Aluminum alloy profile production and processing optimization control method and system

Abstract

The present application relates to the technical field of aluminum alloy profile production and processing control, in particular to an aluminum alloy profile production and processing optimization control method and system, which comprises: collecting multi-source process data in the production process in real time, including smelting furnace temperature distribution, extrusion machine process parameter time sequence and profile surface online monitoring image. The data is subjected to collaborative feature extraction and fusion processing to generate a fusion state feature vector containing material uniformity index, deformation stress distribution feature and surface defect potential factor. The vector is input into a preset process parameter optimization reasoning model to obtain a key process parameter adjustment suggestion set for the next production cycle. In combination with the current equipment operation state constraint, an executable process parameter adjustment instruction sequence is generated and executed. At the same time, the feedback data after adjustment is collected for updating the optimization reasoning model. The method realizes comprehensive perception of the processing state and dynamic intelligent optimization of the process parameters.

Description

Technical Field

[0001] This invention relates to the field of aluminum alloy profile production and processing control technology, and in particular to an optimized control method and system for aluminum alloy profile production and processing. Background Technology

[0002] In current aluminum alloy profile production, process control largely relies on monitoring and adjusting single or a few process parameters. A common practice is to install temperature and pressure sensors in key processes such as melting and extrusion, and set and adjust parameters based on preset process curves or operator experience. Profile quality inspection is typically conducted after extrusion molding through offline sampling or manual visual inspection, with adjustments made to upstream processes only after defects are detected. This approach treats melting, extrusion, and molding quality as relatively independent stages. Existing technologies rely on fragmented data sources and isolated analytical models, lacking effective correlation analysis and collaborative control mechanisms between the temperature uniformity of the melting furnace, the kinetic parameters of the extrusion process, and the final profile surface quality. Parameter adjustments are often based on static rules or local feedback, resulting in delayed responses. Because the multi-dimensional heterogeneous information reflecting the material's internal state, external deformation, and apparent quality is not comprehensively utilized, the judgment of the processing state is incomplete and superficial, making it difficult to predict and suppress defects in advance. This leads to a narrow process window, and product quality stability depends on trial and error and experience accumulation. This invention aims to address how to deeply integrate melting temperature field, extrusion process time dynamics, and online visual information to construct a feature system that can comprehensively and deeply characterize the material processing state. Based on this system, it aims to achieve online forward-looking optimization and closed-loop dynamic evolution of process parameters, thereby overcoming the limitations of existing isolated control and post-event remediation modes. Summary of the Invention

[0003] The purpose of this invention is to overcome the shortcomings of the existing technology and to propose an optimized control method and system for the production and processing of aluminum alloy profiles.

[0004] To achieve the above objectives, the present invention adopts the following technical solution: a method and system for optimizing and controlling the production and processing of aluminum alloy profiles, comprising: Real-time acquisition of multi-source process data in the aluminum alloy profile production and processing flow, including furnace temperature distribution data, extrusion press process parameter timing data, and profile surface online monitoring image data; The multi-source process data is subjected to collaborative feature extraction and fusion processing to generate a fused state feature vector characterizing the current processing state. The fused state feature vector includes material uniformity index, deformation stress distribution characteristics and surface defect potential factors. The fusion state feature vector is input into a preset process parameter optimization inference model, and the process parameter optimization inference model outputs a set of key process parameter adjustment suggestions for the next production cycle. Based on the set of key process parameter adjustment suggestions and the constraints of the current equipment operating status, an executable sequence of process parameter adjustment instructions is generated. The process parameter adjustment instruction sequence is executed, and the adjusted feedback data is collected synchronously to update the process parameter optimization inference model.

[0005] As a further aspect of the present invention, the step of performing collaborative feature extraction and fusion processing on the multi-source process data to generate a fused state feature vector representing the current processing state includes: From the temperature distribution data of the smelting furnace, the temperature values ​​and temperature gradient change rates of multiple key temperature measuring points are extracted, and the temperature uniformity coefficient of the entire molten pool and the overheating risk index of a specific area are calculated. From the time-series data of the extruder process parameters, the curves of extrusion speed, extrusion pressure and die temperature changing with time are analyzed. The main frequency characteristics and abnormal fluctuation modes are extracted through frequency domain analysis, and the equivalent deformation stress distribution inside the profile is calculated. Multi-scale texture analysis and defect region segmentation are performed on the online monitoring image data of the profile surface to extract the quantitative values ​​of surface roughness, area, perimeter and shape complexity features of suspected defect regions, and calculate the potential factor of surface defects. A time-series alignment framework based on production cycle time is established, mapping the temperature uniformity coefficient, overheating risk index, equivalent deformation stress distribution, surface roughness quantification, and defect region characteristics onto a unified time axis. The temperature uniformity coefficient, overheating risk index, equivalent deformation stress distribution, and surface defect potential factors within the same production cycle time period are spliced ​​and normalized to generate the fused state feature vector.

[0006] As a further aspect of the present invention, the step of analyzing the curves of extrusion speed, extrusion pressure, and die temperature changing with time from the time-series data of the extrusion press process parameters, extracting their dominant frequency characteristics and abnormal fluctuation modes through frequency domain analysis, and calculating the equivalent deformation stress distribution inside the profile includes: A fast Fourier transform is performed on the extrusion speed time-series curve to identify the dominant frequency components and amplitudes in its spectrum. The dominant frequency components reflect the periodic characteristics of the equipment operation. A sliding window statistical method is used to analyze the extrusion pressure time series curve, calculate the mean, standard deviation and skewness within each window, and identify abnormal pressure peaks and sustained high pressure ranges that exceed preset thresholds. The temperature time series curve of the mold is decomposed to separate the long-term trend term, the periodic term and the residual term. The abrupt change point in the residual term indicates the instantaneous anomaly of the temperature control system. A comprehensive process stability score is constructed by combining the stability of the dominant frequency components, the frequency of occurrence of the abnormal pressure peaks, and the abrupt change amplitude of the mold temperature residual term. Based on the comprehensive score of process stability and the real-time values ​​of extrusion speed and extrusion pressure, combined with the material rheological stress model, the equivalent deformation stress distribution at each point inside the profile during the extrusion process is calculated.

