A production line product quality traceability method based on industrial vision

Abstract

The present application provides an industrial vision production line product quality traceability method, comprising: extracting key characteristic parameters that can reflect the degree of tool wear from preprocessed texture image data, including texture orientation angle, texture period and surface roughness parameters, and constructing a multi-dimensional feature vector of the product surface microtexture; based on the multi-dimensional feature vector of the product surface microtexture, combined with the correspondence between the degree of tool wear and surface texture characteristics, obtaining an estimated value of the current tool wear degree, and comparing it with a wear warning threshold; for identified suspected defective products, identifying the defect type by analyzing the surface texture, and comprehensively analyzing material properties and design dimensional tolerance factors to determine whether the surface texture abnormality is caused by tool wear or a defect in the product itself.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to an industrial vision production line product quality traceability method. Background Art

[0002] A key technical challenge in product quality traceability on production lines is accurately identifying subtle changes in the microtexture of product surfaces. The texture features of product surfaces are often very subtle, making them difficult to observe and identify with the naked eye. Furthermore, these texture changes are intricately correlated with the wear of production tools. With continued use, production tools gradually wear out, leading to corresponding changes in the surface texture of the processed product. However, these texture changes can be very subtle in the early stages, making them difficult to detect with traditional quality inspection methods. Failure to promptly detect these subtle changes and predict their impact on product quality can lead to missed opportunities for tool replacement, resulting in a decline in product quality. Therefore, using advanced measurement and analysis techniques to accurately identify and quantify changes in product surface microtexture and establish a precise correlation model between these changes and tool wear remains a key technical challenge that needs to be addressed. Summary of the Invention

[0003] The present invention provides a production line product quality traceability method based on industrial vision, which mainly includes:

[0004] Use imaging equipment to perform microscopic imaging of the product surface, obtain texture image data of the machined surface under different cutting parameters, perform image preprocessing to improve image contrast and clarity, and eliminate interference factors such as material defects, clamping instability, and product defects themselves;

[0005] From the preprocessed texture image data, key characteristic parameters that can reflect the degree of tool wear are extracted, including texture orientation angle, texture period and surface roughness parameters, and a multi-dimensional feature vector of the product surface microtexture is constructed;

[0006] Based on the tool's historical usage data, including the number of products processed, cumulative cutting time, and the hardness of the processed material, the tool's surface texture characteristics at different wear stages are collected and analyzed to establish a corresponding relationship between the degree of tool wear and surface texture characteristics;

[0007] Based on the multi-dimensional feature vector of the product surface microtexture and the corresponding relationship between tool wear and surface texture characteristics, an estimated value of the current tool wear degree is obtained and compared with the wear warning threshold;

[0008] If the current tool wear exceeds the preset threshold, a tool replacement warning is triggered, and the batch of products is marked as suspected defective products. The wear critical point is determined by the vibration signal and cutting force change trend, and the optimal time for tool replacement is determined;

[0009] For products suspected of defects that have been identified, the defect type is identified by analyzing the surface texture. A comprehensive analysis of material properties and design dimensional tolerance factors is conducted to determine whether the surface texture abnormality is caused by tool wear or a defect in the product itself.

[0010] Based on the judgment results, the product surface texture detection results and tool wear warning information are correlated and analyzed to build a two-way feedback loop for product quality prediction and tool life optimization. By continuously collecting production site data, the correspondence between tool wear degree and surface texture characteristics is dynamically updated and optimized.

[0011] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0012] The present invention discloses a method for tracing the quality of production line products based on industrial vision. The method acquires product surface texture images through high-resolution imaging, extracts key characteristic parameters reflecting the degree of tool wear, and establishes a correspondence between tool wear and surface texture characteristics. Combining vibration and cutting force monitoring data, the critical point of tool wear is determined, and tool replacement warning is implemented. At the same time, the present invention can also distinguish between surface anomalies caused by tool wear and defects in the product itself, and construct a two-way feedback mechanism for product quality prediction and tool life optimization. By continuously collecting and analyzing production data and dynamically optimizing the correspondence between tool wear and surface texture characteristics, a balance is achieved between product quality improvement and tool cost savings, thereby improving processing efficiency and product yield. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 This is a flow chart of an industrial vision production line product quality traceability method of the present invention.

[0014] Figure 2 This is a schematic diagram of an industrial vision production line product quality traceability method of the present invention.

[0015] Figure 3 This is another schematic diagram of an industrial vision production line product quality traceability method of the present invention. DETAILED DESCRIPTION

[0016] The following will describe the technical solutions in the embodiments of the present invention in detail with reference to the accompanying drawings. The described embodiments are only a part of the embodiments of the present invention.

[0017] like Figure 1-3. In this embodiment, a method for tracing product quality of a production line using industrial vision may specifically include:

[0018] S101. Use imaging equipment to perform microscopic imaging of the product surface, obtain texture image data of the machined surface under different cutting parameters, perform image preprocessing to improve image contrast and clarity, and eliminate interference factors such as material defects, clamping instability, and product defects themselves.

[0019] An imaging device is used to perform microscopic imaging of the product surface to obtain texture image data of the product surface. The texture image data is preprocessed to adjust the pixel grayscale distribution of the texture image data. If there is noise in the texture image data, the noise is removed. The surface texture features of the texture image data are highlighted according to the Canny edge detection algorithm to obtain an edge-enhanced texture image. Morphological operations, including dilation and erosion, are performed on the edge-enhanced texture image to identify and segment the defective areas in the edge-enhanced texture image. Features of the defective area are extracted, including area, perimeter, roundness and grayscale statistics. The features of the defective area are input into a support vector machine classifier to determine whether the defective area is caused by material defects or interference caused by clamping instability. According to the judgment result of the support vector machine classifier, texture images that are greatly affected by interference factors are screened out.

[0020] For example, high-resolution imaging equipment, including scanning electron microscopes, atomic force microscopes, and laser confocal microscopes, is used to microscopically image the product surface. Appropriate imaging equipment is selected based on the material properties and processing method. Imaging parameters, such as the SEM's acceleration voltage, working distance, and magnification, are set to acquire texture image data of the machined surface at different cutting speeds, feed rates, and cutting depths. The acquired texture image data is preprocessed, using an adaptive histogram equalization algorithm to adjust the image pixel grayscale distribution and improve image contrast. A median filter algorithm is used to remove salt and pepper noise from the image and enhance image clarity. The Canny edge detection algorithm is applied to highlight surface texture features, resulting in an edge-enhanced texture image. Morphological operations, including dilation and erosion, are performed on the edge-enhanced texture image to identify and segment possible defect regions. A region growing algorithm is used to further refine the boundaries of the defect regions. Features of the segmented defect regions are extracted, including area, perimeter, roundness, and grayscale statistics. The extracted features are input into a support vector machine classifier to determine whether the defect region is a material defect or interference caused by fixture instability. Based on the support vector machine classification results, the optimal threshold automatically calculated by the Otsu algorithm is set to screen out texture images significantly affected by interference factors. These affected images are then removed, retaining the remaining valid texture image data. A quality assessment is then conducted on the filtered texture images, calculating the signal-to-noise ratio and clarity metrics to ensure the representativeness and reliability of the remaining image data. Texture image data that pass the quality assessment will serve as the basis for subsequent analysis of the relationship between cutting parameters and machined surface quality.

