Aluminum profile cutting parameter self-adaptive regulation and control method and system based on machine learning
By using machine learning technology to collect and analyze vibration characteristic parameters in real time during the aluminum profile cutting process, dynamic matching degree and defect prediction are performed to optimize the cutting strategy, which solves the problem of unstable aluminum profile cutting quality and improves processing efficiency and equipment life.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-02-10
- Publication Date
- 2026-03-27
AI Technical Summary
Existing aluminum profile cutting technology lacks the ability to dynamically adapt to the characteristics of the profile and real-time working conditions, resulting in unstable cutting quality, defects such as burrs, deformation and cracks, and severe equipment wear.
An adaptive control method for aluminum profile cutting parameters based on machine learning is adopted. By collecting vibration characteristic parameter sequences in real time, performing verification analysis and matching degree calculation, and combining defect prediction and optimization strategies, adaptive control of cutting parameters is achieved.
It enables early identification and accurate warning of dynamic anomalies during the cutting process, improves cut quality, reduces scrap rate and equipment wear, and enhances the intelligence of the processing system.
Smart Images
Figure CN121744100A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of aluminum profile processing technology, specifically to an adaptive control method and system for aluminum profile cutting parameters based on machine learning. Background Technology
[0002] Currently, aluminum profiles are widely used in building doors and windows, curtain walls, and other fields. Their cutting quality directly affects the assembly accuracy and final performance of the products. Especially in the customized production of system doors and windows with high requirements for sound insulation and airtightness, the cutting of multi-cavity, multi-specification aluminum profiles faces greater challenges. The dynamic characteristics such as vibration and deformation during the cutting process are closely related to the profile material, structure, and equipment operating conditions. Complex and variable processing conditions require more precise and adaptive control of the cutting process to ensure processing quality and efficiency.
[0003] In existing technologies, aluminum profile cutting largely relies on fixed parameters preset based on experience or simple feedback adjustments. It generally lacks a comprehensive consideration of the profile's structural characteristics and real-time operating conditions. It often fails to identify abnormal vibrations and deformations caused by internal material defects, structural stress, or changes in equipment status during the cutting process, leading to defects such as burrs, deformation, and even cracks on the cut surface. This not only reduces product yield and increases material waste but also exacerbates wear on critical components such as saw blades. Although some research has attempted to introduce sensors for process monitoring, most remain at the stage of data acquisition and simple threshold alarms. They have failed to establish a dynamic mapping relationship between cutting characteristics and optimal parameters, and lack intelligent decision-making and parameter adaptive optimization capabilities based on real-time feedback. This hinders the in-depth development of aluminum profile processing towards digitalization and intelligence. Summary of the Invention
[0004] This invention addresses the technical problem that existing aluminum profile cutting processes rely on fixed parameters and lack dynamic adaptability to profile characteristics and real-time working conditions, by providing a machine learning-based adaptive control method and system for aluminum profile cutting parameters.
[0005] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: In a first aspect, the present invention provides an adaptive control method for aluminum profile cutting parameters based on machine learning, including: Obtain the profile type of the aluminum profile, configure the corresponding basic cutting strategy for cutting, and collect the vibration characteristic parameter sequence during the cutting process; Obtain the benchmark vibration characteristic parameter sequence corresponding to the basic cutting strategy, perform verification analysis on the vibration characteristic parameter sequence to obtain the verification matching degree, and calculate the verification deviation degree; When the verification matching degree is less than the matching degree threshold, the cutting progress coefficient is obtained to predict aluminum profile defects and obtain aluminum profile defect characteristics. Based on the defect characteristics of the aluminum profile, the basic cutting strategy is optimized to obtain an optimized cutting strategy, and cutting control is performed. In this process, divergent optimization parameters are configured according to the verification deviation.
[0006] Secondly, the present invention provides an adaptive control system for aluminum profile cutting parameters based on machine learning, comprising: The feature acquisition module is used to obtain the profile type of aluminum profile, configure the corresponding basic cutting strategy for cutting, and collect the vibration feature parameter sequence during the cutting process. The verification analysis module is used to obtain the benchmark vibration characteristic parameter sequence corresponding to the basic cutting strategy, perform verification analysis on the vibration characteristic parameter sequence, obtain the verification matching degree, and calculate the verification deviation degree. The defect prediction module is used to obtain the cutting progress coefficient when the verification matching degree is less than the matching degree threshold, to predict aluminum profile defects and obtain aluminum profile defect characteristics. The cutting control module is used to optimize the basic cutting strategy based on the defect characteristics of the aluminum profile, obtain an optimized cutting strategy, and perform cutting control, wherein divergent optimization parameters are configured according to the verification deviation.
[0007] The beneficial effects of this invention are: Compared to existing technologies, this invention firstly introduces a real-time acquisition and verification analysis mechanism for vibration characteristic parameter sequences, enabling accurate capture of dynamic anomalies during the cutting process. Secondly, a dual-threshold triggering mechanism is constructed based on verification matching degree and verification deviation degree, achieving early warning and accurate identification of potential defects. Thirdly, by combining the cutting progress coefficient with a machine learning model for aluminum profile defect prediction, the prediction results not only include defect categories but also quantify defect scale, providing precise input for subsequent strategy optimization. Finally, iterative optimization of the cutting strategy is achieved using divergent optimization parameters dynamically configured based on verification deviation degree, realizing adaptive matching between cutting parameters and real-time working conditions and individual profile characteristics, thereby improving cut quality and reducing scrap rate and equipment wear. Attached Figure Description
[0008] Figure 1 A flowchart illustrating the adaptive control method for aluminum profile cutting parameters based on machine learning provided by this invention. Figure 2 This is a schematic diagram of the adaptive control system for aluminum profile cutting parameters based on machine learning provided by the present invention.
