Aluminum alloy round aluminum rod production process optimization control method

Abstract

The application provides an aluminum alloy round aluminum rod production process optimization control method, relates to the technical field of control optimization, and comprises the following steps: obtaining initial process parameters of each production link and constructing a weight matrix; collecting dynamic output characteristics and environmental interference parameters of each production link in real time; constructing a smelting-casting correction model, and predicting output casting temperature correction parameters and casting speed correction parameters; constructing a casting-rolling correction model, and predicting output rolling temperature correction parameters and rolling speed correction parameters; constructing a rolling-heat treatment correction model, and predicting output heat treatment temperature correction parameters and heat preservation time correction parameters; based on the correction parameters output by each model, combining a preset fluctuation early warning threshold, implementing progressive control parameter correction for the casting, rolling and heat treatment links, and dynamically updating the weight matrix according to the correction effect. The technical problems that product performance fluctuates greatly and the qualified rate is low in the production of aluminum alloy round aluminum rods in the prior art are solved.

Description

Technical Field

[0001] This invention relates to the field of control optimization, and in particular to a method for optimizing and controlling the production process of aluminum alloy round aluminum rods. Background Technology

[0002] Aluminum alloy round rods are a core basic material in fields such as power transmission, rail transportation, and machinery manufacturing. Among them, aluminum alloy electrical round rods are specifically designed for the power transmission field, and their core purpose is to serve as conductor raw materials for wires and cables. Aluminum alloy round rods possess high strength, high plasticity and toughness, and high fatigue resistance. At the same time, both their appearance and internal quality must meet high precision, high quality, and lightweight performance standards. Their production process includes multiple stages such as smelting, die casting, rolling, and heat treatment. Current aluminum alloy intelligent manufacturing is an advanced intelligent manufacturing model that reconstructs the entire industrial chain through digitalization, automation, and intelligent technologies. However, there are strong quality fluctuation coupling and transmission effects between each stage.

[0003] However, in the existing technology, each process of aluminum alloy round aluminum rod adopts independent parameter setting, lacks collaborative perception of the transmission of quality fluctuations across processes, and quality control mostly relies on post-processing, making it difficult to proactively compensate for the quality deviations of preceding processes, which in turn leads to large fluctuations in the performance of the final product and a low pass rate.

[0004] Therefore, there is an urgent need for an optimized control method for the production process of aluminum alloy round rods that can achieve collaborative perception of quality fluctuations across multiple stages. This method would utilize digital and intelligent technologies throughout the entire aluminum alloy production process, and through real-time data acquisition and correction, it would control the production process of aluminum alloy round rods. This would overcome the limitations of existing control models, improve product performance stability and yield, maintain the natural lightweight advantage of aluminum alloys while significantly increasing strength and toughness, ensure high-quality requirements for aluminum alloys, and reduce production energy consumption. Summary of the Invention

[0005] This invention addresses the technical problems of large fluctuations in product performance and low pass rates in the production of aluminum alloy round aluminum rods in the prior art, and provides a method for optimizing and controlling the production process of aluminum alloy round aluminum rods.

[0006] The technical solution of the present invention to solve the above-mentioned technical problems is as follows:

[0007] This invention provides a method for optimizing and controlling the production process of aluminum alloy round rods, comprising:

[0008] The initial process parameters for each production stage are obtained, and a weight matrix representing the quality fluctuation transmission relationship between each production stage is constructed based on historical production data.

[0009] Real-time acquisition of dynamic output characteristics and environmental interference parameters of each production stage;

[0010] A smelting-casting correction model is constructed, taking the smelting process parameters in the initial process parameters, the dynamic output characteristics of the smelting process, and the smelting-casting correlation part in the weight matrix as inputs, to predict the output casting temperature correction parameters and casting speed correction parameters.

[0011] A casting-rolling correction model is constructed, using the casting process parameters in the initial process parameters, the dynamic output characteristics of the casting process, and the casting-rolling correlation part in the weight matrix as inputs, to predict the output rolling temperature correction parameters and rolling speed correction parameters.

[0012] A rolling-heat treatment correction model is constructed, taking the rolling process parameters in the initial process parameters, the dynamic output characteristics of the rolling process, and the rolling-heat treatment correlation part in the weight matrix as inputs, to predict the output heat treatment temperature correction parameters and holding time correction parameters.

[0013] Based on the correction parameters output by each model and combined with the preset fluctuation warning threshold, progressive control parameter correction is implemented for the casting, rolling and heat treatment processes, and the weight matrix is ​​dynamically updated according to the correction effect.

[0014] Furthermore, a weight matrix representing the transmission relationship of quality fluctuations between various production stages is constructed based on historical production data, including:

[0015] Extract quality fluctuation characteristic parameters from each production stage in historical production data to construct multiple quality fluctuation feature vectors;

[0016] Calculate the mutual information value between the characteristic vectors of quality fluctuations in adjacent production stages, as the correlation strength of fluctuation transmission between stages;

[0017] Identify the causal pathways through which quality fluctuations are transmitted between different stages;

[0018] Based on the aforementioned correlation strength and causal relationship path, a weight matrix is ​​constructed with production links as nodes and the aforementioned correlation strength as edge weights.

[0019] Furthermore, the dynamic output characteristics and environmental interference parameters of each production stage are collected in real time, including: the dynamic output characteristics of the smelting stage include at least melt temperature uniformity, compositional segregation index, and gas content; the dynamic output characteristics of the casting stage include at least billet surface temperature distribution, solidification front velocity, and shrinkage defect index; the dynamic output characteristics of the rolling stage include at least rolled piece grain size distribution, deformation resistance, and recrystallization ratio; the dynamic output characteristics of the heat treatment stage include at least solid solution saturation, aging kinetic parameters, and precipitated phase morphology; and the environmental interference parameters include at least raw material composition fluctuations, equipment operating condition drift, cooling medium temperature fluctuations, and changes in ambient temperature and humidity.

[0020] Furthermore, a modified smelting-casting model is constructed, including:

[0021] An attention-based long short-term memory network is used as the model architecture;

[0022] The smelting-casting correction model is trained using supervised learning.

[0023] Using time-series data on melt temperature uniformity, composition segregation index, and gas content in the smelting process as input features, the outputs are the updated casting temperature and updated casting speed in the casting process.

[0024] By combining the initial process parameters of the casting process, the casting temperature correction parameters and casting speed correction parameters are calculated.

[0025] Furthermore, a casting-rolling correction model is constructed, including:

[0026] A casting-rolling correction model is constructed using a multi-task learning neural network architecture;

[0027] The casting-rolling correction model is obtained by training through supervised learning.

[0028] Retrieve the pre-trained casting-rolling correction model;

[0029] The main input features are the surface temperature distribution of the billet, the solidification front velocity, and the shrinkage defect index in the casting process. The auxiliary input features are the equipment condition drift and the cooling medium temperature fluctuation in the environmental interference parameters. The output is updated rolling temperature and updated rolling speed.

[0030] Based on the initial process parameters of the rolling process, rolling temperature correction parameters and rolling speed correction parameters are calculated.

[0031] Furthermore, a rolling-heat treatment correction model is constructed, including:

[0032] A rolling-heat treatment correction model was constructed based on neural networks;

[0033] A rolling-heat treatment correction model was obtained by training through supervised learning.

[0034] Retrieve the pre-trained rolling-heat treatment correction model;

[0035] Using the grain size distribution, deformation resistance, and recrystallization ratio of the rolled piece as input features, the output is updated heat treatment temperature and updated holding time.

[0036] Based on the initial process parameters of the heat treatment process, the heat treatment temperature correction parameters and the holding time correction parameters are calculated.

[0037] Furthermore, based on the correction parameters output by each model, progressive control parameter correction is implemented for the casting, rolling, and heat treatment processes, including:

[0038] Set multi-level fluctuation early warning thresholds for each process step, including early warning thresholds, adjustment thresholds, and intervention thresholds;

[0039] When the correction parameters output by the model exceed the warning threshold but do not reach the adjustment threshold, parameter monitoring and trend warning are performed.

[0040] When the correction parameters output by the model exceed the adjustment threshold but do not reach the intervention threshold, the initial process parameters are corrected according to the model output results.

[0041] When the correction parameters output by the model exceed the intervention threshold, a manual intervention mechanism is triggered.

[0042] Furthermore, the weight matrix is ​​dynamically updated based on the correction effect, including:

[0043] The impact of each parameter correction operation on the final product quality is quantified, wherein the impact is a vector, with positive values ​​indicating positive improvement in product quality and negative values ​​indicating negative deterioration in product quality;

[0044] Extract the correlation strength related to the parameter correction operation from the weight matrix, and obtain the historical average change of the correlation strength;

[0045] The ratio of the degree of impact to the corresponding historical average change is calculated and used as the initial value efficiency index for this parameter correction operation.

[0046] Based on the average prediction accuracy of the smelting-casting correction model, casting-rolling correction model, and rolling-heat treatment correction model within a preset time range, the initial value efficiency index is weighted and compensated to obtain the value efficiency index.

[0047] The value efficiency index is compared with a preset update threshold. If the value efficiency index is greater than or equal to the preset update threshold, the weight matrix is ​​dynamically updated.

[0048] Conversely, if the value efficiency index is less than the preset update threshold, the current weight matrix remains unchanged.

[0049] Furthermore, quantify the impact of each parameter correction on the final product quality, including:

[0050] Product quality indicators are selected as the evaluation criteria, wherein the product quality indicators include at least tensile strength, elongation and conductivity;

[0051] The relative change rate of the product quality indicators before and after parameter correction is calculated, and the degree of influence is obtained by weighted calculation.