[0021] A scanning electron microscope (SEM) was used to image the product surface using high-resolution imaging equipment. The acceleration voltage was set to 15 kV, the working distance to 12 mm, and the magnification to 2000x. Texture image data of the machined surface was acquired at a cutting speed of 100 m / min, a feed rate of 0.1 mm / rev, and a cutting depth of 0.5 mm. The acquired texture image was preprocessed. An adaptive histogram equalization algorithm was first applied to divide the image into 8x8 blocks. Histogram equalization was then performed within each block. The blocks were then merged using bilinear interpolation to produce a contrast-enhanced image. A 3x3 median filter was used to remove salt-and-pepper noise and improve image clarity. The Canny edge detection algorithm was applied with a low threshold of 50 and a high threshold of 150 to highlight surface texture features. Morphological operations were performed on the edge-enhanced texture image using a 3x3 structuring element, performing dilation and erosion to identify and segment possible defect regions. A region growing algorithm was used with a growing threshold of 10 to further refine the boundaries of the defect regions. The features of the defective area after segmentation are extracted, including area, perimeter, roundness and grayscale mean. The extracted features are input into the support vector machine classifier, using the radial basis function kernel and the penalty parameter C set to 1.0 to determine whether the defective area is a material defect or interference caused by clamping instability. According to the support vector machine classification results, the Otsu algorithm is used to automatically calculate the optimal threshold to filter out texture images that are greatly affected by interference factors. The signal-to-noise ratio of the filtered texture image is calculated, and the threshold is set to 20dB. The clarity index is calculated using the Laplace operator and the threshold is set to 0.5 to ensure the representativeness and reliability of the remaining image data. The texture image data that passes the quality assessment is used as the data basis for subsequent analysis of the relationship between cutting parameters and machined surface quality.

[0022] S102. Extract key characteristic parameters that can reflect the degree of tool wear from the preprocessed texture image data, including texture orientation angle, texture period and surface roughness parameters, and construct a multi-dimensional feature vector of the product surface microtexture.

[0023] The preprocessed texture image data is subjected to a two-dimensional fast Fourier transform to obtain a two-dimensional power spectrum; the main frequency component is extracted from the two-dimensional power spectrum, and the main direction of the power spectrum is calculated according to the main frequency component to determine the texture orientation angle; the texture period is obtained by calculating the main frequency of the power spectrum; the statistical characteristics of the texture image are calculated using the gray level co-occurrence matrix method, wherein a distance parameter and four directions are set, the contrast, entropy, energy and correlation of the four directions are calculated, and the average value is taken as the surface roughness statistical characteristic parameter; the texture image is subjected to a three-layer decomposition to extract horizontal, vertical and diagonal detail coefficients at different scales; the energy of the detail coefficient at each scale is calculated according to the detail coefficient to obtain the wavelet energy feature; the texture orientation angle, the texture period, the surface roughness statistical characteristic parameter and the wavelet energy feature are combined to construct a feature vector; each feature in the feature vector is normalized to obtain the normalized feature; the normalized features are spliced ​​to form a multi-dimensional feature vector of the product surface microtexture.

[0024] For example, a two-dimensional fast Fourier transform is performed on the preprocessed texture image data to calculate the two-dimensional power spectrum, and the main frequency component is extracted from the power spectrum. By calculating the main direction of the power spectrum, the texture orientation angle is determined. This angle reflects the change in cutting direction caused by cutting tool wear. The main frequency of the power spectrum is calculated to obtain the texture period. The periodic change reflects the periodic change of the machined surface caused by tool wear. The gray-level co-occurrence matrix method is used to calculate the statistical characteristics of the texture image. The distance parameter is set to 1 pixel and the directions are 0°, 45°, 90°, and 135°. The contrast, entropy, energy, and correlation in the four directions are calculated, and the average value is taken as the statistical characteristic parameters of the surface roughness. These parameters change with increasing tool wear, reflecting the deterioration of surface quality. The texture image is decomposed into three layers using the Daubechies wavelet to extract horizontal, vertical, and diagonal detail coefficients at different scales. The energy of the detail coefficient at each scale is calculated to obtain nine wavelet energy features. The wavelet energy features are sensitive to the multi-scale changes in surface texture caused by tool wear and can reflect different degrees of tool wear. The extracted texture orientation angle, texture period, four surface roughness statistical features, and nine wavelet energy features were combined to construct a 15-dimensional feature vector. Each feature was normalized using a minimum-maximum normalization process to scale all eigenvalues ​​to a range of 0-1. The normalized features were then sequentially concatenated to form a multi-dimensional feature vector of the product surface microtexture, which was used for subsequent tool wear analysis.

[0025] A two-dimensional fast Fourier transform (FFT) was performed on the preprocessed 512x512 pixel texture image data to obtain a frequency domain representation. A two-dimensional power spectrum was calculated, and the dominant frequency component was determined by finding the location of the maximum point in the power spectrum. Assuming the coordinates of the maximum point are (100, 200), the texture orientation angle θ = arctan(200 / 100) ≈ 63.4°, and the texture period T = 512 / sqrt(100^2 + 200^2) ≈ 2.3 pixels. An increase in θ indicates a deviation from the cutting direction, while a decrease in T reflects an increase in surface periodicity, both of which indicate increased tool wear. Next, the gray-level co-occurrence matrix method was used to calculate texture statistical features. With a distance parameter of 1 pixel and directions of 0°, 45°, 90°, and 135°, the contrast, entropy, energy, and correlation were calculated for the four directions. For example, the contrast in the 0° direction is 1.5, the entropy is 4.2, the energy is 0.08, and the correlation is 0.85. The average values ​​in the four directions were taken as the surface roughness statistical features. Subsequently, the image was decomposed into three layers using the Daubechies4 wavelet, resulting in 10 subbands: LL3, LH3, HL3, HH3, LH2, HL2, HH2, LH1, HL1, and HH1. The energies of the nine detail subbands (LH3, HL3, HH3, LH2, HL2, HH2, LH1, HL1, and HH1) were calculated, with values ​​such as 0.05 for LH3, 0.06 for HL3, and 0.03 for HH3, forming a 9-dimensional wavelet energy eigenvector. Finally, a 15-dimensional feature vector was constructed by combining one texture orientation angle, one texture period, four surface roughness statistical features, and nine wavelet energy features. Each feature was subjected to minimum-maximum normalization to scale all eigenvalues ​​to the range of 0–1. For example, the normalized texture orientation angle is 0.7, the texture period is 0.4, the contrast is 0.6, the entropy is 0.8, the energy is 0.3, the correlation is 0.9, and the wavelet energy feature is [0.2, 0.3, 0.1, 0.4, 0.5, 0.2, 0.6, 0.7, 0.3]. These normalized features are sequentially concatenated to form the final 15-dimensional feature vector for subsequent tool wear analysis.

