Intelligent control method and system for real-time microdefects of precise metal parts

Abstract

The invention relates to the technical field of high-end equipment, and discloses a real-time micro-defect intelligent control method and system for precision metal parts, and the method comprises the steps: obtaining historical defect detection data, and carrying out the preprocessing of the data to obtain a structured defect data set; determining a defect statistical mode through time sequence analysis and clustering analysis; generating a geometric feature set, a statistical feature set and an associated data set based on the defect statistical mode, and storing by adopting a dynamic retrieval database to form a queriable defect feature library; new defects are detected and matched with the feature library to judge types and potential positions; predicting time points and position coordinates of defects in combination with process parameters; adjusting process parameters according to a prediction result to obtain optimal configuration; optimizing and updating the defect feature library in combination with new defect information; and continuously monitoring and dynamically adjusting process parameters based on the improved feature library, and generating stable production control parameters. According to the method, real-time prediction and dynamic control of microdefects in the precise metal part production process can be achieved, and the product yield is increased.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of high-end equipment, and in particular to a real-time micro-defect intelligent control method and system for precision metal parts. BACKGROUND

[0002] In the technical field of high-end equipment, precision metal part manufacturing is a core link in key industries such as aerospace and automobiles, and the micro-defect control thereof is directly related to the reliability and safety of products. Taking an aero-engine blade as an example, micro-defects are extremely likely to cause part failure under high-load working conditions, resulting in serious consequences.

[0003] The prior art usually adopts an offline detection method, that is, after obtaining a part surface image through industrial CT scanning or an electron microscope, an edge detection algorithm is used to extract defect contour features; the extracted geometric features are compared with a preset defect template library, and when the matching degree exceeds a threshold, the defect type is determined; and according to the defect type, a corresponding adjustment scheme is retrieved from a pre-stored process parameter library, and the machining equipment parameters are corrected through a numerical control system.

[0004] Since the prior art passively adjusts the macro-defects that have already been formed, it cannot capture the micro-defect evolution caused by factors such as material stress and temperature fluctuations in a dynamic production environment in real time, and lacks dynamic correlation analysis between defect features and multi-process parameters, resulting in that the process adjustment lags behind the defect generation, and the false detection rate and the missed detection rate are high. Therefore, the prior art has the problem of low yield of precision metal parts. SUMMARY

[0005] The present application provides a real-time micro-defect intelligent control method and system for precision metal parts to solve the problem of low yield of precision metal parts in the prior art.

[0006] In a first aspect, to solve the above technical problems, the present application provides a real-time micro-defect intelligent control method for precision metal parts, comprising: obtaining historical defect detection data of precision metal parts and preprocessing to obtain a structured defect data set; performing time series analysis on the structured defect data set, extracting the distribution law of defects and performing cluster analysis to determine a defect statistical mode; generating a geometric feature set, a statistical feature set and an associated data set based on the defect statistical mode, and storing them in a preset dynamic retrieval database to obtain a queryable defect feature library; detecting new defects in the production process and performing similarity matching with the defect feature library to determine the type and potential position of the new defects; According to the type and potential position of the new defects, in combination with the process parameters of the current production equipment, the time point and position coordinates of the defects are predicted. According to the time point and the position coordinates, combining a preset parameter adjustment rule, adjusting a process parameter of a production device to obtain an optimized process parameter configuration; According to the optimized process parameter configuration, combining the type of the new defect and the potential position, optimizing and updating the defect feature library to obtain an improved defect feature library; According to the improved defect feature library, continuously detecting defects in a production process and making a prediction, dynamically adjusting a process parameter of the production device, and generating stable precision metal part production control parameters based on a device response time.

[0007] In an optional implementation, the historical defect detection data of the precision metal part is acquired and preprocessed to obtain a structured defect data set, including: The historical detection raw data containing microscopic defects is acquired, and when a data point exceeds a preset mean value threshold range, noise data is removed by a mean filtering method to generate a first data set; The first data set is subjected to edge detection to extract defect boundary coordinates, shape features are analyzed, and size features are calculated based on the Euclidean distance between boundary points to generate a second data set; The second data set is subjected to principal component analysis to calculate feature vectors and eigenvalues, defect distribution positions and frequencies are extracted, and a third data set containing statistical features is generated; The third data set is subjected to data normalization processing, and when the normalized eigenvalues meet preset format requirements, the structured defect data set is output.

[0008] In an optional implementation, the structured defect data set is subjected to time series analysis, the distribution law of defects is extracted, and clustering analysis is performed to determine a defect statistical mode, including: The defect frequency sequence corresponding to the production batch time stamp is extracted from the structured defect data set, an autoregressive moving average method is used to calculate autoregressive parameters and moving average parameters, and a time series is generated; The time series is analyzed to obtain the change of defect frequency between batches, combining the process parameters acquired from a preset production batch database, batch correlation features are extracted, and defects are classified by a clustering method to obtain a defect type distribution; When the defect type distribution meets a preset defect type distribution threshold, the position data is extracted from the batch correlation features, and the distribution law of defects is calculated by statistical analysis; Combining the defect type distribution and the distribution law of defects, a prediction probability of the defect position is calculated by a Bayesian method to obtain the defect statistical mode.

[0009] In an alternative embodiment, the geometric feature set, the statistical feature set and the correlation data set are generated based on the defect statistical pattern, and are stored in a preset dynamic retrieval database to obtain a queryable defect feature library, including: The product surface defect area is selected based on the defect statistical pattern for image segmentation processing, and an edge detection method is used to calculate shape and size features to generate a geometric feature set; When the geometric feature set meets a preset geometric feature threshold, a statistical feature set is generated by statistically analyzing defect distribution density; Process parameters are obtained from a preset production batch database, and the geometric feature set and the statistical feature set are correlated using a relational mapping method to generate a correlation data set; The geometric feature set, the statistical feature set and the correlation data set are stored in a preset dynamic retrieval database, and an index optimization method is used to construct the queryable defect feature library.

[0010] In an alternative embodiment, new defects in the production process are detected, and similarity matching is performed with the defect feature library to determine the type and potential location of the new defects, including: Boundary coordinates are extracted from new defect images in the production process, shape and size features are calculated to generate a new geometric feature set; The new geometric feature set is statistically analyzed for distribution density to generate a new statistical feature set; The new geometric feature set, the new statistical feature set and the defect feature library are similarity matched to generate a similar feature set; The similar feature set is sorted by feature importance and classified by confidence using a random forest method to obtain the type and potential location of the new defects.

[0011] In an alternative embodiment, the type and potential location of the new defects are used to predict the time point and position coordinates of the defects in combination with the process parameters of the current production equipment, including: The real-time process parameters of the current production equipment are normalized and outliers are removed based on the type and potential location of the new defects to generate a defect prediction data set; The time series mean, variance and rate of change of the defect prediction data set are extracted by a sliding window, and the defect prediction feature set is generated in combination with the type and potential location of the new defects; When the defect prediction feature set has a matching degree greater than a preset defect prediction threshold with historical defect data, a support vector machine is used to predict the time point and position coordinates of the defects.

[0012] In an optional implementation, the adjusting the process parameters of the production equipment according to the time point and the position coordinates, in combination with a preset parameter adjustment rule, to obtain an optimized process parameter configuration comprises: According to the time point and the position coordinates, real-time data is collected from the production environment, a deviation of the real-time data from a preset process parameter threshold is calculated, and a deviation data set is generated; The deviation data set is calculated by a linear interpolation method to obtain an adjustment amplitude, and an adjustment parameter set is generated in combination with the time point and the position coordinates; The process parameters of the production equipment are adjusted by the adjustment parameter set, and an optimized process parameter configuration is output.

