Automobile part stamping deformation assessment method based on visual attention mechanism

Abstract

The invention provides an automobile part stamping deformation assessment method based on a visual attention mechanism, and relates to the technical field of computer vision and industrial detection, and the method comprises the steps: collecting a stamping part deformation image, dividing a large-area slow deformation region and a small-area rapid deformation region according to a surface area and a curvature threshold value, and carrying out the differentiation feature extraction; a serious deformation area is identified by comparing curvature radiuses, and an initial attention weight is allocated based on a curvature deviation degree. And a visual attention mechanism is introduced, the attention priority of each area is dynamically adjusted in combination with the overall deformation characteristics and local curvature mutation, and a deformation attention sequence is generated. And finally, comprehensively evaluating the regional bending degree and the blank characteristics, determining the repairability grade, and outputting a repair sequence, a recommended process and an estimated duration according to the repairability grade. The method solves the problems that in the prior art, evaluation features are single, and multi-scale deformation features of large-area slow deformation and small-area rapid deformation are difficult to pay attention to at the same time, so that the evaluation result is inaccurate, and a repair decision is difficult to support.

Description

Technical Field

[0001] This invention relates to the fields of computer vision and industrial inspection technology, and in particular to a method for evaluating stamping deformation of automotive parts based on visual attention mechanisms. Background Technology

[0002] After stamping, accurate assessment of surface deformation areas in automotive parts is crucial for controlling production costs and reducing scrap rates. Current assessment methods mainly rely on manual visual inspection combined with simple measuring tools, or image detection systems with fixed thresholds. When faced with complex deformations, manual methods are inefficient and highly subjective; while automated methods based on fixed thresholds struggle to effectively handle coexisting deformation patterns with vastly different dimensions and risk characteristics on the same part.

[0003] On the same stamped part, there may be both large-area gentle deformation and small-area sharp bending. Large-area gentle deformation has a large area but a slow change in curvature, making it visually noticeable; small-area sharp bending has a small area but an extremely small radius of curvature, resulting in high stress concentration. Existing methods often assign evaluation weights based on only a single feature, such as area ratio or local height difference, leading to excessive attention to large-area gentle deformation while neglecting small-area sharp deformation that truly poses a cracking risk. This results in inaccurate evaluation results that are difficult to support repair decisions. Summary of the Invention

[0004] This invention provides a method for evaluating the stamping deformation of automotive parts based on a visual attention mechanism. It aims to solve the problem that existing technologies cannot simultaneously focus on the multi-scale deformation characteristics of large-area gradual deformation and small-area rapid deformation, resulting in inaccurate evaluation results and difficulty in supporting repair decisions.

[0005] To achieve the above objectives, this invention provides a method for evaluating stamping deformation of automotive parts based on a visual attention mechanism, comprising the following steps: S1. Collect image data of the deformation area of ​​the stamped part, and obtain the surface area and radius of curvature of each deformation area; based on the preset surface area threshold and radius of curvature threshold, divide the deformation area into a large-area slow deformation area and a small-area rapid deformation area. S2. Extract the overall curvature distribution characteristics of the large-area gently deforming region and the local curvature abrupt change locations of the small-area rapidly deforming region to obtain the overall deformation characteristics of each deformation region. S3. Compare the radius of curvature with a preset curvature threshold to determine the degree of bending in each deformation region, and identify severely deformed regions based on the degree of bending. S4. Compare the radius of curvature of the severely deformed region with the preset standard value of radius of curvature to determine the degree of curvature deviation, and assign initial attention weights to each severely deformed region according to the degree of curvature deviation. S5. Based on the initial attention weights, combined with the overall curvature distribution characteristics and local curvature abrupt change locations, the attention priority of each deformation region is obtained through a predetermined visual attention mechanism to generate the final deformation attention order. S6. Based on the final deformation concern order, and taking into account the bending degree of each deformation area and the surface area of ​​the stamped part, evaluate and identify the repairability level of the stamped part blank. S7. Based on the repairability level, determine the repair sequence and recommended repair process for each deformed area, and estimate the repair time required.