[0007] As a further aspect of the present invention, the step of inputting the fused state feature vector into a preset process parameter optimization inference model, wherein the process parameter optimization inference model outputs a set of key process parameter adjustment suggestions for the next production cycle, including: The process parameter optimization reasoning model consists of a quality prediction subnetwork and a parameter optimization subnetwork. The fusion state feature vector is input into the quality prediction sub-network to predict the key quality indicators of the profile produced in the next production cycle under the current process parameter continuation conditions. The key quality indicators include dimensional tolerances, mechanical properties and surface quality scores. The predicted values ​​of the key quality indicators are compared with the preset quality target values, and the deviation of each quality dimension is calculated. The deviation and the current fusion state feature vector are input together into the parameter optimization subnetwork; The parameter optimization subnetwork performs a multi-objective optimization search in the process parameter space consisting of melting temperature, extrusion speed, and die preheating temperature, with the goal of minimizing the deviation, and outputs a set of key process parameter adjustment suggestions that make the predicted quality closest to the target value. The set of key process parameter adjustment suggestions includes the suggested adjustment direction and magnitude for each parameter.

[0008] As a further aspect of the present invention, the step of generating an executable sequence of process parameter adjustment instructions based on the set of key process parameter adjustment suggestions and the constraints of the current equipment operating state includes: Obtain the constraints of the current equipment operating status, including the maximum heating rate of the melting furnace, the maximum pressure limit of the extruder, the power limit of the die heater, and the response delay time of each actuator; Each suggested adjustment range in the set of key process parameter adjustment suggestions is compared and verified with the corresponding equipment constraints. For suggested adjustments that do not exceed the equipment constraints, they will be directly converted into standard control commands that can be recognized by the equipment control interface. For suggested adjustments that exceed equipment constraints, initiate constraint handling strategies: within the allowable constraint boundaries, proportionally reduce the adjustment range to a feasible range, or based on the coupling relationship between process parameters, find alternative parameter adjustment schemes that can partially compensate for the target effect; All verified and processed individual control commands are sorted and combined according to the logical order and time dependency of the production process flow, and necessary command intervals are added to form the executable process parameter adjustment command sequence.

[0009] As a further aspect of the present invention, the step of executing the process parameter adjustment instruction sequence and simultaneously collecting the adjusted feedback data for updating the process parameter optimization inference model includes: The distributed control system of the production line executes the process parameter adjustment command sequence, and issues adjustment commands to the melting furnace temperature controller, the extruder main drive system and the die temperature control system. In the next complete production cycle after the instruction is executed, new multi-source process data is collected synchronously as adjusted feedback data. The adjusted feedback data is subjected to the same collaborative feature extraction and fusion process to generate a new fused state feature vector, denoted as the adjusted state feature vector. Collect actual quality inspection data of the profile samples finally produced in the production cycle, including measured dimensions, mechanical property test results and surface quality assessment report; The fusion state feature vector before execution, the sequence of process parameter adjustment instructions executed, the state feature vector after adjustment, and the actual quality detection data are combined into a new training sample. The new training samples are added to the historical training dataset of the process parameter optimization inference model, and the incremental learning process of the model is triggered to update the model's internal parameters.

[0010] As a further aspect of the present invention, the step of calculating the equivalent deformation stress distribution at various points inside the profile during the extrusion process based on the comprehensive score of process stability and the real-time values ​​of extrusion speed and extrusion pressure, combined with the material rheological stress model, includes: The stored aluminum alloy material rheological stress model is invoked, which describes the flow stress relationship of aluminum alloy at different temperatures and strain rates. The real-time value of the extrusion speed is converted into the average strain rate of the profile in the die deformation zone; The real-time value of the mold temperature is used as an approximate temperature field input for the material during the deformation process; Based on the average strain rate and the approximate temperature field, the instantaneous flow stress reference value of the material is obtained by querying or calculating the material rheological stress model; The instantaneous flow stress benchmark value is corrected based on the comprehensive process stability score. The lower the comprehensive process stability score, the greater the fluctuation. A dynamic fluctuation component negatively correlated with the score is added to the flow stress. Based on the profile cross-sectional shape, mold geometry and friction conditions, a simplified flow simulation is performed using the finite volume method. The corrected flow stress is then distributed to each calculation unit inside the profile, thereby obtaining the equivalent deformation stress distribution.

[0011] As a further aspect of the present invention, for the suggested adjustments exceeding equipment constraints, a constraint handling strategy is initiated: within the allowable constraint boundaries, the adjustment range is proportionally reduced to a feasible range, or, based on the coupling relationship between process parameters, alternative parameter adjustment schemes that can partially compensate for the target effect are sought, including: When the suggested adjustment range of a certain parameter exceeds the maximum allowable adjustment range of the device, the maximum adjustment range that the parameter can achieve is calculated. Calculate the ratio of the suggested adjustment range to the maximum achievable adjustment range, and then globally and proportionally reduce all suggested adjustment ranges that exceed the constraints until all adjustments are within the constraints. When reducing the adjustment magnitude results in the expected quality improvement effect falling below the threshold, an alternative solution search is initiated. The alternative search is based on a process parameter coupling knowledge base, which records the influence weights of different combinations of process parameters on the final quality indicators. Without violating other constraints, other coupled process parameters are adjusted to compensate for the quality improvement effect lost due to the limitation of the main parameter adjustment, forming a new parameter adjustment combination as an alternative parameter adjustment scheme.

[0012] As a further aspect of the present invention, adding the new training samples to the historical training dataset of the process parameter optimization inference model and triggering the incremental learning process of the model to update the model's internal parameters includes: The fusion state feature vector and the process parameter adjustment instruction sequence in the new training samples before execution are used as input features for the incremental learning process. The target quality improvement obtained by fusing the adjusted state feature vector in the new training samples and the actual quality detection data is used as the supervision label for the incremental learning process. Retrieve several historical samples from the historical training dataset that are most similar to the current new sample process scenario to form a temporary local training set; Using the temporary local training set along with the new training samples, a round of local incremental training is performed on the process parameter optimization inference model, especially the parameter optimization sub-network. In local incremental training, a small learning rate is used to update only some network layer parameters of the model in order to avoid catastrophic forgetting of learned knowledge; After training is completed, the new training samples are officially stored in the historical training dataset.

[0013] As a further aspect of the present invention, the present invention also includes an aluminum alloy profile production and processing optimization control system, the system including a processor and a memory, the memory storing a computer program, and the processor executing the computer program to implement the above-mentioned aluminum alloy profile production and processing optimization control method.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are as follows: Instead of analyzing or simply splicing different data sets, this approach employs collaborative feature extraction and fusion processing of furnace temperature distribution, extruder process parameter time series, and online monitoring images of profile surfaces. It analyzes deformation stress variation patterns from time-series data, extracts material uniformity indicators from temperature field data, and mines potential surface defect factors from image data. These heterogeneous features are then fused into a unified, high-dimensional fusion state feature vector. This processing method breaks down the barriers between data from different processes, transforming information characterizing the processing state from single-dimensional, reactive surface parameters into a multi-dimensional set of deep features reflecting the relationship between internal mechanisms and external manifestations. This vector can more fundamentally and comprehensively describe the integrated state from material preparation to plastic forming, providing an unprecedented, unified, and information-rich state description benchmark for subsequent precise decision-making.

