A method and system for testing the tightness of automotive chassis pins

Abstract

This invention discloses a method and system for detecting the fit of automotive chassis pins, belonging to the field of industrial automation inspection technology. The method includes the following steps: fusing high-frequency and low-frequency components of images from different viewpoints using a weighting function to complete pyramid-structured image fusion based on adaptive weights; performing reference position pre-simulation based on a digital twin model, parametrically perturbing the relative position of the pin and slot by introducing assembly tolerance parameters, and registering and comparing the actual image with the reference image; extracting the contact area between the pin and slot based on the registration result, calculating the contact edge offset vector field, thereby obtaining the fit detection result and deviation value; and performing counterfactual reasoning on the fit detection result based on a structural causal model to output the corresponding maintenance decision scheme. This invention can automatically detect the fit status of chassis support arm positioning pins and slots in real time and accurately in complex workshop environments.

Description

Technical Field

[0001] This invention relates to the field of industrial automation testing technology, and in particular to a method and system for testing the tightness of automotive chassis pins. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] In automotive assembly workshops, the chassis is typically secured and supported by support arms using locating pins inserted into locating slots on the chassis. The reliability of this fastening process directly impacts the safety of subsequent assembly operations. Currently, traditional industrial cameras are used for visual inspection of the fit. However, minute gaps and obstructions exist when the pins are inserted into the slots, and under complex lighting conditions and limited viewing angles, traditional vision systems struggle to simultaneously capture both the complete contact contour and assembly deviations. Therefore, based on existing equipment, there is currently a lack of a method for inspecting the fit of automotive chassis pins that can adapt to complex scenarios. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the purpose of this invention is to provide a method and system for detecting the tightness of automotive chassis pins, which can automatically and accurately detect the tightness of chassis support arm positioning pins and slots in real time in complex workshop environments.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solution: To achieve the above objectives, the present invention is implemented through the following technical solution: The first aspect of this invention provides a method for testing the tightness of automotive chassis pins, comprising the following steps: Images of the chassis support arm positioning pin and the slot joint are acquired from multiple angles. The high-frequency and low-frequency components of the images from different perspectives are fused using a weighting function to complete the pyramid structure image fusion based on adaptive weights, thereby obtaining the actual image. Based on the digital twin model, a reference position pre-simulation is performed. By introducing assembly tolerance parameters, the relative position of the pin and the slot is parametrically disturbed to generate a corresponding multi-state reference image library. The actual image and the reference image are then registered and compared to obtain the registration result. Based on the registration results, the contact area between the pin and the slot is extracted, and the offset vector field of the contact edge is calculated to obtain the tightness detection results and deviation values. Based on the structural causal model, counterfactual reasoning is performed on the fit detection results to output the corresponding maintenance decision scheme.

[0006] Preferably, the images of the multi-angle chassis support arm positioning pin and the slot joint include at least the images of the left and right sides of the joint.

[0007] Preferably, the specific steps of pyramid structure image fusion based on adaptive weights are as follows: Gaussian blur and downsampling are applied to images from various viewpoints to construct multi-layer Gaussian pyramids; The high-frequency edge features and low-frequency contour information of each layer of the image are obtained based on the Laplacian operator and are used as high-frequency components and low-frequency components, respectively. Low-frequency components are weighted and fused based on an adaptive weighting function, and high-frequency components are selectively fused based on the principle of local energy maximization and the fusion results of low-frequency components. The fusion results of each layer are reconstructed to obtain the fused actual image.

[0008] Preferably, the specific steps for reference position pre-simulation based on the digital twin method are as follows: Perform 3D geometric modeling of the chassis support arm, positioning pin, and slot; Physical field simulation based on structural mechanics analysis is used to obtain stress and deformation information during the assembly process. Visual simulation is performed based on an optical imaging model to generate corresponding simulated images; By combining physical field simulation results with visual simulation results and introducing assembly tolerance parameters, a multi-state reference image library is generated.

[0009] Preferably, the registration comparison includes coarse registration based on rigid body transformation and fine registration based on non-rigid body deformation model, and the contact edge offset vector field is calculated after registration to quantify the pin insertion depth and coaxiality deviation.

[0010] Preferably, the structural causal model includes nodes for pin wear, slot deformation, support arm stress state, and assembly process parameters. By constructing a causal graph and performing counterfactual reasoning, the causal effects of each potential defect factor are obtained and their probabilities are ranked.

[0011] A second aspect of the present invention provides a vehicle chassis pin tightness testing system, comprising: The data acquisition module is configured to acquire images of the chassis support arm positioning pin and the slot connection area from multiple angles. It uses a weight function to fuse the high-frequency and low-frequency components of the images from different perspectives, and completes the pyramid structure image fusion based on adaptive weights to obtain the actual image. The registration module is configured to perform reference position pre-simulation based on the digital twin model. By introducing assembly tolerance parameters, it parametrically perturbs the relative position of the pin and the slot, generates a corresponding multi-state reference image library, and registers and compares the actual image with the reference image to obtain the registration result. The deviation calculation module is configured to extract the contact area between the pin and the slot based on the registration result, calculate the offset vector field of the contact edge, and thus obtain the tightness detection result and the deviation value. The maintenance decision module is configured to perform counterfactual reasoning on the fit detection results based on a structural causal model and output the corresponding maintenance decision scheme.

[0012] A third aspect of the present invention provides a computer-readable storage medium storing a computer program adapted to be loaded by a processor and to execute the steps of the automobile chassis pin tightness detection method as described in the first aspect of the present invention.