[0009] In the attached diagram, the components represented by each number are as follows: Feature acquisition module 11, verification and analysis module 12, defect prediction module 13, cutting and control module 14. Detailed Implementation
[0010] Example 1, as Figure 1 As shown, this embodiment of the invention provides an adaptive control method for aluminum profile cutting parameters based on machine learning, including: S10: Obtain the profile type of the aluminum profile, configure the corresponding basic cutting strategy for cutting, and collect the vibration characteristic parameter sequence during the cutting process; First, identify the aluminum profile type and configure the corresponding basic cutting strategy. Different aluminum profile types differ in structural characteristics, number of cavities, wall thickness, and material hardness, directly affecting the stress state and dynamic response during cutting. Therefore, the basic cutting strategy must be tailored to avoid quality issues such as cut deformation and increased burrs due to parameter mismatch. Configuring the corresponding basic cutting strategy provides initial processing parameters that conform to the general characteristics of the current profile.
[0011] Simultaneously, a sequence of vibration characteristic parameters is collected during the cutting process. This sequence of vibration characteristic parameters is a time-series signal that reflects the dynamic process of the interaction between the saw blade and the material. It represents the comprehensive response characteristics of the cutting system in the time and frequency domains and is used to characterize the stability, energy distribution, and potential abnormal states of the cutting process.
[0012] Specifically, the profile type of the aluminum profile is obtained, the corresponding basic cutting strategy is configured for cutting, and the vibration characteristic parameter sequence during the cutting process is collected, including: Obtain the profile category of the aluminum profile; Input the profile type into the cutting strategy database, index the corresponding basic cutting strategy, and control the cutting of the aluminum profile; During the cutting process, vibration characteristic parameters are collected to obtain a sequence of vibration characteristic parameters.
[0013] First, determine the profile category of the aluminum profile to be processed. This category is a classification based on the key structural, dimensional, and material characteristics of the aluminum profile. Each profile category is associated with a series of specific characteristic parameters, such as typical cross-sectional shape, specific number of cavities, standardized wall thickness range, specific alloy grade, and conventional surface treatment conditions. These characteristic parameters collectively determine the mechanical behavior and thermal response of the aluminum profile during the cutting process.
[0014] Secondly, the acquired profile categories are used as query criteria and input into the cutting strategy database for retrieval. This database stores a set of initial processing parameters, validated through historical processes, for different profile categories—the basic cutting strategies. The retrieval process indexes the profile categories to obtain the corresponding basic cutting strategies. These strategies include initial key parameters adapted to the general characteristics of this type of aluminum profile, such as the initial saw blade speed setting. For example, for a multi-cavity door and window profile of model 6063-T5, the basic cutting strategy might include an initial saw blade speed setting of 3000 rpm. After obtaining this basic cutting strategy, it serves as the initial instruction to control the cutting equipment to perform the cutting operation on the aluminum profile.
[0015] Simultaneously, during the cutting process based on the fundamental cutting strategy, physical signals reflecting the cutting dynamics are acquired in real time, forming a sequence of vibration characteristic parameters arranged chronologically. Specifically, the acquired vibration characteristic parameters include, but are not limited to, vibration acceleration. Vibration acceleration is the rate of change of vibration velocity per unit time, reflecting the instantaneous state of the interaction between the saw blade and the material. Under normal cutting conditions, the vibration acceleration remains within a certain characteristic range. However, if cutting anomalies occur due to internal material defects, stress concentration, or parameter mismatch, the amplitude or spectral characteristics of the vibration acceleration will deviate significantly. Therefore, this sequence of vibration characteristic parameters provides a direct quantitative characterization of the real-time state of the cutting process, specifically its stability and potential anomalies, offering crucial process data for subsequent analysis and decision-making.
[0016] S20: Obtain the benchmark vibration characteristic parameter sequence corresponding to the basic cutting strategy, perform verification analysis on the vibration characteristic parameter sequence, obtain the verification matching degree, and calculate the verification deviation degree; Secondly, the reference vibration characteristic parameter sequence corresponding to the basic cutting strategy is obtained. This reference vibration characteristic parameter sequence refers to the standard vibration characteristic sequence collected and statistically summarized during the historical processing of the same type of aluminum profile and when the same basic cutting strategy is used for qualified cutting. It is a statistically significant process quality reference benchmark, which characterizes the dynamic characteristics that the profile cutting process should exhibit under ideal or normal working conditions.
[0017] The aforementioned real-time vibration characteristic parameter sequence is verified and analyzed based on the benchmark vibration characteristic parameter sequence. That is, the similarity distance or correlation between the real-time sequence and the benchmark sequence in the multidimensional feature space is calculated by mathematical methods to obtain the verification matching degree and verification deviation degree.
[0018] Specifically, the verification matching degree is used to quantify the consistency between the real-time cutting process dynamics and the historical qualified baseline state, while the verification deviation degree is used to measure the magnitude and direction of the difference between the two. Through the above calculation process, a quantitative assessment of the current cutting process state can be established, providing accurate data criteria for judging whether the process is under control, whether anomalies have occurred, and whether subsequent prediction and optimization steps need to be initiated.
[0019] Specifically, the baseline vibration characteristic parameter sequence corresponding to the basic cutting strategy is obtained, the vibration characteristic parameter sequence is verified and analyzed to obtain the verification matching degree, and the verification deviation degree is calculated, including: Obtain historical cutting vibration records of aluminum profiles of the same type, extract the average vibration characteristic parameter sequence of qualified cutting processes, and obtain the benchmark vibration characteristic parameter sequence. Extract and compare the reference vibration feature parameter sequence within the reference vibration feature parameter sequence; The similarity between the vibration characteristic parameter sequence and the comparison benchmark vibration characteristic parameter sequence is calculated to obtain the verification matching degree; The verification deviation is calculated based on the verification matching degree.
[0020] First, obtain historical cutting vibration records for the same type of aluminum profiles. The historical period refers to a pre-defined statistical timeframe, set based on actual production stability and data volume requirements, such as the past three months or a production cycle for a specific number of batches. The cutting vibration records are stored structured information, including the timestamp of the cutting operation, the corresponding raw vibration sensor signal, the pre-processed vibration characteristic parameter sequence, and the final quality inspection results of that cutting operation.
[0021] From the cutting vibration recording data, the processes whose final cutting quality was deemed acceptable can be selected, and their corresponding vibration characteristic parameter sequences can be extracted. By aligning and arithmetically averaging these acceptable sequences over time, a statistically representative average vibration characteristic parameter sequence is generated. This average vibration characteristic parameter sequence is defined as the benchmark vibration characteristic parameter sequence corresponding to the current profile type and basic cutting strategy. The benchmark vibration characteristic parameter sequence characterizes the typical dynamic characteristics that the profile cutting process should exhibit under ideal process conditions.
