A visual inspection method for stamping of automotive parts

Abstract

This invention discloses a visual inspection method for stamping automotive parts, belonging to the field of visual inspection technology. The method extracts visual features related to the stamping parts and equipment operating status, fits these features with historical equipment operating data, establishes a multivariate time-series coupled model, quantifies and determines the equipment deterioration level, and sets graded early warning thresholds. By comparing real-time features with thresholds, an equipment risk judgment function is constructed to achieve proactive early warning of equipment deterioration and dynamically adjust the inspection frequency. After equipment maintenance, the recovery rate, effect index, and coefficient of variation are calculated by comparing features before and after maintenance to quantitatively verify the maintenance effect. This invention also uses a unique workpiece identifier to achieve full-process data association mapping, realizing linked output of quality inspection and equipment status, enabling early prediction of potential equipment deterioration, improving production continuity and intelligence, and adapting to the continuous and high-precision production needs of automotive parts stamping.

Description

Technical Field

[0001] This invention relates to the field of visual inspection technology, specifically a visual inspection method for stamping automotive parts. Background Technology

[0002] In the field of automotive parts stamping, computer vision inspection technology has been widely used in quality control processes such as defect identification and dimensional inspection of stamped parts. By analyzing the appearance and dimensions of stamped parts through image processing algorithms, problems such as scratches, deformation, and dimensional deviations can be effectively identified, improving product inspection efficiency and accuracy. However, existing visual inspection methods only focus on passive inspection of product quality itself, only checking equipment problems after a non-conforming stamped part is found. This forms a passive response logic of "equipment failure - product defect - inspection discovery," without establishing a correlation analysis between visual inspection data and the operating status of the stamping equipment.

[0003] In existing technologies, the operating status of equipment is mostly monitored by independent equipment monitoring systems, which only record basic data such as operating efficiency, number of failures, and maintenance time. They cannot predict potential deterioration problems of equipment through visual inspection features of stamped parts, which can easily lead to hidden equipment failures not being detected in time, resulting in batches of defective products, production line shutdowns for rectification, and a significant drop in production efficiency.

[0004] At the same time, after the equipment maintenance is completed, there is a lack of quantitative verification methods based on visual inspection data. The maintenance effect is judged only by the no-load test run of the equipment. It is impossible to verify the restoration of the equipment's processing accuracy by combining the actual stamping parts' inspection indicators. This can easily lead to secondary failures caused by substandard maintenance, further increasing production losses and making it difficult to adapt to the continuous and high-precision production requirements of automotive parts stamping. Summary of the Invention

[0005] The purpose of this invention is to provide a visual inspection method for stamping automotive parts to solve the problems mentioned in the background art.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: A visual inspection method for stamping automotive parts includes the following steps: S1. Visual image acquisition and processing of stamped parts after stamping of automotive parts, continuously acquiring visual inspection data of stamped parts, and extracting feature detection indicators related to the operating status of stamping equipment from the visual inspection data. The feature detection indicators include the dimensional deviation fluctuation of stamped parts, the regularity of surface indentation, and the uniformity of edge burrs. S2. Retrieve historical operating data of the stamping equipment, which includes equipment operating efficiency, equipment failure rate, and equipment maintenance records. Perform trend fitting analysis between the extracted feature detection indicators and the historical operating data of the equipment to obtain the equipment condition deterioration judgment threshold corresponding to the feature detection indicators. S3. Compare the real-time collected feature detection indicators with the judgment threshold. If the feature detection indicators exceed the judgment threshold, it is determined that there is a potential fault risk in the stamping equipment, triggering equipment maintenance prompts, and adjusting the collection and analysis frequency of the corresponding feature detection indicators in subsequent visual inspection of stamped parts. S4. After the stamping equipment has completed the maintenance work, continue to collect the corresponding feature detection indicators of subsequent stamped parts, and quantitatively verify the maintenance effect of the stamping equipment based on the changes in the feature detection indicators; S5. Visual inspection data, feature detection indicators, equipment potential fault risk information, equipment maintenance effect verification data and stamping equipment operation status information are associated and mapped and stored in a unified manner. At the same time, visual inspection data is output as the stamping part quality inspection result.

[0007] Furthermore, the specific process of extracting feature detection indicators in S1 includes: Multi-view visual image acquisition is performed on the stamped parts. After obtaining the original grayscale image, filtering and noise reduction processing is performed to eliminate random noise during the image acquisition process and retain the true morphological information of the stamped parts' contours, indentations, and burrs. Edge detection is performed on the filtered image to extract the contour edge point set of the stamping part. Registration and fitting are performed based on the standard CAD contour model of the stamping part to obtain the actual size sequence and theoretical standard size. Statistical analysis is performed on the size deviation of multiple consecutive stamping parts, and the standard deviation of the size deviation sequence is used as the size deviation fluctuation index. The surface area of ​​the stamped part is segmented by grayscale threshold, the indentation area is extracted, and the area, position coordinates and depth grayscale value of each indentation are obtained. Based on the occurrence frequency, area fluctuation degree and position offset of indentations on multiple consecutive stamped parts, the surface indentation regularity index used to characterize the regularity of indentation recurrence is calculated. The burr height sequence is extracted along the edge of the stamped part contour. The uniformity is characterized by the coefficient of variation of the burr height, and the edge burr uniformity index is calculated. The dimensional deviation fluctuation index, surface indentation regularity index, and edge burr uniformity index are output as feature detection indicators related to the operating status of stamping equipment, forming a visual inspection dataset.