[0013] In an optional implementation, the optimizing and updating the defect feature library according to the optimized process parameter configuration, in combination with the type of the new defect and the potential position, to obtain an improved defect feature library comprises: Real-time production data generated according to the optimized process parameter configuration is cleaned and formatted to generate structured real-time production data; According to the structured real-time production data, in combination with the type of the new defect and the potential position, a defect distribution rule is calculated by feature extraction to generate preliminary defect distribution features; When the preliminary defect distribution features do not meet a preset real-time defect threshold, relevant historical data in the defect feature library is queried to generate supplementary feature data; The supplementary feature data and the preliminary defect distribution features are analyzed and fused to update the defect feature library, and the improved defect feature library is obtained.

[0014] In an optional implementation, the continuously detecting defects in a production process and making predictions according to the improved defect feature library, dynamically adjusting the process parameters of the production equipment, and generating stable precision metal part production control parameters based on a device response time comprises: The improved defect feature library is compared with the defect feature library to continuously detect defects in a production process and make predictions, dynamically adjust the process parameters of the production equipment, and obtain adjusted process parameters; Control instructions are sent to the production equipment according to the adjusted process parameters, and a response time is monitored, and an optimized parameter is obtained by iterative optimization of feedback data; When the optimized parameter converges to a preset convergence threshold, the stable precision metal part production control parameter is obtained and output to a designated storage unit.

[0015] In a second aspect, the present application provides a precision metal part real-time micro-defect intelligent control system, comprising: The data acquisition module is configured to acquire historical defect detection data of the precision metal part and perform preprocessing to obtain a structured defect data set. The data analysis module is configured to perform time series analysis on the structured defect data set, extract distribution rules of defects, and perform clustering analysis to determine a defect statistical mode. The feature library construction module is configured to generate a geometric feature set, a statistical feature set and an associated data set based on the defect statistical mode, and store them in a preset dynamic retrieval database to obtain a queryable defect feature library. The classification module is configured to detect a new defect in a production process, and perform similarity matching with the defect feature library to determine a type and a potential position of the new defect. The prediction module is configured to predict a time point and a position coordinate of the defect occurrence based on the type and the potential position of the new defect, and the process parameters of the current production equipment. The optimization module is configured to adjust the process parameters of the production equipment based on the time point and the position coordinate, and a preset parameter adjustment rule to obtain optimized process parameter configurations. The update module is configured to optimize and update the defect feature library based on the optimized process parameter configurations, the type of the new defect and the potential position to obtain an improved defect feature library. The control module is configured to continuously detect defects in the production process and perform prediction based on the improved defect feature library, dynamically adjust the process parameters of the production equipment, and generate stable precision metal part production control parameters based on the response time of the equipment.

[0016] In a third aspect, the present application further provides an electronic device including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, and the processor implements the precision metal part real-time micro-defect intelligent control method and system according to any one of the above aspects when executing the computer program.

[0017] In a fourth aspect, the present application further provides a computer readable storage medium including a stored computer program, wherein the computer readable storage medium controls the device where the computer readable storage medium is located to execute the precision metal part real-time micro-defect intelligent control method and system according to any one of the above aspects when the computer program runs.

[0018] Compared with the prior art, the present application has the following beneficial effects: (1) The present application realizes the systematized integration and dynamic analysis of the whole process data of the production of the precision metal part, generates a high-confidence structured defect data set through the mean filtering denoising, edge feature extraction and principal component analysis of the historical defect detection data, ensures the integrity, consistency and accuracy of the defect features, effectively eliminates the feature distortion caused by the detection environment fluctuation, and lays a reliable foundation for defect prediction and process optimization.

[0019] (2) The present application realizes the modeling of the defect evolution law based on time series clustering and Bayesian probability mapping, dynamically retrieves the database to fuse the geometric feature set, the statistical feature set and the process parameter associated data set, classifies and marks the defect features according to the spatial distribution and evolution trend in multiple dimensions, and stores them in a structured manner, so that the defect data can be intelligently searched according to the type, position and process correlation, and accurate feature basis is provided for real-time prediction.

[0020] (3) The present application realizes the positioning intervention and process parameter self-optimization in the defect initiation stage based on the defect feature library, combined with the support vector machine prediction and closed-loop parameter adjustment mechanism, significantly improves the yield, and provides the manufacturing guarantee for the precision metal parts in the high-reliability field such as aerospace. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 is the flow chart of the real-time micro-defect intelligent control method of the precision metal part provided by the first embodiment of the present application; Figure 2 is the structural schematic diagram of the real-time micro-defect intelligent control system of the precision metal part provided by the second embodiment of the present application. DETAILED DESCRIPTION

[0022] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0023] With reference to Figure 1 , the first embodiment of the present application provides a real-time micro-defect intelligent control method and system for precision metal parts, which comprises the following steps: S11, obtaining the historical defect detection data of the precision metal part and pre-processing to obtain a structured defect data set; S12, performing time series analysis on the structured defect data set, extracting the distribution law of the defects and performing clustering analysis to determine the defect statistical mode; S13, generate a geometric feature set, a statistical feature set and a correlation data set based on the defect statistical mode, and store them in a preset dynamic retrieval database to obtain a queryable defect feature library; S14, detect a new defect in a production process, and perform similarity matching with the defect feature library to determine a type and a potential position of the new defect; S15, predict a time point and a position coordinate of the defect occurrence according to the type and the potential position of the new defect, and in combination with process parameters of a current production equipment; S16, adjust the process parameters of the production equipment according to the time point and the position coordinate, and in combination with a preset parameter adjustment rule to obtain optimized process parameter configurations; S17, optimize and update the defect feature library according to the optimized process parameter configurations, the type of the new defect and the potential position to obtain an improved defect feature library; S18, continuously detect defects in a production process and perform prediction according to the improved defect feature library, dynamically adjust the process parameters of the production equipment, and generate stable precision metal part production control parameters based on equipment response time.

[0024] In step S11, historical defect detection data of a precision metal part is needed to be acquired and preprocessed to obtain a structured defect data set, including: S111, acquire historical detection original data containing micro defects, remove noise data by mean filtering method when data points exceed a preset mean threshold range, and generate a first data set; S112, extract defect boundary coordinates by edge detection on the first data set, analyze to obtain shape features, calculate size features based on Euclidean distance between boundary points, and generate a second data set; S113, perform principal component analysis on the second data set, calculate feature vectors and eigenvalues, extract defect distribution positions and occurrence frequencies, and generate a third data set containing statistical features; S114, perform data normalization processing on the third data set, and output the structured defect data set when normalized eigenvalues meet preset format requirements.

[0025] In step S111, historical detection original data containing micro defects is acquired, and noise data is removed by mean filtering method when data points exceed a preset mean threshold range to generate a first data set.

[0026] It should be noted that the historical detection data is usually derived from industrial CT scanning or electron microscope imaging, and the original data contains a large number of gray pixel values reflecting the microscopic defects on the surface of the metal part, and may also be subject to device jitter or environmental interference to generate noise. The mean filtering method smoothes the noise by calculating the average value of the neighborhood of the pixel point, while retaining the true defect characteristics, which is a key preprocessing step to ensure the accuracy of subsequent feature extraction.

[0027] In an implementation manner, taking the detection of an aero-engine blade as an example, the original data is a 1024x1024 pixel gray image, and the pixel value range is 0-255. When a 3x3 mean filtering window is used, for the center pixel value 220, the surrounding 8 pixel values are 195, 198, 202, 199, 197, 203, 196, and 201, and the average value is 199.5. If the preset mean value threshold is 190±10, the pixel value is updated to 199, effectively suppressing random noise. For abnormal pixel points that are obviously out of the threshold range, they are determined as noise and removed. After filtering, the true topography of the defect area is retained, and the first data set after denoising is generated.

[0028] In step S112, the first data set is subjected to edge detection to extract defect boundary coordinates, shape features are analyzed, size features are calculated based on the Euclidean distance between boundary points, and a second data set is generated.