[0052] Furthermore, the weight matrix is ​​dynamically updated, including:

[0053] The recursive least squares method is used to update the correlation strength in the weight matrix in real time based on recent parameter correction data and the corresponding quality impact.

[0054] Record the timestamp, update content, and update reason for each weight matrix update to form a complete update history log.

[0055] The beneficial effects of this invention are:

[0056] Compared to existing technologies, this application first obtains the initial process parameters of each production stage and constructs a weight matrix based on historical production data to characterize the transmission relationship of quality fluctuations between each production stage. This transforms the coupling relationship of quality fluctuations in each stage into a quantified mathematical matrix, reflecting both the direction of quality fluctuation transmission and the degree of influence, providing reliable data support for the subsequent construction of a correction model. Secondly, it collects the dynamic output characteristics and environmental interference parameters of each production stage in real time, obtaining dynamic output characteristics reflecting the production quality status of this stage and environmental interference parameters reflecting external influencing factors, providing a reliable data foundation for the subsequent construction of a correction model and dynamic optimization control. Thirdly, it constructs a smelting-casting correction model, using the smelting process parameters in the initial process parameters, the dynamic output characteristics of the smelting stage, and the smelting-casting correlation part in the weight matrix as inputs. This model predicts and outputs casting temperature correction parameters and casting speed correction parameters, accurately converting the real-time quality fluctuations in the smelting stage into process parameter corrections for the casting stage, thereby suppressing the transmission of preceding quality fluctuations to subsequent stages. Furthermore, a casting-rolling correction model is constructed. Using the casting process parameters in the initial process parameters, the dynamic output characteristics of the casting stage, and the casting-rolling correlation component in the weight matrix as inputs, it predicts and outputs rolling temperature correction parameters, rolling speed correction parameters, and deformation degree correction parameters. This transforms the quality state of the casting stage and external interference factors into precise process parameter corrections for the rolling stage, preventing casting quality fluctuations from being transmitted or amplified to the rolling stage. Further, a rolling-heat treatment correction model is constructed. Using the rolling process parameters in the initial process parameters, the dynamic output characteristics of the rolling stage, and the rolling-heat treatment correlation component in the weight matrix as inputs, it predicts and outputs heat treatment temperature correction parameters, holding time correction parameters, and cooling rate correction parameters. This transforms the rolled piece state in the rolling stage into precise process parameter corrections for the heat treatment stage, offsetting the impact of rolling quality fluctuations by controlling the heat treatment process. Finally, based on the correction parameters output by each model and combined with the preset fluctuation warning threshold, progressive control parameter correction is implemented for the casting, rolling, and heat treatment processes. The weight matrix is ​​dynamically updated according to the correction effect. Different risk levels are matched with processing strategies through threshold division, and the reliability of the weight matrix update is ensured through multi-dimensional verification. By processing industrial data and applying intelligent technology throughout the entire aluminum alloy production process, long-term stability and adaptability of the entire process control for aluminum alloy round rod production are ultimately achieved, ensuring the high precision, high performance, and high quality of aluminum alloy products.

[0057] Through the aforementioned technical solution, this application effectively breaks through the limitations of traditional independent control of each stage by constructing three cross-stage correction models. This achieves proactive compensation for quality fluctuations in preceding processes on subsequent processes, weakening the cross-stage coupling and transmission effect of quality fluctuations. Combined with a progressive parameter correction mechanism using multi-level fluctuation early warning thresholds, it replaces the traditional post-event remedial mode, achieving differentiated and precise control of process parameters throughout the entire process. Furthermore, the weight matrix is ​​dynamically updated based on the correction effect, ensuring that the model always fits the production dynamics and avoiding control precision decay. Thus, through digital, automated, and intelligent manufacturing technologies, the product performance stability and yield rate of aluminum alloy round aluminum rods are improved. Simultaneously, by reducing repeated process adjustments and parameter redundancy, production energy consumption and production costs are reduced. Attached Figure Description

[0058] Figure 1 A flowchart illustrating an optimized control method for the production process of aluminum alloy round rods provided by this invention;

[0059] Figure 2 This is a flowchart illustrating the progressive control parameter correction for the casting, rolling, and heat treatment stages based on the correction parameters output by each model in the aluminum alloy round rod production process optimization control method provided by this invention. Detailed Implementation

[0060] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0061] In the description of this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0062] In the description of this invention, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this invention is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed herein.

[0063] Examples, such as Figure 1 As shown, this embodiment of the invention provides a method for optimizing and controlling the production process of aluminum alloy round aluminum rods, including:

[0064] S10: Obtain the initial process parameters for each production stage, and construct a weight matrix based on historical production data to characterize the quality fluctuation transmission relationship between each production stage.

[0065] The production process of aluminum alloy round aluminum rods includes smelting, casting, rolling, and heat treatment. The output quality of the preceding process is the input basis for the subsequent process, directly determining the initial state of the subsequent process. Therefore, there is a strong coupling relationship between the quality fluctuations of each process: the quality of the melt after smelting directly serves as the raw material basis for casting, and the quality fluctuation of smelting will directly lead to abnormal solidification of the billet; the billet produced by casting is the processing object of rolling, and the quality fluctuation of casting will directly affect the deformation uniformity of the rolling process; the state of the rolled piece after rolling directly determines the phase transformation kinetics process during heat treatment, and the quality fluctuation of rolling will directly lead to problems such as uneven precipitation of strengthening phases and deviation of solid solution effect.

[0066] To address the aforementioned issues, this application first obtains the initial process parameters for each production stage. For example, in the casting stage, the casting temperature is collected using a heat-resistant K-type thermocouple sensor. The casting speed can be monitored in real time by installing a 1000 lines / revolution incremental encoder on the end of the casting machine's traction roller shaft. Combined with the preset traction roller circumference, the speed is converted to m / min using the formula: casting speed = rotational speed × circumference / 60. In the rolling stage, the rolling temperature is measured using a non-contact infrared thermometer. The rolling speed is measured by installing an absolute encoder with an accuracy of ±0.1% on the end of the rolling mill's work roll drive shaft, and the speed value is calculated using the formula: rolling speed = rotational speed × π × work roll diameter / 60. In the heat treatment stage, the heat treatment temperature is measured using a heat-resistant S-type thermocouple sensor, with 2-3 sensors evenly distributed in each of the furnace heating zone, holding zone, and cooling zone. The holding time is timed by linking the conveyor belt motor encoder with the PLC system.

[0067] Specifically, step S10 of the method further includes:

[0068] Extract quality fluctuation characteristic parameters from each production stage in historical production data to construct multiple quality fluctuation feature vectors;

[0069] Calculate the mutual information value between the characteristic vectors of quality fluctuations in adjacent production stages, as the correlation strength of fluctuation transmission between stages;

[0070] Identify the causal pathways through which quality fluctuations are transmitted between different stages;

[0071] Based on the aforementioned correlation strength and causal relationship path, a weight matrix is ​​constructed with production links as nodes and the aforementioned correlation strength as edge weights.

[0072] In this embodiment, quality fluctuation characteristic parameters for each production stage are first extracted from historical production data to form multiple quality fluctuation characteristic vectors. These quality fluctuation characteristic parameters are key quantitative indicators reflecting the quality stability of each production stage. Examples include temperature fluctuation amplitude, compositional segregation, and gas content fluctuation in the smelting stage; surface temperature gradient fluctuation, solidification rate deviation, and shrinkage defect quantity fluctuation in the casting stage; thickness deviation fluctuation, grain size distribution variation coefficient, and deformation resistance fluctuation in the rolling stage; and holding temperature fluctuation, cooling rate deviation, and precipitate size distribution fluctuation in the heat treatment stage. For example, quality fluctuation characteristic parameters for the smelting, casting, rolling, and heat treatment stages are extracted from historical production data. Multiple quality fluctuation characteristic parameters for the same process stage are arranged in a preset order to form a quality fluctuation characteristic vector for that process stage. For instance, the characteristic vector for the smelting stage can be represented as [temperature fluctuation amplitude, compositional segregation, gas content fluctuation value], and the dimension of the quality fluctuation characteristic vector is equal to the total number of quality fluctuation characteristic parameters selected for that process stage.

[0073] Secondly, the mutual information value between the characteristic vectors of quality fluctuations in adjacent production stages is calculated as the correlation strength of fluctuation transmission between stages. The mutual information value, derived from information theory, measures the degree of correlation between two random variables. For example, for adjacent production stages such as smelting and casting, or casting and rolling, the characteristic vectors of the two stages in historical data are used as input. The mutual information value is obtained through a mutual information calculation formula, such as integration based on the probability density function. The mutual information value is a non-negative value; the larger the mutual information value, the stronger the correlation between the quality fluctuations of the preceding and subsequent stages in adjacent production stages. For example, the mutual information value is used as the correlation strength of fluctuation transmission between stages. For instance, if the mutual information value of the characteristic vector of quality fluctuations in the smelting and casting stage is 0.8, and the mutual information value of the characteristic vector of quality fluctuations in the casting and rolling stage is 0.6, then 0.8 is taken as the correlation strength between the smelting and casting stages, and 0.6 is taken as the correlation strength between the casting and rolling stages.

