutes based on the current temperature state.

[0084] like Figure 4 As shown, at the 38th minute of the aging process, the real-time temperature of the thick-walled zone was 162℃, while the target temperature was 175℃, resulting in a current temperature deviation of 13℃, exceeding the 2℃ threshold. Simultaneously, the expected time for this zone to reach the target temperature was 54 minutes, while the scheduled arrival time according to the zoned heating power allocation scheme was 49 minutes, resulting in a target temperature arrival time deviation of 5 minutes, exceeding the 3-minute threshold. The model predictive controller immediately activated the furnace thermal inertia compensation function for calculation. The normalized temperature deviation value was 6.5, the normalized time deviation value was 1.67, the normalized heat capacity value was 0.97, the normalized current power value of the thick-walled zone was 2.34, and the heating power correction was 27.2kW. The controller added the correction to the current power of the thick-walled zone (117kW), forming a new power setpoint of 144.2kW. After synchronous adjustment, the real-time temperature of the thin-walled zone reached 178℃ at the 46th minute, exceeding the target temperature of 175℃ by 3℃. The controller then negatively corrected the power of the thin-walled zone, reducing the power density to 0.32. To avoid exceeding the time limit.

[0085] After the profile was held in an aging furnace for 4 hours, a microstructure uniformity assessment model predicted the distribution of precipitated phases in each part. This model was constructed based on a scale-invariant learning mechanism trained at multiple resolutions. The input layer received the predicted temperature field evolution, the alloy composition Al-5.8Zn-2.3Mg-1.6Cu-0.12Zr, and the aging time of 4 hours. The feature extraction module converted the input data into a 128-dimensional feature vector, which was then nonlinearly mapped through three fully connected layers with 256, 128, and 64 neurons respectively, using a modified linear unit as the activation function. The output layer showed a microstructure uniformity index of 0.87, exceeding the 0.85 threshold, indicating that the differences in microstructure among different parts of the profile were within acceptable limits. Figure 5 As shown, the prediction results indicate that the average size of the precipitated phase in the thin-walled region is 14.8 nm, the average size of the precipitated phase in the thick-walled region is 15.6 nm, the standard deviation of the size distribution is 1.2 nm, the volume fraction of the precipitated phase fluctuates between 2.38% and 2.46%, and the spatial distribution uniformity is good.

[0086] After the profiles were produced from the furnace, their mechanical properties were tested. Tensile tests were conducted on samples taken from six different thicknesses. The test results showed that the yield strength values ​​were 485 MPa, 488 MPa, 492 MPa, 487 MPa, 490 MPa, and 489 MPa, with an average yield strength of 488.5 MPa and a standard deviation of 2.6 MPa. The calculated yield strength dispersion coefficient was 0.053, which is less than the threshold of 0.06, indicating that the fluctuation in mechanical properties is within the allowable range, and the pre-aging treatment has been completed.

[0087] The technological advancements of this invention compared to traditional uniform heating methods are reflected in multiple aspects. Traditional methods place the entire profile in a uniform temperature environment, neglecting the different heat transfer rates caused by variations in cross-sectional thickness. This inevitably leads to thin-walled regions reaching the aging temperature before thick-walled regions and undergoing over-aging, resulting in severe coarsening of the precipitated phases. Conversely, thick-walled regions experience insufficient aging, and the precipitation strengthening effect is not fully realized. This invention couples the macroscopic temperature field with microscopic precipitation kinetics through a multi-scale asymptotic expansion homogenization algorithm, achieving cross-scale simulation from millimeter-level cross-sections to nanometer-level precipitated phases, accurately predicting the temperature response characteristics of regions with different thicknesses. The thermodynamic path integral Monte Carlo algorithm, based on Feynman path integral theory, represents the phase transition behavior of the aging process as the path evolution of the system in configuration space. Through extensive Monte Carlo sampling, it explores the energy landscape, finding the equilibrium state of the precipitated phase with minimum free energy, providing reliable phase transition kinetic boundary conditions for the optimizer. The two-level game theory model simultaneously optimizes both temperature uniformity and microstructure consistency. By linking the macroscopic heating process with the microstructure evolution through a temperature history coupling term, it overcomes the limitation of single-objective optimization, which cannot simultaneously consider temperature field distribution and precipitate state. The model predictive controller incorporates a furnace thermal inertia compensation mechanism, adjusting the heating power in advance through multi-step prediction and rolling optimization to offset the hysteresis effect of the heating system, ensuring that the temperature of each part reaches the target value synchronously according to a predetermined trajectory. The microstructure uniformity evaluation model employs a scale-invariant learning mechanism, exhibiting excellent generalization ability to temperature field data of different resolutions. It can accurately predict the precipitate distribution state based on real-time temperature data, achieving closed-loop control of the aging process. The entire technical solution forms a complete closed loop from physical mechanism modeling, numerical algorithm optimization, control strategy design to online evaluation feedback, fundamentally solving the problems of uneven temperature field and inconsistent microstructure state during the pre-aging process of complex cross-section aluminum profiles.

[0088] It should be noted that the variables involved in this invention are explained in detail in Table 3.