[0013] A fourth aspect of the present invention provides a computer device comprising: A processor, adapted to execute computer programs; A computer-readable storage medium storing a computer program, which, when executed by the processor, implements the automobile chassis pin tightness detection method as described in the first aspect of the present invention.

[0014] A fifth aspect of the present invention provides a computer program product or computer program comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the steps in the automobile chassis pin tightness detection method as described in the first aspect of the present invention.

[0015] Compared with the prior art, the present invention has the following beneficial effects: This invention achieves high-precision detection and intelligent diagnosis of the sealing state of chassis pins and slots by combining multi-angle image fusion, digital twin modeling, and causal reasoning.

[0016] In terms of image processing, by introducing an adaptive weighting function based on contrast, gradient, and brightness consistency, high-frequency and low-frequency information of multi-view images are fused in layers, which effectively suppresses the influence of complex lighting, reflection, and occlusion on the detection results and significantly improves the imaging integrity and detail clarity of the contact area.

[0017] In terms of the detection mechanism, by constructing a digital twin model that incorporates assembly tolerances, generating a multi-state reference image library, and combining rigid and non-rigid body registration strategies, the system achieves accurate quantification of key parameters such as pin insertion depth, contact area, and coaxiality deviation, thereby improving the robustness and adaptability of the detection results.

[0018] In terms of diagnostic analysis, by constructing a structural causal model that includes multiple assembly defect factors, and using counterfactual reasoning to achieve the mapping from detection results to root cause analysis, the detection system is made interpretable and predictive in its maintenance capabilities, thereby improving the accuracy of fault diagnosis and decision-making efficiency.

[0019] Therefore, this invention enables highly reliable testing and intelligent maintenance support for critical connection points of automotive chassis in complex industrial environments. The advantages of additional aspects of this invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a flowchart of the automobile chassis pin tightness testing method in Embodiment 1 of the present invention. Detailed Implementation

[0022] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0023] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof. The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0024] Example 1: Embodiment 1 of the present invention provides a method for detecting the tightness of automotive chassis pins, such as... Figure 1 As shown, it includes the following steps: S1: Obtain images of the chassis support arm positioning pin and the slot connection area from multiple angles. Use a weighting function to fuse the high-frequency and low-frequency components of the images from different perspectives to complete the pyramid structure image fusion based on adaptive weights, thereby obtaining the actual image.

[0025] S1.1: Obtain images of the engagement area between the chassis support arm positioning pin and the slot from multiple angles.

[0026] In one specific implementation, on the actual assembly line, images of the engagement area between the positioning pin and its mating slot in the chassis support arm are acquired by using multiple industrial cameras or a single camera at multiple workstations. To ensure coverage of key geometric features, the images of the engagement area of ​​the chassis support arm positioning pin and its slot from multiple angles should include at least images of the left and right sides of the engagement area, with top or oblique views added if necessary. The images should have sufficient resolution and consistent lighting to facilitate subsequent frequency domain separation. Camera parameters (focal length, aperture, exposure time) should be fixed during shooting to avoid excessive differences in brightness and sharpness between viewing angles.

[0027] S1.2: The high-frequency and low-frequency components of images from different perspectives are fused using a weighting function to complete the pyramid structure image fusion based on adaptive weights.

[0028] S1.2.1: Perform Gaussian blur and downsampling on images from various viewpoints to construct a multi-layer Gaussian pyramid.

[0029] In one specific implementation, the original image acquired from each viewpoint is first subjected to Gaussian blurring to eliminate high-frequency noise, and then the resolution is reduced by interlaced downsampling to obtain the first layer of the Gaussian pyramid image. The above Gaussian blurring and downsampling operations are repeated on the first layer image to form a multi-layer pyramid structure (usually 4 to 6 layers), with the image size of each layer being halved layer by layer. This process preserves the brightness and structural information of the images at different scales from each viewpoint.

[0030] S1.2.2: Based on the Laplacian operator, the high-frequency edge features and low-frequency contour information of each layer of the image are obtained as high-frequency components and low-frequency components, respectively.

[0031] In one specific implementation, the Laplacian operator is applied to spatial filtering of each Gaussian pyramid image layer to obtain the Laplacian pyramid for that layer. The high-frequency components in the Laplacian pyramid correspond to detailed features such as edges and textures in the image, i.e., high-frequency components, while the Gaussian pyramid itself contains slowly varying information such as the image's contours and illumination distribution, i.e., low-frequency components. Thus, each viewpoint image is decomposed into a set of high-frequency components and a set of low-frequency components at different scales.

[0032] S1.2.3: Low-frequency components are weighted and fused based on an adaptive weighting function, and high-frequency components are selectively fused based on the principle of local maximum energy and the fusion result of low-frequency components.

[0033] S1.2.3.1: Weighted fusion of low-frequency components based on adaptive weighting function.

[0034] In one specific implementation, for each scale layer, an adaptive weighting function is constructed based on the contrast, gradient magnitude, and brightness consistency of local image regions. Contrast is calculated using the gray-level range within a local window. The magnitude of the local gradient vector is calculated as the gradient magnitude, reflecting structural strength. Brightness consistency is characterized using the brightness variance within the local window, measuring uniformity.

[0035] By combining these three metrics, an adaptive fusion weight is generated for each pixel location. The weight is set higher in regions with high contrast, large gradient, and good brightness consistency. The low-frequency components of all viewpoint images at the same scale are weighted and superimposed according to this weight to obtain the low-frequency component fusion result.