[0074] Secondly, the causal paths of quality fluctuations between each stage are identified. Specifically, the output quality of the preceding stage is the input basis for the subsequent stage, directly determining the initial state of the subsequent stage. Therefore, the causal path mainly exists in adjacent stages. Based on the aluminum alloy round aluminum rod production process, the causal path of quality fluctuations is determined as follows: smelting → casting, casting → rolling, rolling → heat treatment. In this way, the direct direction of fluctuation transmission is clarified.

[0075] Finally, based on the correlation strength and causal path, a weight matrix is ​​constructed with production stages as nodes and correlation strength as edge weights. For example, production process stages (smelting, casting, rolling, heat treatment) are used as the rows and columns of the weight matrix W. The element W[i][j] of the weight matrix represents the correlation strength between stage i and stage j. If a causal path exists from stage i to stage j, then W[i][j] equals the correlation strength; if no causal path exists, then W[i][j] equals 0. Since there is no causal path for the same stage, the diagonal element W[i][i] is also 0. For example, smelting, casting, rolling, and heat treatment are used as the rows and columns of the weight matrix respectively, forming a 4×4 weight matrix. If the correlation strength between smelting and casting is 0.8, and the correlation strength between casting and rolling is 0.6, then in the weight matrix, W[smelting][casting] = 0.8, and W[casting][rolling] = 0.6.

[0076] In summary, compared to existing technologies, this application obtains the initial process parameters for each production stage and constructs a weight matrix representing the transmission relationship of quality fluctuations between each production stage based on historical production data. In this way, the coupling relationship of quality fluctuations in each stage is transformed into a quantified mathematical matrix, which not only reflects the direction of quality fluctuation transmission but also quantifies the degree of influence, providing reliable data support for the subsequent construction of a correction model.

[0077] S20: Real-time acquisition of dynamic output characteristics and environmental interference parameters of each production stage.

[0078] In the continuous production process of aluminum alloy round aluminum rods, the output quality of the preceding production stage is the input basis for the subsequent stage. Moreover, there are uncontrollable or difficult-to-control external interference factors in the production process, such as fluctuations in the temperature and humidity of the workshop environment. These factors can indirectly affect the quality by interfering with the stability of the process parameters of each stage.

[0079] To address the aforementioned issues, this application collects dynamic output characteristics and environmental interference parameters of each production stage in real time.

[0080] Specifically, step S20 in the method includes:

[0081] The dynamic output characteristics of the smelting process include at least melt temperature uniformity, compositional segregation index, and gas content; the dynamic output characteristics of the casting process include at least billet surface temperature distribution, solidification front velocity, and shrinkage defect index; the dynamic output characteristics of the rolling process include at least rolled grain size distribution, deformation resistance, and recrystallization ratio; the dynamic output characteristics of the heat treatment process include at least solution saturation, aging kinetic parameters, and precipitated phase morphology; the environmental interference parameters include at least raw material composition fluctuations, equipment operating condition drift, cooling medium temperature fluctuations, and environmental temperature and humidity changes.

[0082] For example, the dynamic output characteristics of the smelting process include at least melt temperature uniformity, composition segregation index, and gas content. Melt temperature uniformity refers to the temperature difference between different areas of the aluminum melt in the smelting furnace. Poor melt temperature uniformity can lead to inconsistent solidification rates during subsequent casting, causing composition segregation or internal stress in the billet. Composition segregation index refers to the uniformity of the distribution of alloying elements (such as silicon, magnesium, copper, etc.) in the aluminum melt. Abnormal composition segregation can lead to uneven mechanical properties of the billet and make it prone to local cracking during rolling. Gas content refers to the content of dissolved gases such as hydrogen in the melt, usually expressed as the gas volume per 100 grams of aluminum melt. Excessive gas content can cause the billet to form pores, resulting in surface pinholes or internal porosity after rolling, which seriously affects the density of the product.

[0083] Among these methods, melt temperature uniformity can be assessed by arranging multiple thermocouples (accuracy ±1℃) in different areas of the melting furnace (such as the furnace wall, center, and bottom) to collect the temperature at each point in real time, and calculating the temperature standard deviation as the melt temperature uniformity. The composition segregation index can be assessed using a direct-reading spectrometer (such as a spark discharge spectrometer), by inserting a probe into the melt to sample and quickly analyze the content distribution of alloying elements such as Si, Mg, and Cu, and calculating the ratio of the element standard deviation to the average value as the composition segregation index. The gas content can be assessed using a melt hydrogen analyzer (such as a vacuum decompression hydrogen analyzer), by extracting a small amount of melt into a sealed cavity and calculating the gas content based on the amount of gas released under vacuum.

[0084] For example, the dynamic output characteristics of the casting process include at least the surface temperature distribution of the billet, the solidification front velocity, and the shrinkage defect index. The surface temperature distribution of the billet refers to the temperature gradient at different locations on the surface of the freshly formed billet. Uneven surface temperature distribution can lead to inconsistent cooling and shrinkage of the billet, resulting in surface cracks or warping, increasing the risk of breakage during subsequent rolling. The solidification front velocity refers to the speed at which the aluminum melt transitions from a liquid to a solid state in the crystallizer. An excessively fast solidification front velocity can lead to insufficient solidification at the center of the billet, forming shrinkage cavities. An excessively slow solidification front velocity can reduce production efficiency and easily lead to surface oxidation. The shrinkage defect index refers to the severity of defects (such as shrinkage cavities and porosity) caused by solidification shrinkage inside the billet. It can be represented by the proportion of defect area obtained from ultrasonic testing. Shrinkage defects weaken the material strength and may expand into fracture during rolling.

[0085] Among these methods, the surface temperature distribution of the billet can be measured by installing an infrared thermal imager at the outlet of the casting machine to perform real-time thermal imaging of the newly formed billet surface, and extracting the surface temperature field distribution through image processing; the solidification front velocity can be measured by installing infrared temperature sensors at different heights in the crystallizer to monitor the movement speed of the phase transition temperature point of the melt from liquid to solid, and calculating the solidification front advancement speed in combination with the billet casting speed; the shrinkage defect index can be measured by installing an ultrasonic flaw detector in the later stage of casting to perform a full-length scan of the billet, and quantifying the area ratio of shrinkage cavities and porosity through defect echo signals as the shrinkage defect index.

[0086] For example, the dynamic output characteristics of the rolling process include at least the grain size distribution, deformation resistance, and recrystallization ratio of the rolled piece. The grain size distribution refers to the grain size and distribution state of the cross-section of the aluminum rod after rolling, usually expressed as the average grain size grade or dimensional standard deviation. Excessively coarse grains will reduce the material strength, and uneven distribution will lead to fluctuations in mechanical properties. Deformation resistance refers to the reaction force of the rolled piece on the rolls during the rolling process. A sudden increase in deformation resistance may cause roll wear or aluminum rod breakage, while too low deformation resistance may lead to abnormal grain growth due to excessively high rolling temperature. The recrystallization ratio refers to the proportion of broken grains generated by plastic deformation during the rolling process that regenerate into equiaxed grains. Insufficient recrystallization will lead to poor plasticity and easy brittle fracture of the aluminum rod, while excessive recrystallization may reduce strength.

[0087] Among them, the grain size distribution of the rolled piece can be determined by using an online metallographic imaging system to perform real-time microscopic imaging of the surface of the rolled aluminum rod, calculate the grain size distribution through image recognition, and output the average grain size grade as the grain size distribution of the rolled piece; the deformation resistance can be determined by installing pressure sensors (such as piezoelectric sensors) on the rolling mill roll system to collect the rolling force F in real time during the rolling process, and combine it with the contact area S of the rolled piece to obtain the deformation resistance by F / S; the recrystallization ratio can be determined by using an online electron backscatter diffraction (EBSD) detection module to perform crystal orientation analysis on the cross section of the rolled piece and statistically analyze the proportion of recrystallized grains to the total grains.

[0088] For example, the dynamic output characteristics of the heat treatment process include at least solution saturation, aging kinetic parameters, and precipitate morphology. Solution saturation refers to the degree to which alloying elements (such as Mg and Si) dissolve into the aluminum matrix in the aluminum alloy, usually expressed as the solubility rate. Insufficient solution saturation will lead to poor subsequent aging strengthening effect and insufficient material strength. Insufficient solution saturation may also cause grain boundary corrosion. Aging kinetic parameters are parameters that describe the precipitation rate of strengthening phases (such as Mg2Si) during the aging process, such as the rate of change of the volume fraction of precipitates over time. Abnormal aging kinetic parameters will lead to uneven size of strengthening phases, affecting the balance between the strength and toughness of the material. Precipitate morphology refers to the shape, size, and distribution of the strengthening phase after aging, which can be quantified by electron microscopy. Abnormal precipitate morphology will lead to increased brittleness of the material, making it prone to cracking during subsequent processing.

[0089] Among them, solid solution saturation can be determined by analyzing the surface composition of the heat-treated aluminum rod using X-ray fluorescence spectrometry (XRF), and the solubility rate of alloying elements can be calculated by the ratio of dissolved amount to total content, which is used as solid solution saturation. The aging kinetic parameters can be obtained by installing a differential scanning calorimeter (DSC) probe in the aging furnace to monitor the heat release rate of the aluminum rod in real time during the aging process and fitting the parameters. The morphology of precipitated phases can be determined by sampling the aluminum rod with an online scanning electron microscope (SEM) to observe the shape, size, and distribution density of the precipitated phases, and the morphology of the precipitated phases can be determined by image analysis.