[0015] The pre-defined process parameter optimization inference model receives the aforementioned fused state feature vector as input and outputs a set of key process parameter adjustment suggestions for the next production cycle. This model is not a fixed rule base; its core mechanism lies in simultaneously collecting feedback data from the production system after executing the adjustment instruction sequence and updating the model itself using these new input-output correspondences. This means that the model's decision-making logic can continuously learn and evolve as the production process continues, absorbing the positive or negative feedback experience from each adjustment. This transforms process optimization from a static mapping process based on fixed rules or historical experience into a dynamic closed-loop intelligent system with self-learning and adaptive capabilities. The model's inference ability continuously enhances with the accumulation of production data, enabling process parameter adjustment suggestions to better reflect real-time operating conditions, approach the optimal process window, and achieve proactive tracking and matching of the dynamic characteristics of the production process. Attached Figure Description

[0016] Figure 1 This is a flowchart of the aluminum alloy profile production and processing optimization control method described in this invention; Figure 2 A flowchart for analyzing extrusion press process parameters and calculating equivalent deformation stress; Figure 3The diagram shows the adjustment range of five sets of process parameters in the production of aluminum alloy profiles and the corresponding quality improvement effect. Figure 4 A diagram showing the changes in different process parameters during the production of aluminum alloy profiles; Figure 5 A bar chart comparing the aluminum alloy profile production process before and after adjustments. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0018] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0019] See Figure 1 A method for optimizing and controlling the production and processing of aluminum alloy profiles includes the following steps: Real-time acquisition of multi-source process data in the aluminum alloy profile production and processing flow, including furnace temperature distribution data, extrusion press process parameter time-series data, and online monitoring image data of the profile surface. Collaborative feature extraction and fusion processing are performed on the multi-source process data to generate a fused state feature vector characterizing the current processing state. The fused state feature vector includes material uniformity indices, deformation stress distribution characteristics, and potential surface defect factors. The fused state feature vector is input into a preset process parameter optimization inference model, which outputs a set of key process parameter adjustment suggestions for the next production cycle. Based on the key process parameter adjustment suggestion set and the constraints of the current equipment operating state, an executable process parameter adjustment instruction sequence is generated. The process parameter adjustment instruction sequence is executed, and feedback data after adjustment is collected synchronously to update the process parameter optimization inference model.

[0020] In one embodiment of the present invention, when implementing the aluminum alloy profile production and processing optimization control method, the process of collaborative feature extraction and fusion processing of multi-source process data to generate a fused state feature vector is achieved through a dedicated data fusion module. This module receives real-time data streams from the melting furnace temperature sensor array, the extruder process parameter historical database, and the profile surface online monitoring image system. The melting furnace temperature distribution data includes temperature readings at different depths and horizontal positions within the molten pool. The extruder process parameter time series data records the numerical sequence of extrusion speed, extrusion pressure, and die temperature changes over time. The profile surface online monitoring image data is continuously acquired by a high-speed industrial camera installed at the discharge port.

[0021] Temperature values ​​and temperature gradient change rates of multiple key temperature measurement points are extracted from the temperature distribution data of the smelting furnace. The key temperature measurement points are pre-set according to the furnace structure and the flow characteristics of the aluminum alloy melt. The temperature gradient change rate is obtained by calculating the temperature difference between adjacent temperature measurement points per unit time. Calculating the temperature uniformity coefficient of the entire molten pool involves statistical analysis of the temperature values ​​of all effective temperature measurement points. The temperature uniformity coefficient can be characterized by the ratio of the standard deviation to the mean. Calculating the overheating risk index of a specific area requires identifying areas in the temperature distribution that are continuously higher than the ideal holding temperature threshold of the melt. The overheating risk index is obtained by weighted accumulation based on the overheating amplitude and duration. The curves of extrusion speed, extrusion pressure, and die temperature over time are analyzed from the time-series data of extrusion press process parameters. The analysis process includes filtering and denoising the original time-series data and removing outliers. Frequency domain analysis is used to extract the dominant frequency characteristics and abnormal fluctuation patterns. This involves spectral transformation and pattern recognition of the preprocessed time-series curves. Frequency domain analysis can reveal the periodic patterns hidden in the changes of process parameters. Calculating the equivalent deformation stress distribution inside the profile requires a comprehensive analysis of the dynamic changes of extrusion speed and extrusion pressure as well as the thermal state of the die. The equivalent deformation stress distribution reflects the approximate spatial distribution of internal stress in the material during deformation.

[0022] Multi-scale texture analysis and defect region segmentation are performed on online monitoring image data of profile surfaces. Multi-scale texture analysis uses different scale filter kernels to convolve the image to extract texture features. Defect region segmentation separates suspected defect areas from normal background based on the gray level, gradient, and texture differences of image pixels. Quantitative values ​​of surface roughness are extracted by analyzing the statistical characteristics of image gray level changes within a specified area. The area, perimeter, and shape complexity features of suspected defect areas are extracted by directly calculating geometric properties from the segmented binary region image. Calculating the surface defect potential factor requires comprehensive evaluation of multiple features of suspected defect areas. The surface defect potential factor is a comprehensive scalar value used to characterize the probability of visible defects appearing on the profile surface within the current production cycle.

[0023] A time-series alignment framework based on production cycle time is established. Production cycle time refers to the standard time interval corresponding to the production of a fixed-length profile product. The time-series alignment framework maps feature sequences generated from different data sources with potentially asynchronous timestamps to a unified production cycle time axis. Temperature uniformity coefficient, overheating risk index, equivalent deformation stress distribution, surface roughness quantification, and defect area features are mapped to a unified time axis. The mapping process assigns data to the corresponding production cycle time window based on the timestamp of data acquisition. For features that generate multiple data points within a production cycle, the aggregation method of taking the average or maximum value is used to generate representative feature values ​​for that cycle. The temperature uniformity coefficient, overheating risk index, equivalent deformation stress distribution, and surface defect potential factors within the same production cycle time period are combined and normalized. Feature concatenation involves combining features from different data sources into a high-dimensional vector in a predetermined order. Normalization scales the values ​​of each dimension in the concatenated high-dimensional vector to a common numerical range, such as between zero and one, generating a fused state feature vector. This fused state feature vector serves as a digital summary of the current production cycle processing state and is fed into a subsequent process parameter optimization inference model. In some embodiments, the calculation of the surface defect potential factor can be based on the following relationship: ; in: Indicates the potential factor of surface defects. This indicates the area of ​​the suspected defect. Indicates the area reference value. This indicates the perimeter of the suspected defective area. Represents the shape complexity eigenvalue. , , These are the weight coefficients for the corresponding features. It can be understood that the specific values ​​of the feature weights can be learned from historical data.