[0008] Furthermore, the specific process for obtaining the equipment condition degradation judgment threshold in S2 includes: A multi-dimensional historical dataset is constructed, which integrates the feature detection indicators with the historical operating data of the stamping equipment. The historical operating data is divided into equipment health status stage, equipment deterioration status stage, and equipment failure status stage according to the time series. At the same time, the equipment operating efficiency, equipment failure rate and equipment maintenance event identifiers corresponding to each timestamp are recorded to form a complete historical associated dataset. The historical correlation dataset is cleaned and normalized to eliminate the interference of outliers and differences in units on the fitting results, resulting in a standardized historical dataset. Based on the standardized historical dataset, the equipment condition deterioration level is defined and divided into four levels according to the failure risk from low to high. A multivariate time series regression fitting method is used to establish a coupled fitting relationship between the feature detection index and the equipment condition deterioration level. The fitting model includes the influence weights of each feature index and its rate of change on the equipment deterioration state. Based on the optimal multi-index coupled time series fitting model, the mean and standard deviation of each feature detection index under different equipment state deterioration levels are statistically analyzed. Based on the mean and standard deviation of each level, the corresponding level 2 warning threshold, level 3 warning threshold, and level 4 warning threshold are calculated for each feature index. A multi-indicator joint judgment logic is constructed. When any feature detection indicator reaches the corresponding level 2 warning threshold, the equipment is judged to have potential failure risk. When any feature detection indicator reaches the level 4 warning threshold, the equipment is judged to have entered a severely degraded state and needs to be shut down for maintenance immediately. The judgment thresholds of each feature detection indicator are associated with the corresponding equipment status level and historical operating data to form a complete threshold mapping table. The effectiveness of the fitted model and the judgment thresholds is verified. The cross-validation method is used to calculate the prediction accuracy of the equipment status level of the model. After meeting the preset accuracy requirements, the threshold mapping table is output.

[0009] Furthermore, when establishing the coupled fitting relationship between feature detection indicators and equipment condition deterioration level, the fitting model includes each feature indicator and its rate of change over time. The regression coefficients are solved by the least squares method, and the optimization is carried out with the actual deterioration level calibrated by historical maintenance records and fault records as the target, so as to obtain the optimal multi-indicator coupled time series fitting model.

[0010] Furthermore, when validating the fitted model and the decision threshold, leave-one-out cross-validation is used to calculate the prediction accuracy. If the accuracy is lower than the preset threshold, the regression coefficients are re-optimized until the accuracy requirement is met.

[0011] Furthermore, the specific process of comparing real-time feature detection metrics with the decision threshold in S3 includes: Real-time acquisition of the feature detection index corresponding to the stamped part at the current moment, and retrieval of the secondary warning threshold, tertiary warning threshold and quaternary warning threshold corresponding to each feature detection index in the threshold mapping table; The real-time feature detection indicators are compared with the corresponding early warning thresholds one by one to construct a function for judging equipment deterioration risk. This function takes the maximum value of the risk level of each feature indicator as the overall risk level. The risk level of each feature indicator is determined based on the comparison results between its real-time value and the early warning thresholds at each level, and is divided into four levels: normal indicator, level 2 early warning, level 3 early warning, and level 4 early warning, which correspond to the equipment health status, potential failure risk, moderate deterioration status, and severe deterioration status, respectively.

[0012] Furthermore, the specific process for triggering device maintenance prompts in S3 includes: When the overall risk level reaches Level 2 warning, the equipment is determined to have potential failure risk, and a warning-level maintenance prompt is generated. The prompt includes the type of abnormal characteristic indicator and the corresponding warning level. When the overall risk level reaches Level 3 warning, the equipment is determined to be in a moderately deteriorated state, and a reminder-level maintenance prompt is generated. The prompt includes the degree of abnormality. When the overall risk level reaches Level 4 warning, the equipment is determined to be in a severely deteriorated state, and an emergency shutdown maintenance prompt is generated, forcibly triggering a shutdown signal and outputting alarm information.

[0013] Furthermore, the specific process for adjusting the acquisition and analysis frequency in S3 includes: Define the basic acquisition frequency and basic analysis frequency of each characteristic index under normal working conditions, and dynamically adjust the acquisition frequency and analysis frequency of subsequent visual inspection of stamped parts according to the current overall risk level; The adjusted collection frequency is the base collection frequency multiplied by an adjustment factor that is positively correlated with the risk level, and is subject to a preset maximum collection frequency limit. The adjustment factor increases with the risk level, so that for each increase in risk level, the collection frequency increases by a preset ratio until the limit is reached. The adjusted analysis frequency is the larger value between the base analysis frequency and the adjusted collection frequency.

[0014] Furthermore, the specific process for quantifying and verifying the maintenance effect in S4 includes: After the stamping equipment completes maintenance, a preset number of subsequent stamping parts feature detection indicators are continuously collected to form a post-maintenance feature sequence. At the same time, the same number of stamping parts feature detection indicators are retrieved before the maintenance prompt is triggered to form a pre-maintenance baseline feature sequence. For each feature detection index, the maintenance recovery rate is calculated based on the mean value after maintenance, the mean value before maintenance, and the reference mean value under the health status of the equipment. The maintenance recovery rate is used to characterize the degree to which the index value after maintenance regresses to the reference value under the health status. The maintenance recovery rates of each feature are weighted and fused to construct a maintenance effectiveness index, where the weight of each feature is preset according to its importance in representing equipment deterioration; the ratio of the standard deviation to the mean of the feature sequence after maintenance is calculated as the coefficient of variation of the feature index; the maintenance effectiveness index and the coefficient of variation of each feature are used as the quantitative evaluation results of maintenance effectiveness and are stored in association with equipment maintenance event information.