[0090] For example, environmental interference parameters are uncontrollable or difficult-to-control external factors in the production process. Their fluctuations indirectly affect the dynamic output characteristics of each stage. Environmental interference parameters include at least raw material composition fluctuations, equipment operating condition drift, cooling medium temperature fluctuations, and environmental temperature and humidity changes. Among them, raw material composition fluctuations refer to the deviation of alloy element content in incoming aluminum ingots or scrap; equipment operating condition drift refers to the performance degradation of production equipment after long-term operation, such as decreased temperature control accuracy of smelting furnaces and dimensional deviations caused by wear of rolling mill roll gaps. Equipment drift can cause different output qualities for the same process parameters, and can be fed back by sensors with equipment operating data; cooling medium temperature fluctuations refer to the temperature changes of casting cooling water, rolling emulsion, and heat treatment quenching media. The temperature of the cooling medium directly affects the solidification rate, rolling lubrication effect, or heat treatment cooling rate; environmental temperature and humidity changes refer to the temperature and humidity fluctuations of the workshop environment. High temperature and high humidity will accelerate the oxidation of the billet surface and affect the stability of the rolling emulsion.

[0091] Among these measures, raw material composition fluctuations can be detected by rapid composition analysis of aluminum ingots / scrap using a laser-induced breakdown spectroscopy (LIBS) instrument upon raw material arrival at the plant. The deviation between the measured values ​​and standard values ​​of key elements such as Si and Fe can be recorded as raw material composition fluctuations. Equipment operating condition drift can be detected by collecting key parameters, such as the heating tube current fluctuations of the smelting furnace and the motor vibration frequency of the rolling mill, through built-in sensors in the equipment control system. The percentage deviation from the rated state can be output in real time as equipment operating condition drift. Cooling medium temperature fluctuations can be detected by installing platinum resistance temperature sensors on the cooling water / emulsion pipelines to monitor the medium temperature in real time and calculate the deviation from the set value as cooling medium temperature fluctuations. Ambient temperature and humidity can be detected by deploying temperature and humidity transmitters in different areas of the workshop to record ambient temperature and relative humidity in real time.

[0092] In summary, compared to existing technologies, this application collects dynamic output characteristics and environmental disturbance parameters of each production stage in real time. This provides dynamic output characteristics reflecting the production quality status of each stage and environmental disturbance parameters reflecting external influencing factors, offering a reliable data foundation for subsequent model building and dynamic optimization control.

[0093] S30: Construct a smelting-casting correction model, using the smelting process parameters in the initial process parameters, the dynamic output characteristics of the smelting process, and the smelting-casting correlation part in the weight matrix as inputs, to predict and output casting temperature correction parameters and casting speed correction parameters.

[0094] The quality of the melt produced in the smelting stage is the direct raw material basis for the casting stage. Therefore, fluctuations in the smelting quality directly affect the casting solidification process. For example, uneven melt temperature can lead to differences in solidification rates in different areas of the cast billet, causing internal stress and surface cracks. Compositional segregation can cause local mechanical property imbalances in the cast billet, and excessive gas content can easily lead to internal porosity. Furthermore, if the casting stage is controlled solely based on preset initial process parameters and cannot respond in real time to dynamic quality fluctuations in the smelting stage, quality fluctuations at the smelting end will be directly transmitted to the cast billet, and may even amplify quality problems due to parameter mismatch.

[0095] To address the aforementioned issues, this application constructs a smelting-casting correction model. Using the smelting process parameters in the initial process parameters, the dynamic output characteristics of the smelting process, and the smelting-casting correlation part in the weight matrix as inputs, it predicts and outputs casting temperature correction parameters and casting speed correction parameters.

[0096] Specifically, step S30 in the method includes:

[0097] An attention-based long short-term memory network is used as the model architecture;

[0098] The smelting-casting correction model is trained using supervised learning.

[0099] Using time-series data on melt temperature uniformity, composition segregation index, and gas content in the smelting process as input features, the outputs are the updated casting temperature and updated casting speed in the casting process.

[0100] By combining the initial process parameters of the casting process, the casting temperature correction parameters and casting speed correction parameters are calculated.

[0101] In this embodiment, a Long Short-Term Memory (LSTM) network based on an attention mechanism is first adopted as the model architecture. The attention mechanism assigns different weights to the dynamic output features of the smelting process. The LSTM network, through a gating mechanism, can effectively memorize key fluctuation information over long time periods, such as multiple abnormal fluctuations in melt temperature within one hour, avoiding correction biases caused by data temporal loss. Exemplarily, the smelting-casting correction model mainly consists of an input layer, an LSTM feature extraction layer, an attention mechanism layer, and an output layer. The input layer receives temporal data on melt temperature uniformity, compositional segregation index, and gas content, forming a raw tensor of dimension (30,3), and eliminates dimensional differences through Z-score normalization. The LSTM feature extraction layer adopts a two-layer stacked design: the first layer has 64 hidden units, focusing on capturing local temporal correlations; the second layer has 32 hidden units, focusing on fusing global fluctuation patterns, ultimately outputting a 32-dimensional feature vector to achieve a temporal dependence on smelting quality fluctuations. The system accurately captures the signal; the attention mechanism layer adopts a 16-head additive attention architecture, which learns the importance weights of features through a feedforward neural network with 16 neurons, and automatically adjusts the weight allocation for melt temperature uniformity, composition segregation index and gas content. For example, when the gas content time series data shows an outlier value >0.2mL / 100g, the corresponding feature weight is increased from the usual 0.2 to 0.5, giving priority to responding to the parameter fluctuations that have the most significant impact on the porosity of the billet; the output layer adopts a 24-neuron, ReLU activated fully connected layer and 2-neuron linearly activated output layer structure to generate casting temperature correction parameters and casting speed correction parameters.

[0102] Secondly, the smelting-casting correction model is trained through supervised learning. For example, the smelting-casting correction model can be trained through the following technical path: 1. Data preparation: Collect time-series data on the historical melt temperature uniformity, composition segregation index, and gas content of multiple sets of smelting processes from historical production data as input sample sets. Simultaneously acquire the historical optimal casting temperature and historical optimal casting speed corresponding to each set of input samples, which have been verified by actual production effects (such as billet quality meeting standards and being suitable for subsequent processes), as label sample sets. Then, divide the input sample set and label sample set into training set, validation set, and test set according to a ratio of 7:1.5:1.5. 2. Model Training: The historical melt temperature uniformity, compositional segregation index, and gas content time-series data in the training set are used as input features. The corresponding historical best casting temperature and historical best casting speed are used as supervision labels. The mean squared error loss function is adopted, and the network parameters, such as the gating weights of LSTM and the feature weights of the attention mechanism, are adjusted iteratively through backpropagation of the Adam optimizer to continuously minimize the error between the model's predicted casting temperature and casting speed and the label values. When the value of the validation set loss function is stable below the threshold for a preset number of rounds (e.g., 10 rounds) without significant decrease, the model is considered to have converged, and the trained smelting-casting correction model is obtained.

[0103] Furthermore, using time-series data on melt temperature uniformity, compositional segregation index, and gas content from the smelting process as input features, the outputs are updated casting temperature and updated casting speed for the casting process. The time-series data reflects the dynamic trends of the smelting process; for example, melt temperature uniformity, compositional segregation index, and gas content are collected at a certain sampling frequency and arranged into corresponding time-series data according to the sampling timestamps. The updated casting temperature and updated casting speed are the optimal control parameters for the casting process based on predictions of the current smelting quality.

[0104] Finally, based on the initial process parameters of the casting stage, the casting temperature correction parameters and casting speed correction parameters are calculated. The initial process parameters of the casting stage include the initial casting temperature and the initial casting speed. The casting temperature correction parameter = updated casting temperature - initial casting temperature, and the casting speed correction parameter = updated casting speed - initial casting speed. For example, if the initial casting temperature is 720℃ and the initial casting speed is 0.8 m / min, and the updated casting temperature output by the melting-casting correction model is 717℃ and the updated casting speed is 0.78 m / min, then the casting temperature correction parameter = 717 - 720 = -3℃, indicating that the initial casting temperature needs to be reduced by 3℃, and the casting speed correction parameter = 0.78 - 0.8 = -0.02 m / min, indicating that the initial casting speed needs to be reduced by 0.02 m / min. In this way, the initial process parameters of the casting stage are dynamically adjusted based on the product quality fluctuations output from the preceding stages, resulting in process parameters that are more suitable for the actual situation, avoiding the transmission and amplification of quality fluctuations.

[0105] In summary, compared to existing technologies, this application constructs a smelting-casting correction model. Using the smelting process parameters in the initial process parameters, the dynamic output characteristics of the smelting stage, and the smelting-casting correlation component in the weight matrix as input, it predicts and outputs casting temperature correction parameters and casting speed correction parameters. In this way, real-time quality fluctuations in the smelting stage are accurately converted into process parameter corrections for the casting stage, thereby suppressing the propagation of preceding quality fluctuations to subsequent stages.

[0106] S40: Construct a casting-rolling correction model, using the casting process parameters in the initial process parameters, the dynamic output characteristics of the casting process, and the casting-rolling correlation part in the weight matrix as inputs, to predict and output rolling temperature correction parameters and rolling speed correction parameters.

[0107] The quality of the billet produced in the casting stage is the direct raw material basis for the rolling stage. Its quality fluctuations will directly affect the entire rolling process. For example, uneven surface temperature of the billet will lead to differences in local deformation resistance during rolling, affecting the uniformity of deformation of the rolled piece. Abnormal solidification front speed will cause the grain structure of the billet to be unbalanced, which will interfere with the grain size control of the rolling process. Excessive shrinkage defect index will easily cause defects to expand into cracks in the rolled piece during rolling, ultimately directly affecting the dimensional accuracy and mechanical properties of the rolled piece.