[0006] Further, in step S1, dividing the deformation region into a large-area gently deforming region and a small-area rapidly deforming region specifically includes: The image data of the deformation area of ​​the stamped part is processed to identify the initial deformation area; If the surface area of ​​the initial deformation region is greater than a preset surface area threshold and the radius of curvature is greater than a preset curvature threshold, it is determined to be a large-area gently deformable region. If the surface area of ​​the initial deformation region is less than a preset surface area threshold and the radius of curvature is less than a preset curvature threshold, it is determined to be a small-area rapid deformation region.

[0007] Furthermore, in step S2, the overall curvature distribution characteristics are represented by a curvature distribution histogram; the locations of local curvature abrupt changes are represented by the coordinates of the curvature abrupt change points.

[0008] Further, in step S3, the step of identifying severely deformed areas based on the degree of bending specifically includes: comparing the radius of curvature of the deformed area with a preset severity curvature threshold, and marking areas below the threshold as high bending areas; and combining the deformation strength indices of adjacent areas to perform boundary optimization and merging of high bending areas to form the final set of severely deformed areas.

[0009] Furthermore, in step S4, the allocation of initial attention weights to each severely deformed region based on the degree of curvature deviation specifically involves: the degree of curvature deviation being quantified by the percentage of relative deviation between the actual curvature radius and the standard value; and the initial attention weights being positively correlated with this percentage of relative deviation.

[0010] Further, in step S5, obtaining the attention priority of each deformation region through a predetermined visual attention mechanism and generating the final deformation attention order includes: integrating the initial attention weights, overall deformation features, and local curvature change positions through a predetermined weighting function, outputting the dynamic priority score of each deformation region, and sorting them to generate the final deformation attention order.

[0011] Further, in step S6, the assessment and identification of the repairability level of the stamping blank specifically involves: establishing an assessment model, taking the deformation concern sequence, bending degree, region type, and blank surface area as input features, and outputting discrete repairability levels through rule matching or model inference.

[0012] Furthermore, in step S7, determining the repair sequence for each deformed region, recommending corresponding repair processes, and estimating the repair time specifically includes: Based on the repairability level and the characteristics of the deformed area, the repair process knowledge base is queried to obtain a recommended repair process; Based on the complexity of the recommended repair process and the area, combined with historical operation data, the repair time for a single area is estimated. Based on the deformation concern order, all areas requiring repair are sorted to form a repair operation sequence, and the estimated total time is summarized.

[0013] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: By acquiring images of stamped parts undergoing deformation, the system divides the area into regions of large-area gradual deformation and small-area rapid deformation based on surface area and curvature thresholds, and extracts differentiated features. Secondly, it identifies severely deformed areas by comparing curvature radii and assigns initial attention weights based on the degree of curvature deviation. Then, a visual attention mechanism is introduced, combining overall deformation features and local curvature abrupt changes to dynamically adjust the attention priority of each region, generating a deformation attention order. Finally, the system comprehensively evaluates the degree of bending in each region and the characteristics of the blank to determine the repairability level, and outputs the repair sequence, recommended process, and estimated time accordingly. This invention solves the problem of existing technologies using only single evaluation features, making it difficult to simultaneously focus on multi-scale deformation features of large-area gradual deformation and small-area rapid deformation, leading to inaccurate evaluation results and difficulty in supporting repair decisions. Attached Figure Description

[0014] Figure 1 This is a flowchart of a method for evaluating stamping deformation of automotive parts based on a visual attention mechanism, according to the present invention.

[0015] Figure 2 This is a flowchart illustrating the determination of the repair sequence, recommended repair process, and estimated repair time for each deformed region in one embodiment of the present invention. Detailed Implementation

[0016] To further understand the content of this invention, a detailed description of the invention is provided in conjunction with the accompanying drawings and embodiments. The specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention. It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0017] like Figures 1-2 This embodiment of a method for evaluating stamping deformation of automotive parts based on visual attention mechanisms specifically includes: S1. Collect image data of the deformation area of ​​the stamped part, and obtain the surface area and radius of curvature of each deformation area; based on the preset surface area threshold and radius of curvature threshold, divide the deformation area into a large-area slow deformation area and a small-area rapid deformation area. In one implementation, image data of the deformed area of ​​the stamped part is acquired. For example, on an automotive stamping production line, the stamped metal sheet is placed on a fixed platform, and a camera is used to capture images from multiple angles, ensuring that the images cover the entire deformed area. This acquisition method can capture the geometric features of the deformed area, including surface contours and local bending conditions, thereby providing a reliable data basis for subsequent segmentation.