[0015] Furthermore, the specific process of association mapping and quality output in S5 includes: Using the unique identifier of the workpiece as the core association key, visual inspection data, various feature inspection indicators, equipment risk information, maintenance effect verification data and equipment operation status information are linked and integrated to establish the correspondence between various types of data and workpiece identifiers and timestamps; The visual inspection data is associated with the stamping part quality judgment standard, and the quality inspection results are output. The judgment logic is based on the comparison results of feature detection indicators and warning thresholds, and is divided into three levels: qualified, slightly abnormal, and seriously unqualified. The full amount of data after association and mapping is stored uniformly according to the preset structure. At the same time, the quality inspection results are bound with the workpiece identification and output to the production monitoring terminal.

[0016] Compared with existing technologies, the beneficial effects of this invention are as follows: By extracting visual feature indicators of stamped parts related to the operating status of stamping equipment, this invention breaks through the limitation of traditional visual inspection focusing only on product quality, establishes a deep correlation between visual inspection data and equipment operating status, and transforms the traditional passive response mode into an active prevention and control mode. This allows for accurate prediction of potential equipment deterioration risks, effectively avoids batch defects caused by hidden faults, significantly reduces the probability of production line downtime for rectification, and improves production continuity. This invention constructs a multivariate time-series coupled fitting model, integrating static feature values ​​and dynamic change rates to achieve quantitative judgment of equipment deterioration levels. Combined with a graded early warning threshold and a risk-driven adaptive adjustment mechanism for detection frequency, it achieves refined and intelligent monitoring of equipment status, improving the accuracy and timeliness of equipment fault early warning. Simultaneously, this invention designs a maintenance effect quantitative verification system based on visual inspection data. By comparing features before and after maintenance, it calculates the recovery rate, effect index, and coefficient of variation, verifying the maintenance effect from both the degree of recovery and stability dimensions. This replaces the traditional no-load trial operation method, accurately judging the recovery of equipment processing accuracy, avoiding secondary faults caused by substandard maintenance, and reducing production losses. Furthermore, this invention uses a unique workpiece identifier to achieve full-process data association mapping, enabling linked output of quality inspection and equipment status. The data is traceable and iterative, providing data support for optimizing equipment maintenance strategies and comprehensively improving the intelligence level and production efficiency of automotive parts stamping processing, perfectly adapting to the needs of continuous and high-precision production. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a visual inspection method for stamping automotive parts according to the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Please see Figure 1 The present invention provides the following technical solution: A visual inspection method for stamping automotive parts includes the following steps: S1. Visual image acquisition and processing of stamped parts after stamping of automotive parts, continuously acquiring visual inspection data of stamped parts, and extracting feature detection indicators related to the operating status of stamping equipment from the visual inspection data. The feature detection indicators include the dimensional deviation fluctuation of stamped parts, the regularity of surface indentation, and the uniformity of edge burrs. S2. Retrieve historical operating data of the stamping equipment, which includes equipment operating efficiency, equipment failure rate, and equipment maintenance records. Perform trend fitting analysis between the extracted feature detection indicators and the historical operating data of the equipment to obtain the equipment condition deterioration judgment threshold corresponding to the feature detection indicators. S3. Compare the real-time collected feature detection indicators with the judgment threshold. If the feature detection indicators exceed the judgment threshold, it is determined that there is a potential fault risk in the stamping equipment, triggering equipment maintenance prompts, and adjusting the collection and analysis frequency of the corresponding feature detection indicators in subsequent visual inspection of stamped parts. S4. After the stamping equipment has completed the maintenance work, continue to collect the corresponding feature detection indicators of subsequent stamped parts, and quantitatively verify the maintenance effect of the stamping equipment based on the changes in the feature detection indicators; S5. Visual inspection data, feature detection indicators, equipment potential fault risk information, equipment maintenance effect verification data and stamping equipment operation status information are associated and mapped and stored in a unified manner. At the same time, visual inspection data is output as the stamping part quality inspection result.

[0020] The specific process of extracting feature detection indicators from S1 includes: Multi-view visual image acquisition is performed on the stamped parts. After obtaining the original grayscale image, filtering and noise reduction processing is performed to eliminate random noise during the image acquisition process and retain the true morphological information of the stamped parts' contours, indentations, and burrs. Edge detection is performed on the filtered image to extract the contour edge point set of the stamping part. Registration and fitting are performed based on the standard CAD contour model of the stamping part to obtain the actual size sequence and theoretical standard size. Statistical analysis is performed on the size deviation of multiple consecutive stamping parts, and the standard deviation of the size deviation sequence is used as the size deviation fluctuation index. The surface area of ​​the stamped part is segmented by grayscale threshold, the indentation area is extracted, and the area, position coordinates and depth grayscale value of each indentation are obtained. Based on the occurrence frequency, area fluctuation degree and position offset of indentations on multiple consecutive stamped parts, the surface indentation regularity index used to characterize the regularity of indentation recurrence is calculated. The burr height sequence is extracted along the edge of the stamped part contour. The uniformity is characterized by the coefficient of variation of the burr height, and the edge burr uniformity index is calculated. The dimensional deviation fluctuation index, surface indentation regularity index, and edge burr uniformity index are output as feature detection indicators related to the operating status of stamping equipment, forming a visual inspection dataset.