[0026] S103. Based on the historical usage data of the tool, including the number of processed products, the cumulative cutting time, and the hardness of the processed material, the surface texture characteristics of the tool processing at different wear stages are collected and analyzed to establish a corresponding relationship between the degree of tool wear and the surface texture characteristics.

[0027] Based on the tool's historical usage data, the number of processed products, the cumulative cutting time, and the hardness of the processed material are obtained, and a quantitative indicator of the tool wear degree is calculated through weighted summation. The surface of the product processed by the tool is microscopically imaged to obtain a surface texture image. The texture orientation angle, texture period, and surface roughness characteristic parameters are extracted from the surface texture image, and the trend of the characteristic parameters changing with the degree of wear is recorded. A mapping relationship between the tool wear index and the surface texture characteristic parameters is established, and the radial basis function is selected as the kernel function for model training. For the newly processed product surface, the texture characteristic parameters are extracted and substituted into the model to calculate the wear index of the current tool. The degree of wear of the tool is judged according to the preset wear degree classification standard.

[0028] For example, a tool wear index is constructed based on the tool's historical usage data, including the number of processed products, the cumulative cutting time, and the hardness of the processed material. The weight of the number of processed products is set to 0.3, the weight of the cumulative cutting time is set to 0.4, and the weight of the processed material hardness is set to 0.3. The quantitative index of the degree of tool wear is calculated by weighted summation. A scanning electron microscope is used to perform microscopic imaging of the surface of products processed by tools with different degrees of wear to obtain surface texture images. The scanning electron microscope acceleration voltage is set to 15kV, the working distance is set to 10mm, and the magnification is set to 2000 times. 50 surface texture images of each of the mild, moderate, and severe wear stages are collected. Feature parameters such as texture orientation angle, texture period, and surface roughness are extracted from the collected surface texture images, and the trend of these parameters changing with the degree of wear is recorded. The support vector regression algorithm is used to establish the mapping relationship between the tool wear index and the surface texture feature parameters. The radial basis function is selected as the kernel function, the penalty factor C is set to 10, and the epsilon parameter is set to 0.1 for model training. The model performance is evaluated using the cross-validation method, and the model with the smallest mean square error is selected as the final model. For newly processed product surfaces, texture feature parameters are extracted and substituted into a trained support vector regression model to calculate the tool wear index. A wear severity grading standard is set: a wear index less than 0.3 is considered light wear, 0.3 to 0.7 is considered moderate wear, and greater than 0.7 is considered heavy wear. Tool wear severity is determined based on this grading. For a high-speed steel tool, historical usage data is collected, including 5,000 processed products, 100 hours of cumulative cutting time, and an average hardness of HRC45 for the processed material. After setting a weighting coefficient, the tool wear index is calculated to be 0.62. The surface of the product processed by this tool is imaged using a scanning electron microscope with an accelerating voltage of 15 kV, a working distance of 10 mm, and a magnification of 2,000x. Fifty surface texture images are collected, and feature parameters are extracted using an image processing algorithm, including an average texture orientation angle of 35°, an average texture period of 2.5 μm, and an average surface roughness Ra of 1.2 μm. These feature parameters, along with a tool wear index of 0.62, serve as training data. This process was repeated to collect tool data with different degrees of wear, and a total of 150 sets of training data were obtained. The support vector regression algorithm was used to establish the mapping relationship, the radial basis function was selected as the kernel function, the penalty factor C was set to 10, and the epsilon was set to 0.1. The model was trained using the 5-fold cross-validation method, and the optimal model with a mean square error of 0.025 was obtained. The surface of the newly processed product was imaged and feature extracted, and the texture orientation angle was 40°, the texture period was 2.8μm, and the surface roughness Ra was 1.5μm. These parameters were input into the trained model, and the wear index of the current tool was calculated to be 0.75. According to the preset grading standard, the tool was judged to be in a state of severe wear, and it was recommended to replace the tool.

[0029] S104. Based on the multi-dimensional feature vector of the product surface microtexture and the corresponding relationship between the tool wear degree and the surface texture characteristics, an estimated value of the current tool wear degree is obtained, and compared with the wear warning threshold.

[0030] A microtexture image of the product surface is acquired, and characteristic parameters such as texture orientation angle, texture period, and surface roughness are extracted from the microtexture image. These characteristic parameters are then normalized to obtain standardized characteristic parameters. A multidimensional feature vector is constructed based on the standardized characteristic parameters, combined in a predetermined order. A K-nearest neighbor algorithm is used to calculate the Euclidean distance between the multidimensional feature vector and all records in a pre-established database of tool wear and surface texture feature correspondences. The K records with the smallest distances are selected as the most similar samples. The K selected most similar samples are weighted according to their similarity to the multidimensional feature vector, with the similarity weights calculated using a Gaussian kernel function. A weighted average method is used to calculate the weighted average of the tool wear corresponding to the K samples to obtain an estimated wear value for the current tool. The estimated tool wear value is compared with a preset wear warning threshold. If the estimated tool wear value exceeds the warning threshold, the current tool is determined to have reached a warning state.