[0024] In practical implementation, when calculating the temperature uniformity coefficient, in addition to using the ratio of standard deviation to mean, some embodiments may also use the entropy value or uniformity function of the temperature field. Optionally, for the calculation of the overheating risk index, a time decay factor can be introduced to gradually weaken the influence of long-standing overheating events on the current index. When calculating the equivalent deformation stress distribution from the time-series data of extruder process parameters, a simplified mechanical model can be combined with real-time process parameters for rapid estimation, rather than performing a complete finite element simulation to reduce computational latency. It is understood that the time-series alignment framework needs to handle the problem of inconsistent data acquisition frequencies from different sensors; high-frequency data requires downsampling, and low-frequency data requires interpolation. Optionally, feature normalization processing can use the min-max normalization method or the Z-score normalization method, and the parameters required for normalization are pre-calculated and stored based on the statistical characteristics of historical data.

[0025] In one embodiment of the present invention, see [reference] Figure 2 In practice, the process of analyzing the curves of extrusion speed, extrusion pressure, and die temperature over time from the extrusion press process parameter time-series data, extracting the dominant frequency characteristics and abnormal fluctuation patterns through frequency domain analysis, and calculating the equivalent deformation stress distribution inside the profile is executed by a dedicated signal processing and mechanical analysis module. The extrusion press process parameter time-series data comes from real-time data acquisition units installed on the extrusion press main drive system, pressure sensors, and die thermocouples. These data are recorded at a fixed sampling frequency and form continuous time-series curves. A Fast Fourier Transform (FFT) is performed on the extrusion speed time-series curve. The FFT converts the speed signal in the time domain to the frequency domain to analyze its frequency composition, identify the dominant frequency components and amplitudes in its spectrum. The dominant frequency components usually correspond to the cycle of the hydraulic system's main pump, the frequency of reciprocating motion, or the refresh frequency of the control system. The dominant frequency components reflect the periodic characteristics of the equipment operation, while the amplitude reflects the energy strength of the frequency component in speed fluctuations. Stable dominant frequency components usually indicate that the equipment's transmission system is operating smoothly.

[0026] The sliding window statistical method is used for the extrusion pressure time series curve. The sliding window statistical method divides the continuous extrusion pressure time series into multiple overlapping or continuous time segments, calculates the mean, standard deviation and skewness in each window. The mean reflects the average load level in the window, the standard deviation measures the degree of pressure fluctuation, and the skewness describes the asymmetry of pressure distribution. Abnormal pressure peaks and sustained high pressure ranges exceeding the preset threshold are identified. Abnormal pressure peaks may be due to defects inside the ingot or instantaneous blockage of the mold, while sustained high pressure ranges may be due to mold wear or low temperature of aluminum alloy billet leading to continuously high deformation resistance.

[0027] Trend decomposition is performed on the mold temperature time series curve. This decomposition separates the original temperature sequence into a long-term trend term that reflects slow changes, a periodic term that reflects cyclical heating and cooling, and a residual term that cannot be explained by the trend and periodicity. The long-term trend term, periodic term, and residual term are separated. Abrupt changes in the residual term indicate instantaneous anomalies in the temperature control system, such as intermittent heater malfunctions or rapid temperature fluctuations caused by sudden changes in cooling water flow. A comprehensive process stability score is constructed by combining the stability of the dominant frequency components, the frequency of abnormal pressure peaks, and the abrupt change amplitude of the mold temperature residual term. The stability of the dominant frequency components can be quantified by calculating their bandwidth or amplitude variance. The frequency of abnormal pressure peaks is the number of peaks exceeding a threshold per unit time. The abrupt change amplitude of the mold temperature residual term is a measure of the range or standard deviation in the residual sequence. The comprehensive process stability score is a scalar value obtained by weighting or combining these quantitative indicators according to rules, used to comprehensively evaluate the stability of the current process.

[0028] Based on the comprehensive score of process stability and the real-time values ​​of extrusion speed and pressure, combined with the material rheological stress model, the equivalent deformation stress distribution at various points inside the profile during extrusion is calculated. In specific implementation, the stored aluminum alloy material rheological stress model is invoked. This model exists in the form of a database or mathematical expression, describing the flow stress relationship of a specific grade of aluminum alloy at different temperatures and strain rates. The real-time value of the extrusion speed is converted into the average strain rate of the profile in the die deformation zone. The conversion process needs to consider the length of the die sizing zone and the ratio of the cross-sectional area of ​​the extrusion cylinder to the cross-sectional area of ​​the profile, i.e., the extrusion ratio. The real-time value of the die temperature is used as the approximate temperature field input of the material during deformation. This is a simplification, as the actual temperature distribution in the deformation zone is more complex, but the die temperature is an important boundary condition and approximate representation. Based on the average strain rate and the approximate temperature field, the instantaneous flow stress reference value of the material is obtained by querying or calculating the material rheological stress model. The query operation involves interpolation in the pre-stored rheological stress data table, while the calculation involves substituting into the constitutive equation of the material.

[0029] The instantaneous flow stress baseline value is corrected based on the comprehensive process stability score. The logic behind this correction is that a lower comprehensive process stability score indicates greater process fluctuations, which can lead to non-uniformity in material deformation and stress concentration. Therefore, a dynamic fluctuation component negatively correlated with the score is added to the flow stress. In some embodiments, the corrected flow stress... It can be calculated as follows: ; in: This represents the corrected flow stress. This represents the instantaneous flow stress reference value obtained through querying or calculation. This represents the calculated overall score for process stability. This represents the theoretical maximum or set upper limit of the overall process stability score. This is a positive adjustment coefficient used to control the sensitivity of process fluctuations to stress effects. Based on the profile cross-sectional shape, die geometry, and friction conditions, a simplified flow simulation is performed using the finite volume method. The finite volume method discretizes the profile extrusion deformation region into multiple control volumes, solving for the mass and momentum conservation equations. The corrected flow stress is then distributed to each computational unit within the profile. This distribution process considers the strain rate and temperature conditions of each unit, resulting in an equivalent deformation stress distribution. The final output of the equivalent deformation stress distribution is a two-dimensional or three-dimensional stress field, characterizing the stress variation along the profile cross-section or length. Optionally, in simplified flow simulations, the material can be assumed to be an incompressible viscoplastic body to reduce computational complexity. It is understood that the calculation results of the equivalent deformation stress distribution are used for subsequent assessment of residual stress risk within the product or as input to the optimization model as part of the state characteristics. In some embodiments, the friction conditions can be simplified to a Coulomb friction model or a shear friction model.