[0021] In this embodiment, S1 includes the following: Multi-view visual image acquisition is performed on the stamped part to obtain the original grayscale image I(x,y) of the surface and contour of the stamped part; Gaussian filtering is applied to the original grayscale image for noise reduction to obtain the filtered image I. g (x,y), and the filtering formula is: , Where σ is the standard deviation of the Gaussian kernel and n is the radius of the filtering window; the filtering eliminates random noise during the image acquisition process and preserves the true morphological information of the stamped parts' contours, indentations, and burrs. For the filtered image I g Edge detection is performed on (x,y) to extract the contour edge point set E of the stamping part, which is represented as: E={(x1,y1),(x2,y2),…,(xm,ym)}, where m represents the total number of contour edge point sets; single-pixel continuous edges are obtained by using the Canny edge detection operator; based on the standard CAD contour model of the stamping part, the extracted contour is registered and fitted with the standard contour to obtain the actual size sequence Di and the theoretical standard size D0; The dimensional deviations di=Di-D0 of N consecutive stamped parts are statistically analyzed. The dimensional deviation fluctuation is the standard deviation of the dimensional deviation sequence, denoted as the dimensional deviation fluctuation F1 of the stamped part. F1 is used to characterize the degree of discrete fluctuation of the stamped part size as the equipment operates. The larger the fluctuation, the worse the stability of the equipment operation. The surface area of ​​the stamped part is segmented by grayscale thresholding to extract the indentation region. The area Ak, position coordinates (xk, yk), and depth grayscale value Gk of each indentation are obtained. The indentation pattern is defined by the probability of recurrence and positional consistency of the indentation features on continuous workpieces. The surface indentation regularity F2 of the stamped part is calculated, and the calculation formula is as follows: , Wherein, ΔAk is the indentation area fluctuation value, μAk is the average indentation area; ΔPk is the indentation position offset, P0 is the maximum allowable position offset; the closer F2 is to 1, the more regular the indentation appears, and the higher the risk of equipment mold or feeding mechanism deterioration. The burr height sequence ha is extracted along the edge of the stamped part contour, where a = 1, 2, ..., M. The uniformity is characterized by the coefficient of variation of the burr height. The edge burr uniformity F3 of the stamped part is calculated using the following formula: , Where μh is the average height of the burrs; the smaller F3 is, the more uniform the burr distribution; the larger F3 is, the more abnormal the edge cutting condition is, and the equipment cutting edge or stamping gap has a deterioration trend. The dimensional deviation fluctuation F1, surface indentation regularity F2, and edge burr uniformity F3 are output as feature detection indicators related to the operating status of the stamping equipment, forming a visual inspection dataset.

[0022] The specific process for obtaining the equipment condition degradation judgment threshold in S2 includes: A multi-dimensional historical dataset is constructed, which integrates the feature detection indicators with the historical operating data of the stamping equipment. The historical operating data is divided into equipment health status stage, equipment deterioration status stage, and equipment failure status stage according to the time series. At the same time, the equipment operating efficiency, equipment failure rate and equipment maintenance event identifiers corresponding to each timestamp are recorded to form a complete historical associated dataset. The historical correlation dataset is cleaned and normalized to eliminate the interference of outliers and differences in units on the fitting results, resulting in a standardized historical dataset. Based on the standardized historical dataset, the equipment condition deterioration level is defined and divided into four levels according to the failure risk from low to high. A multivariate time series regression fitting method is used to establish a coupled fitting relationship between the feature detection index and the equipment condition deterioration level. The fitting model includes the influence weights of each feature index and its rate of change on the equipment deterioration state. Based on the optimal multi-index coupled time series fitting model, the mean and standard deviation of each feature detection index under different equipment state deterioration levels are statistically analyzed. Based on the mean and standard deviation of each level, the corresponding level 2 warning threshold, level 3 warning threshold, and level 4 warning threshold are calculated for each feature index. A multi-indicator joint judgment logic is constructed. When any feature detection indicator reaches the corresponding level 2 warning threshold, the equipment is judged to have potential failure risk. When any feature detection indicator reaches the level 4 warning threshold, the equipment is judged to have entered a severely degraded state and needs to be shut down for maintenance immediately. The judgment thresholds of each feature detection indicator are associated with the corresponding equipment status level and historical operating data to form a complete threshold mapping table. The effectiveness of the fitted model and the judgment thresholds is verified. The cross-validation method is used to calculate the prediction accuracy of the equipment status level of the model. After meeting the preset accuracy requirements, the threshold mapping table is output.

[0023] When establishing the coupled fitting relationship between feature detection indicators and equipment condition deterioration level, the fitting model includes each feature indicator and its rate of change over time. The regression coefficients are solved by the least squares method, and the optimization is carried out with the actual deterioration level calibrated by historical maintenance records and fault records as the target, so as to obtain the optimal multi-indicator coupled time series fitting model.

[0024] When validating the fitted model and the decision threshold, leave-one-out cross-validation is used to calculate the prediction accuracy. If the accuracy is lower than the preset threshold, the regression coefficients are re-optimized until the accuracy requirement is met.