[0036] S1.2.3.2: Selectively fuse high-frequency components based on the principle of maximizing local energy and the fusion results of low-frequency components.

[0037] In one specific implementation, the local high-frequency energy of each viewpoint is calculated. For the high-frequency components of the image at the current pyramid layer (i.e., the corresponding layer in the Laplacian pyramid), the local energy within a local window near each pixel location is calculated. The local energy is defined as the sum of squares (or the sum of absolute values) of the Laplacian response values ​​within the window, reflecting the activity level of details at that location. Higher energy indicates more pronounced edges and textures at that location.

[0038] Traditional methods, based solely on the principle of maximizing local energy, directly select the high-frequency value from the viewpoint with the highest local energy as the fusion result. However, this method ignores the consistency of low-frequency structures across different viewpoints, potentially leading to mismatches between high and low frequencies in regions with similar energies. Therefore, it is necessary to perform high-frequency component fusion correction based on low-frequency components.

[0039] First, the spatial structure information of the low-frequency component fusion result is obtained. The low-frequency component fusion result contains the optimal contour, illumination, and overall structure information at this scale. The following two spatial feature extraction methods are performed on the low-frequency component fusion result: The principal direction of the local gradient. Calculate the gradient vector (horizontal and vertical directions) of each pixel in the low-frequency image to obtain the magnitude and direction angle of the gradient. The principal direction reflects the orientation of the local structure at that location (e.g., edge direction, texture flow direction).

[0040] Local brightness consistency weight. This weight uses the brightness consistency index (i.e., brightness variance within a local window) calculated during low-frequency fusion. A high value indicates rich texture and clear structure, while a low value indicates a flat region. This weight can be used to adjust the intensity of high-frequency correction.

[0041] Then, based on the consistency of low-frequency spatial characteristics, the high-frequency components of each viewpoint are corrected. For each viewpoint... high frequency components At each pixel position At this point, calculate its spatial consistency score with the low-frequency fusion result. .

[0042] The specific steps include: calculate High-frequency components Local gradient principal direction , The local structure is obtained based on high-frequency components.

[0043] Calculate low-frequency fused images in The principal direction of the local gradient at the location .

[0044] Calculate the consistency of their directions: =cos(θv(p) θlow(p)) in The score represents the consistency of direction. The closer the value is to 1, the more consistent the directions. Simultaneously, the original local energy Ev(p) of the high-frequency components is calculated.

[0045] Then, the corrected candidate high-frequency component perspectives are calculated: .

[0046] in, Ev(p) represents the candidate high-frequency component viewpoint, and Ev(p) represents the original local energy, which is used to ensure the clarity of details. Used to ensure that the high-frequency edge direction is consistent with the low-frequency structure direction; wlow(p) is the low-frequency brightness consistency weight, used to reduce the correction intensity in flat areas and avoid noise interference; α and β are preset constants used to balance the energy term and the direction term.

[0047] Finally, the final high-frequency component is selected from the candidate high-frequency component perspective.

[0048] For each pixel position Compare candidate high-frequency component views from all perspectives Score, select the view with the highest score. The original high-frequency components of this perspective This serves as the high-frequency component after fusion at that location.

[0049] It should be noted that if the scores of all viewpoints are below a low threshold (indicating that there is no reliable high-frequency information, such as overexposure or darkness, at that location across all viewpoints), then a weighted average is calculated for the high-frequency values ​​from multiple viewpoints, with each viewpoint having its own weight. To avoid creating voids.

[0050] After the above corrections, the fused high-frequency component of the current pyramid layer is obtained. This high-frequency component retains the sharpest details while maintaining consistency with the low-frequency fusion result in edge orientation and texture flow, thus ensuring that the high and low frequencies can be naturally superimposed during subsequent pyramid reconstruction, eliminating artifacts and structural conflicts.

[0051] S1.2.4: Reconstruct the fusion results of each layer to obtain the fused actual image.

[0052] In one specific implementation, starting from the top of the pyramid (smallest size), the fused low-frequency layer is added to the high-frequency layer to obtain the reconstructed image of that layer; then it is upsampled (interpolated and amplified) and added to the fused low-frequency component of the next layer, while simultaneously superimposing the high-frequency component of that layer. This process is repeated until the original image size is restored, ultimately resulting in a real image that incorporates multi-view information, has clear edges, and a complete outline.

[0053] S2: Based on the digital twin model, a reference position pre-simulation is performed. By introducing assembly tolerance parameters, the relative position of the pin and the slot is parametrically disturbed to generate a corresponding multi-state reference image library. The actual image and the reference image are then registered and compared to obtain the registration result.

[0054] S2.1: Reference position pre-simulation based on digital twin model.

[0055] S2.1.1: Perform three-dimensional geometric modeling of the chassis support arm, positioning pin, and slot.

[0056] In one specific implementation, a high-fidelity three-dimensional geometric model is established in a digital twin platform based on the design drawings and material properties of the chassis support arm, positioning pin, and slot. The model accurately reflects the geometric features such as the pin diameter, chamfer, slot width, depth, and guide slope, while preserving the nominal assembly position relationships.

[0057] S2.1.2: Perform physical field simulation based on structural mechanics analysis to obtain stress and deformation information during the assembly process.

[0058] In one specific implementation, typical loads and constraints from the actual assembly process (such as indentation force, torque, and temperature changes) are applied, and the stress distribution and elastic / plastic deformation in the contact area between the pin and the slot are calculated using the finite element method. The simulation results are output in the form of a displacement field, reflecting the microscopic offsets that may occur under ideal assembly positions.