[0031] For example, characteristic parameters such as texture orientation angle, texture period, and surface roughness are extracted from the microtexture image of the product surface. The extracted characteristic parameters are normalized, converting each parameter value to a range of 0-1. The normalized characteristic parameters are combined in a predetermined order to construct a multidimensional feature vector. Matching is performed using a pre-established database of tool wear levels and surface texture features. This database contains historical processing data, with each record consisting of a feature vector and a corresponding tool wear level. The database is updated regularly by collecting new processing data to maintain data currency. Using the K-nearest neighbor algorithm, the Euclidean distance between the current feature vector and all records in the database is calculated, and the K records with the smallest distances are selected as the most similar samples. The K value is set to 5 to balance computational efficiency and result accuracy. The K most similar samples are weighted based on their similarity to the current feature vector. The similarity weight is calculated using a Gaussian kernel function, with closer distances giving a greater weight. Using a weighted average method, the weighted average of the tool wear levels corresponding to the K samples is calculated to obtain an estimated tool wear level. This calculated tool wear level estimate is compared with a preset wear warning threshold. The warning threshold is determined by analyzing the statistical distribution of historical data and is set at the 75th percentile of the wear level. If the estimated value exceeds the warning threshold, the tool is considered to have reached a warning state. Based on the degree of threshold exceeding, the warning is categorized into three levels: mild, moderate, and severe, corresponding to threshold exceeding 0-10%, 10-20%, and above 20%, respectively. For a high-speed steel milling cutter, for example, feature parameters extracted from the microtexture image of the workpiece machined by the cutter include a texture orientation angle of 35°, a texture period of 2.5 μm, and a surface roughness Ra of 1.2 μm. These parameters are normalized by dividing the texture orientation angle by 360° to obtain 0.0972, the texture period by 10 μm to obtain 0.25, and the surface roughness by 5 μm to obtain 0.24. These parameters form the feature vector [0.0972, 0.25, 0.24]. Using a database containing 10,000 historical records, the K-nearest neighbor algorithm (K=5) is used to calculate the Euclidean distance and identify the five most similar samples. Assume that the eigenvectors of these five samples and the corresponding tool wear degrees are [0.1, 0.26, 0.23] corresponding to 0.65, [0.09, 0.24, 0.25] corresponding to 0.68,

[0032] [0.11, 0.25, 0.24] corresponds to 0.70, [0.095, 0.27, 0.22] corresponds to 0.66,

[0033] [0.10, 0.23, 0.26] corresponds to 0.69. Using a Gaussian kernel function to calculate similarity weights, the resulting weights are 0.25, 0.30, 0.15, 0.20, and 0.10, respectively. A weighted average calculation yields an estimated value for the current tool wear level of 0.6725. Historical data analysis shows that the 75th percentile for wear level is 0.65, which is set as the warning threshold. The current estimate of 0.6725 exceeds the threshold of 0.65 by approximately 9%, indicating a mild warning state and prompting close attention to tool wear.

[0034] S105. If the current tool wear exceeds the preset threshold, a tool replacement warning is triggered, and the batch of products is marked as suspected defective products. The wear critical point is determined by the vibration signal and the cutting force change trend to determine the optimal time for tool replacement.

[0035] Real-time monitored tool wear data is obtained and compared with a preset wear threshold, which is determined by analyzing the statistical distribution of historical processing data; if the tool wear exceeds the wear threshold, a tool replacement warning signal is triggered; based on the tool replacement warning signal, a tool replacement request is generated; based on the tool replacement request, inventory is automatically checked, and if the inventory is sufficient, the tool preparation process is started; vibration signals and cutting force data are collected and preprocessed in real time; a tool status score is calculated based on the tool wear, the vibration signal and the cutting force data, and when the tool status score is lower than the preset threshold, it is determined to be the best time to replace the tool, and the tool replacement process is triggered.

[0036] For example, real-time tool wear data is compared with a preset wear threshold. The wear threshold is determined by analyzing the statistical distribution of historical processing data and is set at the 75th percentile of the wear threshold. If the wear exceeds the threshold, a tool change warning signal is triggered and transmitted to the production management system. The production management system automatically generates a tool change request and automatically checks inventory. If sufficient inventory is available, the tool preparation process is initiated, including tool removal, presetting, and transportation to the processing area. Product batch tracking is performed, marking the currently processed product batch as potentially defective and recording relevant information. For marked batches, the quality inspection frequency is increased from one per 100 pieces to one per 50 pieces, and a special mark is added to the product for subsequent tracking. Vibration sensors and dynamometers installed on the processing equipment collect vibration signals and cutting force data in real time. Fast Fourier transform is performed on the vibration signals to extract the main frequency components and amplitudes. Wavelet analysis is applied to the cutting force data to extract energy characteristics at different scales. The extracted characteristic parameters are compared with a pre-established database of critical wear points to determine whether the tool has reached critical wear point. A weighted scoring method is used to calculate a tool condition score, taking into account wear, vibration signals, and cutting force data. When the score falls below a preset threshold, the optimal time for tool replacement is determined, triggering an automated replacement process. For example, real-time monitoring of a high-speed steel milling cutter revealed a wear level of 0.76. By analyzing historical data, the 75th percentile was determined to be 0.70, which was set as the wear threshold. Because 0.76 exceeded 0.70, the system triggered a tool replacement warning signal and sent it to the production management system. The production management system generated a tool replacement request and sent it to the tool library management module. The tool library automatically checked and found that there were five high-speed steel milling cutters in stock, meeting the replacement requirement, and initiated the preparation process. Simultaneously, the product batch tracking system marked the 100 products currently being processed as potentially defective, recorded this information in the quality control database, and increased the inspection frequency to one in every 50 pieces. The signal collected by the vibration sensor, after fast Fourier transform, showed a dominant frequency of 1200 Hz and an amplitude of 0.5 g. The cutting force data collected by the dynamometer was subjected to wavelet analysis, and the energy eigenvalue in the third-level wavelet decomposition was 0.8. Comparing these characteristics with the wear critical point database revealed that the vibration amplitude exceeded the critical value of 0.4g, and the cutting force energy characteristic approached the critical value of 0.85. Using a weighted scoring method with a weight of 0.4 for wear level, 0.3 for vibration signal, and 0.3 for cutting force data, the calculated tool status score was 68 out of 100. This score was below the preset threshold of 70, indicating that the optimal time for tool replacement had been reached, automatically triggering the tool replacement process.

[0037] Vibration sensors and cutting force sensors are used to monitor the machining process in real time, extract the characteristic frequencies and amplitudes of vibration and cutting force, and establish a corresponding relationship between the vibration amplitude, spindle current, cutting force and the degree of tool wear. When the vibration amplitude suddenly increases and the cutting force continues to rise and exceeds the preset threshold, it is determined that the tool has reached the critical point of wear.

[0038] The vibration and cutting force signals collected in real time during machining are de-noised to obtain denoised vibration and cutting force signals. Fast Fourier transform is used to extract the characteristic frequencies and amplitudes of the vibration and cutting force from these denoised vibration and cutting force signals. Spindle current is collected using a Hall effect current sensor and processed through low-pass filtering to obtain stable current data. A correlation is established between tool wear and vibration amplitude, spindle current, and cutting force. Based on this correlation data, a support vector regression algorithm is used to train a model. The vibration amplitude, spindle current, and cutting force data are input to the model to determine the corresponding tool wear estimate. Warning thresholds for vibration amplitude and cutting force are set based on tool type and workpiece material using multivariate statistical analysis. Current vibration amplitude and cutting force data are compared in real time with pre-set thresholds, and the trained support vector regression model is used to estimate the current tool wear. A weighted scoring method is used to assign weights to vibration amplitude, cutting force, and estimated wear, and a comprehensive score is calculated. If the comprehensive score exceeds a preset critical value, or the vibration amplitude exceeds a sudden increase threshold, or the continuous measurement value of the cutting force exceeds a continuous increase threshold, it is determined that the tool has reached a critical point of wear.