[0029] It should be noted that edge detection refers to using the Canny edge detection algorithm to accurately identify the contour boundary of the microscopic defect through gradient calculation and double-threshold screening. The boundary point coordinate sequence not only defines the defect shape, but also directly reflects key size parameters such as crack length and hole diameter, and these geometric features are important basis for defect classification.

[0030] In an implementation manner, taking the detection of a blade surface crack as an example, the edge detection outputs the boundary point set [(100, 150), (102, 151), (105, 153)]. The Euclidean distance between adjacent points is calculated, such as the distance between (100, 150) and (102, 151) is pixels, and the actual length is 1.12 μm according to the image resolution of 0.5 μm / pixel. By traversing all the boundary points, the maximum length of the crack is 50 pixels (25 μm), and the average width is 20 pixels (10 μm), which are quantitative parameters and together with the subsequent principal component analysis constitute the second data set.

[0031] In step S113, the second data set is subjected to principal component analysis to calculate the feature vector and the feature value, extract the distribution position and the occurrence frequency of the defect, and generate a third data set containing statistical features.

[0032] It should be noted that the principal component analysis refers to calculating the eigenvectors and eigenvalues by covariance matrix decomposition, and converting high-dimensional features such as length, width, area, etc. into linearly independent principal components. The eigenvector is used to represent the main direction of the defect feature, and the eigenvalue is used to reflect the variance contribution of each principal component. The first three principal components usually accumulate more than 90% of the eigenvalue proportion, and the corresponding eigenvector indicates the main spatial mode of the defect distribution, effectively highlighting the key statistical features.

[0033] In an implementation manner, when 1000 defect samples are analyzed, the principal component analysis result shows that the first principal component (explaining 60% of the variance) corresponds to the defects concentrated in the leading edge area of the blade, and the second principal component (25%) reflects the trend of the crack type defects extending along the axial direction. Statistics show that the defect frequency in the leading edge area is 65%, and the crack proportion is more than 70%, and these spatial distribution rules and frequency characteristics are integrated into the third data set to provide statistical basis for subsequent defect prediction.

[0034] In step S114, the third data set is subjected to data normalization processing, and when the normalized eigenvalue meets the preset format requirement, the structured defect data set is output.

[0035] It should be noted that the Z-score normalization converts features of different dimensions (such as μm-level size and percentage frequency) to a unified scale, eliminating the influence of numerical differences on the machine learning model. The preset format requirements include that the eigenvalue is in the interval [-3, 3] and has no missing value, etc., to ensure data quality.

[0036] In an implementation manner, the original value of a crack length feature is 5 μm, the mean value in the data set is 4 μm, and the standard deviation is 1 μm. After normalization, it is (5-4) / 1=1. At the same time, all eigenvalues are checked, and abnormal data outside the range [-3, 3] are removed. The finally generated structured defect data set can be trained and directly input, realizing automatic classification of defect types and improving classification accuracy.

[0037] Preferably, a defect distribution heat map is generated in combination with the edge detection result and the principal component analysis, to visually display the defect density of different regions of the blade by color gradient, to guide the process personnel to preferentially process the high-risk regions. For example, the red region (defect density > 30 pieces / cm²) in the heat map corresponds to a machine feed speed that needs to be reduced by 15%, and such adjustment can reduce the defect occurrence rate and improve the quality control efficiency.

[0038] In step S12, the structured defect data set is subjected to time series analysis, the distribution rule of the defect is extracted and cluster analysis is performed, and the defect statistical mode is determined, including: S121, extracting a defect frequency sequence corresponding to a production batch timestamp from the structured defect data set, calculating autoregressive parameters and moving average parameters by using an autoregressive moving average method to generate a time series; S122, analyzing the time series to obtain a defect frequency change between batches, extracting batch-related features in combination with process parameters obtained from a preset production batch database, and classifying defect types by a clustering method to obtain a defect type distribution; S123, when the defect type distribution meets a preset defect type distribution threshold, extracting position data from the batch-related features, and calculating a distribution rule of the defects by statistical analysis; S124, combining the defect type distribution and the distribution rule of the defects, calculating a prediction probability of the defect position by using a Bayesian method to obtain the defect statistical mode.

[0039] In step S121, a defect frequency sequence corresponding to a production batch timestamp is extracted from the structured defect data set, autoregressive parameters and moving average parameters are calculated by using an autoregressive moving average method to generate a time series.

[0040] It should be noted that the autoregressive parameter a represents the degree of dependence of the current defect frequency on the historical data, and the moving average parameter b reflects the influence weight of the historical noise on the current data. By fitting the time series data, short-term fluctuations and long-term trends of the defect frequency can be effectively captured, providing a quantitative basis for subsequent analysis.

[0041] In one implementation, a defect frequency sequence corresponding to a production batch timestamp is extracted from the structured defect data set, which is a set of discrete observation values arranged in chronological order such as [3, 5, 2, 4, 6, 3, 7, 4, 2, 5] times per thousand pieces, representing the defect occurrence frequency of unit parts in different production batches. The frequency sequence is fitted by using an autoregressive moving average method, and the autoregressive parameter a (representing the linear dependence strength of the current defect frequency on the historical data) and the moving average parameter b (representing the smoothing effect weight of the historical random noise on the current value) are calculated by maximum likelihood estimation to generate a continuous time series. This processing process converts discrete batch data into a continuous time series with time series prediction capability. For example, based on 10 batches of data, a=0.6, b=0.3, indicating that the current defect frequency is 60% determined by the previous batch and 30% affected by the historical noise smoothing, and the time series can predict the next batch defect frequency as 4.2 times per thousand pieces, with an error of only 5% from the actual observation value of 4 times, verifying the effective conversion from the original frequency sequence to the predicted time series.

[0042] In step S122, the time series is analyzed to obtain the inter-batch defect frequency variation, the process parameters obtained from the preset production batch database are combined, the batch-related features are extracted, and the defect types are classified through clustering methods to obtain the defect type distribution.

[0043] It should be noted that the analysis of the time series includes trend decomposition and periodicity detection, the long-term trend and short-term fluctuation are separated by moving average method, and the potential period is identified by Fourier transform. The specific performance of the inter-batch defect frequency variation is the difference between the defect numbers of adjacent batches and its statistical characteristics such as mean and variance. The preset production batch database refers to a relational database that stores the complete process parameters (such as temperature, pressure, speed, etc.) of each production batch, which is associated with the defect data through batch ID. The defect frequency variation law includes characteristics such as trend and periodicity. The correlation analysis between process parameters and defect frequency uses Pearson correlation coefficient calculation, and when |r|>0.7, it is considered that there is strong correlation. K-means clustering determines the optimal number of clusters according to the silhouette coefficient, usually selects 3-5 clusters to balance the classification granularity and applicability.

[0044] In an implementation, the data of a certain bearing production line shows that the short-term fluctuation variance of the defect frequency sequence is 1.2 times per thousand, and the long-term trend shows that every 10 batches increase by 1.5 times per thousand. When the lathe speed is >3000 rpm, the defect frequency and the speed are strongly positively correlated (r=0.82). Through K-means clustering (k=3), the defects are divided into: type A (crack, 45%), type B (porosity, 30%), and type C (inclusion, 25%). Crack defects are mainly distributed in the bearing raceway area (72%), which is consistent with the analysis conclusion.

[0045] In step S123, when the defect type distribution meets the preset defect type distribution threshold, the position data is extracted from the batch-related features, and the distribution law of the defect is calculated through statistical analysis.

[0046] It should be noted that the preset threshold is usually set to not more than 60% of the proportion of single-class defects to ensure the diversity of the classification results. The extracted position data refers to the coordinate set of the defects on the part surface, which is unified to the standard reference system through geometric transformation. The position distribution law is calculated by kernel density estimation (KDE) algorithm, and the bandwidth parameter h is automatically optimized according to Scott rule.