[0022] Secondly, in order to conduct effective real-time comparison, it is necessary to extract the segment corresponding to the time interval of the current real-time cutting process from the complete reference vibration characteristic parameter sequence to form a comparison reference vibration characteristic parameter sequence for direct comparison, so as to ensure that the comparison process is carried out in the same or comparable processing stage.
[0023] Specifically, extracting and comparing the reference vibration feature parameter sequence within the reference vibration feature parameter sequence includes: The vibration characteristic parameter sequence and the reference vibration characteristic parameter sequence are time-aligned; Obtain the segmentation time zone of the vibration characteristic parameter sequence, extract the corresponding vibration characteristic parameters within the reference vibration characteristic parameter sequence, and obtain the comparison reference vibration characteristic parameter sequence.
[0024] First, the real-time acquired vibration characteristic parameter sequence is time-aligned with a reference vibration characteristic parameter sequence. Specifically, time alignment eliminates the initial phase shift on the time axis caused by minor deviations in the detection of the cutting start point or different data acquisition trigger times, ensuring that subsequent comparisons are performed on the same time baseline. Optionally, time alignment can employ methods from the field of signal processing, such as finding the maximum correlation point based on cross-correlation analysis, or performing nonlinear alignment using dynamic time warping algorithms.
[0025] Furthermore, after time alignment is completed, the specific cutting time zone is determined based on the actual cutting process corresponding to the real-time vibration characteristic parameter sequence. Here, the cutting time zone refers to the time interval divided according to the physical stages of the entire cutting process, such as the entry stage, the stable cutting stage, and the exit stage, or according to the percentage of the completed cutting length.
[0026] After obtaining the segmented time zone information, the identical time zone segment can be located and extracted from the time-aligned reference vibration characteristic parameter sequence. This extracted segment becomes the reference vibration characteristic parameter sequence for direct comparison with the real-time sequence.
[0027] In summary, the time alignment and sequence extraction based on the cutting time zone described above ensure the relevance and effectiveness of the comparison, allowing the verification analysis to focus on the cutting dynamics of a specific stage that is currently occurring, thereby improving the accuracy of state judgment.
[0028] Furthermore, the similarity between the real-time acquired vibration characteristic parameter sequence and the aforementioned comparison benchmark vibration characteristic parameter sequence is calculated. Specifically, this similarity calculation process can be implemented using signal similarity measurement algorithms, such as those based on Euclidean distance, dynamic time warping distance, or cosine similarity. The calculated value is the verification matching degree, and its value range is usually normalized between 0 and 1, where 1 represents a complete match, meaning the real-time process is consistent with the benchmark state. The higher the verification matching degree, the higher the degree of agreement between the dynamic characteristics of the real-time cutting process and the qualified historical benchmark, meaning the current cutting process is more stable and the possibility of anomalies is lower.
[0029] Finally, based on the calculated verification matching degree, the verification deviation degree is further obtained through mathematical transformation. Preferably, the verification deviation degree = 1 - verification matching degree. The higher the verification matching degree, the closer the real-time process is to the normal baseline state, and the lower the verification deviation degree; conversely, the lower the verification matching degree, the higher the verification deviation degree, indicating a greater difference between the real-time process and the normal state. The verification matching degree and the verification deviation degree together constitute a quantitative description of the deviation of the current cutting process from the standard state.
[0030] S30: When the verification matching degree is less than the matching degree threshold, obtain the cutting progress coefficient, perform aluminum profile defect prediction, and obtain aluminum profile defect characteristics. Then, the calculated verification matching degree is compared with a preset matching degree threshold. This matching degree threshold is a critical value preset based on historical data analysis and process requirements, representing the boundary for determining whether the cutting process status significantly deviates from the normal range. When the calculated verification matching degree is less than this matching degree threshold, it indicates that there is a significant difference between the real-time acquired vibration characteristics and the qualified benchmark state, meaning that the current cutting process has exhibited abnormal dynamic characteristics or potential risks, and further diagnostic and prediction processes need to be initiated immediately to avoid final workpiece scrapping or equipment damage due to continuous abnormal cutting.
[0031] Specifically, in this case, a cutting progress coefficient representing the degree to which the current cutting operation has been completed is first obtained. Subsequently, based on this cutting progress coefficient and the verification deviation calculated above, aluminum profile defect prediction is performed, and finally, the defect characteristics of the aluminum profile are obtained.
[0032] Specifically, when the verification matching degree is less than the matching degree threshold, the cutting progress coefficient is obtained to predict aluminum profile defects and obtain aluminum profile defect characteristics, including: Determine whether the verification matching degree is less than or equal to the matching degree threshold; If not, continue to collect vibration characteristic parameters; if yes, calculate the ratio of the time length of the vibration characteristic parameter sequence to the time length of the reference vibration characteristic parameter sequence to obtain the cutting progress coefficient. Based on the cutting progress coefficient and the verification deviation, aluminum profile defects are predicted to obtain aluminum profile defect characteristics.
[0033] First, the calculated verification matching degree is compared with a preset matching degree threshold. This threshold is a critical discrimination value set based on long-term historical qualified cutting data statistical analysis and process stability requirements. It is set according to the processing sensitivity and quality control level of different types of aluminum profiles; for example, it can be set to 0.85. If the verification matching degree is greater than the threshold, the current cutting process is considered to be within an acceptable normal fluctuation range, requiring no intervention or adjustment. Therefore, subsequent vibration characteristic parameter acquisition and monitoring continue. If the verification matching degree is less than or equal to the threshold, the current cutting process is considered to have a clear anomaly. This anomaly is usually associated with internal defects inherent in the aluminum profile itself, such as impurities, pores, shrinkage cavities, or uneven structure within the material. These defects alter local resistance during cutting, thereby triggering abnormal vibration characteristics.