[0039] For example, vibration sensors and cutting force sensors are used to collect vibration and cutting force signals during machining in real time. Bandpass filters are used to remove noise, and fast Fourier transforms are used to extract the characteristic frequencies and amplitudes of vibration and cutting forces. A Hall effect current sensor is used to collect spindle current, which is then low-pass filtered to obtain stable current data. A database is established that correlates tool wear with vibration amplitude, spindle current, and cutting force, and a support vector regression model is trained. A radial basis function is selected as the kernel function, with the regularization parameter C set to 1.0 and the epsilon parameter set to 0.1. Vibration amplitude, spindle current, and cutting force data are input and the corresponding tool wear estimate is output. Warning thresholds for vibration amplitude and cutting force are set, and multivariate statistical analysis is used to determine the thresholds based on tool type and machining material. For high-speed steel tools machining 45-grade steel, the 90th percentile of the vibration amplitude is used as the sudden increase threshold, and a continuous increase is defined as five consecutive cutting force measurements exceeding the 85th percentile. The current vibration amplitude and cutting force data are compared with preset thresholds in real time, and the trained support vector regression model is used to estimate the current tool wear. A weighted scoring method was used, with a weight of 0.3 for vibration amplitude, 0.3 for cutting force, and 0.4 for estimated wear severity, to calculate a comprehensive score. If the comprehensive score exceeds a preset threshold, or if the vibration amplitude exceeds a sudden increase threshold, or if the cutting force exceeds a sustained increase threshold for five consecutive measurements, the tool is deemed to have reached critical wear, triggering a tool replacement process. For example, the raw signal collected by the vibration sensor was bandpass filtered from 50 to 5000 Hz and then, through a fast Fourier transform, determined to have a dominant frequency of 1200 Hz and an amplitude of 0.5 g. The signal collected by the cutting force sensor was bandpass filtered from 100 to 2000 Hz, resulting in a dominant frequency of 800 Hz and an amplitude of 500 N. The spindle current collected by the Hall effect current sensor was low-pass filtered at 20 Hz and stabilized at 15 A. These data were input into a trained support vector regression model with an RBF kernel function, C = 1.0, and epsilon = 0.1, resulting in an estimated current tool wear severity of 0.68. Based on historical data statistics, the threshold for sudden increase in vibration amplitude is set at 0.8g (90% percentile), and the threshold for continuous increase in cutting force is set at 550N (85% percentile). Real-time data shows that the vibration amplitude of 0.5g does not exceed the threshold, but the cutting force is measured for five consecutive times, namely 520N, 530N, 535N, 542N, and 551N, with the last value exceeding the threshold. Using the weighted scoring method, the vibration amplitude score is (0.5 / 0.8)*30=18.75, the cutting force score is (551 / 550)*30=30.05, the estimated wear degree score is (0.68 / 0.75)*40=36.27, and the comprehensive score is 85.07. The set comprehensive score critical value is 80, and the current score exceeds the critical value. Comprehensively judged that the tool has reached the critical point of wear, the tool replacement process should be automatically triggered.

[0040] S106. For products suspected of being defective, the defect type is identified by analyzing the surface texture, and the material properties and design dimensional tolerance factors are comprehensively analyzed to determine whether the abnormal surface texture is caused by tool wear or a defect in the product itself.

[0041] Perform a surface scan on the identified suspected defective products, and use a gray-level co-occurrence matrix to extract texture features, which include contrast, entropy, energy, and correlation. At the same time, local texture information is extracted through local binary patterns. Surface texture feature vectors are constructed based on the texture features, and the surface texture feature vectors are classified to obtain preliminary defect type identification results and their probability distributions. The material property data of the product is obtained, and the dimensional tolerance requirements in the design drawings are obtained. The material, tolerance, and process feature vectors are established using the material property data, the dimensional tolerance requirements, and the current processing parameters. If the material, tolerance, and process feature vectors are obtained, a comprehensive analysis is performed in combination with the defect type identification results and their probability distribution. Through the comprehensive analysis, it is determined whether the surface texture anomaly is caused by tool wear or a defect in the product itself, and the confidence level of the judgment result is output.

[0042] For example, a high-resolution surface scan of identified suspected defective products is performed, and texture features, including contrast, entropy, energy, and correlation, are extracted using a gray-level co-occurrence matrix. Local texture information is extracted using local binary patterns, and multi-scale texture features are obtained through wavelet transform. All extracted features are normalized to construct a comprehensive surface texture feature vector. The surface texture feature vector is classified using a support vector machine algorithm, selecting the radial basis function as the kernel function and setting the penalty factor C to 1.0. Texture anomalies are classified into two categories: wear and defects, resulting in preliminary defect type identification results and their probability distribution. Material property data for the product, including hardness, strength, and thermal conductivity, is extracted from the product database. Dimensional tolerance requirements from the design drawings are also obtained, and current processing parameters, such as cutting speed, feed rate, and cutting depth, are added to establish a material, tolerance, and process feature vector. Combining the surface texture classification results and their probability distribution with the material, tolerance, and process feature vectors, a random forest algorithm is used for comprehensive analysis. The number of decision trees was set to 100, with a maximum tree depth of 10. A majority voting principle was used to determine whether surface texture anomalies were caused by tool wear or product defects, and the confidence level of the judgment was output. Taking a milled surface of 45 steel as an example, surface topography data was acquired using a 3D surface scanner with a resolution of 0.1 μm. A gray-level co-occurrence matrix was calculated, yielding a contrast of 1.5, an entropy of 4.2, an energy of 0.08, and a correlation of 0.85. Local texture features were extracted using the LBP algorithm, resulting in a 256-dimensional feature vector. Multi-scale texture features were then derived through a five-layer wavelet decomposition. All features were normalized to the range of 0–1, resulting in a 768-dimensional comprehensive surface texture feature vector. Classification was performed using a support vector machine with an RBF kernel function and C = 1.0, yielding a preliminary classification of the surface as wear with a probability of 0.75. Material properties extracted from the database include hardness HRC45, tensile strength 600 MPa, and thermal conductivity 50.2 W / (m·K); the design tolerance requirement is flatness 0.02 mm, and the current machining parameters are cutting speed 100 m / min, feed rate 0.1 mm / r, and cutting depth 0.5 mm. This data is combined into a 15-dimensional material, tolerance, and process feature vector. A random forest algorithm with 100 trees and a maximum depth of 10 is used to comprehensively analyze all features. After majority voting, 75 decision trees determined that the cause was tool wear, and 25 determined that the cause was a product defect. The final output was a surface anomaly caused by tool wear, with a confidence level of 75%.