[0025] In this embodiment, S2 includes the following: A multi-dimensional historical dataset is constructed, integrating the aforementioned feature detection indicators with the historical operating data of the stamping equipment. The historical operating data is divided into three time series: equipment health status stage Th, equipment deterioration status stage Td, and equipment failure status stage Tf. Simultaneously, the equipment operating efficiency Esys(t), equipment failure rate Fsys(t), and equipment maintenance event identifier M(t) corresponding to each timestamp are recorded. M(t) = 1 indicates that the timestamp corresponds to equipment maintenance work, and M(t) = 0 indicates no maintenance work. This forms a complete historical associated dataset Dh = {(F1(t), F2(t), F3(t), Esys(t), Fsys(t), M(t)) | t ∈ Th ∪ Td ∪ Tf}. The historical correlation dataset Dh was cleaned and normalized to eliminate the interference of outliers and differences in units on the fitting results. Among them, the 3σ criterion was used to remove outlier data points in the dataset that deviated from the mean by more than 3 times the standard deviation. The min-max normalization method was used to map each feature data to the [0,1] interval, so as to obtain the processed standardized historical dataset Ds. Based on the standardized historical dataset Ds, a correlation fitting model between feature detection indicators and equipment condition deterioration is constructed, specifically as follows: Define the equipment condition degradation level, and classify it into Level 1, Level 2, Level 3 and Level 4 according to the failure risk from low to high. The set of equipment condition labels corresponding to each level is L={L1,L2,L3,L4}, where L1 represents healthy equipment condition, L2 represents slightly degraded equipment condition, L3 represents moderately degraded equipment condition, and L4 represents severely degraded equipment condition. A multivariate time-series regression fitting method is used to establish a coupled fitting relationship between feature detection indicators and equipment condition deterioration levels. The fitting model expression is as follows: , Where L(t) is the predicted equipment condition deterioration level at time t, with a value range of [1,4], which is rounded to discrete levels L1-L4; w0 is the regression intercept term, and w1-w6 are the regression coefficients, reflecting the weight of each characteristic index and the rate of change of the index on the equipment deterioration state; the regression coefficients w0-w6 are solved using the least squares method, and the least squares objective function is: J(w) = ∑ t∈Ds [Ltr(t)-L(t)] 2 Where Ltr(t) is the actual degradation level of the equipment at time t, determined by historical maintenance records and fault records, and w=[w0,w1,w2,w3,w4,w5,w6] T The regression coefficient vector is obtained by taking the partial derivative of the objective function J(w) and setting the partial derivative to 0. ∗ And the optimal multi-index coupled time series fitting model L ∗ (t); Based on the optimal multi-index coupled time series fitting model L ∗ (t), combining the mapping relationship between equipment status level and feature detection index, calculate the equipment status deterioration judgment threshold corresponding to each feature detection index, specifically: For each characteristic detection index, its distribution characteristics under different equipment condition deterioration levels are calculated. Taking dimensional deviation fluctuation F1 as an example, the mean values ​​μF11, μF12, μF13, μF14 and standard deviations σF11, σF12, σF13, σF14 of F1 under levels L1-L4 are statistically analyzed. The judgment thresholds for dimensional deviation fluctuation F1 are: Level 2 warning threshold TF11 = μF11 + 1.5σF11, Level 3 warning threshold TF12 = μF12 + 1.2σF12, and Level 4 warning threshold TF13 = μF13 + 0.8σF13. Similarly, the judgment thresholds for surface indentation regularity F2 and edge burr uniformity F3 are also calculated in the same way as the judgment thresholds for dimensional deviation fluctuation F1. A multi-indicator joint judgment logic is constructed. When any feature detection indicator reaches the corresponding secondary warning threshold, the equipment is judged to have potential failure risk. When any feature detection indicator reaches the severe degradation threshold, the equipment is judged to have entered a severe degradation state and needs to be shut down for maintenance immediately. At the same time, the judgment thresholds of each feature detection indicator are associated with the corresponding equipment status level and historical operating data and stored to form a complete threshold mapping table Y={(Fc,TFc1,TFc2,TFc3,Ld)|c=1,2,3;d=1,2,3,4}. The effectiveness of the fitted model and the decision threshold was verified using leave-one-out cross-validation on the standardized historical dataset Ds. The accuracy of the model's equipment condition level prediction, Acc, was calculated, and Acc = (1 / S)∑ s∈[1,S] I(L ∗ (ts)=Ltr(ts)), where S is the total number of cross-validations, and I() is an indicator function that takes the value of 1 when the condition in parentheses is true, and takes the value of 0 otherwise; if the prediction accuracy Acc≥90%, the fitted model and the decision threshold are deemed to be valid, and the threshold mapping table Y is output; if Acc<90%, the correlation fitted model is returned to re-optimize the regression coefficients until the accuracy requirement is met.

[0026] The specific process of comparing real-time feature detection metrics with the decision threshold in S3 includes: Real-time acquisition of the feature detection index corresponding to the stamped part at the current moment, and retrieval of the secondary warning threshold, tertiary warning threshold and quaternary warning threshold corresponding to each feature detection index in the threshold mapping table; The real-time feature detection indicators are compared with the corresponding early warning thresholds one by one to construct a function for judging equipment deterioration risk. This function takes the maximum value of the risk level of each feature indicator as the overall risk level. The risk level of each feature indicator is determined based on the comparison results between its real-time value and the early warning thresholds at each level, and is divided into four levels: normal indicator, level 2 early warning, level 3 early warning, and level 4 early warning, which correspond to the equipment health status, potential failure risk, moderate deterioration status, and severe deterioration status, respectively.

[0027] The specific process for triggering device maintenance prompts in S3 includes: When the overall risk level reaches Level 2 warning, the equipment is determined to have potential failure risk, and a warning-level maintenance prompt is generated. The prompt includes the type of abnormal characteristic indicator and the corresponding warning level. When the overall risk level reaches Level 3 warning, the equipment is determined to be in a moderately deteriorated state, and a reminder-level maintenance prompt is generated. The prompt includes the degree of abnormality. When the overall risk level reaches Level 4 warning, the equipment is determined to be in a severely deteriorated state, and an emergency shutdown maintenance prompt is generated, forcibly triggering a shutdown signal and outputting alarm information.

[0028] The specific process for adjusting the acquisition and analysis frequency in S3 includes: Define the basic acquisition frequency and basic analysis frequency of each characteristic index under normal working conditions, and dynamically adjust the acquisition frequency and analysis frequency of subsequent visual inspection of stamped parts according to the current overall risk level; The adjusted collection frequency is the base collection frequency multiplied by an adjustment factor that is positively correlated with the risk level, and is subject to a preset maximum collection frequency limit. The adjustment factor increases with the risk level, so that for each increase in risk level, the collection frequency increases by a preset ratio until the limit is reached. The adjusted analysis frequency is the larger value between the base analysis frequency and the adjusted collection frequency.