[0034] Once an anomaly is detected and the process enters the handling phase, a cutting progress coefficient needs to be calculated. Specifically, this coefficient is obtained by calculating the ratio of the duration of the currently acquired real-time vibration characteristic parameter sequence to the total duration of the complete reference vibration characteristic parameter sequence. For example, if the total duration of the reference sequence is 10 seconds and the currently acquired sequence duration is 3 seconds, the cutting progress coefficient is 0.3, indicating that approximately 30% of the cutting process has been completed. This cutting progress coefficient characterizes the proportion of time the operation has been completed and indirectly maps the relative cutting depth of the saw blade on the current aluminum profile.
[0035] Furthermore, by combining the calculated cutting progress coefficient with the verification deviation obtained in the preceding steps, aluminum profile defect prediction is performed. Since the structure of the aluminum profile along its length, such as cavity layout and wall thickness variations, may not be uniform, vibration anomalies occurring at different cutting progress positions often exhibit certain statistical patterns in their underlying typical defect types and severity. Based on these patterns, the predicted aluminum profile defect characteristics can be obtained through analysis and reasoning.
[0036] Specifically, based on the cutting progress coefficient and verification deviation, aluminum profile defects are predicted to obtain aluminum profile defect characteristics, including: The aluminum profile defect prediction intelligent agent is invoked, wherein the aluminum profile defect prediction intelligent agent uses a set of sample cutting progress coefficients and a set of sample verification deviations as training inputs, and a set of sample aluminum profile defect features as supervision labels. It is trained by machine learning, and the training inputs and supervision labels are obtained based on historical cutting record data of aluminum profiles of the same profile category. The cutting progress coefficient and verification deviation are input into the aluminum profile defect prediction agent to obtain aluminum profile defect features, wherein the aluminum profile defect features include defect category and defect size.
[0037] First, the aluminum profile defect prediction intelligent agent is invoked. This intelligent agent is a computational model trained based on machine learning algorithms. It can predict the type and scale of defects existing inside the aluminum profile based on the cutting progress coefficient and verification deviation data generated in the real-time cutting process, thereby realizing intelligent diagnosis and quantitative evaluation from process abnormal signals to the specific material defect causes.
[0038] Specifically, the aluminum profile defect prediction agent is constructed as follows: First, data was collected and labeled from historical cutting records of aluminum profiles of the same type. Specifically, for each recorded cutting process where an anomaly occurred, two key process parameters were extracted: the cutting progress coefficient at the time of the anomaly and the verification deviation calculated at that time. The resulting set of sample cutting progress coefficients and sample verification deviations together constituted the training input data.
[0039] Simultaneously, based on the post-operative inspection and analysis of the cut workpiece or chips, the actual aluminum profile defects corresponding to the anomaly are determined and quantified into specific defect characteristics. These characteristics include defect category and defect size. The defect category is a qualitative classification of defects according to materials science and process defect standards, such as belonging to categories like "porosity," "slag inclusion," or "crack." The defect size is a quantitative description of the defect severity, specifically the predicted number of defects within the current abnormal cutting section. For example, the prediction might be that the defect category is "porosity approximately 0.5 mm in diameter," and the defect size is "expected to be 3." The above set of sample aluminum profile defect characteristics constitutes the supervisory label data.
[0040] Furthermore, machine learning algorithms, such as support vector machines, random forests, or neural networks, are used to supervise the training of the model. The model is trained by using the set of sample cutting progress coefficients and the set of sample verification deviations as input features and the set of sample aluminum profile defect features as supervision labels, until the model can learn the complex mapping relationship from process abnormal parameters to specific defect types and scales. The trained model is the aluminum profile defect prediction agent.
[0041] For example, since there is a nonlinear mapping relationship between the cutting progress coefficient, the verification deviation and the defect characteristics of aluminum profiles, and the ensemble learning model performs well in combining the advantages of multiple weak learners to improve prediction robustness, the gradient boosting decision tree model can be selected to construct the aluminum profile defect prediction agent.
[0042] Specifically, this aluminum profile defect prediction agent mainly consists of a feature input module, an integrated decision module, and a feature output module. The feature input module receives a two-dimensional feature vector composed of the cutting progress coefficient and the verification deviation. This vector needs to be normalized before input to eliminate dimensional differences. The integrated decision module integrates multiple decision trees in a sequential manner, where the learning objective of each subsequent decision tree is to correct the residual of the prediction results of the preceding tree. The maximum depth of each decision tree is set to 5 to 8 layers, and the minimum number of samples per leaf node is set to 10, thereby controlling model complexity and preventing overfitting. The feature output module uses logistic regression or a direct classification voting mechanism to synthesize the output results of multiple trees, ultimately mapping them to specific defect categories and their scale and quantity.
[0043] During training, key hyperparameters included a learning rate of 0.1, a number of decision trees (i.e., iterations) of 100, and a subsampling ratio of 0.8. The learning rate was set to balance training speed and final performance; the number of iterations ensured the model had sufficient capacity to learn complex patterns in the data; and the subsampling ratio enhanced model diversity and improved generalization ability. Specifically, supervised learning was employed. From historical records of cutting anomalies in the same profile category, a set of sample feature vectors consisting of cutting progress coefficients and verification deviations was collected as the input sample set. Simultaneously, a sample label set consisting of defect category and defect size annotations was collected. The input sample set and the corresponding sample label set were randomly divided into training, validation, and test sets in a 6:2:2 ratio.
[0044] Subsequently, using the sample feature vectors from the training set as model input and the corresponding defect feature labels as supervision signals, the model parameters are iteratively optimized using the gradient descent algorithm. A multi-class log loss function is used to measure the difference between the model's predicted defect category probability distribution and the true labels. For the regression task of defect size and quantity, mean squared error is used as an auxiliary loss term. The training process is monitored using a validation set. Training is terminated when the comprehensive loss function value on the validation set no longer decreases significantly within 20 consecutive iterations, and the model's defect category prediction accuracy on the validation set reaches over 90%, resulting in a converged aluminum profile defect prediction agent. This aluminum profile defect prediction agent effectively learns the complex correlation between abnormal features of the cutting process and internal defects in aluminum profiles, enabling accurate prediction of defect types and sizes.