[0018] In one specific implementation, image data of the deformed area of ​​the stamped part is acquired, and the image data is initially divided using preset surface area thresholds and radius of curvature thresholds to obtain the initial deformation region boundary. For the initial deformation region boundary, the surface area and radius of curvature values ​​within the region are obtained. If the surface area exceeds the surface area threshold and the radius of curvature exceeds a preset curvature threshold, it is determined to be a large-area, gently deformed region. The remaining portion is extracted from the large-area, gently deformed region. If the surface area is lower than the surface area threshold and the radius of curvature is lower than a preset curvature threshold, it is determined to be a small-area, rapidly deformed region. Based on the large-area, gently deformed region and the small-area, rapidly deformed region, adjacent deformation features are obtained. These adjacent deformation features are then fused through boundary adjustment to obtain a multi-target region segmentation result.

[0019] In one specific implementation, the surface area threshold refers to a preset area value used to distinguish the size of the deformation region. It is usually set based on the overall size of the stamped part; for example, for a medium-sized stamped part, the threshold can be set to 50 square centimeters. The radius of curvature threshold is used to measure the degree of deformation; a larger radius of curvature indicates gradual deformation, and vice versa. Using image processing software, the surface area of ​​each deformation region is first calculated by integrating and summing the pixels within the closed contour of the image to obtain the area value; then, the radius of curvature is calculated by fitting an arc to the boundary curve of the region to determine the radius. This threshold application ensures the objectivity and repeatability of the segmentation. In one specific implementation, regions with a surface area exceeding the threshold and a large radius of curvature are identified as large-area, gradually deforming regions; regions with a surface area below the threshold and a small radius of curvature are identified as small-area, rapidly deforming regions.

[0020] S2. Extract the overall curvature distribution characteristics of the large-area gently deforming region and the local curvature abrupt change locations of the small-area rapidly deforming region to obtain the overall deformation characteristics of each deformation region. In one specific implementation, the surface area size and overall curvature distribution of large-area, gently deforming regions are obtained through multi-objective segmentation results, resulting in a large-area feature set. The remaining portion is extracted from the multi-objective segmentation results, and the radius of curvature and local curvature abrupt changes are obtained for small-area, rapidly deforming regions, resulting in a small-area feature set. Based on the large-area and small-area feature sets, adjacent deformation strength indices are fused to obtain differentiated extraction results. These differentiated extraction results are used to map the deformation range and determine the overall deformation characteristics of each region. It should be noted that the surface area size of the stamped part blank is calculated using an integral algorithm, i.e., the surface grids within the region are summed to obtain the total area value. This extraction helps in assessing the overall deformation scale. In another specific implementation, the overall curvature distribution characteristics first divide the region into sub-grids, calculate the local curvature for each grid, and then generate a distribution histogram. This feature extraction highlights the uniformity of the gently deforming region and, in practical applications, can provide a data foundation for subsequent correction. In one specific implementation, if a large area of ​​gradual deformation has been identified, the system automatically scans its boundary, calculates the surface area to quantify the degree of deformation, and further analyzes it in conjunction with the overall curvature distribution characteristics. Further, the extraction of overall curvature distribution characteristics involves statistically calculating the curvature values ​​at multiple points within the region. Curvature distribution characteristics refer to a statistical description of the surface curvature of the region, such as the mean curvature and variance, obtained by fitting a surface curve. For small areas of rapid deformation, the curvature radius and the location of local curvature abrupt changes are extracted to capture local defects. The curvature radius is a value obtained by fitting an arc to the boundary curve, representing the degree of abrupt deformation; a small radius corresponds to rapid deformation. In one possible implementation, a difference method is used to calculate the curvature values ​​at each point on the curve, and the point with the smallest radius is identified as a reference. This extraction method emphasizes the stress concentration points in the rapid deformation zone, which helps prevent crack propagation.