[0029] In this embodiment, S3 includes the following: The feature detection indicators corresponding to the stamped part at the current moment are obtained in real time and recorded as size deviation fluctuation F'1, surface indentation regularity F'2, and edge burr uniformity F'3; the threshold mapping table Y obtained in S2 is retrieved, and the secondary warning threshold, tertiary warning threshold, and quaternary warning threshold corresponding to each feature detection indicator are read; By comparing real-time feature detection indicators with their corresponding early warning thresholds item by item, a device degradation risk judgment function R(t) is constructed, with the expression: R(t) = max{R1(t),R2(t),R3(t)}, where Rc(t) is the risk level judgment sub-function for the c-th feature detection indicator, c = 1, 2, 3; defined as follows: when F'c < TFc1, Rc(t) = 0, indicating that the feature indicator is normal and the device is in a healthy state; when TFc1 ≤ F'c < TFc2, Rc(t) = 1, indicating that the feature indicator reaches the second-level early warning and the device has a potential failure risk; when TFc2 ≤ F'c < TFc3, Rc(t) = 2, indicating that the feature indicator reaches the third-level early warning and the device is in a moderate degradation state; when F'c ≥ TFc3, Rc(t) = 3, indicating that the feature indicator reaches the fourth-level early warning and the device is in a severely degraded state. Based on the output of the equipment degradation risk assessment function R(t), perform graded fault assessment and trigger alerts: If R(t)=1, it is determined that there is a potential fault risk in the stamping equipment, and an early warning level maintenance prompt is generated. The prompt content includes the type of abnormal characteristic indicator, the corresponding early warning level, and the suggested inspection parts. If R(t)=2, the stamping equipment is determined to be in a moderately deteriorated state, and a reminder-level maintenance prompt is generated, which includes the degree of abnormality and the recommended maintenance time limit. If R(t)=3, the stamping equipment is determined to be in a severely deteriorated state, an emergency shutdown maintenance prompt is generated, a shutdown signal is forcibly triggered and an alarm message is output; Under normal operating conditions, the basic acquisition frequency of each characteristic indicator is defined as f0, and the basic analysis frequency is defined as g0. Based on the current risk level R(t), the acquisition frequency f(t) and analysis frequency g(t) of subsequent visual inspection of stamped parts are dynamically adjusted. The adjusted inspection frequency f(t) is: f(t) = min{fmax, f0 × (1 + e·R(t))}, where fmax is the preset maximum acquisition frequency limit, and e is the frequency adjustment coefficient, preset by the system or optimized based on historical data. When R(t) = 0, f(t) = f0, maintaining the basic frequency. For every increase of 1 in R(t), the acquisition frequency increases proportionally until the upper limit is reached. The adjusted analysis frequency g(t) is the larger of the basic analysis frequency g0 and the adjusted acquisition frequency f(t), i.e., g(t) = max{g0, f(t)}, ensuring that the analysis frequency is not lower than the acquisition frequency and avoiding data backlog.

[0030] The specific process of quantitatively verifying the maintenance effect in S4 includes: After the stamping equipment completes maintenance, a preset number of subsequent stamping parts feature detection indicators are continuously collected to form a post-maintenance feature sequence. At the same time, the same number of stamping parts feature detection indicators are retrieved before the maintenance prompt is triggered to form a pre-maintenance baseline feature sequence. For each feature detection index, the maintenance recovery rate is calculated based on the mean value after maintenance, the mean value before maintenance, and the reference mean value under the health status of the equipment. The maintenance recovery rate is used to characterize the degree to which the index value after maintenance regresses to the reference value under the health status. The maintenance recovery rates of each feature are weighted and fused to construct a maintenance effectiveness index, where the weight of each feature is preset according to its importance in representing equipment deterioration; the ratio of the standard deviation to the mean of the feature sequence after maintenance is calculated as the coefficient of variation of the feature index; the maintenance effectiveness index and the coefficient of variation of each feature are used as the quantitative evaluation results of maintenance effectiveness and are stored in association with equipment maintenance event information.

[0031] In this embodiment, S4 includes the following: After the stamping equipment completes maintenance, N characteristic detection indicators of subsequent stamping parts are continuously collected to form a post-maintenance characteristic sequence Fc(m), where N is the preset number of verification samples and c=1,2,3; at the same time, the characteristic detection indicators of N stamping parts before the maintenance prompt are retrieved to form a pre-maintenance baseline characteristic sequence Fc(b). For each feature detection index, the maintenance recovery rate Qc is calculated to characterize the degree to which the index recovers from a deteriorated state to a healthy state. The calculation formula is: Qc = 1 - (μFc(m) - μFca) / (μFc(b) - μFca), where μFc(m) is the mean of the c-th feature after maintenance, μFc(b) is the mean of the c-th feature before maintenance, and μFca is the reference mean of the c-th feature under the equipment health state, obtained from the feature distribution statistics corresponding to health level L1. A multi-feature fusion maintenance effect index E is constructed, and E = ∑ c∈[1,3] λc·Qc, where λc is the weight coefficient of the c-th feature, and ∑ c∈[1,3] λc=1, which can be preset according to the importance of each feature in representing equipment degradation; The coefficient of variation (CVc) of the feature sequence after maintenance is calculated to characterize the fluctuation stability of the feature index, and CVc = σc(m) / μFc(m), where σc(m) is the standard deviation of the c-th feature after maintenance. The maintenance effect index E and the coefficient of variation (CVc) are used as the quantitative evaluation results of the maintenance effect, and are associated with and stored with the equipment maintenance event information for subsequent equipment maintenance strategy optimization and iterative update of historical operation data.