[0030] In one embodiment of the present invention, in a specific implementation, the process of inputting the fused state feature vector into a preset process parameter optimization inference model and outputting a set of key process parameter adjustment suggestions for the next production cycle is completed by an inference engine deployed on an industrial server. The process parameter optimization inference model is an artificial intelligence model trained on historical production data, and its architecture includes two cascaded subnetworks, namely a quality prediction subnetwork and a parameter optimization subnetwork.

[0031] The process parameter optimization inference model consists of a quality prediction subnetwork and a parameter optimization subnetwork. The quality prediction subnetwork is responsible for predicting the future output quality based on the current processing state, while the parameter optimization subnetwork is responsible for finding the optimal process parameter adjustment scheme based on the quality prediction deviation. The fused state feature vector is input into the quality prediction subnetwork, which is a deep neural network whose input layer node number is equal to the dimension of the fused state feature vector. The model predicts the key quality indicators of the profile produced in the next production cycle under the current process parameter conditions. The prediction uses the fused state feature vector of the previous production cycle as input, assuming the process parameters remain unchanged, to infer the quality state the profile will reach at the end of the next production cycle. The predicted key quality indicators include dimensional tolerances, mechanical properties, and surface quality scores. Dimensional tolerance prediction may involve the deviation range of key dimensions of the profile cross-section; mechanical property prediction may include estimated values ​​of tensile strength, yield strength, or elongation; and the surface quality score is a comprehensive evaluation value that integrates appearance factors such as surface roughness, scratches, and color difference.

[0032] The predicted values ​​of key quality indicators are compared with the preset quality target values. The preset quality target values ​​are set according to the product specifications or customer standards and are the target range or specific values ​​that each type of key quality indicator must achieve. The deviation of each quality dimension is calculated. The deviation is the difference between the predicted value of the key quality indicator and the corresponding quality target value. Since different quality indicators have different physical dimensions, the original deviation needs to be normalized to eliminate the influence of dimensions and obtain the normalized relative deviation of each dimension.

[0033] The normalized deviation and the current fusion state feature vector are input into the parameter optimization sub-network. The input to the parameter optimization sub-network includes the normalized quality prediction deviation and the fusion state feature vector reflecting the current production state. This allows the parameter optimization sub-network to perform targeted parameter optimization based on an understanding of the current operating conditions. Within the process parameter space comprised of melting temperature, extrusion speed, and die preheating temperature, the parameter optimization sub-network performs a multi-objective optimization search with the goal of minimizing the normalized deviation. The process parameter space defines the allowable adjustment range for each adjustable process parameter. The multi-objective optimization search needs to simultaneously consider the comprehensive minimization of normalized deviations across multiple quality dimensions. The output is a set of key process parameter adjustment suggestions that makes the predicted quality closest to the target value. This set of key process parameter adjustment suggestions represents one or more Pareto optimal solutions found during the optimization search process. The set of key process parameter adjustment suggestions includes the suggested adjustment direction and magnitude for each parameter; for example, it suggests increasing the melting temperature by a specific degree Celsius, decreasing the extrusion speed by a specific percentage, and adjusting the die preheating temperature to a specific value. In practical implementation, the quality prediction subnetwork can employ a multilayer perceptron or long short-term memory network structure to capture the complex nonlinear relationship or temporal dependency between state features and quality results. It can be understood that the quality target value can be a fixed value or a variable value dynamically set according to production orders. The formula for calculating the normalized relative deviation of each quality dimension can be used: ; in: Indicates the first The normalized relative deviation of each quality indicator The output of the quality prediction subnetwork represents the first... Predicted values ​​for key quality indicators Indicates the first The preset quality target values ​​corresponding to each key quality indicator. When When the value is zero or close to zero, other normalized benchmarks can be used. When the parameter optimization subnetwork performs multi-objective optimization search, it can be understood that the search algorithm can be a gradient-based optimization method, or a gradient-free optimization method based on evolutionary algorithms or reinforcement learning. In some embodiments, the objective function of the multi-objective optimization search is... It can be constructed as a weighted sum of squares of the normalized relative deviations in each dimension: ; in: This represents the value of the objective function that needs to be minimized. Indicates the quantity of key quality indicators. Indicates the first The normalized relative deviation of each quality indicator Indicates the first Weighting coefficients for the normalized bias of each quality dimension. Optional, weighting coefficients. The configuration can be tailored to the quality requirements of different products. In some embodiments, the set of key process parameter adjustment suggestions can include multiple alternatives, along with the predicted quality results for each option, for operators to use as a reference for final decision-making. Optionally, a simplified forward quality prediction model can be integrated within the parameter optimization subnetwork to quickly evaluate the predicted quality under different parameter combinations, thereby accelerating the optimization process.

[0034] See Figure 3 This is a graph showing the adjustment range of five process parameter adjustment schemes in aluminum alloy profile production and their corresponding quality improvement effects. Directly linking the adjustment range of the five process parameters with the quality improvement effect allows production decision-makers to quickly identify scheme 5 as the optimal choice, while also clarifying that the improvement effect of scheme 4 is insufficient and an alternative scheme needs to be initiated. It clearly demonstrates the positive correlation between the "process parameter adjustment range and the quality improvement rate," providing a quantitative basis for subsequent parameter optimization and reducing the blindness of experience-based decisions. A standardized evaluation system for the process adjustment effect is established through a unified normalized quality improvement rate index, facilitating horizontal comparisons between different batches and different schemes. Real-time monitoring of the quality improvement rate threshold can provide early warnings of schemes with insufficient effects, avoiding the generation of defective products and reducing production risks and costs.

[0035] In one embodiment of the present invention, in a specific implementation, the process of generating an executable sequence of process parameter adjustment instructions based on a set of key process parameter adjustment suggestions and the constraints of the current equipment operating status is executed by the instruction planning and verification module. The module obtains the constraints of the current equipment operating status, including the maximum heating rate of the melting furnace, the maximum pressure limit of the extruder, the power limit of the die heater, and the response delay time of each actuator. These constraints are derived from the equipment technical specifications, safety operating procedures, and actual dynamic characteristics identified by the system, and are stored in the control system in the form of configuration files or a database. Each suggested adjustment range in the set of key process parameter adjustment suggestions is compared and verified with the corresponding equipment constraints. The comparison and verification involves comparing the suggested adjustment value or amount with the allowable operating range of the equipment one by one. For suggested adjustments that do not exceed the equipment constraints, they are directly converted into standard control instructions recognizable by the equipment control interface. The conversion process transforms the adjustment amount into specific set values ​​and writes them into instructions according to different equipment control protocols. For suggested adjustments that exceed the equipment constraints, a constraint processing strategy is initiated. The constraint processing strategy aims to find an executable solution that is as close as possible to the original optimization suggestion while ensuring safety.