[0108] To address the aforementioned issues, this application constructs a casting-rolling correction model, using the casting process parameters in the initial process parameters, the dynamic output characteristics of the casting process, and the casting-rolling correlation part in the weight matrix as inputs, to predict and output rolling temperature correction parameters and rolling speed correction parameters.

[0109] Specifically, step S40 in the method includes:

[0110] A casting-rolling correction model is constructed using a multi-task learning neural network architecture;

[0111] The casting-rolling correction model is obtained by training through supervised learning.

[0112] Retrieve the pre-trained casting-rolling correction model;

[0113] The main input features are the surface temperature distribution of the billet, the solidification front velocity, and the shrinkage defect index in the casting process. The auxiliary input features are the equipment condition drift and the cooling medium temperature fluctuation in the environmental interference parameters. The output is updated rolling temperature and updated rolling speed.

[0114] Based on the initial process parameters of the rolling process, rolling temperature correction parameters and rolling speed correction parameters are calculated.

[0115] In this embodiment, a casting-rolling correction model is first constructed using a multi-task learning neural network architecture. The multi-task learning neural network is chosen because rolling temperature and rolling speed are core parameters that are coupled together. For example, increasing the rolling temperature reduces the deformation resistance of the aluminum rod, requiring simultaneous adjustment of the rolling speed to ensure uniform deformation. Predicting only one parameter can easily lead to an imbalance between the two; for instance, if the temperature increases but the speed is not adjusted, it may cause the rolled piece to overheat. By using a multi-task learning neural network to learn multiple related tasks simultaneously, the overall prediction accuracy is improved through information sharing between tasks. For example, the casting-rolling correction model mainly consists of an input preprocessing layer, a shared feature extraction layer, a dual-task-specific feature layer, and an output layer. The input preprocessing layer maps all input features to the [0,1] interval through Min-Max normalization to eliminate dimensional differences. The shared feature extraction layer uses a two-layer fully connected network with 32 neurons in the first layer and 16 neurons in the second layer, both equipped with ReLU activation functions. The dual-task-specific feature layer optimizes the different process requirements of rolling temperature and rolling speed: the rolling temperature-specific layer has a fully connected network with 12 ReLU activation neurons, superimposed with a temperature constraint layer, and the rolling speed-specific layer also has a fully connected network with 12 ReLU activation neurons. The output layer consists of two independent linear activation neuron branches, which output updated rolling temperature and updated rolling speed, respectively.

[0116] Secondly, a casting-rolling correction model is trained through supervised learning. For example, the casting-rolling correction model can be trained using the following technical path: 1. Data preparation: Measured data on the surface temperature distribution of the cast billet, the solidification front velocity, and the shrinkage defect index in historical production data are collected, along with recorded data on equipment condition drift (such as mill roll gap wear) and cooling medium temperature fluctuations (such as the temperature difference of the rolling emulsion) during the same period. These are used as the input sample set. Simultaneously, the historical optimal rolling temperature and historical optimal rolling speed, which have been verified by actual production effects (e.g., after rolling with these parameters, the rolled piece has qualified grain size, uniform deformation, no cracks, and is suitable for subsequent heat treatment), are obtained as the label sample set. The input sample set and the corresponding label sample set are divided into training set, validation set, and test set in a ratio of 7:1.5:1.5. 2. Model Training: The historical surface temperature distribution of billets, solidification front velocity, shrinkage defect index, equipment condition drift, and cooling medium temperature fluctuation in the training set are used as input features. The corresponding historical optimal rolling temperature and historical optimal rolling speed are used as supervision labels. The network parameters, such as the weights of shared layers and the biases of dedicated layers, are optimized through the backpropagation algorithm. A joint loss function, such as temperature prediction MSE and speed prediction MSE, is used to measure the prediction error. The model minimizes the prediction bias of the two parameters simultaneously. When the joint loss function value of the validation set is stable below the process allowable threshold for a preset number of rounds (e.g., 10 rounds) without significant decrease, the model is considered to have converged, and the trained casting-rolling correction model is obtained.

[0117] Next, retrieve the pre-trained casting-rolling correction model.

[0118] Furthermore, using the billet surface temperature distribution, solidification front velocity, and shrinkage defect index from the casting process as primary input features, and equipment condition drift and cooling medium temperature fluctuation from environmental interference parameters as auxiliary input features, the outputs are updated rolling temperature and updated rolling speed. The billet surface temperature distribution, solidification front velocity, and shrinkage defect index reflect the output quality of the casting process, and their values ​​directly determine the adjustment range and direction of parameters in the rolling process. Equipment condition drift, a parameter in the environmental interference parameters, refers to factors such as increased mill roll gap due to wear and decreased motor speed accuracy. Equipment drift can cause different rolling effects with the same process parameters. Cooling medium temperature fluctuation refers to the impact of emulsion temperature changes during rolling on lubrication and cooling effects. The auxiliary inputs serve to eliminate external interference and prevent the model from misjudging rolled product defects caused by equipment drift as casting quality problems, thereby improving correction accuracy. The updated rolling temperature and updated rolling speed are the optimal control parameters for the rolling process calculated based on the current casting quality and environmental interference.

[0119] Finally, combining the initial process parameters of the rolling process, the rolling temperature correction parameters and rolling speed correction parameters are calculated. The initial process parameters of the rolling process include the initial rolling temperature and the initial rolling speed. The rolling temperature correction parameter = updated rolling temperature - initial rolling temperature, and the rolling speed correction parameter = updated rolling speed - initial rolling speed. For example, if the output updated rolling temperature is 480℃ and the updated rolling speed is 1.2 m / min, and the initial rolling temperature in the initial process parameters of the rolling process is 470℃ and the initial rolling speed is 1.3 m / min, then the rolling temperature correction parameter = 480 - 470 = +10℃, indicating that the rolling temperature needs to be increased by 10℃, and the rolling speed correction parameter = 1.2 - 1.3 m = -0.1 m / min, indicating that the rolling speed needs to be decreased by 0.1 m / min. In this way, dynamic matching between the rolling process control parameters and the casting quality is achieved.

[0120] In summary, compared to existing technologies, this application constructs a casting-rolling correction model. Using the casting process parameters in the initial process parameters, the dynamic output characteristics of the casting stage, and the casting-rolling correlation component in the weight matrix as input, it predicts and outputs rolling temperature correction parameters and rolling speed correction parameters. In this way, the quality state of the casting stage and external interference factors are transformed into precise process parameter correction amounts for the rolling stage, preventing casting quality fluctuations from being transmitted or amplified to the rolling stage.

[0121] S50: Construct a rolling-heat treatment correction model, using the rolling process parameters in the initial process parameters, the dynamic output characteristics of the rolling process, and the rolling-heat treatment correlation part in the weight matrix as inputs, to predict and output heat treatment temperature correction parameters and holding time correction parameters.

[0122] The quality of the rolled pieces produced in the rolling process is the direct basis for the raw materials processed in the heat treatment process. Its quality fluctuations will directly affect the entire heat treatment process. For example, if the grain size distribution of the rolled pieces is uneven, fine grains are prone to premature overheating due to rapid grain boundary diffusion during the heat treatment heating stage, while coarse grains will result in insufficient solid solution due to slow atomic diffusion. Ultimately, this will cause localized embrittlement and uneven overall strength of the aluminum rod after heat treatment.

[0123] To address the aforementioned issues, this application constructs a rolling-heat treatment correction model. Using the rolling process parameters in the initial process parameters, the dynamic output characteristics of the rolling process, and the rolling-heat treatment correlation part in the weight matrix as inputs, it predicts and outputs heat treatment temperature correction parameters and holding time correction parameters.

[0124] Specifically, step S50 in the method includes:

[0125] A rolling-heat treatment correction model was constructed based on neural networks;

[0126] A rolling-heat treatment correction model was obtained by training through supervised learning.

[0127] Retrieve the pre-trained rolling-heat treatment correction model;

[0128] Using the grain size distribution, deformation resistance, and recrystallization ratio of the rolled piece as input features, the output is updated heat treatment temperature and updated holding time.

[0129] Based on the initial process parameters of the heat treatment process, the heat treatment temperature correction parameters and the holding time correction parameters are calculated.

[0130] In this embodiment, a rolling-heat treatment correction model is first constructed based on a neural network. The neural network can capture the nonlinear correlation between the rolled piece state and the heat treatment control parameters through multi-layer nonlinear transformations. For example, the rolling-heat treatment correction model mainly consists of an input preprocessing layer, a feature extraction layer, a physical constraint layer, and an output layer. The input preprocessing layer receives the grain size distribution, deformation resistance, and recrystallization ratio of the rolled piece, and then eliminates dimensional differences through Z-score standardization. The feature extraction layer adopts a three-layer fully connected neural network with the number of neurons decreasing layer by layer from 32 to 16 to 8. All layers are equipped with the ReLU activation function, and each layer embeds BatchNorm batch normalization to accelerate training convergence. A Dropout layer with a dropout rate of 0.2 is added between the first and second layers to suppress overfitting. The physical constraint layer corrects the feature mapping direction through two types of core constraints: one is the diffusion dynamics constraint based on Fick's second law, which strengthens the positive correlation that temperature increase → diffusion coefficient increase → holding time can be shortened; the other is the phase transformation temperature constraint based on the solid solution characteristics of aluminum alloy, which clarifies the process boundaries of heat treatment temperature and holding time. The output layer adopts a dual-branch fully connected structure, with each branch containing one linearly activated neuron, which outputs the updated heat treatment temperature and the updated holding time, respectively.