[0047] In an implementation, the proportion of crack defects of a certain precision gear is 58% (lower than the threshold 60%), and the KDE analysis shows that the density peak of cracks at the tooth root is 0.15 / mm², which is 5 times that of the tooth top area (0.03 / mm²). This distribution law is highly consistent with the stress concentration area of finite element analysis, which confirms the effectiveness of the method.

[0048] In step S124, the prediction probability of the defect position is calculated by combining the defect type distribution and the distribution rule of the defects, and a Bayesian method is used to obtain the defect statistical mode.

[0049] It should be noted that the defect statistical mode refers to a "defect type-position-process parameter" joint probability model established by a Bayesian network. The prior probability P(A) comes from the historical defect type distribution statistics, and the likelihood function P(B|A) is calculated by the real-time process parameters and the defect distribution rule. When the posterior probability P(A|B) > 0.7, the statistical mode containing the predicted position, defect type confidence, and key impact parameters is output.

[0050] In an implementation, in the production of a certain titanium alloy blade, the prior probability P(A) of cracks in the blade root area is 0.75. When the hardness of the current batch of materials is detected to decrease by 5 HB, the likelihood function P(B|A) is calculated to be 0.85, and the final posterior probability P(A|B) is 0.91. The system immediately triggers the process optimization of the blade root area, and the crack occurrence rate in this area is reduced from 18% to 7%.

[0051] In step S13, the geometric feature set, the statistical feature set, and the association data set are generated based on the defect statistical mode, and are stored by using a preset dynamic retrieval database to obtain a queryable defect feature library, including: S131, based on the defect statistical mode, the product surface defect area is selected for image segmentation processing, and the shape and size features are calculated by using an edge detection method to generate a geometric feature set; S132, when the geometric feature set meets a preset geometric feature threshold, the statistical feature set is generated by analyzing the defect distribution density; S133, the process parameters are obtained from a preset production batch database, and the geometric feature set and the statistical feature set are associated by using a relationship mapping method to generate an association data set; S134, the geometric feature set, the statistical feature set, and the association data set are stored by using a preset dynamic retrieval database, and the queryable defect feature library is constructed by using an index optimization method.

[0052] In step S131, based on the defect statistical mode, the product surface defect area is selected for image segmentation processing, and the shape and size features are calculated by using an edge detection method to generate a geometric feature set.

[0053] It should be noted that the image segmentation uses an improved U-Net model, which can reach an mIoU of 92% in metal surface defect detection. The Canny algorithm is used for edge detection, the size of the Gaussian filter is set to 5x5, and the high-to-low threshold ratio is 1:3, so as to ensure the integrity of the outline of the small defects.

[0054] In an implementation, in the detection of a certain aerospace valve sealing surface, the U-Net model segments a minimum of 0.1 μm micro-cracks, and after compression of the contour point set extracted by the Canny algorithm, the crack length of 2.15±0.03 mm and the width of 0.18±0.01 mm are calculated, thereby providing accurate geometric features for subsequent analysis.

[0055] In step S132, when the geometric feature set meets the preset geometric feature threshold, the statistical feature set is generated by statistical analysis of defect distribution density.

[0056] It should be noted that the geometric feature threshold is set according to the key region of the part, such as the crack length threshold of 0.5 mm in the root region of the aviation gear. The sliding window method is used for density calculation, and the window size is 5 mm×5 mm and the step is 2 mm.

[0057] In an implementation, the crack length threshold of the mortise and tenon region of a certain turbine disk is 0.3 mm, and the statistical analysis shows that the density in the third quadrant reaches 1.2 per cm² (the density in other regions is less than 0.5 per cm²). Combined with the process log, it is determined that the insufficient cooling rate in this region causes the density to decrease to 0.4 per cm² after adjustment.

[0058] In step S133, the process parameters are obtained from the preset production batch database, the geometric feature set and the statistical feature set are associated by using the relational mapping method, and the associated data set is generated.

[0059] It should be noted that the preset production batch database refers to a relational database that stores complete process parameters (such as temperature, pressure, and speed) of each production batch, and is associated with defect data through batch ID. The relational mapping uses canonical correlation analysis (CCA), takes the geometric feature set such as crack length and width and the statistical feature set such as regional density as the first group of variables, and takes the process parameters (temperature, pressure, etc.) as the second group of variables. The first three pairs of canonical variables (cumulative contribution rate > 85%) are selected to establish a multi-dimensional correlation model. The correlation strength of each feature is quantified by the canonical load coefficient.

[0060] In an implementation, in the production of a certain bearing ring, the CCA analysis shows that in the first canonical variable (contribution rate 62%), the crack length (load 0.72) is strongly correlated with the grinding temperature (load 0.82); and the regional density (load 0.68) is significantly correlated with the cooling time (load 0.75) in the second canonical variable (23%). This result clearly defines the "geometric-statistical-process" three-dimensional correlation feature, and provides a quantitative basis for multi-parameter collaborative optimization.

[0061] In step S134, the geometry feature set, the statistical feature set and the correlation data set are stored in a preset dynamic retrieval database, and an index optimization method is used to construct the queryable defect feature library.

[0062] It should be noted that the preset dynamic retrieval database refers to a distributed database system (such as MongoDB cluster) optimized for fast query, used for raw data storage; and the defect feature library is an upper application database, storing structured feature data. The index strategy includes establishing R-tree spatial index for the geometry feature set, establishing B+ tree range index for the statistical feature set, and using hash index for the correlation data set, so as to realize multi-condition combination query response time < 50 ms.

[0063] In an implementation manner, the system first stores the original detection data into the preset dynamic retrieval database, and constructs the defect feature library after feature extraction. When querying "temperature > 800℃ and crack length > 2mm", through the spatial index of the geometry feature and the combination index of the process parameters, 32 matching results are returned from 100,000 records in 35 ms, and the corresponding statistical features (such as density 0.8 pieces / cm²) and process parameters (temperature 825℃) are displayed.

[0064] In step S14, a new defect in the production process is detected, and similarity matching is performed with the defect feature library to determine the type and potential position of the new defect, including: S141, boundary coordinates are extracted from a new defect image in the production process, shape and size features are calculated to generate a new geometry feature set; S142, the distribution density of the new geometry feature set is statistically analyzed to generate a new statistical feature set; S143, similarity matching is performed between the new geometry feature set, the new statistical feature set and the defect feature library to generate a similar feature set; S144, a random forest method is used to sort the feature importance and calculate the classification confidence of the similar feature set, to obtain the type and potential position of the new defect.

[0065] In step S141, boundary coordinates are extracted from a new defect image in the production process, shape and size features are calculated to generate a new geometry feature set.

[0066] It should be noted that the collection of the new defect image uses a high-resolution industrial camera such as 5 million pixels, 5μm / pixel, and a ring LED light source is used to eliminate reflection interference. The boundary coordinate extraction uses an improved Sobel operator, which adds a Gaussian second derivative to enhance the edge continuity, and the minimum detectable defect size reaches 0.05mm.

[0067] In an implementation, a suspected defect is detected on the surface of an aviation hydraulic valve block. After the image is processed by a Sobel operator, a set of boundary points {(x1, y1),..., (xn, yn)} is obtained. By calculating the minimum circumscribed rectangle, the long axis of the defect is 2.13 mm, the short axis is 0.78 mm, the ellipticity is 0.83, and the area is 1.64 mm². A geometric feature vector containing 7-dimensional features is formed. These features provide shape and size information of the defect, laying a foundation for subsequent classification.

[0068] In step S142, the distribution density of the new geometric feature set is statistically analyzed to generate a new statistical feature set.

[0069] It should be noted that the density analysis adopts a dynamic grid method, and the grid size is adaptively adjusted according to the feature area of the part (the grid of the key area is encrypted to 1 mm x 1 mm). The statistical features include, for example, the number of defects per unit area and the average nearest neighbor distance, which comprehensively represent the spatial distribution characteristics.