[0059] S2.1.3: Perform visual simulation based on the optical imaging model to generate the corresponding simulated image.

[0060] In one specific implementation, images are rendered in a digital twin scene based on virtual camera parameters identical to those of the actual camera. The rendering process takes into account optical effects such as diffuse reflection, specular reflection, shadows, and lens distortion to generate simulated images that closely resemble actual shooting conditions.

[0061] S2.1.4: Combine physical field simulation results with visual simulation results, and introduce assembly tolerance parameters to generate a multi-state reference image library.

[0062] In one specific implementation, based on the nominal assembly position, assembly tolerance parameters are introduced, such as pin insertion depth tolerance ±0.2mm, coaxiality tolerance φ0.1mm, and slot opening width tolerance, to parametrically perturb the relative position of the pin and slot. Each set of perturbation parameters corresponds to an assembly state, and physical field and visual simulations are performed respectively to generate a set of simulated images. This ultimately forms a multi-state reference image library covering common deviation ranges.

[0063] S2.2: Register and compare the actual image with the reference image.

[0064] In one specific implementation, the registration comparison includes coarse registration based on rigid body transformation and fine registration based on non-rigid body deformation model, and calculates the contact edge offset vector field after registration to quantify the pin insertion depth and coaxiality deviation.

[0065] The specific steps are as follows: S2.2.1: Coarse registration based on rigid body transformation.

[0066] Rigid body registration is performed between the actual image and each reference image based on feature points (such as the center of the pin end face and the corner points of the card slot edge), including translation, rotation and scaling, to find the best alignment position.

[0067] S2.2.2: Fine registration based on a non-rigid deformation model.

[0068] Based on coarse registration, a thin plate spline or B-spline free deformation model is used to perform fine-grained alignment of local nonlinear deformations (such as surface wear and micro-plastic deformation).

[0069] After registration, the offset vector field between the reference image and the actual image at the contact edge of the pin and the slot is calculated. This field contains the offset direction and magnitude of each edge point, thereby quantifying the pin insertion depth deviation and coaxiality deviation.

[0070] S3: Based on the registration results, extract the contact area between the pin and the slot, calculate the offset vector field of the contact edge, and thus obtain the tightness detection results and deviation values.

[0071] S3.1: Extract the contact area between the pin and the slot based on the registration results, and calculate the contact edge offset vector field.

[0072] In one specific implementation, based on the registration results, the contact boundary region between the pin and the slot is segmented from the actual image, typically the theoretical contact zone between the cylindrical surface of the pin and the inner wall of the slot. Using the contact edge offset vector field, the contact boundary in the reference model is mapped to the actual image space, and the normal and tangential offsets of the actual boundary relative to the ideal boundary are calculated point by point.

[0073] S3.2: Determine the pin insertion depth, contact area, and coaxiality deviation based on the contact edge offset vector field to obtain the fit test results and deviation values. Based on the contact edge offset vector field, extract the following key indicators: Pin insertion depth: The maximum offset component along the axial direction, reflecting whether it is inserted in place or is too deep / too shallow.

[0074] Contact area: The effective contact area enclosed by the actual contact boundary. The contact rate is obtained by comparing it with the theoretical contact area.

[0075] Coaxiality deviation: The distribution range of the radial offset vector and the roundness error.

[0076] By combining these indicators, the fit test results and specific deviation values ​​are obtained. In this embodiment, the fit test results include qualified, warning, and unqualified (the specific judgment threshold is set according to the actual situation). Specific deviation values ​​include insertion depth deviation of +0.15mm, contact rate of 82%, and coaxiality deviation of 0.08mm.

[0077] S4: Based on the structural causal model, perform counterfactual reasoning on the fit detection results and output the corresponding maintenance decision scheme.

[0078] In one specific implementation, the structural causal model includes nodes such as pin wear, slot deformation, support arm stress state, and assembly process parameters. By constructing a causal graph and performing counterfactual reasoning, the causal effects of each potential defect factor are obtained and their probabilities are ranked.

[0079] Given the currently detected fit deviation value, counterfactual reasoning is performed using a structural causal model. The specific steps are as follows: First, a counterfactual scenario is set up, such as "assuming the pin is not worn and other conditions remain unchanged." Then, the change in the fit under this counterfactual scenario is calculated to obtain the causal effect of each potential defect factor on the current observation bias. Finally, the causal effects of all potential defect factors are normalized to obtain a probability ranking and output a maintenance decision scheme. (For example: pin wear contributes 45%, slot deformation 30%, support arm stress state 15%, process parameters 10%).

[0080] Example of maintenance plan content: If the pin is the most likely to wear out, it is recommended to replace the pin or increase the surface hardness.

[0081] If the card slot is most likely to be deformed, it is recommended to reshape the card slot or replace the card slot component.

[0082] If the stress state of the support arm is abnormal, it is recommended to check the assembly force or add a stress relief structure.

[0083] If the process parameters contribute significantly, it is recommended to optimize the press-fitting parameters or increase lubrication.

[0084] The final result is an actionable maintenance report, which includes a ranking of defect causes, recommended actions, and expected improvement outcomes.

[0085] Example 2: Embodiment 2 of the present invention provides a vehicle chassis pin tightness testing system, comprising: The data acquisition module is configured to acquire images of the chassis support arm positioning pin and the slot connection area from multiple angles. It uses a weighting function to fuse the high-frequency and low-frequency components of the images from different perspectives, thereby completing the pyramid structure image fusion based on adaptive weights to obtain the actual image.