[0131] Secondly, a rolling-heat treatment correction model is trained through supervised learning. For example, the rolling-heat treatment correction model can be trained through the following technical path: 1. Data preparation: Collect measured data on grain size distribution, deformation resistance, and recrystallization ratio of historical rolled pieces from historical production data as input sample sets. Simultaneously acquire the historical optimal heat treatment temperature and historical optimal holding time corresponding to each set of input samples, which have been verified by actual production effects (e.g., after processing with these parameters, the solid solution saturation of aluminum rods is ≥95%, the precipitate size distribution is uniform, and the mechanical properties meet the standards), as label sample sets. Divide the input sample set and the corresponding label sample set into training set, validation set, and test set according to a ratio of 7:1.5:1.5. 2. Model Training: The historical grain size distribution, deformation resistance, and recrystallization ratio of rolled pieces in the training set are used as input features. The corresponding historical best heat treatment temperature and historical best holding time are used as supervision labels. The network parameters, such as the weights of the feature extraction layer and the bias of the physical constraint layer, are iteratively optimized through the backpropagation algorithm (with Adam optimizer). A joint loss function is used, such as the weighted sum of the mean square error of the heat treatment temperature prediction and the mean square error of the holding time prediction, to measure the prediction bias. When the value of the joint loss function on the validation set is stable below the process allowable threshold (such as the loss value corresponding to temperature error ≤ ±5℃ and time error ≤ ±0.1h) for a preset number of rounds (e.g., 10 rounds) without significant decrease, the model is considered to have converged, and the trained rolling-heat treatment correction model is obtained.

[0132] Next, retrieve the pre-trained rolling-heat treatment correction model.

[0133] Furthermore, using the grain size distribution, deformation resistance, and recrystallization ratio of the rolled piece as input features, the output is an updated heat treatment temperature and an updated holding time. The updated heat treatment temperature and updated holding time are the optimal control parameters for the heat treatment process calculated based on the current rolling quality.

[0134] Finally, based on the initial process parameters of the heat treatment stage, the heat treatment temperature correction parameters and holding time correction parameters are calculated. The initial process parameters of the heat treatment stage include the initial heat treatment temperature and the initial holding time. The heat treatment temperature correction parameter = updated rolling temperature - initial heat treatment temperature, and the holding time correction parameter = updated holding time - initial holding time. For example, if the output updated heat treatment temperature is 530℃ and the updated holding time is 2.5h, and the initial heat treatment temperature in the initial process parameters of the heat treatment stage is 520℃ and the initial holding time is 2h, then the heat treatment temperature correction parameter = 530 - 520 = +10℃, indicating that the heat treatment temperature needs to be increased by 10℃, and the holding time correction parameter = 2.5 - 2 = +0.5h, indicating that the holding time needs to be extended by 0.5 hours. In this way, dynamic matching between the control parameters of the heat treatment stage and the rolling quality is achieved.

[0135] In summary, compared to existing technologies, this application constructs a rolling-heat treatment correction model. Using the rolling process parameters in the initial process parameters, the dynamic output characteristics of the rolling stage, and the rolling-heat treatment correlation component in the weight matrix as input, it predicts and outputs correction parameters for heat treatment temperature and holding time. In this way, the state of the rolled piece in the rolling stage is transformed into precise process parameter corrections for the heat treatment stage, thus offsetting the impact of rolling quality fluctuations by controlling the heat treatment process.

[0136] S60: Based on the correction parameters output by each model and combined with the preset fluctuation warning threshold, progressive control parameter correction is implemented for the casting, rolling and heat treatment processes, and the weight matrix is ​​dynamically updated according to the correction effect.

[0137] The aforementioned steps obtain correction parameters for multiple process steps in the production of aluminum alloy round aluminum rods through multiple correction models. Based on this, differentiated early warnings can be issued. Furthermore, by combining the actual quality impact after parameter correction, the weight matrix can be dynamically updated to ensure that the model correlation strength always matches the actual production situation.

[0138] To address the aforementioned issues, this application implements progressive control parameter correction for the casting, rolling, and heat treatment processes based on the correction parameters output by each model and a preset fluctuation warning threshold, and dynamically updates the weight matrix according to the correction effect.

[0139] Specifically, such as Figure 2 As shown, step S60 in the method includes:

[0140] Set multi-level fluctuation early warning thresholds for each process step, including early warning thresholds, adjustment thresholds, and intervention thresholds;

[0141] When the correction parameters output by the model exceed the warning threshold but do not reach the adjustment threshold, parameter monitoring and trend warning are performed.

[0142] When the correction parameters output by the model exceed the adjustment threshold but do not reach the intervention threshold, the initial process parameters are corrected according to the model output results.

[0143] When the correction parameters output by the model exceed the intervention threshold, a manual intervention mechanism is triggered.

[0144] In this embodiment, a multi-level fluctuation warning threshold is first set, including a warning threshold, an adjustment threshold, and an intervention threshold. The multi-level fluctuation warning threshold is a risk level boundary set for the correction parameters of each stage based on the characteristics of the aluminum alloy production process, the equipment's load-bearing limit, and historical quality data. The multi-level fluctuation warning threshold can be dynamically set according to specific product information and process scenarios. For example, it can be determined comprehensively based on process quality boundaries, equipment safety limits, and the risk of fluctuation transmission: historical data with adjusted parameters and a product quality pass rate ≥99% are clustered, and the equipment's physical load-bearing limit and the impact of preceding correction parameters on subsequent stages are considered to jointly determine the multi-level fluctuation warning threshold. For example, taking the casting temperature correction parameter in the casting stage as an example, the multi-level fluctuation warning threshold is set as follows: the warning threshold is ±2℃ (slight fluctuation, temporarily not affecting quality, but the trend needs to be monitored), the adjustment threshold is ±3℃ (moderate fluctuation, which may lead to an increase in the surface temperature difference of the cast billet if not corrected), and the intervention threshold is ±5℃ (severe fluctuation, exceeding the equipment's automatic adjustment capability, which may cause cracks in the cast billet). This achieves a differentiated response strategy.

[0145] Secondly, when the correction parameters output by the model exceed the warning threshold but do not reach the adjustment threshold, it indicates that the quality fluctuation is relatively minor and has not yet had a substantial impact on the quality of the current stage. If automatic correction is initiated immediately, frequent fine-tuning of parameters may lead to process instability. Therefore, only parameter monitoring and trend warnings are performed. For example, if the casting temperature correction parameter in the casting stage is +2.5℃, which is between the warning threshold and the adjustment threshold, the direction of change of the correction parameter is tracked in real time to determine whether the fluctuation is an instantaneous disturbance or a continuous deterioration.

[0146] Secondly, when the correction parameters output by the model exceed the adjustment threshold but do not reach the intervention threshold, it indicates that the fluctuation has reached a point where failure to correct it will directly affect the quality of the current stage, and subsequent transmission will amplify the problem. Therefore, the initial process parameters are corrected according to the model output results. For example, if the casting temperature correction parameter in the casting stage is +4℃, which is between the adjustment threshold and the intervention threshold, then the casting temperature in the initial process parameters is increased by 4℃ based on the casting temperature correction parameter +4℃.

[0147] Finally, when the correction parameters output by the model exceed the intervention threshold, it indicates that the fluctuations have reached a level that automated correction cannot handle, and may even lead to major quality problems or equipment failures. At this point, manual intervention must be initiated to prevent the risks from escalating. For example, if the casting temperature correction parameter in the casting process is +6℃, which exceeds the intervention threshold, a manual intervention mechanism will be triggered, such as sending an emergency intervention notification to the technician or triggering an alarm to perform manual troubleshooting.

[0148] Specifically, step S60 in the method further includes:

[0149] The impact of each parameter correction operation on the final product quality is quantified, wherein the impact is a vector, with positive values ​​indicating positive improvement in product quality and negative values ​​indicating negative deterioration in product quality;

[0150] Extract the correlation strength related to the parameter correction operation from the weight matrix, and obtain the historical average change of the correlation strength;

[0151] The ratio of the degree of impact to the corresponding historical average change is calculated and used as the initial value efficiency index for this parameter correction operation.

[0152] Based on the average prediction accuracy of the smelting-casting correction model, casting-rolling correction model, and rolling-heat treatment correction model within a preset time range, the initial value efficiency index is weighted and compensated to obtain the value efficiency index.

[0153] The value efficiency index is compared with a preset update threshold. If the value efficiency index is greater than or equal to the preset update threshold, the weight matrix is ​​dynamically updated.

[0154] Conversely, if the value efficiency index is less than the preset update threshold, the current weight matrix remains unchanged.

[0155] In this embodiment, the impact of each parameter correction operation on the final product quality is first quantified. This impact is a vector, with positive values ​​indicating positive improvements in product quality, such as increased tensile strength or elongation, and negative values ​​indicating negative deterioration, such as decreased conductivity or the appearance of cracks. This establishes a quantitative correlation between the parameter correction operation and the final product quality, providing a basis for subsequently determining whether the weight matrix needs adjustment.

