A multi-sensor-based connector automatic detection method and system

Abstract

The application relates to the technical field of industrial automation detection, and discloses a connector automatic detection method and system based on multiple sensors, which comprises the following steps: acquiring a point cloud data set, performing a filtering and smoothing operation on the point cloud data set, and obtaining a smooth three-dimensional point cloud model; performing clustering on the smooth three-dimensional point cloud model to obtain geometric shape features; combining the geometric shape features, obtaining a conflict solution path through path optimization and redistribution, and integrating to obtain a sensor deployment scheme; extracting coverage path data from the sensor deployment scheme, determining an adaptive parameter configuration; executing the adaptive parameter configuration, denoising real-time sensor signal data, and obtaining a denoised signal sequence; determining an internal defect position from the denoised signal sequence, and generating a detection result sequence; and determining a final optimization result sequence based on the detection result sequence. The method can solve the problem that the internal defect recognition capability of the prior art for a complex structure connector is insufficient.

Description

Technical Field

[0001] This invention relates to the field of industrial automation testing technology, and in particular to an automatic testing method and system for connectors based on multiple sensors. Background Technology

[0002] Currently, in the field of industrial automation, connectors are core components for transmitting signals and electrical energy in electronic devices, and their testing quality directly affects equipment reliability and system stability. With increasingly complex manufacturing processes, efficient and accurate automatic connector testing has become crucial for industrial upgrading, and intelligent sensing systems, as the core of the testing process, play a decisive role in the testing results due to their adaptability and accuracy.

[0003] In one existing technology, the intelligent sensing system employs a fixed sensor deployment method, combining manual visual inspection with single-device scanning for detection. Sensor positions are first preset based on experience, and data from the connector surface is collected by the sensors. Surface defects are then analyzed with manual assistance or by a single scanning device, finally outputting the detection results. However, this existing technology only inspects the basic surface of the connector and does not consider the diverse geometries of connectors. When faced with complex structures such as bent pins and multi-layered nesting, fixed sensors cannot adapt to regional differences, and a single scan is insufficient to capture internal features.

[0004] In summary, existing technologies cannot meet the automatic detection requirements of connectors with complex geometries, resulting in insufficient ability to identify internal defects in connectors with complex structures. Summary of the Invention

[0005] This invention provides a multi-sensor-based automatic connector inspection method and system to solve the problem that existing technologies cannot adapt to the automatic inspection requirements of complex connector geometries, resulting in insufficient ability to identify internal defects in complex connector structures.

[0006] In a first aspect, to solve the above-mentioned technical problems, the present invention provides an automatic connector detection method based on multiple sensors, comprising: A point cloud dataset containing the original point cloud data of the connector and surface contour information is obtained, and the point cloud dataset is filtered and smoothed to obtain a smoothed 3D point cloud model. Clustering is performed on the smooth 3D point cloud model to divide the connector region into multiple sub-regions, and geometric shape features are extracted and obtained. Based on the geometric features, a preliminary path set is generated. The preliminary path set is then optimized and redistributed to obtain conflict resolution paths. Finally, the conflict resolution paths are extracted and integrated to obtain a sensor deployment scheme. Extract coverage path data from the sensor deployment scheme, adjust sensor parameters based on the coverage path data, and determine adaptive parameter configuration; The adaptive parameter configuration is executed, real-time sensor signal data is collected, and the real-time sensor signal data is denoised to obtain a denoised signal sequence. The signal attenuation value is obtained from the denoised signal sequence to obtain the attenuation distribution sequence. If the attenuation distribution sequence exceeds the preset attenuation threshold, the signal segment in the attenuation distribution sequence that exceeds the preset attenuation threshold is locally amplified to determine the location of the internal defect and generate a detection result sequence. Based on the detection result sequence, the omission rate is calculated. If the omission rate is lower than a preset omission threshold, the final optimized result sequence is determined.

[0007] In one optional implementation, the step of performing a filtering and smoothing operation on the point cloud dataset to obtain a smoothed 3D point cloud model includes: Based on the point cloud dataset, determine whether the distance between points exceeds a preset distance threshold. If it does, perform a filtering and denoising operation on the point cloud dataset to obtain a denoised point cloud dataset. If it does not exceed the threshold, use the point cloud dataset as the denoised point cloud dataset. The denoised point cloud dataset is downsampled to obtain a density point cloud dataset; Smoothing operation is performed on the density point cloud dataset to obtain a smoothed point cloud dataset; Shape feature extraction is performed on the smoothed point cloud dataset to obtain a smoothed 3D point cloud model.

[0008] In one optional implementation, the clustering operation on the smooth 3D point cloud model, dividing the connector region into multiple sub-regions, and extracting and obtaining geometric shape features, includes: Based on the smooth 3D point cloud model, the connector region is clustered into multiple sub-regions to obtain sub-region point cloud subsets; wherein, the multiple sub-regions include pin-dense regions and nested layer regions; Based on the sub-region point cloud subset, calculate the local surface curvature value, calculate the curvature standard deviation based on the local surface curvature value, and if the curvature standard deviation is less than a preset curvature threshold, obtain the curvature feature subset of the pin-dense region; Based on the curvature feature subset, the point spacing distribution of the nested layer region is obtained, and the nested layer region is refined based on the point spacing distribution to determine the geometric shape boundary subset. The integrated density distribution parameters are obtained from the geometric boundary subset to obtain the geometric features of the connector.