[0021] In one specific implementation, the identification of locations where local curvature abruptly changes involves detecting coordinate points where curvature values ​​change drastically. By scanning the region boundary and comparing the curvature difference between adjacent points, if the difference exceeds a preset threshold, it is marked as a location of abrupt change.

[0022] S3. Compare the radius of curvature with a preset curvature threshold to determine the degree of bending in each deformation region, and identify severely deformed regions based on the degree of bending. In one specific implementation, the curvature radius value of each region is obtained through multi-target differential feature extraction results. Each curvature radius value is compared with a preset curvature threshold. If the curvature radius value is lower than the preset curvature threshold, it is marked as high bending, thus obtaining a bending degree index. Using this bending degree index, a deformation intensity index of adjacent regions (obtained from the multi-target differential feature extraction results) is fused to determine the bending gradient distribution of each region, identifying preliminary severely deformed regions. A curvature distribution mapping is extracted from these preliminary severely deformed regions, and the curvature gradient distribution in the curvature distribution mapping is locally adjusted to obtain an adjusted set of severely deformed regions. Based on this adjusted set of severely deformed regions, a differential feature set (obtained from the multi-target differential feature extraction results) is fused to map the overall deformation features, determining the final severely deformed regions.

[0023] It should be noted that the curvature threshold is a value preset based on the mechanical properties of the stamping material and historical deformation data. The preset process includes collecting historical stamping data and performing statistical analysis to determine the threshold range.

[0024] In one specific implementation, the degree of bending can be divided into three levels: slight, moderate, and severe. The greater the deviation between the radius of curvature and the threshold, the higher the degree of bending.

[0025] In one specific implementation, when identifying severely deformed areas, the system comprehensively assesses the degree of bending and other extracted features, such as curvature distribution. When the degree of bending reaches a severe level, the area is automatically marked as a severely deformed area.

[0026] In one specific implementation, the degree of bending can be determined by a weighted calculation based on the area size. Specifically, the degree of bending in a large area is calculated by the ratio of the average radius of curvature to a threshold, while in a small area, the focus is on the local maximum deviation.

[0027] Preferably, the quantification of the degree of bending can also incorporate the material's elastic modulus as a correction factor.

[0028] Specifically, the adjusted bending value is obtained by multiplying the curvature radius deviation by the modulus coefficient. This correction is applicable to high-strength material stamping, ensuring accurate identification. Furthermore, during the identification process, if multiple regions have similar bending degrees, the system uses a clustering algorithm to group severely affected regions.

[0029] S4. Compare the radius of curvature of the severely deformed region with the preset standard value of radius of curvature to determine the degree of curvature deviation, and assign initial attention weights to each severely deformed region according to the degree of curvature deviation. In one specific implementation, curvature radius values ​​are obtained from severely deformed regions. These curvature radius values ​​are then compared one by one with preset standard curvature radius values ​​to obtain a curvature deviation index. Using this curvature deviation index, the regional deformation intensity distribution, obtained through a preset curvature distribution mapping extracted from the curvature radius values ​​of severely deformed regions, is fused to determine the initial allocation of attention weights. For this initial allocation, deformation gradient changes are extracted from a preset differential feature fusion based on the regional deformation intensity distribution, and the curvature deviation index is adjusted. Based on the adjusted curvature deviation index, the overall attention distribution is mapped to determine the final attention weights.

[0030] In one implementation, the radius of curvature values ​​are obtained from previously identified severely deformed areas. This process first involves scanning the three-dimensional model data of the stamped part to locate those areas that have been marked as severely deformed.

[0031] Specifically, the system extracts the local radius of curvature of each region through coordinate system mapping, and these values ​​are calculated based on surface geometry.