[0043] Combined with the product CAD model, key dimensions and tolerance requirements are extracted, and a correlation model between material properties, dimensional tolerances and surface texture characteristics is established. If the product surface texture anomaly is consistent with the material properties and tolerance requirements, it is judged to be a defect of the product itself; if the degree of anomaly exceeds the allowable range of the material and tolerance and is consistent with the tool wear characteristics, it is judged to be caused by tool wear.

[0044] Extract key dimensions and tolerance requirements from the product CAD model to obtain the product's geometric features, dimensional values, and tolerance ranges. Utilize the support vector regression method to establish a correlation model between material properties, dimensional tolerances, and surface texture features. Based on the correlation model, input standardized material properties and tolerance data to obtain the expected surface texture parameter range. Compare the surface texture anomaly with the prediction results of the correlation model, and set the judgment threshold to a multiple of the standard deviation of the prediction range. If the degree of anomaly is within the threshold, it is determined to be a defect in the product itself; if it exceeds the threshold, it is matched with the tool wear feature library. Use cosine similarity to calculate the similarity between the anomaly and the features in the library to determine whether the anomaly is caused by tool wear.

[0045] For example, a material property database is established, recording the hardness, toughness, and thermal conductivity parameters of different materials. This data is stored in a structured data format and indexed to improve query efficiency. Material property data is obtained through standardized experimental testing, and literature data is collected for cross-validation to ensure data accuracy. Key dimensions and tolerance requirements are extracted from the product CAD model. A rule-based feature recognition algorithm is used to identify the product's main geometric features, obtaining their dimensional values ​​and tolerance ranges. For complex shapes, a convolutional neural network is used for feature extraction and classification. A support vector regression method is used to establish a correlation model between material properties, dimensional tolerances, and surface texture features. A radial basis kernel function is selected, with a penalty factor C set to 1.0 and an epsilon parameter set to 0.1. Standardized material property and tolerance data are input, and the expected surface texture parameter range is output. K-fold cross-validation is used to evaluate model performance, and the model with the lowest mean squared error is selected as the final model. Observed surface texture anomalies are compared with the correlation model's predictions, with a threshold set at 1.5 standard deviations of the predicted range. If the anomaly is within the threshold, it is considered a product defect; if it exceeds the threshold, it is matched against the tool wear feature library. A tool wear feature library is established and regularly updated using historical data statistics and expert knowledge. Cosine similarity is used to calculate the degree of similarity between observed anomalies and features in the library. If the similarity exceeds 0.8, the anomaly is determined to be caused by tool wear. For a 45-grade steel part, the material property database retrieves its hardness of HRC 45, toughness of 100 J / cm², and thermal conductivity of 50.2 W / (m·K). Key dimensions are extracted using a CAD model feature recognition algorithm: diameter 50 ± 0.02 mm and length 100 ± 0.05 mm. This data is input into a support vector regression model with C = 1.0 and epsilon = 0.1, predicting a surface roughness Ra range of 1.2 to 1.8 μm. After actual machining, a precision roughness meter measured the surface roughness to be 2.1 μm, exceeding the predicted range by 1.5 standard deviations (0.3 μm). The tool wear feature library is automatically matched to 100 sets of historical wear data. The calculated maximum cosine similarity between the current anomaly and the features in the library is 0.85, exceeding the threshold of 0.8. The overall judgment is that the surface anomaly is caused by tool wear, not a product defect. This triggers the tool replacement process, and this data is added to the wear feature library for subsequent analysis and prediction.

[0046] S107. Based on the judgment results, the product surface texture detection results and the tool wear warning information are correlated and analyzed to establish a two-way feedback of product quality prediction and tool life optimization. By continuously collecting production site data, the correspondence between the tool wear degree and surface texture characteristics is dynamically updated and optimized.

[0047] The system obtains product surface texture detection results and tool wear warning information; measures the degree of tool wear to obtain tool wear data. Based on the tool wear data and processing parameters, a product quality prediction model is established, using a radial basis function as the kernel function. The output of the product quality prediction model is compared with preset quality standards to determine product conformance. If the product fails, tool replacement timing is optimized using a genetic algorithm. The genetic algorithm sets chromosome codes to represent different tool replacement strategies. Based on the optimization results of the tool replacement strategy, an online learning algorithm is used to implement incremental learning. The online learning algorithm continuously collects production site data and updates the parameters of the product quality prediction model.

[0048] For example, a correlation data model is constructed between product surface texture detection results and tool wear warning information. A time-series data storage structure is used to record the surface texture parameters and corresponding tool wear levels during each machining process. Tool wear levels are measured using a combination of optical sensors and vibration signal analysis, enabling real-time monitoring. A product quality prediction model is established using a support vector regression algorithm, selecting a radial basis function as the kernel function, setting the penalty factor C to 1.0, and epsilon to 0.1. Tool wear levels and machining parameters are input, and the predicted product surface texture parameters are output. The predicted results are compared with pre-set quality standards to determine product conformity, and the results are fed back to the tool change strategy optimization module. Tool change timing is optimized using a genetic algorithm, with chromosome encoding representing different tool change strategies. New strategies are generated through crossover and mutation operations. A multi-objective fitness function is constructed, including product qualification rate, tool cost, equipment utilization, and product delivery time, with weight coefficients set to 0.4, 0.3, 0.2, and 0.1, respectively, to balance product quality and tool cost. Iterative optimization is performed to determine the optimal tool change timing, and the results are applied to production scheduling. An online learning algorithm achieves incremental learning by continuously collecting production site data. The parameters of the product quality prediction model and tool life optimization algorithm are updated every 100 new samples. A sliding window method is used to evaluate the performance of the updated model. If the performance degrades by more than 5%, a model retraining process is triggered. The correspondence between tool wear and surface texture features is dynamically adjusted to achieve adaptive model optimization. For example, a high-speed steel milling cutter machining 45-grade steel was used. The associated database records data from nearly 1000 machining runs. An optical sensor measured tool wear of 0.15 mm. Vibration signal analysis revealed a dominant frequency of 1200 Hz and an amplitude of 0.5 g, resulting in a comprehensive assessment of tool wear of 0.68. This data was fed into a support vector regression model with C = 1.0 and epsilon = 0.1. The predicted surface roughness Ra was 1.8 μm, exceeding the preset quality standard of 1.6 μm by 12.5%. A genetic algorithm was used to optimize the tool replacement strategy, with a population size of 100 and 50 iterations. The fitness function calculation results are: product qualification rate 95% (weight 0.4), tool cost reduction 8% (weight 0.3), equipment utilization rate increase 5% (weight 0.2), product delivery time shortened 2%

[0049] (weight 0.1), resulting in an overall score of 0.856. After optimization, it was recommended that tool wear be replaced when the tool wear level reached 0.75. After collecting 100 new samples, the online learning algorithm updated the model parameters. A sliding window method with a window size of 200 was used to evaluate the performance of the updated model. The mean square error decreased from 0.025 to 0.022, a 12% performance improvement. The automatic adjustment of the correspondence between tool wear and surface texture features increased the weight of wear in the roughness Ra prediction model from 0.6 to 0.65, reflecting the enhanced impact of tool condition on machining quality.