[0089] Table 3. Variable Explanation Table

[0090] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for pre-aging stability treatment of aluminum profiles, characterized in that, Includes the following steps: After the aluminum profiles are allowed to rest naturally, they are fed into an aging furnace. A cross-sectional thickness scanning device is used to scan and measure the thickness at various locations along the cross-section of the aluminum profile, establishing a cross-sectional thickness distribution dataset. Based on this dataset, a multi-scale asymptotic expansion homogenization algorithm is used to calculate the predicted temperature field evolution of the aluminum profiles during the pre-aging process. A thermodynamic path integral Monte Carlo simulation algorithm is then used to calculate the equilibrium state of precipitates under different temperature regimes. The predicted temperature field evolution and the equilibrium state of the precipitates are input into an aging path co-optimizer to obtain a zoned heating power allocation scheme. The aging furnace applies differentiated heating power to different cross-sectional thickness regions of the aluminum profile according to the zoned heating power allocation scheme. Simultaneously, a multi-point temperature sensor array collects real-time temperature data from various parts of the aluminum profile. The temperature data is dynamically adjusted by a model prediction controller. The model prediction controller calculates the current temperature deviation and the time deviation from the target temperature based on the temperature data. When the absolute value of the current temperature deviation is greater than the set temperature threshold or the absolute value of the target temperature arrival time deviation is greater than the set time threshold, the heating power correction is calculated using the furnace thermal inertia compensation function and output to each heating zone. After the aluminum profile is held in the aging furnace for a preset time, the distribution of precipitated phases in each part of the aluminum profile is predicted and evaluated using the microstructure uniformity evaluation model to obtain the microstructure uniformity index. When the microstructure uniformity index is less than the first threshold, the heating power is adjusted again. When the microstructure uniformity index is greater than or equal to the first threshold, the next step is initiated. After the aluminum profile is taken out of the furnace, mechanical properties are tested to obtain the yield strength values ​​of different parts and calculate the yield strength dispersion coefficient. When the yield strength dispersion coefficient is less than or equal to the second threshold, the pre-aging treatment is completed. When the yield strength dispersion coefficient is greater than the second threshold, the process parameters for this batch are recorded and the heating power allocation scheme for the next batch is adjusted.

2. The method for pre-aging stability treatment of aluminum profiles according to claim 1, characterized in that, The cross-sectional thickness scanning device uses the laser triangulation principle to perform non-contact scanning of the aluminum profile cross-section. The scanning accuracy is 0.01 mm and the scanning interval is 5 mm. The resulting cross-sectional thickness distribution dataset contains thickness information of each cross-sectional location along the entire length of the aluminum profile.

3. The pre-aging stability treatment method for aluminum profiles according to claim 2, characterized in that, The multi-scale asymptotic expansion homogenization algorithm decouples the macroscopic temperature field and microstructure evolution through asymptotic expansion. It solves the equivalent heat conduction equation at the macroscopic scale and calculates the local phase transition dynamics within a representative volume element at the microscopic scale. The two scales are coupled through a homogenization operator.

4. The pre-aging stability treatment method for aluminum profiles according to claim 3, characterized in that, The multi-scale asymptotic expansion homogenization algorithm establishes a macroscopic finite element mesh for the aluminum profile cross-section, and establishes a microscopic representative volume element within each macroscopic element. The macroscopic temperature field is solved using the Fourier heat conduction equation, and the microscopic phase transition dynamics are described using the phase field equation. The homogenization operator feeds back the changes in the latent heat of microscopic phase transition and diffusion coefficient to the macroscopic temperature field equation.

5. The pre-aging stability treatment method for aluminum profiles according to claim 4, characterized in that, The thermodynamic path integral Monte Carlo simulation algorithm is based on Feynman path integral theory. It represents the atomic diffusion and phase transition behavior in the aging process as the path evolution of the system in the configuration space and calculates the partition function through Monte Carlo sampling.

6. The method for pre-aging stability treatment of aluminum profiles according to claim 5, characterized in that, The thermodynamic path integral Monte Carlo simulation algorithm discretizes the time axis into virtual time slices, each virtual time slice corresponding to a system configuration. It samples all system configuration paths using the Monte Carlo method, calculates the effect of each system configuration path, and obtains the equilibrium state of the precipitated phase with the minimum free energy.

7. The pre-aging stability treatment method for aluminum profiles according to claim 6, characterized in that, The time-dependent path co-optimizer uses a two-level game model for optimization, which includes an upper-level model with the objective of temperature uniformity and a lower-level model with the objective of organizational consistency. The two objective functions influence each other through temperature history coupling terms.

8. The pre-aging stability treatment method for aluminum profiles according to claim 7, characterized in that, The objective function of the upper-level model is used to minimize the weighted combination of the variance of the temperature field at each part of the aluminum profile and the variance of the time to reach the target temperature. The objective function of the lower-level model is used to minimize the product term of the standard deviation of the precipitate size distribution and the deviation of the precipitate volume fraction.

9. The pre-aging stability treatment method for aluminum profiles according to claim 8, characterized in that, The two-layer game model is solved using an alternating iterative method. First, the parameters of the lower-layer model are fixed to optimize the upper-layer model to obtain the optimal heating power allocation. Then, the parameters of the upper-layer model are fixed to optimize the lower-layer model to obtain the optimal aging temperature curve. The iteration is repeated until the changes in the objective functions of both layers are less than the set convergence threshold.

10. The method for pre-aging stability treatment of aluminum profiles according to claim 9, characterized in that, The zoned heating power distribution scheme divides the heating area of ​​the aging furnace into several independent control zones. Each independent control zone corresponds to different cross-sectional thickness ranges of aluminum profiles. The power density of the thick-walled zone is different from that of the thin-walled zone, and the power density of the transition zone is determined by linear interpolation based on the cross-sectional thickness.

Images