[0032] In one specific implementation, the preset standard value for the radius of curvature is a value predetermined based on the material type and stamping process parameters. The process of setting this standard value includes analyzing the curvature distribution of historical stamping samples and calculating the mean and standard deviation using statistical methods.

[0033] In one specific implementation, the degree of curvature deviation in severely deformed regions is determined by quantifying the deviation value. Specifically, the degree of deviation can be calculated using a relative deviation formula, which involves subtracting the standard value from the actual radius of curvature and then dividing by the standard value to obtain a percentage deviation. When the deviation exceeds a certain percentage, the system classifies it into different levels.

[0034] In one specific implementation, the area of ​​the region can be used as a correction factor to determine the degree of curvature deviation. Specifically, for large, severely affected areas, a weighted average is used for deviation calculation, while for small areas, the peak deviation is emphasized.

[0035] Preferably, assigning attention weights based on the degree of curvature deviation involves a mapping mechanism. Specifically, the weight values ​​are proportional to the degree of deviation and are calculated using a linear or exponential function.

[0036] For example, the weight is equal to the percentage of deviation multiplied by an adjustment coefficient. This allocation is suitable for neural network models, where regions with higher deviations receive higher weights to guide the model to focus on problem areas. In precision electronic casing stamping, the system assigns weights of 0.8 or higher to high-deviation regions, ensuring resources are allocated to these areas. This mechanism enhances the targeting of the analysis and enables dynamic adjustments on continuous production lines.

[0037] Understandably, the entire process, from obtaining the radius of curvature to assigning weights, forms a closed-loop evaluation chain. In one specific implementation, for large, severely affected areas, a weighted average is used to calculate the deviation, while for small areas, peak deviation is emphasized.

[0038] S5. Based on the initial attention weights, combined with the overall curvature distribution characteristics and local curvature abrupt change locations, the attention priority of each deformation region is obtained through a predetermined visual attention mechanism to generate the final deformation attention order. In one specific implementation, deformation region identification results are obtained from the surface of the target object. Curvature radius values ​​are extracted from these results, and a deviation quantification index is obtained by comparing with standard values. An intensity distribution, extracted from the deformation region identification results, is fused to this deviation quantification index. A preset curvature distribution is extracted using a distribution mapping, and gradient changes are obtained from this preset curvature distribution. These gradient changes are calculated based on differences in curvature radius values ​​to determine deformation boundary delineation. Feature fusion is adjusted based on the deformation boundary delineation. This feature fusion is obtained from the intensity distribution and gradient changes, acquiring range abrupt change data. An attention distribution is mapped using this range abrupt change data, and this mapping is performed using the preset curvature distribution. An initial weight allocation is determined using the attention distribution, based on the deformation boundary delineation. This initial weight allocation is combined with local curvature abrupt change locations, extracted from the deformation region identification results, to obtain a priority dynamic adjustment sequence. The attention order is optimized based on the priority dynamic adjustment sequence. The optimized attention order is then fused with overall deformation features, obtained from the surface of the target object, to determine the final deformation attention order. In one specific implementation, based on attention weights and the overall deformation characteristics of each region, it is first necessary to understand the concept of overall deformation characteristics. This overall deformation characteristic refers to the total deformation amplitude of each deformation region in the 3D model of the stamped part from its normal state to its deformed state, quantified by measuring the displacement vectors of points within the region. Specifically, the system extracts the region boundaries from the model data, calculates the average displacement value of all points, and forms a range index.

[0039] In one implementation, based on attention weights and the overall deformation range of each region, it is first necessary to understand the concept of overall deformation characteristics. This characteristic refers to the overall deformation amplitude of each deformation region in the three-dimensional model of the stamped part from the normal state to the deformed state, which is quantified by measuring the displacement vector of points within the region.

[0040] Specifically, the system extracts the region boundary from the model data, calculates the average displacement value of all points, and forms a feature index.

[0041] Preferably, the attention priority of each region is dynamically adjusted through a visual attention mechanism. This visual attention mechanism is a framework that simulates visual focusing; applied here in the analysis of stamped parts, it allocates resources based on input data, prioritizing high-risk areas.