[0156] Secondly, the correlation strength related to the parameter correction operation is extracted from the weight matrix, and the historical average variation of this correlation strength is obtained. The historical average variation is the average fluctuation of the correlation strength over a past period, such as ±0.1, representing the reasonable range of variation of the correlation strength under normal production fluctuations. The historical average variation provides a benchmark for assessing whether the current parameter correction operation is abnormal.

[0157] Next, the ratio of the impact level to the corresponding historical average change magnitude is calculated as the initial value efficiency index for this parameter correction operation. The initial value efficiency index = impact level / corresponding historical average change magnitude. This index measures the degree of deviation of the qualitative change brought about by the correction operation from historical normal fluctuations. For example, if the impact level is +4.38% and the corresponding historical average change magnitude is 0.1, then the initial value efficiency index = +4.38% / 0.1 = +43.8%, indicating that the positive effect of the current correction far exceeds the historical average, potentially a valid signal of improved correlation strength.

[0158] Furthermore, due to inherent errors in the predictive capabilities of the models themselves, a weighted compensation is applied to the initial value efficiency index based on the average prediction accuracy of the smelting-casting correction model, the casting-rolling correction model, and the rolling-heat treatment correction model within a preset time range. This weighted compensation yields the value efficiency index. The preset time range can be dynamically set according to actual needs, such as the past month or three months. The value efficiency index = average prediction accuracy × initial value efficiency index. The lower the average prediction accuracy of the model, the lower the value efficiency index. The weighted compensation filters out the interference of invalid corrections caused by inaccurate model predictions on the weight matrix. For example, by testing the prediction accuracy of the smelting-casting correction model, the casting-rolling correction model, and the rolling-heat treatment correction model using independent test sets (e.g., 90%, 95%, 94%), the average prediction accuracy = (90% + 95% + 94%) / 3 = 93%, and the value efficiency index = 93% × (+43.8%) = +40.7%. Thus, weighted compensation avoids updating the matrix due to unreliable correction data.

[0159] Finally, the value efficiency index is compared with a preset update threshold. If the value efficiency index is greater than or equal to the preset update threshold, the weight matrix is ​​dynamically updated; otherwise, if the value efficiency index is less than the preset update threshold, the current weight matrix remains unchanged. The preset update threshold is a critical value set based on historical data for updating the weight matrix. For example, +50% means that the weight matrix is ​​considered to need adjustment only when the value efficiency of parameter correction reaches this level. For instance, if the value efficiency index is +52%, which is greater than the preset update threshold, it indicates that the current correlation strength has significantly deviated from historical patterns, and the correction effect is reliable, requiring an update to the weight matrix. Conversely, if the value efficiency index is less than the preset update threshold, it indicates that the change is within a reasonable range or the correction effect is insufficient to support matrix adjustment, and the current matrix remains unchanged.

[0160] Specifically, the "quantification of the impact of each parameter correction on the final product quality" includes:

[0161] Product quality indicators are selected as the evaluation criteria, wherein the product quality indicators include at least tensile strength, elongation and conductivity;

[0162] The relative change rate of the product quality indicators before and after parameter correction is calculated, and the degree of influence is obtained by weighted calculation.

[0163] In this embodiment of the application, product quality indicators are first selected as the evaluation basis. Among them, product quality indicators include at least tensile strength, elongation and conductivity. Tensile strength can reflect the load-bearing capacity of aluminum alloy round aluminum rod, elongation can reflect the plastic deformation capacity of aluminum alloy round aluminum rod, and conductivity can reflect the conductivity of aluminum alloy round aluminum rod. These three parameters are the key performance indicators of aluminum alloy round aluminum rod, which directly determine the product grade and application scenario.

[0164] Secondly, the relative change rate of product quality indicators before and after parameter correction is calculated, and the degree of influence is obtained by weighted calculation. For example, tensile strength before and after parameter correction can be measured by tensile testing; for instance, the tensile strength before parameter correction is 270 MPa, and the tensile strength after parameter correction is 280 MPa. Elongation before and after parameter correction can be measured by eddy current testing; for instance, the elongation before parameter correction is 8.0%, and the elongation after parameter correction is 8.5%. Conductivity before and after parameter correction can be measured; for instance, the conductivity before parameter correction is 60.0% IACS, and the conductivity after parameter correction is 61.2% IACS. For example, the relative change rate = (product quality indicator after parameter correction - product quality indicator before parameter correction) / product quality indicator before parameter correction. For example, if the tensile strength before parameter correction is 270 MPa, the elongation is 8.0%, and the conductivity is 60.0% IACS, and the tensile strength after parameter correction is 280 MPa, the elongation is 8.5%, and the conductivity is 61.2%. For IACS, the relative change rate of tensile strength = (280-270) / 270 = 3.7%, the relative change rate of elongation = (8.5%-8.0%) / 8.0% = 6.25%, and the relative change rate of conductivity = (61.2-60.0) / 60.0 = 2.0%. Then, the degree of influence is calculated by weighting. The weights need to be dynamically set according to the product application scenario and actual needs. For example, if tensile strength, elongation, and conductivity are assigned weights of 0.4, 0.4, and 0.2 respectively, then the degree of influence = 0.4×3.7%+0.4×6.25%+0.2×2.0%=4.38%. If the degree of influence is positive, it means that the product quality has been improved after correction, such as increased tensile strength, increased elongation, and increased conductivity. Conversely, if the degree of influence is negative, it means that the product quality has been worsened after correction, such as decreased elongation and decreased conductivity. In severe cases, cracks and other appearance defects may occur.

[0165] Specifically, the "dynamic update of the weight matrix" includes:

[0166] The recursive least squares method is used to update the correlation strength in the weight matrix in real time based on recent parameter correction data and the corresponding quality impact.

[0167] Record the timestamp, update content, and update reason for each weight matrix update to form a complete update history log.

[0168] In this embodiment, recursive least squares is first used to update the correlation strength in the weight matrix in real time based on recent parameter correction data and the corresponding quality impact. Recursive least squares is a real-time parameter estimation method suitable for scenarios where data is continuously generated during production and parameters need dynamic optimization. For example, recursive least squares can be used to update the correlation strength in the weight matrix in real time based on recent parameter correction data and the corresponding quality impact. This mainly includes four steps: data screening, algorithm initialization, iterative optimization, and process verification, as follows: First, sample pairs of recent valid parameter correction data and quality impact are screened, and abnormal data caused by sensor malfunctions are removed. Then, the algorithm is initialized. The initial correlation strength θ0 is taken as the historical average, such as 0.05, as the starting point for iteration. The initial covariance matrix P0 is set to a 100× identity matrix, representing a low initial confidence level for θ0, allowing for significant correction of subsequent data. The forgetting factor λ is selected as 0.96, leveraging the higher weight of recent data to weaken the interference of outdated data such as equipment status and raw material batches from one month ago, focusing on current production changes. Then, the gain matrix K is iteratively calculated according to the sample order. k (Determine sample contribution), update association strength θ k (Based on prediction error correction) and covariance matrix P k (To increase confidence), finally check whether θk meets the process boundary (e.g., 0.02≤θ≤0.2). If it exceeds the range, truncate it, replace the corresponding value in the matrix and record the update log to achieve real-time optimization of the correlation strength to fit the current production dynamics.

[0169] Secondly, the timestamp, update content, and update reason for each weight matrix update are recorded to form a complete update history log. The timestamp is accurate to the minute, facilitating the tracking of whether quality fluctuations within a certain time period are related to weight matrix updates. The update content refers to which specific correlation strength was adjusted and the magnitude of the adjustment. The update reason includes the value efficiency indicator that triggered the update and the corresponding correction data. In this way, when subsequent quality anomalies occur, the matrix update process can be traced back to determine whether it was caused by improper updates, providing a basis for model iteration.

[0170] In summary, compared to existing technologies, this application implements progressive control parameter correction for the casting, rolling, and heat treatment processes based on the correction parameters output by each model and a preset fluctuation warning threshold, and dynamically updates the weight matrix according to the correction effect. Thus, based on the correction parameters output by each correction model, and by classifying and matching processing strategies for different risk levels through threshold division, it achieves both automated and precise control, while mitigating extreme risks through manual intervention. Furthermore, multi-dimensional verification ensures the reliability of the weight matrix update, ultimately achieving long-term stability and adaptability of the entire process control for aluminum alloy round aluminum rod production.

[0171] In summary, the embodiments of this application have at least the following technical effects:

[0172] Compared to existing technologies, this application first obtains the initial process parameters for each production stage, and then constructs a weight matrix based on historical production data to characterize the transmission relationship of quality fluctuations between each production stage. In this way, the coupling relationship of quality fluctuations in each stage is transformed into a quantified mathematical matrix, which not only reflects the direction of quality fluctuation transmission but also quantifies the degree of influence, providing reliable data support for the subsequent construction of a correction model.

[0173] Secondly, this application collects dynamic output characteristics and environmental disturbance parameters of each production stage in real time. In this way, it obtains dynamic output characteristics reflecting the production quality status of each stage and environmental disturbance parameters reflecting external influencing factors, providing a reliable data foundation for subsequent construction of correction models and dynamic optimization control.

[0174] Furthermore, this application constructs a smelting-casting correction model, using the smelting process parameters in the initial process parameters, the dynamic output characteristics of the smelting stage, and the smelting-casting correlation component in the weight matrix as inputs, to predict and output casting temperature correction parameters and casting speed correction parameters. In this way, the real-time quality fluctuations in the smelting stage are accurately converted into process parameter corrections for the casting stage, thereby suppressing the propagation of preceding quality fluctuations to subsequent stages.