[0070] In an implementation, a turbine blade basin area is divided into 3 mm x 3 mm grids, and the statistics show that the defect density of the 5th grid reaches 0.8 / cm² (average 0.2 / cm²), and the average nearest neighbor distance is 1.2 mm (other areas > 3 mm). Combined with K function analysis, it is confirmed that there is significant aggregation in this area (p < 0.01), which provides a basis for process abnormalities.

[0071] In step S143, similarity matching is performed between the new geometric feature set, the new statistical feature set, and the defect feature library to generate a similar feature set.

[0072] It should be noted that the similarity calculation adopts a weighted Euclidean distance, with a geometric feature weight of 0.6 and a statistical feature weight of 0.4. The matching process uses the LSH (Locality-Sensitive Hashing) algorithm for acceleration, with a retrieval time of < 10 ms and a recall rate of > 95% in a 1 million-level feature library.

[0073] In an implementation, after the newly detected defect feature vector is compared with the feature library, 5 most similar records are returned, with a minimum distance normalized value of 0.15. Among them, 3 records correspond to "insufficient cooling type cracks", with a confidence degree mark of 0.88, providing a reliable reference for subsequent classification.

[0074] In step S144, the random forest method is used to sort the feature importance of the similar feature set and calculate the classification confidence degree, to obtain the type and potential location of the new defect.

[0075] It should be noted that the random forest contains 500 decision trees, and the Gini coefficient is used to evaluate the importance of features. The classification confidence is calculated by the voting proportion, and the result is reliable when the highest vote proportion is > 70%. The position prediction adopts the KNN algorithm, and the k value is adaptively selected according to the defect density.

[0076] For example, a new defect of a bearing roller is classified, the crack type probability is 82%, the air hole is 12%, and the inclusion is 6%. The main feature importance ranking is ellipticity (0.62) > area (0.21) > density (0.17). The predicted position is in the middle of the raceway ± 2mm area, and the error with the actual detection position is only 0.8mm.

[0077] In step S15, the time point and position coordinates of the defect are predicted according to the type of the new defect and the potential position, combined with the process parameters of the current production equipment, including: S151, according to the type of the new defect and the potential position, the real-time process parameters of the current production equipment are normalized and the abnormal values are removed, and a defect prediction data set is generated; S152, the time series mean, variance and change rate of the defect prediction data set are extracted by sliding window, and the defect prediction feature set is generated combined with the type of the new defect and the potential position; S153, when the defect prediction feature set and the historical defect data match degree is greater than the preset defect prediction threshold, the time point and position coordinates of the defect are predicted by support vector machine.

[0078] In step S151, according to the type of the new defect and the potential position, the real-time process parameters of the current production equipment are normalized and the abnormal values are removed, and a defect prediction data set is generated.

[0079] It should be noted that the normalization adopts linear normalization method to map temperature, pressure, speed and other parameters to 0-1 interval, for example, temperature 150°C is normalized to 0.5, and at the same time, based on standard deviation method, the abnormal values exceeding ± 3 times of the mean value are removed, to ensure the data quality.

[0080] In one implementation, a certain semiconductor production line normalizes temperature 100-200°C, pressure 0.5-2.0 Pa, and speed 10-50 m / min. Through standardization processing, the dimension difference is eliminated to obtain a first data set.

[0081] In step S152, the time series mean, variance and change rate of the defect prediction data set are extracted by sliding window, and the defect prediction feature set is generated combined with the type of the new defect and the potential position.

[0082] It should be noted that the sliding window method calculates the features at fixed time intervals, for example, the mean, variance and rate of change of temperature are calculated every 2 seconds with a window length of 10 seconds to capture the dynamic changes of the process parameters. The defect prediction feature set is a feature vector composed of three parts: time series features (mean / variance / rate of change), defect type encoding (e.g. crack = 01, pore = 10) and position coordinates. The Min-Max normalization method is used to normalize each feature to the [0, 1] interval.

[0083] In one implementation, when the temperature data in a certain production line window is 150°C, 152°C, 149°C and 151°C, the mean is calculated to be 150.5°C, and the normalized value is 0.51; the variance is 1.25, and the normalized value is 0.42; the rate of change is 0.8°C / s, and the normalized value is 0.38. Combined with the current defect type (crack, encoding 01) and the predicted position (x: 12mm, y: 18mm), the defect prediction feature set is constructed as [0.51, 0.42, 0.38, 0, 1, 0.48, 0.60].

[0084] In step S153, when the matching degree of the defect prediction feature set and the historical defect data is greater than the preset defect prediction threshold, the support vector machine is used to predict the time point and position coordinates of the defect occurrence.

[0085] It should be noted that the matching degree is calculated using the cosine similarity method, and the defect prediction feature set is compared with the last 100 samples in the historical defect data. The preset threshold is 0.85 (empirical value). The support vector machine is selected based on the following conditions: 1) feature dimension ≥ 5 dimensions; 2) number of historical samples ≥ 1000 groups; 3) classification confidence ≥ 0.9. The RBF kernel function is used for prediction, the penalty coefficient C = 1.0, gamma = 0.1, the input is the normalized feature set, and the output is the time point and position coordinates of the defect occurrence.

[0086] In one implementation, the historical data shows that when the feature set similarity reaches 0.88 (greater than the threshold 0.85), the support vector machine predicts the time point as the current time + 300 ± 15 seconds (confidence 92%), and the position coordinates as (x: 12.5 ± 1.2mm, y: 22.3 ± 0.8mm). In practical application, this method improves the prediction accuracy through multi-dimensional feature integration.

[0087] In step S16, the process parameters of the production equipment are adjusted according to the time point and the position coordinates, combined with the preset parameter adjustment rule, to obtain the optimized process parameter configuration, including: S161, according to the time point and the position coordinates, real-time data is collected from the production environment, the deviation of the real-time data from the preset process parameter threshold is calculated, and a deviation data set is generated. S162, calculate the adjustment amplitude of the deviation data set by linear interpolation method, combine the time point and the position coordinate, and generate an adjustment parameter set; S163, adjust the process parameters of the production equipment through the adjustment parameter set, and output the optimized process parameter configuration.

[0088] In step S161, real-time data is collected from the production environment according to the time point and the position coordinate, the deviation of the real-time data from the preset process parameter threshold is calculated, and a deviation data set is generated.

[0089] It should be noted that the data collection is continuously monitored at an interval of 1 second by high-precision sensors to monitor parameters such as temperature, pressure, and speed, and the preset threshold is set in a fixed range, such as a semiconductor production line temperature threshold of 100-200°C, a pressure threshold of 0.5-2.0 Pa, and a speed threshold of 10-50 m / min. When the parameter exceeds the threshold, the comparison module calculates the deviation value and records the deviation direction and amplitude, for example, when the temperature is 210°C and exceeds the upper limit by 10°C, a deviation data set containing the specific deviation value is generated.

[0090] In one implementation, a certain production line detects a temperature deviation of 10°C, a pressure deviation of 0.2 Pa, and a speed deviation of 5 m / min, and the system analyzes the deviation data in association with historical records and finds that the high temperature may cause wafer surface defects, providing a basis for subsequent adjustment.

[0091] In step S162, the adjustment amplitude of the deviation data set is calculated by linear interpolation method, and the adjustment parameter set is generated by combining the time point and the position coordinate.

[0092] It should be noted that the adjustment parameter set includes target parameter values, adjustment rates, execution devices, and space-time constraints, and other key information. Linear interpolation is based on historical optimal adjustment records, and by matching the similarity of the current deviation data set and historical cases, the optimal adjustment strategy is selected. For example, when the temperature needs to be reduced from 210°C to 190°C, the system matches historical cases to adopt an adjustment rate of 4°C / min, and the pressure is simultaneously reduced by 0.1 MPa. The adjustment parameter set also specifies the target area (such as x: 5-10 mm, y: 5-10 mm) and the effective time window (such as 300-600 seconds) to ensure the accuracy of the adjustment.