[0086] The data acquisition module is also configured as follows: Obtain images of the engagement area between the chassis support arm positioning pin and the slot from multiple angles.

[0087] In one specific implementation, on the actual assembly line, images of the engagement area between the positioning pin and its mating slot in the chassis support arm are acquired by using multiple industrial cameras or a single camera at multiple workstations. To ensure coverage of key geometric features, the images of the engagement area of ​​the chassis support arm positioning pin and its slot from multiple angles should include at least images of the left and right sides of the engagement area, with top or oblique views added if necessary. The images should have sufficient resolution and consistent lighting to facilitate subsequent frequency domain separation. Camera parameters (focal length, aperture, exposure time) should be fixed during shooting to avoid excessive differences in brightness and sharpness between viewing angles.

[0088] By using a weighting function to fuse the high-frequency and low-frequency components of images from different viewpoints, a pyramid structure image fusion based on adaptive weights is achieved.

[0089] Gaussian blur and downsampling are applied to images from various viewpoints to construct multi-layer Gaussian pyramids.

[0090] In one specific implementation, the original image acquired from each viewpoint is first subjected to Gaussian blurring to eliminate high-frequency noise, and then the resolution is reduced by interlaced downsampling to obtain the first layer of the Gaussian pyramid image. The above Gaussian blurring and downsampling operations are repeated on the first layer image to form a multi-layer pyramid structure (usually 4 to 6 layers), with the image size of each layer being halved layer by layer. This process preserves the brightness and structural information of the images at different scales from each viewpoint.

[0091] The high-frequency edge features and low-frequency contour information of each layer of the image are obtained based on the Laplacian operator and are used as high-frequency components and low-frequency components, respectively.

[0092] In one specific implementation, the Laplacian operator is applied to spatial filtering of each Gaussian pyramid image layer to obtain the Laplacian pyramid for that layer. The high-frequency components in the Laplacian pyramid correspond to detailed features such as edges and textures in the image, i.e., high-frequency components, while the Gaussian pyramid itself contains slowly varying information such as the image's contours and illumination distribution, i.e., low-frequency components. Thus, each viewpoint image is decomposed into a set of high-frequency components and a set of low-frequency components at different scales.

[0093] Low-frequency components are weighted and fused based on an adaptive weighting function, and high-frequency components are selectively fused based on the principle of local maximum energy and the fusion results of low-frequency components.

[0094] Low-frequency components are weighted and fused based on an adaptive weighting function.

[0095] In one specific implementation, for each scale layer, an adaptive weighting function is constructed based on the contrast, gradient magnitude, and brightness consistency of local image regions. Contrast is calculated using the gray-level range within a local window. The magnitude of the local gradient vector is calculated as the gradient magnitude, reflecting structural strength. Brightness consistency is characterized using the brightness variance within the local window, measuring uniformity.

[0096] By combining these three metrics, an adaptive fusion weight is generated for each pixel location. The weight is set higher in regions with high contrast, large gradient, and good brightness consistency. The low-frequency components of all viewpoint images at the same scale are weighted and superimposed according to this weight to obtain the low-frequency component fusion result.

[0097] Based on the principle of maximizing local energy and the fusion results of low-frequency components, high-frequency components are selectively fused.

[0098] In one specific implementation, the local high-frequency energy of each viewpoint is calculated. For the high-frequency components of the image at the current pyramid layer (i.e., the corresponding layer in the Laplacian pyramid), the local energy within a local window near each pixel location is calculated. The local energy is defined as the sum of squares (or the sum of absolute values) of the Laplacian response values ​​within the window, reflecting the activity level of details at that location. Higher energy indicates more pronounced edges and textures at that location.

[0099] Traditional methods, based solely on the principle of maximizing local energy, directly select the high-frequency value from the viewpoint with the highest local energy as the fusion result. However, this method ignores the consistency of low-frequency structures across different viewpoints, potentially leading to mismatches between high and low frequencies in regions with similar energies. Therefore, it is necessary to perform high-frequency component fusion correction based on low-frequency components.

[0100] First, the spatial structure information of the low-frequency component fusion result is obtained. The low-frequency component fusion result contains the optimal contour, illumination, and overall structure information at this scale. The following two spatial feature extraction methods are performed on the low-frequency component fusion result: The principal direction of the local gradient. Calculate the gradient vector (horizontal and vertical directions) of each pixel in the low-frequency image to obtain the magnitude and direction angle of the gradient. The principal direction reflects the orientation of the local structure at that location (e.g., edge direction, texture flow direction).

[0101] Local brightness consistency weight. This weight uses the brightness consistency index (i.e., brightness variance within a local window) calculated during low-frequency fusion. A high value indicates rich texture and clear structure, while a low value indicates a flat region. This weight can be used to adjust the intensity of high-frequency correction.

[0102] Then, based on the consistency of low-frequency spatial characteristics, the high-frequency components of each viewpoint are corrected. For each viewpoint... high frequency components At each pixel position At this point, calculate its spatial consistency score with the low-frequency fusion result. .

[0103] The specific steps include: calculate High-frequency components Local gradient principal direction , The local structure is obtained based on high-frequency components.

[0104] Calculate low-frequency fused images in The principal direction of the local gradient at the location .