[0042] Specifically, the mechanism includes an attention module that receives attention weights, the overall deformation range, and the locations of abrupt changes in local curvature as input, and calculates a priority score through weighted summation. In one possible implementation, the attention module employs a soft attention function to normalize the score for each region.

[0043] For example, in the stamping of precision mechanical parts, the system dynamically updates scores, prioritizing areas with multiple abrupt changes. This mechanism ensures real-time adjustments and adapts to different parts in mass production. In one implementation, area density can be introduced as an auxiliary factor. Specifically, area density refers to the concentration of deformation points, calculated together with overall deformation characteristics to enhance the accuracy of focus. For instance, in the stamping of electronic device housings, if an area has high density and abrupt changes, the visual attention mechanism amplifies its priority. This extension supports the handling of complex deformations and can be applied on continuous production lines. Furthermore, the dynamic adjustment of the visual attention mechanism can be implemented in a layered manner. First, the global scope is processed, then local abrupt changes are focused. This layering is reflected in the stamping analysis as a two-stage calculation: the global stage integrates weights and scope, while the local stage refines the abrupt change locations.

[0044] In one specific implementation, the final output of the deformation priority order includes coordinate annotations to aid in positioning. After the system generates a list, it can be exported as a report for use in workshop operations. During stamping on the instrument panel, the coordinates of high-priority areas are highlighted, supporting rapid intervention. This output, based on the technical features of the claims, provides support for practical applications.

[0045] S6. Based on the final deformation concern order, and taking into account the bending degree of each deformation area and the surface area of ​​the stamped part, evaluate and identify the repairability level of the stamped part blank. In one specific implementation, data of large-area gently deforming regions are extracted through the deformation attention sequence. Regional deformation classification and attention priority are then used to obtain an initial value for the degree of bending. Based on this initial value, small-area rapidly deforming regions are analyzed to obtain deformation boundary identification results and determine the material stress distribution map. Combining the material stress distribution map with the surface area size, a comprehensive evaluation and fusion of repair potential analysis and mapping is used to obtain a deformation repair simulation sequence. If the deformation repair simulation sequence exceeds a preset threshold, the surface detection parameters of the blank are adjusted. The regional bending degree evaluation is integrated using the detection parameters to determine the optimal repair path. The repairability level of the optimized repair path is obtained, and path-level matching is used to determine the repairability level of the stamping blank. In one implementation, based on the final deformation concern order and priority, the degree of bending is first assessed for large-area, gently deforming regions. These regions typically refer to the parts of a stamped part where deformation is gradual and widespread, and are quantified by measuring the average bending angle of the surface curve within these areas.

[0046] Specifically, the system extracts the region boundaries from the 3D model, calculates the angular changes between points, and forms a bending index.

[0047] For example, in sheet metal stamping production, for automotive door panel components, the degree of bending in large, gently deforming areas is determined by comparing the slopes of the initial and post-deformation curves, ensuring the uniformity of the assessment across the entire area. This assessment, combined with a priority order, treats high-priority areas first, supporting subsequent comprehensive judgment. Furthermore, the assessment of the degree of bending in small, rapidly deforming areas focuses on locations with severe and locally concentrated deformation. These areas often represent potential defects, such as sharp bending points. The system identifies the locations of curvature peaks by scanning the model mesh and calculates the peak difference in local bending angles.

[0048] It should be noted that this assessment is based on geometric analysis principles, examining the maximum bending change within a region, rather than the overall average. In the stamping scenario of home appliance casings, for edge bending areas, the system locates areas of rapid deformation and quantifies their degree of bending to reveal potential breakage risks. In this way, the assessment process enhances sensitivity to detail and combines it with priority considerations, resulting in targeted analysis. Preferably, the surface area of ​​the stamped part blank is included in the comprehensive assessment. Surface area refers to the original surface dimensions of the blank before deformation, calculated through coordinate integration.

[0049] Specifically, the system extracts the blank outline from the model data and calculates the total surface area value as a reference factor for evaluation.

[0050] For example, in the stamping of precision mechanical parts, if the surface area is large, the degree of bending in the large, gently deforming areas is weighted to reflect the overall difficulty of repair. This incorporation process ensures comprehensive evaluation and adapts to blanks of different sizes in mass production.