[0093] In one implementation, a certain production line generates an adjustment parameter set for a predicted defect area (x: 5-10 mm, y: 5-10 mm), including reducing the temperature from 210°C to 190°C at a rate of 4°C / min and reducing the pressure from 2.0 Pa to 1.8 Pa at a rate of 0.02 MPa / min, which is achieved by adjusting the heater power and controlling the air valve, and the adjustment is completed within 5 minutes.

[0094] In step S163, the process parameters of the production equipment are adjusted by the adjusted parameter set, and an optimized process parameter configuration is output.

[0095] It should be noted that the real-time control module adopts a closed-loop feedback mechanism to perform parameter adjustment, such as stabilizing the temperature at 190°C by adjusting the heater power, reducing the pressure to 1.8 Pa by adjusting the gas valve, and maintaining the speed at 45 meters / minute by controlling the motor. The system continuously monitors the adjustment effect and makes fine adjustments to ensure that the optimal parameter configuration is reached at the predicted time point and position coordinates.

[0096] In an implementation manner, after a certain production line performs adjustment, the process parameters reach a stable state at the wafer center of the target area, and the system dynamically optimizes the threshold setting according to historical data, such as automatically correcting the pressure threshold range according to the environmental humidity, to improve the adaptability of the system.

[0097] In step S17, the defect feature library is updated and improved according to the optimized process parameter configuration, the type of the new defect, and the potential position, including: The real-time production data stream generated according to the optimized process parameter configuration is cleaned and formatted to generate structured real-time production data; S171, according to the structured real-time production data, the type of the new defect, and the potential position, the defect distribution rule is calculated by feature extraction to generate preliminary defect distribution features; S172, when the preliminary defect distribution features do not meet the preset real-time defect threshold, relevant historical data in the defect feature library is queried to generate supplementary feature data; S173, the supplementary feature data and the preliminary defect distribution features are analyzed and fused to update the defect feature library, and the improved defect feature library is obtained.

[0098] In step S171, according to the structured real-time production data, the type of the new defect, and the potential position, the defect distribution rule is calculated by feature extraction to generate preliminary defect distribution features.

[0099] It should be noted that the feature extraction adopts a multi-scale analysis method. In the time dimension, the statistical features (mean, variance, gradient) of the process parameters (temperature, pressure, etc.) are calculated through a sliding window. In the spatial dimension, the kernel density estimation (KDE) algorithm is used to calculate the hotspot distribution of the defect position. The defect distribution law refers to the spatiotemporal pattern of the occurrence of defects, including time correlation (such as an increase in the defect rate within 30 minutes after temperature fluctuation) and spatial aggregation (such as an excessive defect density in a specific coordinate region). The preliminary defect distribution feature is a feature vector quantified by these spatiotemporal patterns, and the difference from the final distribution law is that it has not yet been verified by historical data.

[0100] In an implementation manner, a certain production line discovers, through analysis of real-time data, that when the temperature of the edge region of a wafer fluctuates in the interval of 195-200°C (variance > 5°C), the crack defect density of the region within the next 30 minutes reaches 0.15 pieces / cm² (the normal value is 0.05 pieces / cm²). The system extracts the "temperature variance-crack density" feature pair, combines the equipment vibration frequency spectrum feature (50-100Hz energy increase), and forms a preliminary defect distribution feature containing time delay and spatial positioning.

[0101] In step S172, when the preliminary defect distribution feature does not meet the preset real-time defect threshold, relevant historical data in the defect feature library is queried to generate supplementary feature data.

[0102] It should be noted that the preset real-time defect threshold is a multi-dimensional standard set according to the production process requirements, including a defect density threshold, a spatiotemporal correlation threshold, and a feature stability threshold. When the preliminary defect distribution feature exceeds these preset real-time defect thresholds, the system retrieves historical feature data of similar working conditions (process parameter deviation ± 5%, same defect type) from the feature library as supplementary data.

[0103] In an implementation manner, a certain production line detects that the defect density of the center region of a wafer currently reaches 0.12 pieces / cm² (exceeding the threshold of 0.1 pieces / cm²), the system retrieves historical records under similar temperature conditions (200±10°C) in the last 30 days, and finds that when the cooling rate <2°C / s, the defects in the region are mostly distributed in a radial manner (accounting for 68%), and the crack feature of the radial distribution is taken as the supplementary feature data.

[0104] In step S173, the supplementary feature data and the preliminary defect distribution feature are subjected to data analysis and fusion, the defect feature library is updated, and the improved defect feature library is obtained.

[0105] It should be noted that the data fusion adopts a weighted average method based on confidence, and the supplementary feature data and the preliminary feature are integrated according to the confidence ratio. For the repeated feature mode, the feature parameters are updated through the exponential weighted moving average algorithm; for the newly appeared feature mode, it is formally included in the feature library after three batches of production verification (the prediction accuracy is greater than 90%). The improved feature library will optimize the feature index structure, improve the retrieval efficiency of new defects, and at the same time retain the historical feature data for comparison and analysis.

[0106] In an implementation manner, a certain turbine disc production line fuses the "high temperature gradient-radial crack" feature (confidence 85%) found in real time with the historical "low speed cooling-circumferential crack" feature (confidence 80%), and the improved feature library can more accurately predict the defect distribution of the mortise and tenon area. After three batches of verification, the defect prediction accuracy of this area is improved from 82% to 93%, and without establishing a new prediction model, only by optimizing the existing feature parameters and weights, the performance is improved. The system synchronizes the improved feature library to all detection terminals, ensuring the consistency of the quality control standard.

[0107] In step S18, the improved defect feature library is used to continuously detect defects in the production process and make predictions, dynamically adjust the process parameters of the production equipment, and generate stable precision metal part production control parameters based on the response time of the equipment, including: S181, compare the improved defect feature library with the defect feature library, continuously detect defects in the production process and make predictions, dynamically adjust the process parameters of the production equipment, and obtain the adjusted process parameters; S182, according to the adjusted process parameters, send control instructions to the production equipment and monitor the response time, and through iterative optimization of feedback data, obtain the optimized parameters; S183, when the optimized parameters converge to a preset convergence threshold, the stable precision metal part production control parameters are obtained, and output to a designated storage unit.

[0108] In step S181, the improved defect feature library is compared with the defect feature library, the defects in the production process are continuously detected and predicted, and the process parameters of the production equipment are dynamically adjusted to obtain the adjusted process parameters.

[0109] It should be noted that the system compares and analyzes the real-time defect data on the production line, such as scratch, particle adhesion and other defect types and location information on the semiconductor wafer, with the historical patterns in the defect feature library. When the defect distribution pattern changes are detected, the parameter adjustment suggestions are generated based on the process parameter optimization rule library built in the defect feature library. The rule library contains hundreds of verified "defect feature-process parameter" mapping rules, such as "pore density rises 0.1 / cm²→reduce cooling rate 10%+increase mold temperature 5°C".

[0110] In an implementation, a certain turbine blade production line detects that the crack density in the tip region rises from 0.08 / cm² to 0.15 / cm², the system matches the optimization rule "crack density rises→increase preheating temperature+reduce punching speed" in the process parameter optimization rule library, and generates an adjustment scheme of increasing the preheating temperature from 180°C to 195°C and reducing the punching speed from 2.5m / s to 2.2m / s.

[0111] In step S182, control instructions are sent to the production equipment according to the adjusted process parameters, and the response time is monitored, and the optimization parameters are obtained through iterative optimization of feedback data.

[0112] It should be noted that the control instructions are sent to the production equipment in real time through a high-speed industrial network, such as sending temperature adjustment instructions to heating devices and pressure adjustment instructions to hydraulic systems. The system collects equipment response data at an interval of 1 second, and monitors the consistency of the actual change curve of the parameters with the target value.