[0045] Finally, in practical applications, the cutting progress coefficient and verification deviation calculated in real time are input into the aluminum profile defect prediction agent to obtain the aluminum profile defect characteristics. These aluminum profile defect characteristics clearly include defect categories and defect scales, which can provide a precise basis for subsequently developing targeted cutting strategy optimization schemes.
[0046] S40: Based on the defect characteristics of the aluminum profile, optimize the basic cutting strategy to obtain an optimized cutting strategy and perform cutting control, wherein divergent optimization parameters are configured according to the verification deviation.
[0047] Finally, based on the aforementioned aluminum profile defect characteristics, the basic cutting strategy is optimized to obtain an optimized cutting strategy. Since the aluminum profile defect characteristics, such as specific defect categories and sizes, are directly related to the abnormal resistance sources encountered during the cutting process and their physical effects, these aluminum profile defect characteristics can provide a clear physical basis and optimization direction for adjusting cutting dynamics parameters.
[0048] Specifically, by reducing the saw blade speed to address localized hard inclusions, or adjusting the feed rate to smoothly traverse internal pore areas, sudden changes in cutting force can be specifically reduced, thereby suppressing vibration, improving cut quality, and protecting the saw blade. Cutting control based on optimized cutting strategies allows for adaptive and refined processing of aluminum profile sections with specific defects, thus improving the success rate of single cuts and the quality of finished products.
[0049] Specifically, based on the defect characteristics of the aluminum profile, the basic cutting strategy is optimized to obtain an optimized cutting strategy, and cutting control is performed, including: The basic cutting strategy is randomly adjusted to obtain multiple first adjusted cutting strategies, wherein each cutting strategy includes cutting parameters; Based on the multiple first adjustment cutting strategies and aluminum profile defect characteristics, the cutting performance is predicted to obtain multiple first cutting deformation parameters; Based on the verification deviation, a divergence optimization ratio is configured, and the multiple first adjustment cutting strategies are diverged to obtain multiple divergence cutting strategies. Continue iterative optimization until convergence, and output the cutting strategy with the minimum cutting deformation parameter as the optimized cutting strategy for cutting control.
[0050] First, starting with the initial basic cutting strategy, the basic cutting strategy is randomly adjusted to generate multiple first adjusted cutting strategies. Specifically, for example, the random adjustment can be performed according to the following quantification rules: for the key adjustable parameters in the basic cutting strategy, a maximum allowable positive and negative adjustment ratio around the initial value is set. For example, for the saw blade speed, a random adjustment of ±5% to ±15% is allowed based on the initial value. The random adjustment process is as follows: for each parameter, an adjustment coefficient is randomly generated in a uniform distribution within its corresponding ratio range, and the adjustment coefficient is multiplied by the initial value of the parameter to obtain the adjusted parameter value. For example, if the initial value of the saw blade speed in the basic strategy is 2800 rpm, and the set random adjustment range is ±10%, then the adjusted speed will be randomly determined within the range of 2520 rpm to 3080 rpm.
[0051] Through the above process, the multiple first-adjustment cutting strategies obtained are all results of independent random sampling within the feasible adjustment range of each parameter of the basic cutting strategy. They represent a series of candidate solutions obtained through preliminary exploration around the initial parameter point in its process neighborhood. Each cutting strategy contains a complete set of adjustable cutting parameters after the above-mentioned quantified random adjustment, such as the adjusted saw blade speed. The number of first-adjustment cutting strategies can be preset according to computing resources and optimization requirements, for example, generating 20 to 50.
[0052] Secondly, multiple first-adjustment cutting strategies, along with the predicted aluminum profile defect features, are input into a pre-trained cutting prediction agent. This cutting prediction agent can simulate the mechanical effects that may occur when cutting the current defective aluminum profile under different strategies and output cutting deformation parameters. Specifically, these cutting deformation parameters are used to quantify the predicted degree of cut deformation; the smaller the value, the better the expected cutting quality. For example, it may include the cut straightness deviation, with the unit being millimeters. Thus, a first cutting deformation parameter corresponding to each first-adjustment cutting strategy can be obtained.
[0053] Specifically, based on the multiple first adjustment cutting strategies and the defect characteristics of the aluminum profile, cutting performance is predicted to obtain multiple first cutting deformation parameters, including: The cutting prediction agent trained by machine learning is invoked, wherein the cutting prediction agent uses a set of sample cutting strategies and a set of sample aluminum profile defect features as training inputs, and uses a set of sample cutting deformation parameters as supervision labels for supervised training. The multiple first adjustment cutting strategies are respectively combined with the aluminum profile defect features and input into the cutting prediction agent to obtain multiple first cutting deformation parameters.
[0054] First, the cutting prediction agent is invoked. This cutting prediction agent is a machine learning model that can be used to quickly predict the cut deformation parameters that may be generated after the cutting strategy is executed, based on the combination of the cutting strategy parameters to be evaluated and the defect characteristics of the aluminum profile. This enables efficient and accurate virtual performance evaluation and screening of multiple candidate strategies without the need for physical trial cutting.
[0055] Specifically, the construction of this cutting prediction agent is based on a supervised learning paradigm. Its training data comes from historical processing records. The specific data collection method is as follows: for historically completed cutting operations with complete quality inspections, the parameters of the specific cutting strategies used are collected to form a sample cutting strategy set. Simultaneously, the corresponding aluminum profile defect features are collected to form a sample aluminum profile defect feature set. Both serve as the training input features for the model. Furthermore, based on the precise measurement results of the cut surface after the cutting, the degree of cut deformation caused by the cutting is quantified, such as the angle deviation of the cut, forming a sample cutting deformation parameter set, which is used as the label for supervised training. Finally, using the above data, a cutting prediction agent is constructed and trained through machine learning algorithms.
[0056] The cutting prediction agent can be constructed using a variety of machine learning algorithms suitable for regression prediction, such as deep neural networks, support vector regression, or gradient boosting regression trees.
[0057] Finally, in practical application, the generated multiple first adjustment cutting strategies are combined with the identified defects in the current aluminum profile to form a complete input feature pair. This feature pair is then input one by one into the trained cutting prediction agent. Based on its internally learned mapping relationships, the cutting prediction agent calculates and outputs a corresponding predicted value for each input feature pair, namely the first cutting deformation parameter. This first cutting deformation parameter is a quantified value used to estimate the severity of cut deformation that may occur if the adjustment strategy is used to cut the defective aluminum profile. After all the first adjustment cutting strategies undergo this process, multiple first cutting deformation parameters are obtained, providing a performance evaluation basis for subsequent strategy selection and optimization iterations.