[0105] Calculate the consistency of their directions: =cos(θv(p) θlow(p)) in The score represents the consistency of direction. The closer the value is to 1, the more consistent the directions. Simultaneously, the original local energy Ev(p) of the high-frequency components is calculated.

[0106] Then, the corrected candidate high-frequency component perspectives are calculated: .

[0107] in, Ev(p) represents the candidate high-frequency component viewpoint, and Ev(p) represents the original local energy, which is used to ensure the clarity of details. Used to ensure that the high-frequency edge direction is consistent with the low-frequency structure direction; wlow(p) is the low-frequency brightness consistency weight, used to reduce the correction intensity in flat areas and avoid noise interference; α and β are preset constants used to balance the energy term and the direction term.

[0108] Finally, the final high-frequency component is selected from the candidate high-frequency component perspective.

[0109] For each pixel position Compare candidate high-frequency component views from all perspectives Score, select the view with the highest score. The original high-frequency components of this perspective This serves as the high-frequency component after fusion at that location.

[0110] It should be noted that if the scores of all viewpoints are below a low threshold (indicating that there is no reliable high-frequency information, such as overexposure or darkness, at that location across all viewpoints), then a weighted average is calculated for the high-frequency values ​​from multiple viewpoints, with each viewpoint having its own weight. To avoid creating voids.

[0111] After the above corrections, the fused high-frequency component of the current pyramid layer is obtained. This high-frequency component retains the sharpest details while maintaining consistency with the low-frequency fusion result in edge orientation and texture flow, thus ensuring that the high and low frequencies can be naturally superimposed during subsequent pyramid reconstruction, eliminating artifacts and structural conflicts.

[0112] The fusion results of each layer are reconstructed to obtain the fused actual image.

[0113] In one specific implementation, starting from the top of the pyramid (smallest size), the fused low-frequency layer is added to the high-frequency layer to obtain the reconstructed image of that layer; then it is upsampled (interpolated and amplified) and added to the fused low-frequency component of the next layer, while simultaneously superimposing the high-frequency component of that layer. This process is repeated until the original image size is restored, ultimately resulting in a real image that incorporates multi-view information, has clear edges, and a complete outline.

[0114] The registration module is configured to perform reference position pre-simulation based on the digital twin model. By introducing assembly tolerance parameters, it parametrically perturbs the relative position of the pin and the slot, generates a corresponding multi-state reference image library, and compares the actual image with the reference image to obtain the registration result.

[0115] The registration module is also configured as follows: Reference position simulation based on digital twin model.

[0116] Three-dimensional geometric modeling was performed on the chassis support arm, positioning pin, and slot.

[0117] In one specific implementation, a high-fidelity three-dimensional geometric model is established in a digital twin platform based on the design drawings and material properties of the chassis support arm, positioning pin, and slot. The model accurately reflects the geometric features such as the pin diameter, chamfer, slot width, depth, and guide slope, while preserving the nominal assembly position relationships.

[0118] Physical field simulations based on structural mechanics analysis are used to obtain stress and deformation information during the assembly process.

[0119] In one specific implementation, typical loads and constraints from the actual assembly process (such as indentation force, torque, and temperature changes) are applied, and the stress distribution and elastic / plastic deformation in the contact area between the pin and the slot are calculated using the finite element method. The simulation results are output in the form of a displacement field, reflecting the microscopic offsets that may occur under ideal assembly positions.

[0120] Visual simulation is performed based on an optical imaging model to generate corresponding simulated images.

[0121] In one specific implementation, images are rendered in a digital twin scene based on virtual camera parameters identical to those of the actual camera. The rendering process takes into account optical effects such as diffuse reflection, specular reflection, shadows, and lens distortion to generate simulated images that closely resemble actual shooting conditions.

[0122] By combining physical field simulation results with visual simulation results and introducing assembly tolerance parameters, a multi-state reference image library is generated.

[0123] In one specific implementation, based on the nominal assembly position, assembly tolerance parameters are introduced, such as pin insertion depth tolerance ±0.2mm, coaxiality tolerance φ0.1mm, and slot opening width tolerance, to parametrically perturb the relative position of the pin and slot. Each set of perturbation parameters corresponds to an assembly state, and physical field and visual simulations are performed respectively to generate a set of simulated images. This ultimately forms a multi-state reference image library covering common deviation ranges.

[0124] The actual image is registered and compared with the reference image.

[0125] In one specific implementation, the registration comparison includes coarse registration based on rigid body transformation and fine registration based on non-rigid body deformation model, and calculates the contact edge offset vector field after registration to quantify the pin insertion depth and coaxiality deviation.

[0126] The specific steps are as follows: Coarse registration based on rigid body transformation.

[0127] Rigid body registration is performed between the actual image and each reference image based on feature points (such as the center of the pin end face and the corner points of the card slot edge), including translation, rotation and scaling, to find the best alignment position.

[0128] Fine registration based on a non-rigid deformation model.

[0129] Based on coarse registration, a thin plate spline or B-spline free deformation model is used to perform fine-grained alignment of local nonlinear deformations (such as surface wear and micro-plastic deformation).

[0130] After registration, the offset vector field between the reference image and the actual image at the contact edge of the pin and the slot is calculated using the offset calculation module. This field contains the offset direction and magnitude of each edge point, thereby quantifying the pin insertion depth deviation and coaxiality deviation.

[0131] The deviation calculation module is configured to extract the contact area between the pin and the slot based on the registration result, calculate the offset vector field of the contact edge, and thus obtain the tightness detection result and deviation value.