[0175] Furthermore, this application constructs a casting-rolling correction model, using the casting process parameters in the initial process parameters, the dynamic output characteristics of the casting process, and the casting-rolling correlation component in the weight matrix as inputs, to predict and output rolling temperature correction parameters and rolling speed correction parameters. In this way, the quality status of the casting process and external interference factors are transformed into precise process parameter correction amounts for the rolling process, preventing casting quality fluctuations from being transmitted or amplified to the rolling process.

[0176] Furthermore, this application constructs a rolling-heat treatment correction model, using the rolling process parameters in the initial process parameters, the dynamic output characteristics of the rolling stage, and the rolling-heat treatment correlation component in the weight matrix as inputs, to predict and output heat treatment temperature correction parameters and holding time correction parameters. In this way, the rolled piece state in the rolling stage is transformed into precise process parameter correction amounts for the heat treatment stage, and the impact of rolling quality fluctuations is offset by controlling the heat treatment process.

[0177] Finally, based on the correction parameters output by each model and combined with preset fluctuation warning thresholds, this application implements progressive control parameter correction for the casting, rolling, and heat treatment processes, and dynamically updates the weight matrix according to the correction effect. In this way, based on the correction parameters output by each correction model, and by classifying and matching different risk levels through threshold divisions, it achieves both automated and precise control, while mitigating extreme risks through manual intervention. Furthermore, multi-dimensional verification ensures the reliability of the weight matrix update, ultimately achieving long-term stability and adaptability of the entire process control for aluminum alloy round aluminum rod production.

[0178] Through the above technical solution, this application effectively breaks through the limitations of traditional independent control of each stage by constructing three cross-stage correction models. It achieves proactive compensation for quality fluctuations in preceding processes on subsequent processes, weakening the cross-stage coupling and transmission effect of quality fluctuations. Combined with a progressive parameter correction mechanism using multi-level fluctuation early warning thresholds, it replaces the traditional post-event remedial mode, achieving differentiated and precise control of process parameters throughout the entire process. Furthermore, the weight matrix is ​​dynamically updated based on the correction effect, ensuring that the model always fits the production dynamics and avoids control accuracy decay. This improves the product performance stability and yield rate of aluminum alloy round aluminum rods, while reducing repeated process adjustments and parameter redundancy, thus lowering production energy consumption and costs.

[0179] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0180] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0181] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0182] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0183] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0184] Although preferred embodiments of the invention have been described, those skilled in the art, once they have learned the basic inventive concept, can make other changes and modifications to these embodiments.

[0185] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of this invention and its equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for optimizing control of a production process of aluminum alloy round aluminum bars, the production process comprising, in sequence, a melting, a casting, a rolling and a heat treatment step, characterized in that, The method includes: The initial process parameters for each production stage are obtained, and a weight matrix representing the quality fluctuation transmission relationship between each production stage is constructed based on historical production data. Real-time acquisition of dynamic output characteristics and environmental interference parameters of each production stage; A smelting-casting correction model is constructed, taking the smelting process parameters in the initial process parameters, the dynamic output characteristics of the smelting process, and the smelting-casting correlation part in the weight matrix as inputs, to predict the output casting temperature correction parameters and casting speed correction parameters. A casting-rolling correction model is constructed, using the casting process parameters in the initial process parameters, the dynamic output characteristics of the casting process, and the casting-rolling correlation part in the weight matrix as inputs, to predict the output rolling temperature correction parameters and rolling speed correction parameters. A rolling-heat treatment correction model is constructed, taking the rolling process parameters in the initial process parameters, the dynamic output characteristics of the rolling process, and the rolling-heat treatment correlation part in the weight matrix as inputs, to predict the output heat treatment temperature correction parameters and holding time correction parameters. Based on the correction parameters output by each model and combined with the preset fluctuation warning threshold, progressive control parameter correction is implemented for the casting, rolling and heat treatment processes, and the weight matrix is ​​dynamically updated according to the correction effect.

2. The method of claim 1, wherein the aluminum alloy round rod production process optimization control method is characterized by, A weight matrix representing the transmission relationship of quality fluctuations between various production stages is constructed based on historical production data, including: Extract quality fluctuation characteristic parameters from each production stage in historical production data to construct multiple quality fluctuation feature vectors; Calculate the mutual information value between the characteristic vectors of quality fluctuations in adjacent production stages, as the correlation strength of fluctuation transmission between stages; Identify the causal pathways through which quality fluctuations are transmitted between different stages; Based on the aforementioned correlation strength and causal relationship path, a weight matrix is ​​constructed with production links as nodes and the aforementioned correlation strength as edge weights.

3. The method of claim 1, wherein the aluminum alloy round rod production process optimization control method is characterized by, The system collects dynamic output characteristics and environmental interference parameters for each production stage in real time, including: dynamic output characteristics for the smelting stage, including at least melt temperature uniformity, compositional segregation index, and gas content; dynamic output characteristics for the casting stage, including at least billet surface temperature distribution, solidification front velocity, and shrinkage defect index; dynamic output characteristics for the rolling stage, including at least rolled piece grain size distribution, deformation resistance, and recrystallization ratio; dynamic output characteristics for the heat treatment stage, including at least solution saturation, aging kinetic parameters, and precipitated phase morphology; and environmental interference parameters, including at least raw material composition fluctuations, equipment operating condition drift, cooling medium temperature fluctuations, and changes in ambient temperature and humidity.

4. The method of claim 1, wherein the aluminum alloy round rod production process optimization control method is characterized by, Construct a modified smelting-casting model, including: An attention-based long short-term memory network is used as the model architecture; The smelting-casting correction model is trained using supervised learning. Using time-series data on melt temperature uniformity, composition segregation index, and gas content in the smelting process as input features, the outputs are the updated casting temperature and updated casting speed in the casting process. By combining the initial process parameters of the casting process, the casting temperature correction parameters and casting speed correction parameters are calculated.

5. The method for optimizing and controlling the production process of aluminum alloy round aluminum rods according to claim 1, characterized in that, Constructing a casting-rolling correction model, including: A casting-rolling correction model is constructed using a multi-task learning neural network architecture; The casting-rolling correction model is obtained by training through supervised learning. Retrieve the pre-trained casting-rolling correction model; The main input features are the surface temperature distribution of the billet, the solidification front velocity, and the shrinkage defect index in the casting process. The auxiliary input features are the equipment condition drift and the cooling medium temperature fluctuation in the environmental interference parameters. The output is updated rolling temperature and updated rolling speed. Based on the initial process parameters of the rolling process, rolling temperature correction parameters and rolling speed correction parameters are calculated.

6. The method of claim 1, wherein the aluminum alloy round rod production process optimization control method is characterized by, Construct a rolling-heat treatment modified model, including: A rolling-heat treatment correction model was constructed based on neural networks; A rolling-heat treatment correction model was obtained by training through supervised learning. Retrieve the pre-trained rolling-heat treatment correction model; Using the grain size distribution, deformation resistance, and recrystallization ratio of the rolled piece as input features, the output is updated heat treatment temperature and updated holding time. Based on the initial process parameters of the heat treatment process, the heat treatment temperature correction parameters and the holding time correction parameters are calculated.

7. The method of claim 1, wherein the aluminum alloy round rod production process optimization control method is characterized by, Based on the correction parameters output by each model, progressive control parameter correction is implemented for the casting, rolling, and heat treatment processes, including: Set multi-level fluctuation early warning thresholds for each process step, including early warning thresholds, adjustment thresholds, and intervention thresholds; When the correction parameters output by the model exceed the warning threshold but do not reach the adjustment threshold, parameter monitoring and trend warning are performed. When the correction parameters output by the model exceed the adjustment threshold but do not reach the intervention threshold, the initial process parameters are corrected according to the model output results. When the correction parameters output by the model exceed the intervention threshold, a manual intervention mechanism is triggered.

8. The method of claim 1, wherein the aluminum alloy round rod production process optimization control method is characterized by, The weight matrix is ​​dynamically updated based on the correction effect, including: The impact of each parameter correction operation on the final product quality is quantified, wherein the impact is a vector, with positive values ​​indicating positive improvement in product quality and negative values ​​indicating negative deterioration in product quality; Extract the correlation strength related to the parameter correction operation from the weight matrix, and obtain the historical average change of the correlation strength; The ratio of the degree of impact to the corresponding historical average change is calculated and used as the initial value efficiency index for this parameter correction operation. Based on the average prediction accuracy of the smelting-casting correction model, casting-rolling correction model, and rolling-heat treatment correction model within a preset time range, the initial value efficiency index is weighted and compensated to obtain the value efficiency index. The value efficiency index is compared with a preset update threshold. If the value efficiency index is greater than or equal to the preset update threshold, the weight matrix is ​​dynamically updated. Conversely, if the value efficiency index is less than the preset update threshold, the current weight matrix remains unchanged.

9. The method of claim 8, wherein the optimization control is performed by a computer program. Quantify the impact of each parameter correction on the final product quality, including: Product quality indicators are selected as the evaluation criteria, wherein the product quality indicators include at least tensile strength, elongation and conductivity; The relative change rate of the product quality indicators before and after parameter correction is calculated, and the degree of influence is obtained by weighted calculation.

10. The method of claim 8, wherein the optimization control is performed by a computer program. Dynamically update the weight matrix, including: The correlation strength in the weight matrix is updated in real time based on the recent parameter correction data and corresponding quality influence degree by using the recursive least square method; The time stamp, update content and update reason of each weight matrix update are recorded to form a complete update history log.

Images