[0058] Furthermore, based on the verification deviation calculated above, the divergence optimization ratio is configured. The divergence optimization ratio is defined as a numerical parameter between 0 and 1, used to quantitatively control the proportion of excellent strategies selected from the current candidate strategy set to enter the next stage of divergence exploration during each round of iterative optimization. A larger verification deviation indicates a more significant difference between the real-time cutting process and the normal state, i.e., a more severe degree of vibration anomaly. Therefore, a larger divergence optimization ratio needs to be set to perform a more refined and broader neighborhood search on high-performing strategies, improving the globality and accuracy of optimization under abnormally severe conditions.
[0059] Specifically, based on the verification deviation, a divergence optimization ratio is configured, and the plurality of first adjustment cutting strategies are diverged to obtain a plurality of divergent cutting strategies, including: Configure the verification deviation as a divergence optimization ratio; Arrange multiple first cutting deformation parameters in ascending order, and select the first adjustment cutting strategy corresponding to the first cutting deformation parameter with the previous divergent optimization ratio. Perform divergent random adjustment to obtain multiple divergent cutting strategies.
[0060] First, the calculated validation deviation value is directly used as the divergence optimization ratio. This divergence optimization ratio is a real-valued parameter limited to the interval between zero and one, used to dynamically determine what percentage of the top-performing candidate policies should be selected for more in-depth parameter space exploration in each policy iteration.
[0061] Secondly, the first cutting deformation parameters corresponding to all first adjustment cutting strategies are sorted according to the predicted deformation severity from smallest to largest; that is, the smaller the deformation parameter value, the better the predicted cutting performance. After sorting, based on the previously set divergence optimization ratio, the corresponding proportion of high-quality strategies are selected from the front end. For example, if the divergence optimization ratio is set to 0.3, then the top 30%, i.e., the first adjustment cutting strategies with the smallest cutting deformation parameters, are selected.
[0062] Finally, the selected first adjustment and cutting strategies are subjected to divergent random adjustment. Specifically, the divergent random adjustment operation refers to applying a new, small-amplitude random perturbation near the parameter vector of each first adjustment and cutting strategy, thereby exploring and generating new parameter combinations in its neighboring parameter space.
[0063] Optionally, the specific implementation method is as follows: For key parameters in the cutting strategy, such as saw blade speed, the preset allowable adjustment range is ±5% to ±10% of the original parameter value. Divergent random adjustment uses this range as a boundary, employing a uniformly distributed random number generation method to generate a new random value for each parameter within its corresponding adjustment range. For example, for a saw blade speed of 3000 rpm, if its adjustment range is set to ±8%, the new speed value will be randomly generated within the range of 2760 rpm to 3240 rpm. Simultaneously, to ensure the adaptability of the disturbance range, this adjustment range can be correlated with the verification deviation; for every 0.1 increase in verification deviation, the adjustment range increases by 2% of the original parameter value.
[0064] After this step, each selected first adjustment cutting strategy may generate several new parameter variants, and all the newly generated strategies are collectively referred to as multiple divergent cutting strategies.
[0065] This divergent stochastic adjustment process enables a refined search of the region where the best strategy is located, aiming to find a better solution than the current best strategy, thereby driving the optimization process to evolve towards a higher quality solution space.
[0066] Furthermore, iterative optimization continues, forming an iterative loop. This involves predicting the cutting performance of the newly generated divergent cutting strategy again, evaluating its cutting deformation parameters, and further filtering and divergent optimization based on the results. This iterative process continues until the algorithm converges. The convergence condition is set based on the degree of improvement of the optimal cutting deformation parameters between adjacent iterations. For example, convergence is determined when the reduction in the optimal cutting deformation parameters obtained in three consecutive iterations is less than a preset threshold, such as 1‰.
[0067] Finally, after convergence, the cutting strategy with the minimum cutting deformation parameter is selected and determined as the final optimized cutting strategy. The control system will then adjust the subsequent cutting process in real time based on this optimized cutting strategy, thereby achieving high-quality adaptive processing of aluminum profile sections with specific defects.
[0068] In summary, the embodiments of this application have at least the following technical effects: Compared to existing technologies, this invention achieves an improvement from static parameter setting to dynamic process adaptation. First, by acquiring vibration characteristics in real time and performing quantitative comparison and analysis with a benchmark sequence, process monitoring is upgraded from simple threshold alarms to precise numerical evaluation, enabling early and accurate identification of anomalies and potential risks during the cutting process. Second, it introduces aluminum profile defect prediction based on a machine learning model, correlating abnormal signals with the specific type and severity of internal material defects, achieving a leap from anomaly detection to root cause diagnosis. Third, it designs an adaptive iterative optimization mechanism that integrates verification deviation, dynamically adjusting the scope and accuracy of the optimization search according to the severity of the current anomaly, ensuring efficient and stable finding of high-quality cutting strategies under various complex and abnormal working conditions.
[0069] Ultimately, the entire technical solution constructs a complete intelligent control closed loop for perception, diagnosis, decision-making, and execution, which improves the cutting quality stability of aluminum profiles in multi-specification customized production, while reducing material scrap rate and tool wear, and enhancing the overall robustness and intelligence of the processing system.
[0070] Example 2, as Figure 2 As shown, based on the same inventive concept as the machine learning-based adaptive control method for aluminum profile cutting parameters provided in Embodiment 1, this embodiment of the invention also provides a machine learning-based adaptive control system for aluminum profile cutting parameters, including: The feature acquisition module 11 is used to obtain the profile type of aluminum profile, configure the corresponding basic cutting strategy for cutting, and collect the vibration feature parameter sequence during the cutting process. The verification analysis module 12 is used to obtain the benchmark vibration characteristic parameter sequence corresponding to the basic cutting strategy, perform verification analysis on the vibration characteristic parameter sequence, obtain the verification matching degree, and calculate the verification deviation degree. The defect prediction module 13 is used to obtain the cutting progress coefficient when the verification matching degree is less than the matching degree threshold, perform aluminum profile defect prediction, and obtain aluminum profile defect characteristics. The cutting control module 14 is used to optimize the basic cutting strategy based on the defect characteristics of the aluminum profile, obtain an optimized cutting strategy, and perform cutting control, wherein divergent optimization parameters are configured according to the verification deviation.