[0132] The deviation calculation module is also configured as follows: Based on the registration results, the contact area between the pin and the slot is extracted, and the offset vector field of the contact edge is calculated.

[0133] In one specific implementation, based on the registration results, the contact boundary region between the pin and the slot is segmented from the actual image, typically the theoretical contact zone between the cylindrical surface of the pin and the inner wall of the slot. Using the contact edge offset vector field, the contact boundary in the reference model is mapped to the actual image space, and the normal and tangential offsets of the actual boundary relative to the ideal boundary are calculated point-by-point. The pin insertion depth, contact area, and coaxiality deviation are determined based on the contact edge offset vector field, thus obtaining the fit detection results and deviation values. Based on the contact edge offset vector field, the following key indicators are extracted: Pin insertion depth: The maximum offset component along the axial direction, reflecting whether it is inserted in place or is too deep / too shallow.

[0134] Contact area: The effective contact area enclosed by the actual contact boundary. The contact rate is obtained by comparing it with the theoretical contact area.

[0135] Coaxiality deviation: The distribution range of the radial offset vector and the roundness error.

[0136] By combining these indicators, the fit test results and specific deviation values ​​are obtained. In this embodiment, the fit test results include qualified, warning, and unqualified (the specific judgment threshold is set according to the actual situation). Specific deviation values ​​include insertion depth deviation of +0.15mm, contact rate of 82%, and coaxiality deviation of 0.08mm.

[0137] The maintenance decision module is configured to perform counterfactual reasoning on the fit detection results based on a structural causal model and output the corresponding maintenance decision scheme.

[0138] The maintenance decision module is also configured as follows: In one specific implementation, the structural causal model includes nodes such as pin wear, slot deformation, support arm stress state, and assembly process parameters. By constructing a causal graph and performing counterfactual reasoning, the causal effects of each potential defect factor are obtained and their probabilities are ranked.

[0139] Given the currently detected fit deviation value, counterfactual reasoning is performed using a structural causal model. The specific steps are as follows: First, a counterfactual scenario is set up, such as "assuming the pin is not worn and other conditions remain unchanged." Then, the change in the fit under this counterfactual scenario is calculated to obtain the causal effect of each potential defect factor on the current observation bias. Finally, the causal effects of all potential defect factors are normalized to obtain a probability ranking and output a maintenance decision scheme. (For example: pin wear contributes 45%, slot deformation 30%, support arm stress state 15%, process parameters 10%).

[0140] Example of maintenance plan content: If the pin is the most likely to wear out, it is recommended to replace the pin or increase the surface hardness.

[0141] If the card slot is most likely to be deformed, it is recommended to reshape the card slot or replace the card slot component.

[0142] If the stress state of the support arm is abnormal, it is recommended to check the assembly force or add a stress relief structure.

[0143] If the process parameters contribute significantly, it is recommended to optimize the press-fitting parameters or increase lubrication.

[0144] The final result is an actionable maintenance report, which includes a ranking of defect causes, recommended actions, and expected improvement outcomes.

[0145] Example 3: Embodiment 3 of the present invention provides a computer-readable storage medium storing a computer program adapted for loading by a processor and executing the steps of the automobile chassis pin tightness detection method as described in Embodiment 1 of the present invention, wherein the steps are: Images of the chassis support arm positioning pin and the slot joint are acquired from multiple angles. The high-frequency and low-frequency components of the images from different perspectives are fused using a weighting function to complete the pyramid structure image fusion based on adaptive weights, thereby obtaining the actual image. Based on the digital twin model, a reference position pre-simulation is performed. By introducing assembly tolerance parameters, the relative position of the pin and the slot is parametrically disturbed to generate a corresponding multi-state reference image library. The actual image and the reference image are then registered and compared to obtain the registration result. Based on the registration results, the contact area between the pin and the slot is extracted, and the offset vector field of the contact edge is calculated to obtain the tightness detection results and deviation values. Based on the structural causal model, counterfactual reasoning is performed on the fit detection results to output the corresponding maintenance decision scheme.

[0146] The detailed steps are the same as the automobile chassis pin tightness testing method provided in Example 1, and will not be repeated here.

[0147] Example 4: Embodiment 4 of the present invention provides a computer device, the device comprising: A processor, adapted to execute computer programs; A computer-readable storage medium storing a computer program, which, when executed by the processor, implements the steps in the automobile chassis pin tightness detection method as described in Embodiment 1 of the present invention, wherein the steps are: Images of the chassis support arm positioning pin and the slot joint are acquired from multiple angles. The high-frequency and low-frequency components of the images from different perspectives are fused using a weighting function to complete the pyramid structure image fusion based on adaptive weights, thereby obtaining the actual image. Based on the digital twin model, a reference position pre-simulation is performed. By introducing assembly tolerance parameters, the relative position of the pin and the slot is parametrically disturbed to generate a corresponding multi-state reference image library. The actual image and the reference image are then registered and compared to obtain the registration result. Based on the registration results, the contact area between the pin and the slot is extracted, and the offset vector field of the contact edge is calculated to obtain the tightness detection results and deviation values. Based on the structural causal model, counterfactual reasoning is performed on the fit detection results to output the corresponding maintenance decision scheme.

[0148] The detailed steps are the same as the automobile chassis pin tightness testing method provided in Example 1, and will not be repeated here.