[0036] Within the permissible constraints, the adjustment range is proportionally reduced to a feasible range, or alternative parameter adjustment schemes that can partially compensate for the target effect are sought based on the coupling relationship between process parameters. When the suggested adjustment range of a parameter exceeds its maximum permissible adjustment range of the equipment, the maximum achievable adjustment range of the parameter is calculated. The maximum achievable adjustment range is determined by the difference between the current setpoint of the equipment and the permissible limit value of the equipment. The ratio of the suggested adjustment range to the maximum achievable adjustment range is calculated, and all suggested adjustment ranges exceeding the constraints are globally and proportionally reduced according to this ratio until all adjustments are within the constraints. Global proportional reduction means multiplying all parameter adjustment ranges exceeding the constraints by a common scaling factor less than 1. scaling factor The most stringent constraints determine that the scaled adjustment amount exactly satisfies all constraints. For example, for a suggested adjustment set containing three adjustable parameters, the equipment constraints and adjustment verification process are shown in Table 1.

[0037] Table 1 Equipment Constraints and Suggested Adjustments and Verifications: When the reduction in adjustment results in the expected quality improvement falling below a threshold, an alternative solution search is initiated. The expected quality improvement is estimated by rapidly forward-engineering the parameter optimization sub-network of the process parameter optimization inference model after re-inputting the reduced adjustment amount. The alternative solution search is based on a process parameter coupling knowledge base, which records the influence weights of different process parameter combinations on the final quality indicators. This knowledge base is constructed by analyzing historical production data or using physics-based process models, representing the interactive effects between parameters in the form of matrices or rule sets. Other coupled process parameters are adjusted to compensate for the quality improvement lost due to the limitation on the main parameter adjustment, provided that other constraints are not violated. This forms new parameter adjustment combinations as alternative parameter adjustment schemes. For example, when the extrusion speed cannot be increased to the recommended value due to equipment limitations, the die temperature can be moderately increased to partially compensate for the impact on the profile surface quality.

[0038] In practical implementation, response latency constraints require inserting necessary waiting time between operational steps that require device stability when orchestrating instruction sequences. It can be understood that global scaling is a conservative but safe strategy, ensuring that all adjustments are synchronized and scaled down proportionally. In some embodiments, the scaling factor... The calculation must consider the adjustment direction simultaneously; for positive and negative adjustments, their feasible space relative to the upper and lower limits must be calculated separately. Optionally, the threshold can be dynamically set according to the tolerance of the product quality level. Alternative solution search can be formalized as a quadratic optimization problem under constraints, with the goal of minimizing the deviation between the final predicted quality and the original optimization objective by adjusting secondary parameters when the adjustment of the primary parameters is limited. All verified and processed individual control instructions are sorted and combined according to the logical order and time dependency of the production process flow, and necessary instruction intervals are added to form an executable sequence of process parameter adjustment instructions. The instruction interval is used to ensure that the equipment state change caused by the previous instruction has basically stabilized before executing the next instruction. In some embodiments, the instruction sequence can be distributed to the distributed control system in the form of a script file or instruction list. Optionally, for adjustments that do not exceed the constraints, the instruction generation timestamp can be accurately calculated based on the production cycle time and response delay.

[0039] See Figure 4This is a diagram illustrating the changes in different process parameters during aluminum alloy profile production. It clearly presents the complete process: "parameter optimization sub-network outputs suggested values ​​→ system checks constraints → generates executable adjusted values," verifying the closed-loop effectiveness of the process optimization model in the patent. By comparing the three types of values, production personnel can quickly confirm the feasibility of the adjustment range without manually checking equipment constraints, reducing decision-making time and the risk of errors. The complete overlap between the "suggested values" and "adjusted values" of the two types of parameters indicates that the optimization suggestion did not trigger the constraint handling strategy, reflecting sufficient adjustment space for process parameters under the current production conditions. Through intuitive numerical comparison, production personnel can quickly confirm the rationality of the adjustment range without manually checking the equipment parameter manual, significantly improving decision-making efficiency. This diagram clearly records the adjustment process of key parameters and can serve as a standardized process document for production batches, facilitating subsequent traceability and review.

[0040] In one embodiment of the present invention, the process of executing a sequence of process parameter adjustment instructions and simultaneously collecting adjusted feedback data to update the process parameter optimization inference model constitutes a closed-loop learning link in the optimization control. The process parameter adjustment instruction sequence is executed through the distributed control system of the production line. The distributed control system parses the received instruction sequence and transforms it into specific operation commands for the underlying equipment controllers, issuing adjustment commands to the furnace temperature controller, the extruder main drive system, and the die temperature control system. These commands drive the actuators to change their setpoints or operating modes to achieve process parameter adjustment. In the next complete production cycle after the instruction execution, new multi-source process data is simultaneously collected as adjusted feedback data. Simultaneous collection means that the triggering of data collection is strictly synchronized with the start and end of the production cycle. The new multi-source process data includes furnace temperature distribution data, extruder process parameter time-series data, and profile surface online monitoring image data from the same sources as before the adjustment. The adjusted feedback data undergoes the same collaborative feature extraction and fusion processing to generate a new fused state feature vector, which is recorded as the adjusted state feature vector. The adjusted state feature vector characterizes the actual state during processing after the process parameter adjustment.

[0041] The actual quality inspection data of the profile samples produced at the end of the production cycle are collected. The profile samples are taken from the production line according to the preset sampling frequency. The actual quality inspection data includes measured dimensions, mechanical property test results and surface quality assessment report. The measured dimensions are obtained by offline measuring tools such as calipers and profilometers. The mechanical property test results are obtained by testing standard specimens with a tensile testing machine. The surface quality assessment report is generated by quality inspectors or an automated rating system based on visual inspection standards.

[0042] A new training sample is created by combining the pre-execution fusion state feature vector, the executed process parameter adjustment command sequence, the post-adjustment state feature vector, and actual quality inspection data. The pre-execution fusion state feature vector records the state before adjustment, the process parameter adjustment command sequence records the control action taken, and the post-adjustment state feature vector and actual quality inspection data together reflect the state transition and final result caused by the control action. The new training sample completely encapsulates a causal instance of a decision and feedback. The new training sample is added to the historical training dataset of the process parameter optimization inference model, triggering the model's incremental learning process to update the model's internal parameters. The historical training dataset is a structured database used to store all historical training samples. The incremental learning process is designed to enable the model to continuously optimize its decision-making performance using newly generated production data without requiring a complete retraining. In specific implementation, the incremental learning process involves using the pre-execution fusion state feature vector and the process parameter adjustment command sequence from the new training sample as input features. These input features represent the state upon which the model bases its decision and the output action. The target quality improvement, calculated by fusing the adjusted state feature vector from the new training samples with the actual quality detection data, serves as the supervision label for the incremental learning process. This target quality improvement quantifies the quality change brought about by the adjustment, and its calculation can be based on a comparison between the actual quality detection data and the product quality benchmark before adjustment. In some embodiments, the target quality improvement... It can be defined as an actual quality index vector. With reference quality index vector The sum of weighted improvement scores across all dimensions: ; in: Indicates the target quality improvement amount. Indicates the quantity of quality indicators. Indicates the first The actual measured values ​​of each quality indicator This indicates the corresponding reference value (such as the quality value of the previous cycle or the minimum acceptable standard value). It is the first The weighting coefficients of each quality indicator, It is a function used to calculate a score for the improvement of a single indicator value relative to its reference value. The scoring rule can be defined based on the indicator characteristics as higher is better, lower is better, or closer to the target value is better. This can be understood as a reference quality indicator vector. The specific value can be set according to the update strategy.