[0071] Specifically, the feature acquisition module 11 is used for: Obtain the profile type of the aluminum profile, configure the corresponding basic cutting strategy for cutting, and collect the vibration characteristic parameter sequence during the cutting process, including: Obtain the profile category of the aluminum profile; Input the profile type into the cutting strategy database, index the corresponding basic cutting strategy, and control the cutting of the aluminum profile; During the cutting process, vibration characteristic parameters are collected to obtain a sequence of vibration characteristic parameters.
[0072] Specifically, the verification and analysis module 12 is used for: Obtain the benchmark vibration characteristic parameter sequence corresponding to the basic cutting strategy, perform verification analysis on the vibration characteristic parameter sequence to obtain the verification matching degree, and calculate the verification deviation degree, including: Obtain historical cutting vibration records of aluminum profiles of the same type, extract the average vibration characteristic parameter sequence of qualified cutting processes, and obtain the benchmark vibration characteristic parameter sequence. Extract and compare the reference vibration feature parameter sequence within the reference vibration feature parameter sequence; The similarity between the vibration characteristic parameter sequence and the comparison benchmark vibration characteristic parameter sequence is calculated to obtain the verification matching degree; The verification deviation is calculated based on the verification matching degree.
[0073] Specifically, extracting and comparing the reference vibration feature parameter sequence within the reference vibration feature parameter sequence includes: The vibration characteristic parameter sequence and the reference vibration characteristic parameter sequence are time-aligned; Obtain the segmentation time zone of the vibration characteristic parameter sequence, extract the corresponding vibration characteristic parameters within the reference vibration characteristic parameter sequence, and obtain the comparison reference vibration characteristic parameter sequence.
[0074] Specifically, the defect prediction module 13 is used for: When the matching degree is less than the matching degree threshold, the cutting progress coefficient is obtained to predict aluminum profile defects and obtain aluminum profile defect characteristics, including: Determine whether the verification matching degree is less than or equal to the matching degree threshold; If not, continue to collect vibration characteristic parameters; if yes, calculate the ratio of the time length of the vibration characteristic parameter sequence to the time length of the reference vibration characteristic parameter sequence to obtain the cutting progress coefficient. Based on the cutting progress coefficient and the verification deviation, aluminum profile defects are predicted to obtain aluminum profile defect characteristics.
[0075] Specifically, based on the cutting progress coefficient and verification deviation, aluminum profile defects are predicted to obtain aluminum profile defect characteristics, including: The aluminum profile defect prediction intelligent agent is invoked, wherein the aluminum profile defect prediction intelligent agent uses a set of sample cutting progress coefficients and a set of sample verification deviations as training inputs, and a set of sample aluminum profile defect features as supervision labels. It is trained by machine learning, and the training inputs and supervision labels are obtained based on historical cutting record data of aluminum profiles of the same profile category. The cutting progress coefficient and verification deviation are input into the aluminum profile defect prediction agent to obtain aluminum profile defect features, wherein the aluminum profile defect features include defect category and defect size.
[0076] Specifically, the cutting control module 14 is used for: Based on the defect characteristics of the aluminum profile, the basic cutting strategy is optimized to obtain an optimized cutting strategy, and cutting control is performed, including: The basic cutting strategy is randomly adjusted to obtain multiple first adjusted cutting strategies, wherein each cutting strategy includes cutting parameters; Based on the multiple first adjustment cutting strategies and aluminum profile defect characteristics, the cutting performance is predicted to obtain multiple first cutting deformation parameters; Based on the verification deviation, a divergence optimization ratio is configured, and the multiple first adjustment cutting strategies are diverged to obtain multiple divergence cutting strategies. Continue iterative optimization until convergence, and output the cutting strategy with the minimum cutting deformation parameter as the optimized cutting strategy for cutting control.
[0077] Specifically, based on the multiple first adjustment cutting strategies and the defect characteristics of the aluminum profile, cutting performance is predicted to obtain multiple first cutting deformation parameters, including: The cutting prediction agent trained by machine learning is invoked, wherein the cutting prediction agent uses a set of sample cutting strategies and a set of sample aluminum profile defect features as training inputs, and uses a set of sample cutting deformation parameters as supervision labels for supervised training. The multiple first adjustment cutting strategies are respectively combined with the aluminum profile defect features and input into the cutting prediction agent to obtain multiple first cutting deformation parameters.
[0078] Specifically, based on the verification deviation, a divergence optimization ratio is configured, and the plurality of first adjustment cutting strategies are diverged to obtain a plurality of divergent cutting strategies, including: Configure the verification deviation as a divergence optimization ratio; Arrange multiple first cutting deformation parameters in ascending order, and select the first adjustment cutting strategy corresponding to the first cutting deformation parameter with the previous divergent optimization ratio. Perform divergent random adjustment to obtain multiple divergent cutting strategies.
Claims
1. A machine learning-based adaptive control method for aluminum profile cutting parameters, characterized in that, The method includes: Obtain the profile type of the aluminum profile, configure the corresponding basic cutting strategy for cutting, and collect the vibration characteristic parameter sequence during the cutting process; Obtain the benchmark vibration characteristic parameter sequence corresponding to the basic cutting strategy, perform verification analysis on the vibration characteristic parameter sequence to obtain the verification matching degree, and calculate the verification deviation degree; When the verification matching degree is less than the matching degree threshold, the cutting progress coefficient is obtained to predict aluminum profile defects and obtain aluminum profile defect characteristics. Based on the defect characteristics of the aluminum profile, the basic cutting strategy is optimized to obtain an optimized cutting strategy, and cutting control is performed. In this process, divergent optimization parameters are configured according to the verification deviation.