[0149] Example 5: Embodiment 5 of the present invention provides a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the steps in the automobile chassis pin tightness detection method described in Embodiment 1 of the present invention, wherein the steps are: Images of the chassis support arm positioning pin and the slot joint are acquired from multiple angles. The high-frequency and low-frequency components of the images from different perspectives are fused using a weighting function to complete the pyramid structure image fusion based on adaptive weights, thereby obtaining the actual image. Based on the digital twin model, a reference position pre-simulation is performed. By introducing assembly tolerance parameters, the relative position of the pin and the slot is parametrically disturbed to generate a corresponding multi-state reference image library. The actual image and the reference image are then registered and compared to obtain the registration result. Based on the registration results, the contact area between the pin and the slot is extracted, and the offset vector field of the contact edge is calculated to obtain the tightness detection results and deviation values. Based on the structural causal model, counterfactual reasoning is performed on the fit detection results to output the corresponding maintenance decision scheme.

[0150] The detailed steps are the same as the automobile chassis pin tightness testing method provided in Example 1, and will not be repeated here.

[0151] The steps and methods involved in Examples 2, 3, 4 and 5 above correspond to those in Example 1. For specific implementation methods, please refer to the relevant description section of Example 1.

[0152] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0153] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of this application is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in or transmitted through a computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless means. The computer-readable storage medium can be any available medium that a computer can access or a data processing device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium, an optical medium, or a semiconductor medium, etc.

[0154] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for detecting the adhesion of an automobile chassis pin, characterized by, Includes the following steps: Images of the chassis support arm positioning pin and the slot joint are acquired from multiple angles. The high-frequency and low-frequency components of the images from different perspectives are fused using a weighting function to complete the pyramid structure image fusion based on adaptive weights, thereby obtaining the actual image. Based on the digital twin model, a reference position pre-simulation is performed. By introducing assembly tolerance parameters, the relative position of the pin and the slot is parametrically disturbed to generate a corresponding multi-state reference image library. The actual image and the reference image are then registered and compared to obtain the registration result. Based on the registration results, the contact area between the pin and the slot is extracted, and the offset vector field of the contact edge is calculated to obtain the tightness detection results and deviation values. Based on the structural causal model, counterfactual reasoning is performed on the fit detection results to output the corresponding maintenance decision scheme.

2. The method for detecting the tightness of automobile chassis pins as described in claim 1, characterized in that, The images of the multi-angle chassis support arm positioning pin and the slot connection area should include at least the images of the left and right sides of the connection area.

3. The method for detecting the tightness of automotive chassis pins as described in claim 1, characterized in that, The specific steps of pyramid structure image fusion based on adaptive weights are as follows: Gaussian blur and downsampling are applied to images from various viewpoints to construct multi-layer Gaussian pyramids; The high-frequency edge features and low-frequency contour information of each layer of the image are obtained based on the Laplacian operator and are used as high-frequency components and low-frequency components, respectively. Low-frequency components are weighted and fused based on an adaptive weighting function, and high-frequency components are selectively fused based on the principle of local energy maximization and the fusion result of low-frequency components. The adaptive weighting function is constructed based on the contrast, gradient magnitude and brightness consistency of local image regions. The fusion results of each layer are reconstructed to obtain the fused actual image.

4. The method for detecting the tightness of automotive chassis pins as described in claim 1, characterized in that, The specific steps for reference location rehearsal based on the digital twin method are as follows: Perform 3D geometric modeling of the chassis support arm, positioning pin, and slot; Physical field simulation based on structural mechanics analysis is used to obtain stress and deformation information during the assembly process. Visual simulation is performed based on an optical imaging model to generate corresponding simulated images; By combining physical field simulation results with visual simulation results and introducing assembly tolerance parameters, a multi-state reference image library is generated.

5. The method for detecting the tightness of automotive chassis pins as described in claim 1, characterized in that, The registration comparison includes coarse registration based on rigid body transformation and fine registration based on non-rigid body deformation model. After registration, the contact edge offset vector field is calculated to quantify the pin insertion depth and coaxiality deviation.

6. The method for detecting the tightness of automotive chassis pins as described in claim 1, characterized in that, The structural causal model includes nodes such as pin wear, slot deformation, support arm stress state, and assembly process parameters. By constructing a causal graph and performing counterfactual reasoning, the causal effects of each potential defect factor are obtained and their probabilities are ranked.

7. A system for detecting the tightness of automotive chassis pins, characterized in that, include: The data acquisition module is configured to acquire images of the chassis support arm positioning pin and the slot connection area from multiple angles. It uses a weight function to fuse the high-frequency and low-frequency components of the images from different perspectives, and completes the pyramid structure image fusion based on adaptive weights to obtain the actual image. The registration module is configured to perform reference position pre-simulation based on the digital twin model. By introducing assembly tolerance parameters, it parametrically perturbs the relative position of the pin and the slot, generates a corresponding multi-state reference image library, and registers and compares the actual image with the reference image to obtain the registration result. The deviation calculation module is configured to extract the contact area between the pin and the slot based on the registration result, calculate the offset vector field of the contact edge, and thus obtain the tightness detection result and the deviation value. The maintenance decision module is configured to perform counterfactual reasoning on the fit detection results based on a structural causal model and output the corresponding maintenance decision scheme.

8. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the automobile chassis pin tightness detection method as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program adapted to be loaded by a processor and executed as described in any one of claims 1-6.

10. A computer device, characterized in that, include: A processor, adapted to execute computer programs; A computer-readable storage medium storing a computer program, which, when executed by the processor, implements the method for detecting the tightness of automobile chassis pins as described in any one of claims 1-6.

Images