[0050] The above description is merely a preferred embodiment of the present application and an illustration of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in this application is not limited to the technical solutions formed by the specific combination of the above-mentioned technical features, but also encompasses other technical solutions formed by any combination of the above-mentioned technical features or their equivalents without departing from the concept of this application. For example, a technical solution formed by replacing the above-mentioned features with (but not limited to) technical features with similar functions disclosed in this application.

Claims

1. A production line product quality traceability method based on industrial vision, characterized in that: The method comprises: using an imaging device to perform microscopic imaging on the surface of a product, obtaining texture image data of the machined surface under different cutting parameters, performing image preprocessing to improve image contrast and clarity, and eliminating interference factors such as material defects, clamping instability, and defects of the product itself; extracting key characteristic parameters that can reflect the degree of tool wear from the preprocessed texture image data, including texture orientation angle, texture period, and surface roughness parameters, and constructing a multi-dimensional feature vector of the micro-texture of the product surface; collecting and analyzing the texture features of the tool machined surface at different wear stages based on the historical usage data of the tool, including the number of processed products, the cumulative cutting time, and the hardness of the processed material, and establishing a corresponding relationship between the degree of tool wear and the surface texture features; combining the multi-dimensional feature vector of the micro-texture of the product surface with the degree of tool wear and the surface texture features The corresponding relationship between them is used to obtain the estimated value of the wear degree of the current tool and compare it with the wear warning threshold; if the current tool wear degree exceeds the preset threshold, the tool replacement warning is triggered, and the current batch of products is identified as suspected defective products. The wear critical point is judged by the vibration signal and the cutting force change trend, and the best time to replace the tool is determined; for the identified suspected defective products, the defect type is identified by analyzing the surface texture, and the material properties and design dimensional tolerance factors are comprehensively analyzed to determine whether the surface texture abnormality is caused by tool wear or the product itself defect; according to the judgment result, the product surface texture detection result is correlated with the tool wear warning information to construct a two-way feedback of product quality prediction and tool life optimization, and by continuously collecting production site data, the correspondence between tool wear degree and surface texture characteristics is dynamically updated and optimized.

2. The industrial vision production line product quality traceability method according to claim 1, wherein: The method uses an imaging device to perform microscopic imaging on the product surface, obtains texture image data of the machined surface under different cutting parameters, performs image preprocessing to improve image contrast and clarity, and eliminates interference factors such as material defects, clamping instability, and product defects themselves, including: using an imaging device to perform microscopic imaging on the product surface to obtain texture image data of the product surface; preprocessing the texture image data to adjust the pixel grayscale distribution of the texture image data; if there is noise in the texture image data, removing the noise; highlighting the surface texture features of the texture image data according to the Canny edge detection algorithm to obtain an edge-enhanced texture image; performing morphological operations on the edge-enhanced texture image, including dilation and erosion, to identify and segment the defective area in the edge-enhanced texture image; extracting features of the defective area, including area, perimeter, roundness, and grayscale statistics; inputting the features of the defective area into a support vector machine classifier to determine whether the defective area is caused by material defects or clamping instability; and screening out texture images that are greatly affected by interference factors based on the judgment results of the support vector machine classifier.

3. The industrial vision production line product quality traceability method according to claim 1, wherein: The method extracts key characteristic parameters that can reflect the degree of tool wear from the preprocessed texture image data, including texture orientation angle, texture period and surface roughness parameters, and constructs a multi-dimensional feature vector of the microtexture of the product surface, including: performing a two-dimensional fast Fourier transform on the preprocessed texture image data to obtain a two-dimensional power spectrum; extracting a main frequency component from the two-dimensional power spectrum, calculating the main direction of the power spectrum according to the main frequency component, and determining the texture orientation angle; obtaining the texture period by calculating the main frequency of the power spectrum; and using a gray level co-occurrence matrix method to calculate the statistical characteristics of the texture image, wherein a distance parameter and four directions are set, and the four directions are calculated. The contrast, entropy, energy and correlation of the texture orientation are calculated, and the average value is taken as the statistical feature parameter of surface roughness; the texture image is decomposed into three layers to extract the horizontal, vertical and diagonal detail coefficients at different scales; the energy of the detail coefficient at each scale is calculated according to the horizontal, vertical and diagonal detail coefficients to obtain the wavelet energy feature; the texture orientation angle, the texture period, the surface roughness statistical feature parameter and the wavelet energy feature are combined to construct a feature vector; each feature in the feature vector is normalized to obtain the normalized feature; the normalized features are spliced ​​to form a multi-dimensional feature vector of the product surface microtexture.

4. The industrial vision production line product quality traceability method according to claim 1, wherein: According to the historical usage data of the tool, including the number of processed products, the cumulative cutting time and the hardness of the processed material, the texture features of the tool processing surface at different wear stages are collected and analyzed, and a corresponding relationship between the degree of tool wear and the surface texture features is established, including: according to the historical usage data of the tool, the number of processed products, the cumulative cutting time and the hardness of the processed material are obtained, and a quantitative index of the degree of tool wear is obtained by weighted summation; microscopic imaging of the surface of the product processed by the tool is obtained to obtain a surface texture image; texture orientation angle, texture period and surface roughness characteristic parameters are extracted from the surface texture image, and the trend of the characteristic parameters changing with the degree of wear is recorded; a mapping relationship between the tool wear index and the surface texture characteristic parameters is established, and a radial basis function is selected as the kernel function for model training; for the surface of the newly processed product, the texture characteristic parameters are extracted, substituted into the model, the wear index of the current tool is calculated, and the degree of tool wear is judged according to a preset wear degree grading standard.