[0113] In an implementation, after a certain precision metal part production line executes the temperature adjustment instruction, the sensor data shows that the temperature is stably reduced from 1200°C to 1150°C within 5 minutes, and the pressure is adjusted to the target value of 45 megapascals within 3 minutes. The system compares the difference between the real-time defect density and the target value, and uses an iterative calculation method to fine-tune the parameters, such as further adjusting the temperature to 1145°C, which reduces the defect density from the initial 0.05% to the target value of 0.01% after 3 iterations.

[0114] In step S183, when the optimization parameters converge to a preset convergence threshold, the stable precision metal part production control parameters are obtained and output to a designated storage unit.

[0115] It should be noted that the preset convergence threshold includes process parameter stability, defect density compliance and device response consistency requirements. The parameter combination reaching the steady state needs to be verified by real-time production, simulation and historical data. The system sets strict convergence standards, requiring that the parameter fluctuation amplitude is less than 0.5% and the defect density change is less than 0.1% in 5 consecutive iterations. The process parameters reaching the steady state are stored in encrypted form, including complete timestamp and version information.

[0116] In an implementation manner, after 7 rounds of parameter optimization of an aero-engine component production line, the laser power is stabilized in the range of 1580W±5W, and the defect density is maintained at 0.05%±0.008%, the system automatically marks the parameter combination as a golden standard process scheme, and safely stores it for subsequent production batch calling. The storage process uses blockchain technology to ensure that the data cannot be tampered with, and each parameter version generates a unique hash value identifier, such as a1b9...f2c4, which provides a reliable basis for product quality traceability.

[0117] To sum up, the application discloses a kind of precision metal parts real-time micro-defect intelligent control method, including obtaining the historical defect detection data of precision metal parts, and pre-processing, obtain structured defect data set;Time series analysis is carried out on the structured defect data set, the distribution law of defect is extracted and cluster analysis is carried out, to determine the defect statistical mode;Geometric feature set, statistical feature set and associated data set are generated based on the defect statistical mode, and stored using a preset dynamic retrieval database, to obtain a queryable defect feature library;New defect in production process is detected, and similarity matching is carried out with the defect feature library, to judge the type and potential position of new defect;According to the type of new defect and the potential position, combine the process parameters of current production equipment, predict the time point and position coordinates of defect appearance;According to the time point and the position coordinates, combine the preset parameter adjustment rule, adjust the process parameters of production equipment, to obtain the optimized process parameter configuration;According to the optimized process parameter configuration, combine the type of new defect and the potential position, optimize and update the defect feature library, to obtain an improved defect feature library;According to the improved defect feature library, continuously detect defects in production process and make prediction, dynamically adjust the process parameters of the production equipment, generate stable precision metal part production control parameters based on device response time.

[0118] Referring Figure 2 , the second embodiment of the application provides a precision metal part real-time micro-defect intelligent control system, comprising: A data acquisition module is configured to acquire historical defect detection data of precision metal parts and pre-process the data to obtain a structured defect data set. a data analysis module configured to perform time series analysis on the structured defect data set, extract distribution rules of defects, and perform cluster analysis to determine a defect statistical mode; a feature library construction module configured to generate a geometric feature set, a statistical feature set, and an association data set based on the defect statistical mode, and store the sets in a preset dynamic retrieval database to obtain a queryable defect feature library; a classification module configured to detect a new defect in a production process, and perform similarity matching with the defect feature library to determine a type and a potential location of the new defect; a prediction module configured to predict a time point and a coordinate of the new defect based on the type and the potential location of the new defect, and a process parameter of a current production equipment; an optimization module configured to adjust the process parameter of the production equipment based on the time point and the coordinate, and a preset parameter adjustment rule to obtain an optimized process parameter configuration; an updating module configured to optimize and update the defect feature library based on the optimized process parameter configuration, the type of the new defect, and the potential location to obtain an improved defect feature library; a control module configured to continuously detect defects in the production process and perform prediction based on the improved defect feature library, dynamically adjust the process parameter of the production equipment, and generate stable production control parameters of the precision metal part based on a device response time.

[0119] It should be noted that the precision metal part real-time micro-defect intelligent control device provided by the embodiments of the present application is used to execute all process steps of the precision metal part real-time micro-defect intelligent control method and system provided by the embodiments described above, and the working principles and beneficial effects of the two are one-to-one corresponding, and thus will not be repeated.

[0120] The embodiments of the present application further provide an electronic device. The electronic device includes a processor, a memory, and a computer program stored in the memory and executable on the processor, for example, a precision metal part real-time micro-defect intelligent control system program. The processor implements the steps in the above-mentioned various precision metal part real-time micro-defect intelligent control method embodiments when executing the computer program, for example Figure 1 the step S11 shown. Alternatively, the processor implements the functions of each module / unit in the above-mentioned various device embodiments when executing the computer program, for example, the data acquisition module.

[0121] For example, the computer program can be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present application. The one or more modules / units can be a series of computer program instruction segments capable of completing a specific function, which are used to describe the execution process of the computer program in the electronic device.

[0122] The electronic device can be a computing device such as a desktop computer, a notebook computer, a palm computer, and a smart tablet. The electronic device can include, but is not limited to, a processor and a memory. Those skilled in the art can understand that the above components are only examples of the electronic device and do not constitute a limitation on the electronic device, and the electronic device can include more or fewer components than the above, or combine certain components, or different components, for example, the electronic device can also include an input / output device, a network access device, a bus, etc.

[0123] The processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic components, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The processor is the control center of the electronic device, which connects all parts of the electronic device through various interfaces and lines.

[0124] The memory can be used to store the computer program and / or modules, and the processor realizes various functions of the electronic device by running or executing the computer program and / or modules stored in the memory, and calling data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required by a function (such as a sound playing function, an image playing function, etc.), and the like; and the data storage area can store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), and the like. In addition, the memory can include a high-speed random access memory, and can also include a nonvolatile memory, for example, a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state memory devices.

[0125] The modules / units integrated in the electronic device, if realized in the form of software function units and sold or used as independent products, can be stored in a computer readable storage medium. Based on this understanding, all or part of the processes in the above-mentioned embodiment methods can also be completed by a computer program instructing related hardware, and the computer program can be stored in a computer readable storage medium. The computer program can realize the steps of the above-mentioned various method embodiments when executed by a processor. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or some intermediate forms, etc. The computer readable medium can include any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc. It should be noted that the contents included in the computer readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction, for example, in some jurisdictions, according to legislation and patent practice, the computer readable medium does not include electrical carrier signals and telecommunication signals.

[0126] It should be noted that the apparatus embodiments described above are merely illustrative, and the units described as separate units can or can not be physically separate, and the units displayed as units can or can not be physical units, i.e. can be located in one place, or can be distributed to multiple network units. Part or all of the modules can be selected to achieve the purpose of the embodiment scheme according to actual needs. In addition, the connection relationship between the modules in the apparatus embodiment provided by the present application indicates that there is a communication connection between them, which can be implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement it without creative labor.

[0127] The above specific embodiments further illustrate the purpose, technical scheme and beneficial effects of the present application. It should be understood that the above description is only a specific embodiment of the present application and is not intended to limit the protection scope of the present application. It is particularly pointed out that any modification, equivalent replacement, improvement, etc. made by those skilled in the art within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for real-time intelligent control of microscopic defects in precision metal parts, characterized in that, include: Historical defect detection data of precision metal parts are acquired and preprocessed to obtain a structured defect dataset; Time series analysis was performed on the structured defect dataset to extract the distribution patterns of defects and perform cluster analysis to determine the statistical patterns of defects. Based on the aforementioned defect statistical pattern, a geometric feature set, a statistical feature set, and a related dataset are generated and stored using a preset dynamic retrieval database to obtain a queryable defect feature library. New defects in the production process are detected and matched with the defect feature library to determine the type and potential location of the new defects. Based on the type of the new defect and its potential location, and in conjunction with the process parameters of the current production equipment, the time point and location coordinates of the defect's occurrence are predicted. Based on the time point and the location coordinates, and combined with the preset parameter adjustment rules, the process parameters of the production equipment are adjusted to obtain the optimized process parameter configuration; Based on the optimized process parameter configuration, combined with the type of the new defect and the potential location, the defect feature library is optimized and updated to obtain an improved defect feature library; Based on the improved defect feature library, defects in the production process are continuously detected and predicted, and the process parameters of the production equipment are dynamically adjusted. Based on the equipment response time, stable production control parameters for precision metal parts are generated.