[0032] The specific process of correlation mapping and quality output in S5 includes: Using the unique identifier of the workpiece as the core association key, visual inspection data, various feature inspection indicators, equipment risk information, maintenance effect verification data and equipment operation status information are linked and integrated to establish the correspondence between various types of data and workpiece identifiers and timestamps; The visual inspection data is associated with the stamping part quality judgment standard, and the quality inspection results are output. The judgment logic is based on the comparison results of feature detection indicators and warning thresholds, and is divided into three levels: qualified, slightly abnormal, and seriously unqualified. The full amount of data after association and mapping is stored uniformly according to the preset structure. At the same time, the quality inspection results are bound with the workpiece identification and output to the production monitoring terminal.

[0033] In this embodiment, S5 includes the following: Using the unique identifier IDp of the workpiece as the core association key, the entire process data from S1 to S4 is integrated, including visual inspection data Ip, various feature inspection indicators, equipment risk information, maintenance effect verification data, and equipment operating status information; where p is the production serial number of the stamped part. Construct a data association mapping model to establish a bidirectional mapping relationship between various types of data and IDp and timestamp tp. The mapping expression is: M(IDp,tp)={Ip,F'cp,Fcp(b),Fcp(m),R(tp),Rcp,Ep,CVcp,f(tp),g(tp)}, where M() is the data association mapping function; The visual inspection data is associated with the stamping part quality judgment standard, and the quality inspection result Zp is output. The judgment logic is as follows: if all F'cp < TFc1, then Zp = 0; if there exists TFc1 ≤ F'cp < TFc3, then Zp = 1; if there exists F'cp ≥ TFc3, then Zp = 2. Among them, Zp = 0 is quality qualified, Zp = 1 is minor abnormality, and Zp = 2 is serious non-compliance. The full data after association and mapping is stored uniformly according to the preset structure, and the quality inspection result Zp is bound with IDp and output to the production monitoring terminal.

[0034] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0035] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A visual inspection method for stamping automotive parts, characterized in that: The method includes the following steps: S1. Visual image acquisition and processing of stamped parts after stamping of automotive parts, continuously acquiring visual inspection data of stamped parts, and extracting feature detection indicators related to the operating status of stamping equipment from the visual inspection data. The feature detection indicators include the dimensional deviation fluctuation of stamped parts, the regularity of surface indentation, and the uniformity of edge burrs. S2. Retrieve historical operating data of the stamping equipment, which includes equipment operating efficiency, equipment failure rate, and equipment maintenance records. Perform trend fitting analysis between the extracted feature detection indicators and the historical operating data of the equipment to obtain the equipment condition deterioration judgment threshold corresponding to the feature detection indicators. S3. Compare the real-time collected feature detection indicators with the judgment threshold. If the feature detection indicators exceed the judgment threshold, it is determined that there is a potential fault risk in the stamping equipment, triggering equipment maintenance prompts, and adjusting the collection and analysis frequency of the corresponding feature detection indicators in subsequent visual inspection of stamped parts. S4. After the stamping equipment has completed the maintenance work, continue to collect the corresponding feature detection indicators of subsequent stamped parts, and quantitatively verify the maintenance effect of the stamping equipment based on the changes in the feature detection indicators; S5. Visual inspection data, feature detection indicators, equipment potential fault risk information, equipment maintenance effect verification data and stamping equipment operation status information are associated and mapped and stored in a unified manner. At the same time, visual inspection data is output as the stamping part quality inspection result.

2. The visual inspection method for stamping of automotive parts according to claim 1, characterized in that: The specific process of extracting feature detection indicators in S1 includes: Multi-view visual image acquisition is performed on the stamped parts. After obtaining the original grayscale image, filtering and noise reduction processing is performed to eliminate random noise during the image acquisition process and retain the true morphological information of the stamped parts' contours, indentations, and burrs. Edge detection is performed on the filtered image to extract the contour edge point set of the stamping part. Registration and fitting are performed based on the standard CAD contour model of the stamping part to obtain the actual size sequence and theoretical standard size. Statistical analysis is performed on the size deviation of multiple consecutive stamping parts, and the standard deviation of the size deviation sequence is used as the size deviation fluctuation index. The surface area of ​​the stamped part is segmented by grayscale threshold, the indentation area is extracted, and the area, position coordinates and depth grayscale value of each indentation are obtained. Based on the occurrence frequency, area fluctuation degree and position offset of indentations on multiple consecutive stamped parts, the surface indentation regularity index used to characterize the regularity of indentation recurrence is calculated. The burr height sequence is extracted along the edge of the stamped part contour. The uniformity is characterized by the coefficient of variation of the burr height, and the edge burr uniformity index is calculated. The dimensional deviation fluctuation index, surface indentation regularity index, and edge burr uniformity index are output as feature detection indicators related to the operating status of stamping equipment, forming a visual inspection dataset.

3. The visual inspection method for stamping of automotive parts according to claim 1, characterized in that: The specific process for obtaining the equipment condition degradation judgment threshold in S2 includes: A multi-dimensional historical dataset is constructed, which integrates the feature detection indicators with the historical operating data of the stamping equipment. The historical operating data is divided into equipment health status stage, equipment deterioration status stage, and equipment failure status stage according to the time series. At the same time, the equipment operating efficiency, equipment failure rate and equipment maintenance event identifiers corresponding to each timestamp are recorded to form a complete historical associated dataset. The historical correlation dataset is cleaned and normalized to eliminate the interference of outliers and differences in units on the fitting results, resulting in a standardized historical dataset. Based on the standardized historical dataset, the equipment condition deterioration level is defined and divided into four levels according to the failure risk from low to high. A multivariate time series regression fitting method is used to establish a coupled fitting relationship between the feature detection index and the equipment condition deterioration level. The fitting model includes the influence weights of each feature index and its rate of change on the equipment deterioration state. Based on the optimal multi-index coupled time series fitting model, the mean and standard deviation of each feature detection index under different equipment state deterioration levels are statistically analyzed. Based on the mean and standard deviation of each level, the corresponding level 2 warning threshold, level 3 warning threshold, and level 4 warning threshold are calculated for each feature index. A multi-indicator joint judgment logic is constructed. When any feature detection indicator reaches the corresponding level 2 warning threshold, the equipment is judged to have potential failure risk. When any feature detection indicator reaches the level 4 warning threshold, the equipment is judged to have entered a severely degraded state and needs to be shut down for maintenance immediately. The judgment thresholds of each feature detection indicator are associated with the corresponding equipment status level and historical operating data to form a complete threshold mapping table. The effectiveness of the fitted model and the judgment thresholds is verified. The cross-validation method is used to calculate the prediction accuracy of the equipment status level of the model. After meeting the preset accuracy requirements, the threshold mapping table is output.