[0009] In one optional implementation, the step of optimizing and reallocating the initial path set to obtain conflict resolution paths includes: Based on the initial path set, the path coverage rate is calculated. If the path coverage rate is lower than a preset path coverage threshold, path branches are added to obtain an adjusted path set. Based on the adjusted path set, the scan density distribution of the pin-dense region is analyzed to obtain a subset of the density distribution. The intersection points of the covered paths are obtained from the density distribution subset. The overlap of the intersection points of the covered paths is analyzed. If the overlap is higher than a preset overlap threshold, the intersection points of the covered paths are merged to obtain a simplified path set. For the simplified path set, combined with the geometric shape boundary subset, the path length is optimized to obtain the length-optimized path; Based on the optimized path of the length, the sensors are simulated and deployed. If there is a location conflict, the locations are reassigned to obtain a conflict resolution path.

[0010] In one optional implementation, adjusting sensor parameters based on the coverage path data to determine the adaptive parameter configuration includes: The curvature value distribution is calculated using the coverage path data, and the curvature mapping set is determined based on the curvature value distribution. If the curvature value of the curvature mapping set is higher than the preset curvature threshold, the sensor gain parameter is adjusted to determine the enhanced signal strength configuration. The parameter correlation is obtained from the enhanced signal strength configuration, and an adaptive parameter configuration that adapts to the shape characteristics of the connector is determined based on the parameter correlation.

[0011] In one optional implementation, the step of obtaining signal attenuation values ​​from the denoised signal sequence to obtain an attenuation distribution sequence, and if the attenuation distribution sequence exceeds a preset attenuation threshold, then locally amplifying the signal segments in the attenuation distribution sequence that exceed the preset attenuation threshold to determine the location of internal defects and generate a detection result sequence, includes: The signal attenuation value is obtained from the denoised signal sequence, and the signal is segmented according to the signal characteristics to obtain a denoised signal segment. The attenuation degree of the denoised signal segment is calculated based on the denoised signal segment. The attenuation levels are integrated according to the signal segmentation order to obtain an attenuation distribution sequence; If the attenuation distribution sequence exceeds a preset attenuation threshold, the denoised signal segment in the attenuation distribution sequence that exceeds the preset attenuation threshold is enhanced to obtain an amplified signal sequence. Local feature values ​​are extracted from the amplified signal sequence, the local feature values ​​are classified, the internal defect locations are determined, the internal defect locations are integrated, and a detection result sequence is generated.

[0012] In one optional implementation, the step of calculating the omission rate based on the detection result sequence, and determining the final optimized result sequence if the omission rate is lower than a preset omission threshold, includes: Based on the detection result sequence, extract the feature value sequence; The feature value sequence is classified to obtain the missing region sequence; If the sequence of missing regions is lower than a preset missing threshold, the detection result sequence is weighted and fused to generate a full-coverage report dataset. Using the full coverage report dataset, the omission rate distribution is calculated to determine the final optimized result sequence.

[0013] Secondly, the present invention provides an automatic connector inspection system based on multiple sensors, comprising: The point cloud model construction module acquires a point cloud dataset containing the original point cloud data of the connector and surface contour information, and performs filtering and smoothing operations on the point cloud dataset to obtain a smoothed 3D point cloud model. The sub-region feature extraction module performs clustering operations on the smooth 3D point cloud model, dividing the connector region into multiple sub-regions and extracting and obtaining geometric shape features. The sensor deployment optimization module, in conjunction with the geometric features, generates a preliminary path set, optimizes and reallocates the preliminary path set to obtain conflict resolution paths, and extracts and integrates the conflict resolution paths to obtain a sensor deployment scheme. The sensor parameter configuration module extracts coverage path data from the sensor deployment scheme, adjusts sensor parameters according to the coverage path data, and determines adaptive parameter configuration. The signal acquisition and denoising module executes the adaptive parameter configuration, acquires real-time sensor signal data, performs denoising processing on the real-time sensor signal data, and obtains a denoised signal sequence. The sub-region detection module obtains the signal attenuation value from the denoised signal sequence to obtain the attenuation distribution sequence. If the attenuation distribution sequence exceeds a preset attenuation threshold, the signal segment in the attenuation distribution sequence that exceeds the preset attenuation threshold is locally magnified to determine the location of the internal defect and generate a detection result sequence. The result integration and optimization module calculates the omission rate based on the detection result sequence. If the omission rate is lower than a preset omission threshold, the final optimized result sequence is determined.

[0014] Compared with the prior art, the present invention has the following beneficial effects: (1) This invention effectively removes noise and outliers from the point cloud dataset of the connector by filtering and smoothing it, and obtains a high-precision smooth three-dimensional point cloud model, which lays a reliable data foundation for subsequent region division and feature extraction. It solves the problem that the existing technology is unable to adapt to the automatic detection requirements of the complex geometry of the connector, resulting in insufficient ability to identify internal defects of complex structure connectors.