2. The intelligent control method for real-time microscopic defects of precision metal parts according to claim 1, characterized in that, The process involves acquiring historical defect detection data of precision metal parts and preprocessing it to obtain a structured defect dataset, including: Obtain historical raw data containing micro-defects. When the data points exceed the preset mean threshold range, remove the noise data using the mean filtering method to generate the first dataset. Edge detection is used to extract the coordinates of defect boundaries in the first dataset, shape features are analyzed, and size features are calculated based on the Euclidean distance between boundary points to generate the second dataset. Principal component analysis is performed on the second dataset to calculate eigenvectors and eigenvalues, extract the location and frequency of defects, and generate a third dataset containing statistical features. The third dataset is normalized, and when the normalized feature values ​​meet the preset format requirements, the structured defect dataset is output.

3. The intelligent control method for real-time microscopic defects of precision metal parts according to claim 1, characterized in that, The step of performing time series analysis on the structured defect dataset, extracting the distribution patterns of defects, and performing cluster analysis to determine the statistical patterns of defects includes: Extract the defect frequency sequence corresponding to the production batch timestamp from the structured defect dataset, and calculate the autoregressive parameters and moving average parameters using the autoregressive moving average method to generate a time series. The time series is analyzed to obtain the defect frequency changes between batches. Combined with the process parameters obtained from the preset production batch database, batch correlation features are extracted, and the defect types are classified by clustering method to obtain the defect type distribution. When the defect type distribution meets the preset defect type distribution threshold, location data is extracted from the batch association features, and the distribution pattern of defects is calculated through statistical analysis. By combining the defect type distribution and the distribution pattern of the defects, the Bayesian method is used to calculate the predicted probability of the defect location, thereby obtaining the defect statistical pattern.

4. The intelligent control method for real-time microscopic defects of precision metal parts according to claim 1, characterized in that, The process involves generating a geometric feature set, a statistical feature set, and a related dataset based on the defect statistical pattern, and storing them in a preset dynamic retrieval database to obtain a queryable defect feature library, including: Based on the aforementioned defect statistical pattern, the defective areas on the product surface are selected for image segmentation processing. The shape and size features are calculated using an edge detection method to generate a geometric feature set. When the geometric feature set meets the preset geometric feature threshold, a statistical feature set is generated by statistically analyzing the density of defect distribution. Process parameters are obtained from a pre-set production batch database, and the geometric feature set and the statistical feature set are associated using a relational mapping method to generate an associated dataset. The geometric feature set, the statistical feature set, and the associated dataset are stored in a preset dynamic retrieval database, and the queryable defect feature library is constructed using an index optimization method.

5. The intelligent control method for real-time microscopic defects of precision metal parts according to claim 1, characterized in that, The process of detecting new defects in the production process and matching them with the defect feature database to determine the type and potential location of the new defects includes: Extract boundary coordinates from images of new defects in the production process, calculate shape and size features to generate a new geometric feature set; The distribution density of the new geometric feature set is statistically analyzed to generate a new statistical feature set; The new geometric feature set, the new statistical feature set, and the defect feature library are matched for similarity to generate a similar feature set; The random forest method is used to rank the similar feature sets by feature importance and calculate the classification confidence to obtain the type and potential location of the new defect.

6. The intelligent control method for real-time microscopic defects of precision metal parts according to claim 1, characterized in that, The step of predicting the time point and location coordinates of the defect based on the type and potential location of the new defect, combined with the process parameters of the current production equipment, includes: Based on the type and potential location of the new defect, the real-time process parameters of the current production equipment are normalized and outliers are removed to generate a defect prediction dataset. The time series mean, variance, and rate of change of the defect prediction dataset are extracted by a sliding window, and a defect prediction feature set is generated by combining the type of the new defect and the potential location. When the matching degree between the defect prediction feature set and the historical defect data is greater than the preset defect prediction threshold, a support vector machine is used to predict the time point and location coordinates of the defect occurrence.

7. The intelligent control method for real-time microscopic defects of precision metal parts according to claim 1, characterized in that, The step of adjusting the process parameters of the production equipment based on the time point and the location coordinates, combined with preset parameter adjustment rules, to obtain an optimized process parameter configuration includes: Based on the time point and the location coordinates, real-time data is collected from the production environment, the deviation between the real-time data and the preset process parameter threshold is calculated, and a deviation dataset is generated. The adjustment range is calculated using a linear interpolation method on the deviation dataset, and an adjustment parameter set is generated by combining the time point and the location coordinates. The process parameters of the production equipment are adjusted by adjusting the set of adjustment parameters, and the optimized process parameter configuration is output.

8. The intelligent control method for real-time microscopic defects of precision metal parts according to claim 1, characterized in that, The step of optimizing and updating the defect feature library based on the optimized process parameters, combined with the type of the new defect and its potential location, to obtain an improved defect feature library includes: The real-time production data stream generated according to the optimized process parameter configuration is cleaned and formatted to generate structured real-time production data; Based on the structured real-time production data, combined with the type of the new defect and its potential location, the defect distribution pattern is calculated through feature extraction to generate preliminary defect distribution features; When the preliminary defect distribution characteristics do not meet the preset real-time defect threshold, relevant historical data in the defect feature library is queried to generate supplementary feature data. The supplementary feature data and the preliminary defect distribution features are analyzed and fused to update the defect feature library, resulting in the improved defect feature library.

9. The intelligent control method for real-time microscopic defects of precision metal parts according to claim 1, characterized in that, The process of continuously detecting and predicting defects in the production process based on the improved defect feature library, dynamically adjusting the process parameters of the production equipment, and generating stable precision metal part production control parameters based on the equipment response time includes: The improved defect feature library is compared with the original defect feature library to continuously detect and predict defects in the production process, and dynamically adjust the process parameters of the production equipment to obtain the adjusted process parameters. Based on the adjusted process parameters, control commands are sent to the production equipment and the response time is monitored. Iterative optimization is performed using feedback data to obtain optimized parameters. When the optimized parameters converge to the preset convergence threshold, the stable precision metal parts production control parameters are obtained and output to the designated storage unit.

10. A real-time intelligent control system for microscopic defects in precision metal parts, characterized in that, include: The data acquisition module is used to acquire historical defect detection data of precision metal parts and preprocess it to obtain a structured defect dataset. The data analysis module is used to perform time series analysis on the structured defect dataset, extract the distribution patterns of defects, perform cluster analysis, and determine the statistical patterns of defects. The feature library construction module is used to generate geometric feature sets, statistical feature sets and related datasets based on the defect statistical patterns, and store them using a preset dynamic retrieval database to obtain a queryable defect feature library. The classification module is used to detect new defects in the production process and perform similarity matching with the defect feature library to determine the type and potential location of the new defects; The prediction module is used to predict the time point and location coordinates of the defect based on the type of the new defect and its potential location, combined with the process parameters of the current production equipment. The optimization module is used to adjust the process parameters of the production equipment according to the time point and the location coordinates, combined with preset parameter adjustment rules, to obtain the optimized process parameter configuration. An update module is used to optimize and update the defect feature library based on the optimized process parameter configuration, combined with the type of the new defect and the potential location, to obtain an improved defect feature library; The control module is used to continuously detect and predict defects in the production process based on the improved defect feature library, dynamically adjust the process parameters of the production equipment, and generate stable production control parameters for precision metal parts based on the equipment response time.

Images