Intelligent optimization method for heat treatment process parameters of aluminum alloy sheet strip

Abstract

The invention discloses an intelligent optimization method for heat treatment process parameters of an aluminum alloy sheet strip, relates to the technical field of aluminum alloys, and greatly improves the accuracy and real-time performance of prediction through the combination of a multi-physics field coupling model and a data-driven deep neural network model, gets rid of a traditional low-efficiency mode depending on manual trial and error, and improves the prediction accuracy and real-time performance. And the production requirements of new materials and new specifications can be quickly responded. Compared with a single-point optimal result in the prior art, due to the introduction of a multi-objective optimization algorithm, global optimization of comprehensive performance is realized through collaborative optimization among performance indexes which are mutually restricted. And finally, through real-time closed-loop control, uncertain interference in the production process is effectively overcome, the consistency of product performance and the yield are remarkably improved, the research and development period is greatly shortened, and the production cost is reduced.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of aluminum alloy, and particularly relates to an intelligent optimization method for heat treatment process parameters of aluminum alloy sheet strip. BACKGROUND

[0002] As a key lightweight high-strength material, the heat treatment process of aluminum alloy sheet strip is the core link to determine the performance of the final product. At present, there are significant technical bottlenecks in the production mode in this field: on the one hand, the traditional process highly depends on manual experience and trial-and-error, and when facing diversified alloy components and product specifications, not only is the production efficiency low, but also the consistency of product performance is difficult to guarantee; on the other hand, although some researches have tried to introduce intelligent technology, the existing schemes generally have limitations. These limitations mainly manifest in that the used model is usually poor in adaptability and difficult to accurately simulate the physical process under complex working conditions; the real-time performance of the optimization process is insufficient and cannot cope with dynamic changes in production; it is difficult to effectively trade off and cooperatively optimize multiple performance indicators such as strength, toughness and residual stress. SUMMARY

[0003] In view of the defects in the prior art, the application provides an intelligent optimization method for heat treatment process parameters of aluminum alloy sheet strip to solve the above technical problems.

[0004] An intelligent optimization method for heat treatment process parameters of aluminum alloy sheet strip, comprising the following steps:

[0005] Based on finite element analysis, a multi-physical field coupling model is constructed to cover temperature field, phase transition dynamics and residual stress evolution, and the multi-physical field coupling model is used to simulate the microstructure evolution and performance change of the aluminum alloy sheet strip under different heat treatment process parameters;

[0006] Based on deep neural network and multi-objective optimization algorithm, a multi-objective process parameter optimization model is established, which is used to represent the mapping relationship between the heat treatment process parameters and the performance indicators of the aluminum alloy sheet strip, and according to a plurality of preset optimization objectives, the optimal combination of heat treatment process parameters is solved;

[0007] During the heat treatment process, process state data is collected in real time, and the process state data is input into the intelligent optimization model to dynamically adjust the heat treatment process parameters, so as to realize closed-loop control.

[0008] As a preferred, the following steps are specifically adopted when establishing the multi-physical field coupling model:

[0009] The initial microstructure data of the aluminum alloy sheet strip is obtained by electron backscatter diffraction;

[0010] The microstructure evolution at the grain level is modeled by combining the phase field equation and the dislocation density evolution model.

[0011] As preferably, the heat treatment process parameters include at least one of heating rate, holding temperature, holding time and cooling rate.

[0012] As preferably, the intelligent optimization model is constructed, further comprising the following steps:

[0013] The performance index data of the aluminum alloy sheet strip samples under different combinations of heat treatment process parameters are collected through heat treatment experiments to construct a training data set for model training and verification.

[0014] As preferably, it further comprises an adaptive optimization step, which is specifically:

[0015] The alloy composition feature vector and the thickness feature factor are introduced in the data set training process;

[0016] The intelligent optimization model is fine-tuned using the transfer learning method to achieve adaptive optimization of aluminum alloy sheet strips with different alloy compositions or thicknesses.

[0017] As preferably, the alloy composition feature vector is represented by adjusting the diffusion activation energy parameter and the aging precipitation kinetics parameter;

[0018] The thickness feature factor is used to adjust the cooling rate and holding time.