5. The industrial vision production line product quality traceability method according to claim 1, wherein: The method comprises: obtaining a product surface microtexture image, extracting texture orientation angle, texture period, and surface roughness characteristic parameters from the microtexture image, and normalizing the characteristic parameters to obtain standardized characteristic parameters; constructing a multidimensional feature vector based on the standardized characteristic parameters in a predetermined order; calculating the Euclidean distance between the multidimensional feature vector and all records in a pre-established database of tool wear degree and surface texture feature correspondence using a K-nearest neighbor algorithm, and selecting K records with the smallest distance as the most similar samples; weighting the selected K most similar samples according to their similarity to the multidimensional feature vector, wherein the similarity weight is calculated using a Gaussian kernel function; calculating the weighted average of the tool wear degrees corresponding to the K samples using a weighted average method to obtain the current tool wear degree estimate; and comparing the tool wear degree estimate with a preset wear warning threshold. If the tool wear degree estimate exceeds the warning threshold, it is determined that the current tool has reached a warning state.

6. The industrial vision production line product quality traceability method according to claim 1, wherein: If the current tool wear exceeds a preset threshold, a tool replacement warning is triggered, and the current batch of products is marked as suspected defective products. The wear critical point is judged by the vibration signal and the cutting force change trend, and the best time to replace the tool is determined, including: obtaining real-time monitoring of tool wear data, and comparing it with a preset wear threshold. The wear threshold is determined by analyzing the statistical distribution of historical processing data; if the tool wear exceeds the wear threshold, a tool replacement warning signal is triggered; based on the tool replacement warning signal, a tool replacement request is generated; based on the tool replacement request, the inventory is automatically checked, and if the inventory is sufficient, then The tool preparation process is started; vibration signals and cutting force data are collected and pre-processed in real time; a tool status score is calculated based on the degree of tool wear, the vibration signal and the cutting force data; when the tool status score is lower than a preset threshold, the best time to replace the tool is determined, and the tool replacement process is triggered; the process also includes: using vibration sensors and cutting force sensors to monitor the machining process in real time, extracting the characteristic frequency and amplitude of vibration and cutting force, establishing a corresponding relationship between the vibration amplitude, spindle current and cutting force and the degree of tool wear; when the vibration amplitude suddenly increases and the cutting force continues to rise and exceeds the preset threshold, it is determined that the tool has reached the critical point of wear.

7. The industrial vision production line product quality traceability method according to claim 6, wherein: The method adopts vibration sensors and cutting force sensors to monitor the machining process in real time, extracts the characteristic frequencies and amplitudes of vibration and cutting force, establishes the corresponding relationship between vibration amplitude, spindle current and cutting force and the degree of tool wear, and determines that the tool has reached the critical point of wear when the vibration amplitude suddenly increases and the cutting force continues to rise and exceeds the preset threshold value, including: removing noise from the vibration signal and cutting force signal collected in real time during the machining process to obtain the denoised vibration signal and cutting force signal; extracting the characteristic frequencies and amplitudes of vibration and cutting force through fast Fourier transform based on the denoised vibration signal and cutting force signal; using Hall current sensor to collect the spindle current value, and obtaining stable current data through low-pass filtering; establishing the corresponding relationship between the degree of tool wear and vibration amplitude, spindle current and cutting force The method comprises the following steps: a first step is to determine the tool wear critical point, a second step is to determine the tool wear critical point, and a third step is to determine the tool wear critical point; a first step is to determine the tool wear critical point, and a second ...

8. The industrial vision production line product quality traceability method according to claim 1, wherein: The method for identifying the type of defect of the identified suspected defective product by analyzing the surface texture, comprehensively analyzing the material properties and design dimensional tolerance factors, and judging whether the surface texture abnormality is caused by tool wear or the product itself defect, includes: scanning the surface of the identified suspected defective product, extracting texture features using a grayscale co-occurrence matrix, the texture features including contrast, entropy, energy and correlation, and extracting local texture information through a local binary pattern; constructing a surface texture feature vector based on the texture features, classifying the surface texture feature vector, and obtaining a preliminary defect type identification result and its probability distribution; obtaining the material property data of the product, and obtaining the dimensional tolerance requirements in the design drawing; using the material property data to extract the texture features; and obtaining the surface texture feature vector based on the texture features. According to the material, tolerance and process feature vector, the dimensional tolerance requirements and the current processing parameters are established; if the material, tolerance and process feature vector is obtained, a comprehensive analysis is performed in combination with the defect type identification result and its probability distribution; through the comprehensive analysis, it is judged whether the surface texture abnormality is caused by tool wear or a defect of the product itself, and the confidence level of the judgment result is output; it also includes: combining the product CAD model, extracting key dimensions and tolerance requirements, and establishing a correlation model between material properties, dimensional tolerances and surface texture characteristics. If the surface texture abnormality of the product is consistent with the material properties and tolerance requirements, it is determined to be a defect of the product itself; if the degree of abnormality exceeds the allowable range of the material and tolerance, and is consistent with the tool wear characteristics, it is determined to be caused by tool wear.

9. The industrial vision production line product quality traceability method according to claim 8, wherein: The method combines the product CAD model to extract key dimensions and tolerance requirements, and establishes a correlation model between material properties, dimensional tolerances, and surface texture features. If the product surface texture anomaly is consistent with the material properties and tolerance requirements, it is determined to be a product defect. If the degree of abnormality exceeds the allowable range of materials and tolerances and is consistent with tool wear characteristics, it is determined to be caused by tool wear, including: extracting key dimensions and tolerance requirements from the product CAD model to obtain the product's geometric features and their dimensional values ​​and tolerance ranges; using the support vector regression method to establish a correlation model between material properties, dimensional tolerances, and surface texture characteristics; based on the correlation model, inputting standardized material properties and tolerance data to obtain the expected surface texture parameter range; Compare the surface texture anomaly with the prediction results of the correlation model, and set the judgment threshold to a multiple of the standard deviation of the prediction range; if the degree of anomaly is within the judgment threshold, it is determined to be a defect in the product itself; if it exceeds the judgment threshold, it is matched with the tool wear feature library; use cosine similarity to calculate the similarity between the anomaly and the features in the library to determine whether the anomaly is caused by tool wear.

10. The industrial vision production line product quality traceability method according to claim 1, wherein: According to the judgment result, the product surface texture detection result and the tool wear warning information are correlated and analyzed to construct a two-way feedback of product quality prediction and tool life optimization. By continuously collecting production site data, the correspondence between the tool wear degree and the surface texture characteristics is dynamically updated and optimized, including: obtaining the product surface texture detection result and the tool wear warning information; measuring the tool wear degree to obtain tool wear data; establishing a product quality prediction model based on the tool wear data and processing parameters, and the product quality prediction model selects radial basis function as the kernel function; comparing the output result of the product quality prediction model with the preset quality standard to determine whether the product is qualified; if the product is unqualified, optimizing the tool replacement timing based on the genetic algorithm, and the genetic algorithm sets the chromosome encoding to represent different tool replacement strategies; according to the optimization result of the tool replacement strategy, using the online learning algorithm to realize incremental learning, and the online learning algorithm continuously collects production site data to update the parameters of the product quality prediction model.

Images