[0051] In one possible implementation, the comprehensive evaluation is achieved through a weighted summation mechanism. Lower weights are assigned to the degree of bending in large, gently deforming areas, while higher weights are assigned to small, rapidly deforming areas. These weights are then multiplied by a normalization coefficient based on the surface area to generate the total evaluation score.

[0052] S7. Based on the repairability level, determine the repair sequence and recommended repair process for each deformed area, and estimate the repair time required.

[0053] In one specific implementation, deformation region data is extracted based on the repairability level, and material deformation types are fused using region priority classification to obtain a repair sequence. The boundary stress distribution is analyzed for the repair sequence, and potential assessments are integrated using process matching rules to determine recommended repair processes. The recommended repair processes are combined with time estimation criteria to obtain repair path optimization results and determine the estimated repair time. If the estimated repair time exceeds a threshold, the deformation region data is adjusted to incorporate the boundary stress distribution, resulting in an adjusted repair sequence. In another specific implementation, the repair order of deformation regions is determined based on the repairability level. This order is achieved by sorting the levels, first processing high-level regions, i.e., those assessed as difficult to repair. The system extracts the level values ​​of each region from the assessment results and arranges them in descending order to ensure priority repair of regions with high potential risk.

[0054] In one specific implementation, the estimated repair time is calculated by integrating repairability level and regional characteristics. Specifically, the time estimate is based on an empirical model, taking into account factors such as regional area, degree of deformation and bending, and the average execution time of the recommended process. The system first quantifies these factors, for example, by converting the level into coefficients, and then multiplies them by a standard time value to form a total estimate.

[0055] In one possible implementation, determining the repair order can be aided by incorporating region dependencies. Region dependencies refer to the mutual influence between adjacent deformed regions, identified through the analysis model mesh. If two regions are of similar levels but have dependencies, their order is adjusted to process them simultaneously.

[0056] For example, during the stamping process of vehicle chassis components, the system detects dependencies and groups the relevant areas for repair. This expansion enhances the flexibility of the sequence and adapts to complex deformation situations.

[0057] Specifically, the matching of recommended repair processes can be further refined with process parameters, such as pressure values ​​or temperature settings. These parameters are dynamically adjusted according to the grade to ensure the process is targeted. In the stamping of electronic device housings, for high-grade areas with rapid deformation, the process parameters are set to higher pressures to effectively unfold the bending points. This refinement process stems from a mapping table between grades and parameters, supporting precise execution.

[0058] In one specific implementation, determining the repair sequence can be aided by incorporating regional dependencies. Regional dependencies refer to the mutual influence between adjacent deformed regions, identified through the analysis model mesh. If two regions are of similar levels but have dependencies, the sequence is adjusted to process them simultaneously. For example, in the stamping process of vehicle chassis components, after detecting dependencies, the system groups the relevant regions for repair. This expansion enhances the flexibility of the sequence, adapting to complex deformation situations. Specifically, the matching of recommended repair processes can be further refined using process parameters, such as pressure values ​​or temperature settings. These parameters are dynamically adjusted according to the level, ensuring the process is targeted. In the stamping of electronic device housings, for high-level areas with rapid deformation, the process parameters are set to higher pressures to effectively unfold the bending points. This refinement process originates from a mapping table between levels and parameters, supporting precise execution.

[0059] In one embodiment, the calculation of the estimated repair time can incorporate adjustments based on operator experience. After providing an initial estimate, the system allows input of correction factors to adjust based on historical data.