[0019] As preferably, the performance indicators of the aluminum alloy sheet strip include at least one of hardness, tensile strength, elongation and surface oxidation degree.

[0020] As preferably, the multi-objective optimization algorithm is NSGA-II algorithm;

[0021] The multiple optimization objectives include at least two of maximizing tensile strength and elongation, controlling residual stress below a preset threshold, and minimizing energy consumption and process cycle.

[0022] As preferably, the process state data is obtained by at least one of infrared temperature measurement sensor, online hardness detector or residual stress measurement instrument.

[0023] As preferably, the dynamic adjustment is realized by fuzzy PID control algorithm to adjust the heating power or the cooling medium flow rate, and the PID control rule is as follows:

[0024] If the measured temperature is greater than the set value + 3℃, reduce the heating power;

[0025] If the hardness is less than the target value - 5HB and the elastic modulus decreases, extend the aging time;

[0026] If the residual stress is greater than 0.3 then the cooling rate is reduced.

[0027] The beneficial effects of the present application are: the scheme greatly improves the accuracy and real-time performance of the prediction by combining the multi-physical field coupling model with the data-driven deep neural network model, breaks away from the traditional inefficient mode of relying on artificial trial and error, and can quickly respond to the production needs of new materials and new specifications. Compared with the single-point optimal result of the prior art, through collaborative optimization between mutually restrictive performance indicators, the global optimal comprehensive performance is achieved. Finally, through real-time closed-loop control, the uncertainty interference in the production process is effectively overcome, the consistency and yield of product performance are significantly improved, the research and development cycle is greatly shortened, and the production cost is reduced. BRIEF DESCRIPTION OF DRAWINGS

[0028] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0029] Figure 1 A flowchart of an aluminum alloy sheet strip heat treatment process parameter intelligent optimization method provided by the present application. DETAILED DESCRIPTION

[0030] In the description of the present application, it should be understood that the terms "center", "longitudinal", "transverse", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise", "axial", "radial", "circumferential" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and therefore cannot be understood as indicating or implying that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as limiting the present application.

[0031] The following disclosure provides many different embodiments or examples for implementing different structures of the present application. In order to simplify the disclosure of the present application, the components and settings of a specific example are described below. Of course, they are only examples and the purpose is not to limit the present application.

[0032] The embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0033] As Figure 1 shown, an aluminum alloy sheet strip heat treatment process parameter intelligent optimization method includes the following steps:

[0034] Based on finite element analysis, a multi-physical field coupling model is constructed to cover temperature field, phase change dynamics and residual stress evolution, which is used to simulate the microstructure evolution and performance change of aluminum alloy sheet strip under different heat treatment process parameters;

[0035] Based on deep neural network and multi-objective optimization algorithm, a multi-objective process parameter optimization model is established to represent the mapping relationship between heat treatment process parameters and aluminum alloy sheet strip performance indicators, and to solve the optimal combination of heat treatment process parameters according to the preset multiple optimization objectives;

[0036] During the heat treatment process, real-time process state data is collected and input into the intelligent optimization model to dynamically adjust the heat treatment process parameters, realizing closed-loop control.

[0037] Among them, the deep neural network model takes heating rate, holding temperature, holding time and cooling rate as input variables, and outputs performance indicators such as tensile strength , elongation , hardness HB and residual stress , etc. The training data of the deep neural network model comes from multi-physical field simulation and a small amount of experimental data, and the network structure is a four-layer fully connected network with ReLU activation function.