[0043] A temporary local training set is formed by retrieving several historical samples from the historical training dataset that are most similar to the current new sample's process scenario. The similarity metric can be based on the Euclidean distance or cosine similarity of the fused state feature vectors before execution. This temporary local training set, along with the new training samples, is used to perform a round of local incremental training on the process parameter optimization inference model, particularly the parameter optimization sub-network. The goal of the training is to enable the model to learn the decision-making experience represented by the new samples. In the local incremental training, a small learning rate is used to update only some network layer parameters to avoid catastrophic forgetting of learned knowledge; for example, only the weights of the last few fully connected layers of the parameter optimization sub-network can be updated. After training, the new training samples are formally stored in the historical training dataset, completing the iterative accumulation of model knowledge. In some embodiments, triggering the incremental learning process can be set to execute automatically and periodically or after accumulating a certain number of new training samples. Optionally, the number of similar historical samples retrieved can be set according to computational resources. It is understood that the optimizer used for local incremental training can be stochastic gradient descent or its variants.

[0044] See Figure 5 This is a bar chart comparing the aluminum alloy profile manufacturing process before and after adjustments. All four feature dimensions showed significant improvements after the process adjustments, validating the patent's technical logic that "the processing state is significantly optimized after executing the process parameter adjustment command." The surface quality feature improved by nearly 100%, reflecting the most significant effect of the process adjustment in suppressing surface defects in the profile. The adjusted process stability feature value reached 0.9, perfectly matching the target value, indicating that fluctuations in the production process have been effectively controlled. The deformation stress optimization degree still has a gap of 0.08 from the target value, which can serve as a key optimization direction for subsequent process iterations. The significant improvements in each feature dimension prove that the adjustment suggestions output by the process parameter optimization inference model have practical value, rather than being merely theoretical optimizations. By comparing feature values ​​with target values, dimensions that fail to meet the standards can be identified in advance, avoiding final product quality problems caused by processing state defects.

[0045] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for optimizing and controlling the production and processing of aluminum alloy profiles, characterized in that, Includes the following steps: Real-time acquisition of multi-source process data in the aluminum alloy profile production and processing flow, including furnace temperature distribution data, extrusion press process parameter timing data, and profile surface online monitoring image data; The multi-source process data is subjected to collaborative feature extraction and fusion processing to generate a fused state feature vector characterizing the current processing state. The fused state feature vector includes material uniformity index, deformation stress distribution characteristics and surface defect potential factors. The fusion state feature vector is input into a preset process parameter optimization inference model, and the process parameter optimization inference model outputs a set of key process parameter adjustment suggestions for the next production cycle. Based on the set of key process parameter adjustment suggestions and the constraints of the current equipment operating status, an executable sequence of process parameter adjustment instructions is generated. The process parameter adjustment instruction sequence is executed, and the adjusted feedback data is collected synchronously to update the process parameter optimization inference model.

2. The method for optimizing and controlling the production and processing of aluminum alloy profiles according to claim 1, characterized in that, The step of performing collaborative feature extraction and fusion processing on the multi-source process data to generate a fused state feature vector representing the current processing state includes: From the temperature distribution data of the smelting furnace, the temperature values ​​and temperature gradient change rates of multiple key temperature measuring points are extracted, and the temperature uniformity coefficient of the entire molten pool and the overheating risk index of a specific area are calculated. From the time-series data of the extruder process parameters, the curves of extrusion speed, extrusion pressure and die temperature changing with time are analyzed. The main frequency characteristics and abnormal fluctuation modes are extracted through frequency domain analysis, and the equivalent deformation stress distribution inside the profile is calculated. Multi-scale texture analysis and defect region segmentation are performed on the online monitoring image data of the profile surface to extract the quantitative values ​​of surface roughness, area, perimeter and shape complexity features of suspected defect regions, and calculate the potential factor of surface defects. A time-series alignment framework based on production cycle time is established, mapping the temperature uniformity coefficient, overheating risk index, equivalent deformation stress distribution, surface roughness quantification, and defect region characteristics onto a unified time axis. The temperature uniformity coefficient, overheating risk index, equivalent deformation stress distribution, and surface defect potential factors within the same production cycle time period are spliced ​​and normalized to generate the fused state feature vector.

3. The method for optimizing and controlling the production and processing of aluminum alloy profiles according to claim 2, characterized in that, The process involves analyzing the time-series data of the extrusion press process parameters, including the curves of extrusion speed, extrusion pressure, and die temperature changing over time. Frequency domain analysis is used to extract the dominant frequency characteristics and abnormal fluctuation patterns. The equivalent deformation stress distribution inside the profile is then calculated. A fast Fourier transform is performed on the extrusion speed time-series curve to identify the dominant frequency components and amplitudes in its spectrum. The dominant frequency components reflect the periodic characteristics of the equipment operation. A sliding window statistical method is used to analyze the extrusion pressure time series curve, calculate the mean, standard deviation and skewness within each window, and identify abnormal pressure peaks and sustained high pressure ranges that exceed preset thresholds. The temperature time series curve of the mold is decomposed to separate the long-term trend term, the periodic term and the residual term. The abrupt change point in the residual term indicates the instantaneous anomaly of the temperature control system. A comprehensive process stability score is constructed by combining the stability of the dominant frequency components, the frequency of occurrence of the abnormal pressure peaks, and the abrupt change amplitude of the mold temperature residual term. Based on the comprehensive score of process stability and the real-time values ​​of extrusion speed and extrusion pressure, combined with the material rheological stress model, the equivalent deformation stress distribution at each point inside the profile during the extrusion process is calculated.