2. The adaptive control method for aluminum profile cutting parameters based on machine learning according to claim 1, characterized in that, Obtain the profile type of the aluminum profile, configure the corresponding basic cutting strategy for cutting, and collect the vibration characteristic parameter sequence during the cutting process, including: Obtain the profile category of the aluminum profile; Input the profile type into the cutting strategy database, index the corresponding basic cutting strategy, and control the cutting of the aluminum profile; During the cutting process, vibration characteristic parameters are collected to obtain a sequence of vibration characteristic parameters.
3. The adaptive control method for aluminum profile cutting parameters based on machine learning according to claim 1, characterized in that, Obtain the benchmark vibration characteristic parameter sequence corresponding to the basic cutting strategy, perform verification analysis on the vibration characteristic parameter sequence to obtain the verification matching degree, and calculate the verification deviation degree, including: Obtain historical cutting vibration records of aluminum profiles of the same type, extract the average vibration characteristic parameter sequence of qualified cutting processes, and obtain the benchmark vibration characteristic parameter sequence. Extract and compare the reference vibration feature parameter sequence within the reference vibration feature parameter sequence; The similarity between the vibration characteristic parameter sequence and the comparison benchmark vibration characteristic parameter sequence is calculated to obtain the verification matching degree; The verification deviation is calculated based on the verification matching degree.
4. The adaptive control method for aluminum profile cutting parameters based on machine learning according to claim 3, characterized in that, Extracting and comparing the reference vibration feature parameter sequence within the reference vibration feature parameter sequence includes: The vibration characteristic parameter sequence and the reference vibration characteristic parameter sequence are time-aligned; Obtain the segmentation time zone of the vibration characteristic parameter sequence, extract the corresponding vibration characteristic parameters within the reference vibration characteristic parameter sequence, and obtain the comparison reference vibration characteristic parameter sequence.
5. The adaptive control method for aluminum profile cutting parameters based on machine learning according to claim 1, characterized in that, When the matching degree is less than the matching degree threshold, the cutting progress coefficient is obtained to predict aluminum profile defects and obtain aluminum profile defect characteristics, including: Determine whether the verification matching degree is less than or equal to the matching degree threshold; If not, continue to collect vibration characteristic parameters; if yes, calculate the ratio of the time length of the vibration characteristic parameter sequence to the time length of the reference vibration characteristic parameter sequence to obtain the cutting progress coefficient. Based on the cutting progress coefficient and the verification deviation, aluminum profile defects are predicted to obtain aluminum profile defect characteristics.
6. The adaptive control method for aluminum profile cutting parameters based on machine learning according to claim 5, characterized in that, Based on the cutting progress coefficient and verification deviation, aluminum profile defects are predicted to obtain aluminum profile defect characteristics, including: The aluminum profile defect prediction intelligent agent is invoked, wherein the aluminum profile defect prediction intelligent agent uses a set of sample cutting progress coefficients and a set of sample verification deviations as training inputs, and a set of sample aluminum profile defect features as supervision labels. It is trained by machine learning, and the training inputs and supervision labels are obtained based on historical cutting record data of aluminum profiles of the same profile category. The cutting progress coefficient and verification deviation are input into the aluminum profile defect prediction agent to obtain aluminum profile defect features, wherein the aluminum profile defect features include defect category and defect size.
7. The adaptive control method for aluminum profile cutting parameters based on machine learning according to claim 1, characterized in that, Based on the defect characteristics of the aluminum profile, the basic cutting strategy is optimized to obtain an optimized cutting strategy, and cutting control is performed, including: The basic cutting strategy is randomly adjusted to obtain multiple first adjusted cutting strategies, wherein each cutting strategy includes cutting parameters; Based on the multiple first adjustment cutting strategies and aluminum profile defect characteristics, the cutting performance is predicted to obtain multiple first cutting deformation parameters; Based on the verification deviation, a divergence optimization ratio is configured, and the multiple first adjustment cutting strategies are diverged to obtain multiple divergence cutting strategies. Continue iterative optimization until convergence, and output the cutting strategy with the minimum cutting deformation parameter as the optimized cutting strategy for cutting control.
8. The adaptive control method for aluminum profile cutting parameters based on machine learning according to claim 7, characterized in that, Based on the aforementioned multiple first adjustment cutting strategies and aluminum profile defect characteristics, cutting performance is predicted to obtain multiple first cutting deformation parameters, including: The cutting prediction agent trained by machine learning is invoked, wherein the cutting prediction agent uses a set of sample cutting strategies and a set of sample aluminum profile defect features as training inputs, and uses a set of sample cutting deformation parameters as supervision labels for supervised training. The multiple first adjustment cutting strategies are respectively combined with the aluminum profile defect features and input into the cutting prediction agent to obtain multiple first cutting deformation parameters.
9. The adaptive control method for aluminum profile cutting parameters based on machine learning according to claim 7, characterized in that, Based on the verification deviation, a divergence optimization ratio is configured, and the multiple first adjustment cutting strategies are diverged to obtain multiple divergence cutting strategies, including: Configure the verification deviation as a divergence optimization ratio; Arrange multiple first cutting deformation parameters in ascending order, and select the first adjustment cutting strategy corresponding to the first cutting deformation parameter with the previous divergent optimization ratio. Perform divergent random adjustment to obtain multiple divergent cutting strategies.
10. An adaptive control system for aluminum profile cutting parameters based on machine learning, characterized in that, The method for adaptive control of aluminum profile cutting parameters based on machine learning as described in any one of claims 1-9 includes: The feature acquisition module is used to obtain the profile type of aluminum profile, configure the corresponding basic cutting strategy for cutting, and collect the vibration feature parameter sequence during the cutting process. The verification analysis module is used to obtain the benchmark vibration characteristic parameter sequence corresponding to the basic cutting strategy, perform verification analysis on the vibration characteristic parameter sequence, obtain the verification matching degree, and calculate the verification deviation degree. The defect prediction module is used to obtain the cutting progress coefficient when the verification matching degree is less than the matching degree threshold, to predict aluminum profile defects and obtain aluminum profile defect characteristics. The cutting control module is used to optimize the basic cutting strategy based on the defect characteristics of the aluminum profile, obtain an optimized cutting strategy, and perform cutting control, wherein divergent optimization parameters are configured according to the verification deviation.