[0038] In this scheme, first, a multi-physical field coupling model is constructed through finite element analysis, which tightly couples heat conduction, phase change process in the material and residual stress evolution caused thereby. When a set of heat treatment process parameters is input, the model can simultaneously calculate the temperature distribution, microstructure and final residual stress state of the entire sheet strip in time and space dimensions, providing accurate physical simulation data for subsequent optimization. Subsequently, using these simulation data and historical data accumulated in actual production, a deep neural network is trained to learn and build the mapping relationship between process parameters and final performance indicators, which is used to predict the product performance corresponding to any set of process parameters, avoiding repeated time-consuming and lengthy finite element simulation. On this basis, combined with a multi-objective optimization algorithm, the entire parameter space is explored according to the preset performance requirements to generate a scheme combination, allowing engineers to make trade-off choices among different performance indicators according to actual requirements. Finally, this optimization model is deployed to the actual production line. By installing sensors on the heat treatment equipment, the system can real-time collect actual process state data such as furnace temperature, strip speed, cooling water temperature, etc. These data are fed back to the optimization model in real time, and the model will judge whether the current working condition deviates from the optimal trajectory. Once a deviation occurs, the model will immediately recalculate and dynamically fine-tune the subsequent process parameters, thereby forming a closed-loop control system to ensure that the product performance is always stable within the target range.

[0039] Compared with the prior art, the scheme greatly improves the accuracy and real-time performance of the prediction by combining the multi-physical field coupling model with the data-driven deep neural network model, gets rid of the low-efficiency mode of traditional dependence on artificial trial and error, and can quickly respond to the production needs of new materials and new specifications. Moreover, compared with the single-point optimal result of the prior art, the multi-objective optimization algorithm realizes the global optimization of the comprehensive performance by collaborative optimization between the mutually restrictive performance indicators. Finally, through real-time closed-loop control, the uncertainty interference in the production process is effectively overcome, the consistency of product performance and the yield are significantly improved, and the research and development cycle is greatly shortened and the production cost is reduced.

[0040] More specifically, the following steps are used when establishing the multi-physical field coupling model:

[0041] The initial microstructure data of the aluminum alloy sheet strip is obtained by electron backscatter diffraction;

[0042] The grain-level organization evolution is modeled by combining the phase field equation with the dislocation density evolution model.

[0043] The grain orientation analysis of the original plate is performed by electron backscatter diffraction to obtain the grain size, grain boundary distribution and misorientation data, the grains are divided into three types of equiaxed grains, deformed grains and recrystallized grains, and the diffusion anisotropy tensor of each grain is calculated, and the specific calculation formula is as follows:

[0044] wherein, is the diffusion coefficient of the i-th grain in the j-th direction, is the reference diffusion coefficient, is the diffusion activation energy, is the gas constant, is the temperature, is the anisotropy factor, is the angle between the grain image and the diffusion direction; A three-dimensional finite element model is established in COMSOL Multiphysics, which embeds the phase field equation to describe the grain growth and phase change evolution, and the three-dimensional finite element model is specifically:

[0045]

[0046]

[0047] wherein, is the phase field variable, is the mobility, is the free energy density function, is the interfacial energy coefficient, and the Thermo-Calc database is used to calculate the free energy of each phase to simulate the dissolution behavior of aluminum-copper alloy in the solid-liquid process.​​

[0048] The dislocation density is defined as As a field variable, its evolution equation is

[0049]

[0050] where, is the dislocation density, is the dislocation production term, is the dislocation annihilation term, represents the rate of change of dislocation density with respect to time, The model is used to predict the residual stress distribution after solid solution.

[0051] First, sample on the actual production line, after cutting, grinding, electrolytic polishing and a series of sample preparation processes, the sample is put into the scanning electron microscope, by using electron backscatter diffraction technology, the real initial microstructure data of the aluminum alloy sheet strip to be processed is obtained, the initial microstructure data includes the crystallographic orientation, size, shape of each grain, and the distribution of various grain boundaries and texture information. Then input the initial microstructure data into a grain level calculation framework combined with phase field equation and dislocation density evolution model. In this model, the phase field equation plays a role in accordance with the thermodynamic principle, simulates the dynamics of phase transition such as solid solution and precipitation, describes how new phases nucleate, grow, coarsen, and the dynamic migration of interfaces between different phases. At the same time, the dislocation density evolution model is based on dislocation dynamics theory, which tracks the complex behaviors of dislocation multiplication, annihilation, slip and climb in the microstructure, which directly determines the plastic deformation and work hardening / softening characteristics of the material. The two models are not independent, but are tightly coupled through key physical mechanisms. Through this coupling, the model can accurately capture the complex and dynamic interactions and feedback between temperature field, stress field and microstructure at the grain scale during heat treatment.