4. The method for optimizing and controlling the production and processing of aluminum alloy profiles according to claim 1, characterized in that, The process parameter optimization inference model inputs the fused state feature vector into a preset process parameter optimization model, and the process parameter optimization inference model outputs a set of key process parameter adjustment suggestions for the next production cycle, including: The process parameter optimization reasoning model consists of a quality prediction subnetwork and a parameter optimization subnetwork. The fusion state feature vector is input into the quality prediction sub-network to predict the key quality indicators of the profile produced in the next production cycle under the current process parameter continuation conditions. The key quality indicators include dimensional tolerances, mechanical properties and surface quality scores. The predicted values ​​of the key quality indicators are compared with the preset quality target values, and the deviation of each quality dimension is calculated. The deviation and the current fusion state feature vector are input together into the parameter optimization subnetwork; The parameter optimization subnetwork performs a multi-objective optimization search in the process parameter space consisting of melting temperature, extrusion speed, and die preheating temperature, with the goal of minimizing the deviation, and outputs a set of key process parameter adjustment suggestions that make the predicted quality closest to the target value. The set of key process parameter adjustment suggestions includes the suggested adjustment direction and magnitude for each parameter.

5. The method for optimizing and controlling the production and processing of aluminum alloy profiles according to claim 4, characterized in that, The step of generating an executable sequence of process parameter adjustment instructions based on the set of key process parameter adjustment suggestions and the constraints of the current equipment operating status includes: Obtain the constraints of the current equipment operating status, including the maximum heating rate of the melting furnace, the maximum pressure limit of the extruder, the power limit of the die heater, and the response delay time of each actuator; Each suggested adjustment range in the set of key process parameter adjustment suggestions is compared and verified with the corresponding equipment constraints. For suggested adjustments that do not exceed the equipment constraints, they will be directly converted into standard control commands that can be recognized by the equipment control interface. For suggested adjustments that exceed equipment constraints, initiate constraint handling strategies: within the allowable constraint boundaries, proportionally reduce the adjustment range to a feasible range, or based on the coupling relationship between process parameters, find alternative parameter adjustment schemes that can partially compensate for the target effect; All verified and processed individual control commands are sorted and combined according to the logical order and time dependency of the production process flow, and necessary command intervals are added to form the executable process parameter adjustment command sequence.

6. The method for optimizing and controlling the production and processing of aluminum alloy profiles according to claim 1, characterized in that, The execution of the process parameter adjustment instruction sequence, and the simultaneous collection of adjusted feedback data for updating the process parameter optimization inference model, includes: The distributed control system of the production line executes the process parameter adjustment command sequence, and issues adjustment commands to the melting furnace temperature controller, the extruder main drive system and the die temperature control system. In the next complete production cycle after the instruction is executed, new multi-source process data is collected synchronously as adjusted feedback data. The adjusted feedback data is subjected to the same collaborative feature extraction and fusion process to generate a new fused state feature vector, denoted as the adjusted state feature vector. Collect actual quality inspection data of the profile samples finally produced in the production cycle, including measured dimensions, mechanical property test results and surface quality assessment report; The fusion state feature vector before execution, the sequence of process parameter adjustment instructions executed, the state feature vector after adjustment, and the actual quality detection data are combined into a new training sample. The new training samples are added to the historical training dataset of the process parameter optimization inference model, and the incremental learning process of the model is triggered to update the model's internal parameters.

7. The method for optimizing and controlling the production and processing of aluminum alloy profiles according to claim 3, characterized in that, The process involves calculating the equivalent deformation stress distribution at various points inside the profile during extrusion based on the comprehensive score of process stability, the real-time values ​​of extrusion speed and extrusion pressure, and the material rheological stress model. The stored aluminum alloy material rheological stress model is invoked, which describes the flow stress relationship of aluminum alloy at different temperatures and strain rates. The real-time value of the extrusion speed is converted into the average strain rate of the profile in the die deformation zone; The real-time value of the mold temperature is used as an approximate temperature field input for the material during the deformation process; Based on the average strain rate and the approximate temperature field, the instantaneous flow stress reference value of the material is obtained by querying or calculating the material rheological stress model; The instantaneous flow stress benchmark value is corrected based on the comprehensive process stability score. The lower the comprehensive process stability score, the greater the fluctuation. A dynamic fluctuation component negatively correlated with the score is added to the flow stress. Based on the profile cross-sectional shape, mold geometry and friction conditions, a simplified flow simulation is performed using the finite volume method. The corrected flow stress is then distributed to each calculation unit inside the profile, thereby obtaining the equivalent deformation stress distribution.

8. The method for optimizing and controlling the production and processing of aluminum alloy profiles according to claim 5, characterized in that, For suggested adjustments exceeding equipment constraints, a constraint handling strategy is initiated: within the allowable constraint boundaries, the adjustment range is proportionally reduced to a feasible range, or, based on the coupling relationship between process parameters, alternative parameter adjustment schemes that can partially compensate for the target effect are sought, including: When the suggested adjustment range of a certain parameter exceeds the maximum allowable adjustment range of the device, the maximum adjustment range that the parameter can achieve is calculated. Calculate the ratio of the suggested adjustment range to the maximum achievable adjustment range, and then globally and proportionally reduce all suggested adjustment ranges that exceed the constraints until all adjustments are within the constraints. When reducing the adjustment magnitude results in the expected quality improvement effect falling below the threshold, an alternative solution search is initiated. The alternative search is based on a process parameter coupling knowledge base, which records the influence weights of different combinations of process parameters on the final quality indicators. Without violating other constraints, other coupled process parameters are adjusted to compensate for the quality improvement effect lost due to the limitation of the main parameter adjustment, forming a new parameter adjustment combination as an alternative parameter adjustment scheme.

9. The method for optimizing and controlling the production and processing of aluminum alloy profiles according to claim 6, characterized in that, The step of adding the new training samples to the historical training dataset of the process parameter optimization inference model and triggering the incremental learning process of the model to update the model's internal parameters includes: The fusion state feature vector and the process parameter adjustment instruction sequence in the new training samples before execution are used as input features for the incremental learning process. The target quality improvement obtained by fusing the adjusted state feature vector in the new training samples and the actual quality detection data is used as the supervision label for the incremental learning process. Retrieve several historical samples from the historical training dataset that are most similar to the current new sample process scenario to form a temporary local training set; Using the temporary local training set along with the new training samples, a round of local incremental training is performed on the process parameter optimization inference model, especially the parameter optimization sub-network. In local incremental training, a small learning rate is used to update only some network layer parameters of the model in order to avoid catastrophic forgetting of learned knowledge; After training is completed, the new training samples are officially stored in the historical training dataset.

10. A production and processing optimization control system for aluminum alloy profiles, characterized in that, It includes a processor and a memory, the memory storing a computer program, and the processor executing the computer program to implement the aluminum alloy profile production and processing optimization control method according to any one of claims 1 to 9.

Images