4. The visual inspection method for stamping of automotive parts according to claim 3, characterized in that: When establishing the coupled fitting relationship between feature detection indicators and equipment condition deterioration level, the fitting model includes each feature indicator and its rate of change over time. The regression coefficients are solved by the least squares method, and the optimization is carried out with the actual deterioration level calibrated by historical maintenance records and fault records as the target, so as to obtain the optimal multi-indicator coupled time series fitting model.

5. The visual inspection method for stamping of automotive parts according to claim 3, characterized in that: When validating the fitted model and the decision threshold, leave-one-out cross-validation is used to calculate the prediction accuracy. If the accuracy is lower than the preset threshold, the regression coefficients are re-optimized until the accuracy requirement is met.

6. The visual inspection method for stamping of automotive parts according to claim 1, characterized in that: The specific process of comparing the real-time feature detection index with the judgment threshold in S3 includes: Real-time acquisition of the feature detection index corresponding to the stamped part at the current moment, and retrieval of the secondary warning threshold, tertiary warning threshold and quaternary warning threshold corresponding to each feature detection index in the threshold mapping table; The real-time feature detection indicators are compared with the corresponding early warning thresholds one by one to construct a function for judging equipment deterioration risk. This function takes the maximum value of the risk level of each feature indicator as the overall risk level. The risk level of each feature indicator is determined based on the comparison results between its real-time value and the early warning thresholds at each level, and is divided into four levels: normal indicator, level 2 early warning, level 3 early warning, and level 4 early warning, which correspond to the equipment health status, potential failure risk, moderate deterioration status, and severe deterioration status, respectively.

7. The visual inspection method for stamping of automotive parts according to claim 6, characterized in that: The specific process for triggering the device maintenance prompt in S3 includes: When the overall risk level reaches Level 2 warning, the equipment is determined to have potential failure risk, and a warning-level maintenance prompt is generated. The prompt includes the type of abnormal characteristic indicator and the corresponding warning level. When the overall risk level reaches Level 3 warning, the equipment is determined to be in a moderately deteriorated state, and a reminder-level maintenance prompt is generated. The prompt includes the degree of abnormality. When the overall risk level reaches Level 4 warning, the equipment is determined to be in a severely deteriorated state, and an emergency shutdown maintenance prompt is generated, forcibly triggering a shutdown signal and outputting alarm information.

8. The visual inspection method for stamping of automotive parts according to claim 6, characterized in that: The specific process of adjusting the acquisition and analysis frequency in S3 includes: Define the basic acquisition frequency and basic analysis frequency of each characteristic index under normal working conditions, and dynamically adjust the acquisition frequency and analysis frequency of subsequent visual inspection of stamped parts according to the current overall risk level; The adjusted collection frequency is the base collection frequency multiplied by an adjustment factor that is positively correlated with the risk level, and is subject to a preset maximum collection frequency limit. The adjustment factor increases with the risk level, so that for each increase in risk level, the collection frequency increases by a preset ratio until the limit is reached. The adjusted analysis frequency is the larger value between the base analysis frequency and the adjusted collection frequency.

9. The visual inspection method for stamping of automotive parts according to claim 1, characterized in that: The specific process of quantifying and verifying the maintenance effect in S4 includes: After the stamping equipment completes maintenance, a preset number of subsequent stamping parts feature detection indicators are continuously collected to form a post-maintenance feature sequence. At the same time, the same number of stamping parts feature detection indicators are retrieved before the maintenance prompt is triggered to form a pre-maintenance baseline feature sequence. For each feature detection index, the maintenance recovery rate is calculated based on the mean value after maintenance, the mean value before maintenance, and the reference mean value under the health status of the equipment. The maintenance recovery rate is used to characterize the degree to which the index value after maintenance regresses to the reference value under the health status. The maintenance recovery rates of each feature are weighted and fused to construct a maintenance effectiveness index, where the weight of each feature is preset according to its importance in representing equipment deterioration; the ratio of the standard deviation to the mean of the feature sequence corresponding to each feature after maintenance is calculated as the coefficient of variation of the corresponding feature; the maintenance effectiveness index and the coefficient of variation of each feature are used as the quantitative evaluation results of maintenance effectiveness and are stored in association with equipment maintenance event information.

10. The visual inspection method for stamping of automotive parts according to claim 1, characterized in that: The specific process of correlation mapping and quality output in S5 includes: Using the unique identifier of the workpiece as the core association key, visual inspection data, various feature inspection indicators, equipment risk information, maintenance effect verification data and equipment operation status information are linked and integrated to establish the correspondence between various types of data and workpiece identifiers and timestamps; The visual inspection data is associated with the stamping part quality judgment standard, and the quality inspection results are output. The judgment logic is based on the comparison results of feature detection indicators and warning thresholds, and is divided into three levels: qualified, slightly abnormal, and seriously unqualified. The full amount of data after association and mapping is stored uniformly according to the preset structure. At the same time, the quality inspection results are bound with the workpiece identification and output to the production monitoring terminal.

Images