[0052] More specifically, the heat treatment process parameters include at least one of a heating rate, a holding temperature, a holding time, and a cooling rate.

[0053] More specifically, when constructing the intelligent optimization model, the following steps are further included:

[0054] Collect the performance index data of the aluminum alloy sheet strip sample under different combinations of heat treatment process parameters through heat treatment experiments, and construct a training data set for model training and verification.

[0055] First, statistical methods such as response surface methodology or orthogonal experimental design are needed to systematically plan the combination of heat treatment process parameters. A series of representative parameter points are selected to ensure that these experimental points can efficiently and uniformly cover the entire process parameter space. Subsequently, according to these designed parameter combinations, sample preparation and heat treatment execution are carried out. Standardized samples are cut from the same batch of aluminum alloy sheet strip to eliminate differences in the original state of the material. Then, in the experimental furnace, each group of samples is subjected to the preset heat treatment process, and high-precision thermocouples and control systems are used to ensure temperature uniformity and the accuracy of the heating and cooling rates. After completing the heat treatment, each group of samples is subjected to comprehensive performance testing. This includes using a universal material testing machine for standard tensile testing to accurately determine key mechanical performance indicators such as tensile strength, yield strength, and elongation. At the same time, non-destructive testing techniques such as X-ray diffraction are used to quantify the size and distribution of residual stresses in the surface and interior. These data from process parameters to performance results are recorded one by one and collected into a structured, high-quality training dataset.

[0056] More specifically, it also includes an adaptive optimization step, which is specifically:

[0057] Introducing alloy composition feature vectors and thickness feature factors in the dataset training process;

[0058] Using transfer learning methods to fine-tune the intelligent optimization model to achieve adaptive optimization of aluminum alloy sheet strips with different alloy compositions or thicknesses.

[0059] In the specific application process, the thickness feature factor is guided where is the current thickness, is the reference thickness (generally taken as 1.2 mm); and for new alloys such as 7075 aluminum alloy, when training with the existing model, only the following parameters are fine-tuned: (1) the diffusion activation energy is adjusted to 150 kj / mol; (2) the aging precipitation kinetics parameter is adjusted to 1.2;

[0060] First, in the process of constructing the training dataset, an alloy composition feature vector and a thickness feature factor are introduced into each data sample. The alloy composition feature vector represents the specific mass percentage of each main alloying element and key trace element in the aluminum alloy sample. The thickness feature factor represents the nominal thickness value of the sample sheet. Then, the pre-trained base model is used through transfer learning, which is trained on the physical laws and process knowledge of aluminum alloy heat treatment. For new alloys or new thicknesses, a small amount of new data is obtained through supplementary experiments. Then, the new data is fed into the pre-trained model, and the model is fine-tuned. During the fine-tuning process, the weights near the input layer of the deep neural network are usually frozen, and only the weights near the output layer are retrained and adjusted.

[0061] More specifically, the alloy composition feature vector is characterized by adjusting the diffusion activation energy parameter and the aging precipitation kinetics parameter.

[0062] The thickness feature factor is used to adjust the cooling rate and holding time.

[0063] Adjusting the cooling rate and holding time:

[0064]

[0065]

[0066] wherein, is the adjusted holding time, is the holding time in the reference state, is the adjusted cooling rate, is the cooling rate in the reference state.

[0067] When the system receives a new alloy composition vector, the diffusion activation energy parameter and the aging precipitation kinetics parameter in the multi-physical field coupling model are dynamically adjusted through the pre-set conversion rules based on material science principles. Then, for sheet, the model will use a higher equivalent heat transfer coefficient to simulate its rapid and uniform cooling process; for thick plate, a lower coefficient will be used to capture the significant difference in cooling rate and temperature gradient between the core and the surface. In the simulation of the holding process, the thickness factor is used to adjust the effective diffusion time required for homogenization. The model will determine whether a given holding time is sufficient for the current thickness of the plate based on the relationship between the diffusion distance and the time derived from Fick's second law, combined with the thickness factor, to ensure that the simulation is a true representation of the core organization reaching the target solid solution state.