[0060] The above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. The present invention has been described in detail with reference to preferred embodiments. Those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications and substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for evaluating stamping deformation of an automobile part based on a visual attention mechanism, characterized by, The method comprises: S1, collecting image data of the stamping part deformation area, obtaining the surface area and the radius of curvature of each deformation area; based on the preset surface area threshold and the curvature radius threshold, the deformation area is divided into a large-area slow deformation area and a small-area rapid deformation area; S2, extracting the overall curvature distribution characteristics of the large-area slow deformation area and the local curvature mutation position of the small-area rapid deformation area, and obtaining the overall deformation characteristics of each deformation area; S3, comparing the curvature radius with the preset curvature threshold, determining the bending degree of each deformation area, and identifying the deformation serious area according to the bending degree; S4, comparing the curvature radius of the deformation serious area with the preset curvature radius standard value, determining the curvature deviation degree, and distributing the initial attention weight to each deformation serious area according to the curvature deviation degree; S5, according to the initial attention weight, combining the overall curvature distribution characteristics and the local curvature mutation position, obtaining the attention priority of each deformation area through the predetermined visual attention mechanism, and generating the final deformation attention sequence; S6, according to the final deformation attention sequence, comprehensively evaluating and identifying the repairability grade of the stamping part blank according to the bending degree of each deformation area and the surface area of the stamping part; S7, according to the repairability grade, determining the repair sequence of each deformation area, recommending the repair process, and estimating the required repair time.

2. The method for evaluating the stamping deformation of an automobile part based on the visual attention mechanism according to claim 1, characterized in that, In step S1, the deformation area is divided into a large-area slow deformation area and a small-area rapid deformation area, which specifically includes: processing the image data of the stamping part deformation area, identifying the initial deformation area; when the surface area value of the initial deformation area is greater than the preset surface area threshold and the curvature radius is greater than the preset curvature threshold, it is determined as a large-area slow deformation area; when the surface area value of the initial deformation area is less than the preset surface area threshold and the curvature radius is less than the preset curvature threshold, it is determined as a small-area rapid deformation area.

3. The method for evaluating the stamping deformation of an automobile part based on the visual attention mechanism according to claim 1, characterized in that, In step S2, the overall curvature distribution characteristics are represented by a curvature distribution histogram; the local curvature mutation position is represented by the curvature mutation point coordinates.

4. The method for evaluating the stamping deformation of an automobile part based on the visual attention mechanism according to claim 1, characterized in that, In step S3, the deformation serious area is identified according to the bending degree, which specifically includes: comparing the curvature radius of the deformation area with the preset severity curvature threshold, and the area below the threshold is marked as a high-bending area; the deformation intensity index of the adjacent area is comprehensively considered to optimize and merge the high-bending area, forming the final deformation serious area set.

5. The method for evaluating the stamping deformation of an automobile part based on the visual attention mechanism according to claim 1, characterized in that, In step S4, the initial attention weight of each deformation serious area is distributed according to the curvature deviation degree, which is specifically: the curvature deviation degree is quantified by the relative deviation percentage of the actual curvature radius and the standard value; the initial attention weight and the relative deviation percentage are in a positive correlation function relationship.

6. The method for evaluating the stamping deformation of an automobile part based on the visual attention mechanism according to claim 1, characterized in that, In step S5, the attention priority of each deformation area is obtained through the predetermined visual attention mechanism, and the final deformation attention sequence is generated, which includes: the initial attention weight, the overall deformation characteristics and the local curvature mutation position are comprehensively considered through the predetermined weighting function, and the dynamic priority score of each deformation area is outputted, and the final deformation attention sequence is generated by sorting.

7. The method for evaluating the stamping deformation of an automobile part based on the visual attention mechanism according to claim 1, characterized in that, In step S6, the repairability level of the stamping blank is evaluated and identified, specifically: an evaluation model is established, the deformation attention sequence, the bending degree, the region type and the surface area of the stamping blank are taken as input features, and a discrete repairability level is output through predetermined rule matching or model reasoning. 8.The method of claim 1, wherein the method further comprises: determining a deformation of the automobile part based on the attention mechanism. In step S7, the repair sequence of each deformation region is determined, the corresponding repair process is recommended, and the required repair time is estimated, specifically including: According to the repairability level and the deformation region characteristics, the repair process knowledge base is queried to match the recommended repair process; According to the complexity of the recommended repair process and the area of the region, combined with historical operation data, the single-region repair time is estimated; According to the deformation attention sequence, all repair regions are sorted to form a repair operation sequence, and the total estimated time is summarized.

Images