[0068] More specifically, the performance indicators of the aluminum alloy sheet include at least one of hardness, tensile strength, elongation, and surface oxidation level.

[0069] More specifically, the multi-objective optimization algorithm is NSGA-II algorithm.

[0070] The multiple optimization objectives include at least two of maximizing tensile strength and elongation, controlling residual stress below a preset threshold, and minimizing energy consumption and process cycle.

[0071] When the NSGA-II algorithm is used for multi-objective optimization, the objective function is:

[0072] maximizing wherein is a weight coefficient;

[0073] minimizing with the constraint condition being wherein, is the yield strength of the material;

[0074] minimizing unit energy consumption wherein is a change curve of heating / cooling power varying with time;

[0075] The Pareto optimal solution set is output, and the parameter combination with the optimal comprehensive performance is selected: solid solution temperature 500℃, holding time 45min, cooling rate 30℃ / min, aging temperature 190℃, and aging time 10h.

[0076] When the deep neural network proxy model is constructed, the NSGA-II algorithm starts its optimization iteration process. The NSGA-II algorithm first randomly generates an initial population containing numerous different process parameter combinations. Then, it inputs each individual in the population, i.e., each set of process parameter combinations, into the neural network proxy model to quickly obtain its corresponding multiple performance index prediction values.

[0077] More specifically, the process state data is obtained by at least one of an infrared temperature measurement sensor, an online hardness detector, or a residual stress measurement instrument.

[0078] Wherein, the temperature measurement accuracy of the infrared temperature measurement sensor is ±1℃, the accuracy of the online hardness detector is ±2HB, and the elastic modulus measurement accuracy of the laser ultrasonic system is ±1GPa.

[0079] More specifically, the dynamic adjustment is realized by a fuzzy PID control algorithm, which is used to adjust the heating power or the cooling medium flow rate. The PID control rules are as follows:

[0080] If the measured temperature > set value + 3℃, reduce the heating power;

[0081] If the hardness < target value - 5HB and the elastic modulus decreases, extend the aging time;

[0082] If residual stress > 0.3 Then reduce the cooling rate.

[0083] When the controller receives the measured temperature of a certain heating zone exceeds the set value by a threshold of 3℃, it immediately identifies the risk of overheating. At this time, the fuzzy PID algorithm is started, which first calculates the power down ratio according to the overshoot amplitude and temperature change rate through fuzzy reasoning. Subsequently, this command is quickly conveyed to the actuator to instantaneously reduce the input power of the heating element in that area by adjusting the conduction angle of the thyristor or the output frequency of the frequency converter, effectively avoiding abnormal grain growth or performance degradation due to overheating.

[0084] When the online hardness detection finds that the product hardness is lower than the target value by 5HB, and the controller also notices that the elastic modulus is also showing a downward trend, at this time, the fuzzy PID controller calculates a suitable adjustment according to the hardness deviation and modulus change, such as first adjusting the residence time of the current batch of products in the aging furnace by 5%. Then closely monitor the hardness feedback of the next few batches, if the hardness is still not up to standard, continue to adjust in proportion on the new basis, forming a stable and continuous feedback loop, until the hardness stably enters the target interval, both ensuring product quality and avoiding unnecessary energy waste.

[0085] When the residual stress detection value exceeds 30% of the material yield strength, the controller will immediately start the cooling rate intervention program. A new, more gentle target cooling rate curve is calculated. The command will drive the actuator to respond: if air cooling is used, the system will accurately reduce the flow rate of high-pressure gas through the proportional valve; if quenching liquid cooling is used, the flow of the cooling liquid will be reduced through the frequency pump, or the temperature of the quenching medium will be increased by adjusting the mixing valve to increase the proportion of warm water. This controlled intervention to actively slow down the cooling effectively suppresses the generation of excessive thermal gradients and phase transformation stresses, thereby effectively controlling the peak value of the residual stress of the final product below a safe level.

[0086] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or make equivalent substitutions for part or all of the technical features; and these modifications or substitutions do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application, and they should be covered in the scope of the claims and the specification of the present application.

Claims

1. A method for intelligent optimization of heat treatment process parameters for aluminum alloy thin sheets and strips, characterized in that, Includes the following steps: Based on finite element analysis, a multiphysics coupling model is constructed, encompassing temperature field, phase transformation dynamics, and residual stress evolution. This model is used to simulate the microstructure evolution and performance changes of aluminum alloy thin sheets and strips under different heat treatment process parameters. Based on deep neural networks and multi-objective optimization algorithms, a multi-objective process parameter optimization model is established. This model represents the mapping relationship between heat treatment process parameters and the performance indicators of aluminum alloy thin sheets and strips. Based on multiple preset optimization objectives, the optimal combination of heat treatment process parameters is obtained. During the heat treatment process, process status data is collected in real time and input into the intelligent optimization model to dynamically adjust the heat treatment process parameters and achieve closed-loop control.

2. The intelligent optimization method for heat treatment process parameters of aluminum alloy thin sheet and strip according to claim 1, characterized in that, The specific steps for establishing a multiphysics coupling model are as follows: initial microstructure data of aluminum alloy thin plates and strips are obtained by electron backscatter diffraction; the evolution of the microstructure at the grain level is modeled by combining the phase field equation and the dislocation density evolution model.

3. The intelligent optimization method for heat treatment process parameters of aluminum alloy thin sheet and strip according to claim 1, characterized in that, The heat treatment process parameters include at least one of heating rate, holding temperature, holding time, and cooling rate.

4. The intelligent optimization method for heat treatment process parameters of aluminum alloy thin sheet and strip according to claim 3, characterized in that, The intelligent optimization model construction also includes the following steps: collecting performance index data of aluminum alloy thin plate and strip samples under different combinations of heat treatment process parameters through heat treatment experiments, and constructing a training dataset for model training and verification.

5. The intelligent optimization method for heat treatment process parameters of aluminum alloy thin sheet and strip according to claim 4, characterized in that, It also includes an adaptive optimization step, which specifically involves: introducing alloy composition feature vectors and thickness feature factors during the training process of the dataset; and using transfer learning to fine-tune the intelligent optimization model to achieve adaptive optimization for aluminum alloy thin plates and strips with different alloy compositions or thicknesses.

6. The intelligent optimization method for heat treatment process parameters of aluminum alloy thin sheet and strip according to claim 5, characterized in that, The alloy composition feature vector is characterized by adjusting the diffusion activation energy parameter and the aging precipitation kinetic parameter; the thickness feature factor is used to adjust the cooling rate and holding time.

7. The intelligent optimization method for heat treatment process parameters of aluminum alloy thin sheet and strip according to claim 1, characterized in that, The performance indicators of the aluminum alloy sheet and strip include at least one of hardness, tensile strength, elongation, and surface oxidation degree.

8. The intelligent optimization method for heat treatment process parameters of aluminum alloy thin sheet and strip according to claim 1, characterized in that, The multi-objective optimization algorithm is the NSGA-II algorithm; the multiple optimization objectives include maximizing tensile strength and elongation, controlling residual stress below a preset threshold, and minimizing at least two of the following: energy consumption and process cycle.

9. The intelligent optimization method for heat treatment process parameters of aluminum alloy thin sheet and strip according to claim 1, characterized in that, The process status data is obtained through at least one of the following devices: an infrared temperature sensor, an online hardness tester, or a residual stress measuring instrument.

10. The intelligent optimization method for heat treatment process parameters of aluminum alloy thin sheet and strip according to claim 1, characterized in that, The dynamic adjustment is achieved through a fuzzy PID control algorithm, used to adjust the heating power or the cooling medium flow rate. The PID control rules are as follows: if the measured temperature > set value + 3℃, then reduce the heating power; if the hardness < target value - 5HB and the elastic modulus decreases, then extend the aging time; if the residual stress > 0.3... This will reduce the cooling rate.

Images