[0015] (2) By using a clustering algorithm to divide the smooth three-dimensional point cloud model into regions, this invention can accurately identify different functional sub-regions such as pin-dense areas and nested layer areas, and extract the geometric features such as curvature and density distribution of each sub-region, thus achieving accurate characterization of the complex geometric structure of the connector.

[0016] (3) This invention plans and optimizes the sensor coverage path by combining the geometric features of the connector. Through a series of operations such as adding path branches, merging overlapping intersections, optimizing path length and resolving position conflicts, an optimized sensor deployment scheme is obtained, which ensures full coverage detection of each area of ​​the connector by the sensor and effectively reduces the detection blind spot. Attached Figure Description

[0017] Figure 1 This is a schematic flowchart of an automatic connector detection method based on multiple sensors provided in the first embodiment of the present invention; Figure 2 This is a schematic diagram of a multi-sensor-based automatic connector detection system provided in the second embodiment of the present invention. Detailed Implementation

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

[0019] Reference Figure 1 The first embodiment of the present invention provides an automatic connector detection method based on multiple sensors, comprising the following steps: S11, acquire a point cloud dataset containing the original point cloud data of the connector and surface contour information, and perform a filtering and smoothing operation on the point cloud dataset to obtain a smoothed three-dimensional point cloud model. S12, perform clustering operation on the smooth 3D point cloud model to divide the connector region into multiple sub-regions and extract and obtain geometric shape features; S13, Combine the geometric features to generate a preliminary path set, perform path optimization and reallocation on the preliminary path set to obtain conflict resolution paths, and perform extraction and integration operations on the conflict resolution paths to obtain a sensor deployment scheme; S14, extract coverage path data from the sensor deployment scheme, adjust sensor parameters according to the coverage path data, and determine adaptive parameter configuration; S15, execute the adaptive parameter configuration, collect real-time sensor signal data, perform denoising processing on the real-time sensor signal data, and obtain a denoised signal sequence; S16, obtain the signal attenuation value from the denoised signal sequence to obtain the attenuation distribution sequence. If the attenuation distribution sequence exceeds the preset attenuation threshold, perform local amplification processing on the signal segment in the attenuation distribution sequence that exceeds the preset attenuation threshold to determine the location of the internal defect and generate a detection result sequence. S17. Based on the detection result sequence, calculate the omission rate. If the omission rate is lower than the preset omission threshold, determine the final optimized result sequence.

[0020] In step S11, a point cloud dataset containing the original point cloud data of the connector and surface contour information is obtained, and the point cloud dataset is filtered and smoothed to obtain a smoothed three-dimensional point cloud model.

[0021] A laser beam is emitted by a 3D scanner to capture the geometric information of the connector surface, obtaining raw point cloud data and surface contour information to obtain a point cloud dataset. It is worth noting that the point cloud dataset may contain noise and isolated outliers due to ambient light interference and scanning jitter. The point cloud dataset consists of a large number of 3D coordinate points, each corresponding to a sampling position on the connector surface. Simultaneously, the point cloud dataset contains surface contour information for each point, such as the normal vector and grayscale value at that position, presenting the overall 3D shape of the connector.

[0022] In one implementation, the step of performing a filtering and smoothing operation on the point cloud dataset to obtain a smoothed 3D point cloud model includes: Based on the point cloud dataset, determine whether the distance between points exceeds a preset distance threshold. If it does, perform a filtering and denoising operation on the point cloud dataset to obtain a denoised point cloud dataset. If it does not exceed the threshold, use the point cloud dataset as the denoised point cloud dataset. The denoised point cloud dataset is downsampled to obtain a density point cloud dataset; Smoothing operation is performed on the density point cloud dataset to obtain a smoothed point cloud dataset; Shape feature extraction is performed on the smoothed point cloud dataset to obtain a smoothed 3D point cloud model.

[0023] It should be noted that the preset distance threshold is determined based on the connector's design precision and the scanning device's resolution, and is typically 1.5-2 times the scanning device's minimum sampling interval. For example, when using a 3D scanner with a resolution of 0.3mm and a minimum sampling interval of 0.3mm, the preset distance threshold can be set to 0.5mm.

[0024] The filtering and denoising operation is implemented using a statistical filter. First, a neighborhood range is defined for each point, with a neighborhood radius of 2mm. The mean and standard deviation of the distance from all points in the neighborhood to the point are calculated. If the distance from a point to the mean of the neighborhood exceeds twice the standard deviation, the point is determined to be an isolated outlier and is removed. The filtering is completed by traversing all points to obtain a denoised point cloud dataset.

[0025] The downsampling operation employs a voxel grid downsampling method. First, the 3D space containing the denoised point cloud dataset is divided into equal-sized voxel grids, with each voxel having a side length of 1 mm. Each voxel grid is a cube with a side length of 1 mm. For each voxel grid, the mean 3D coordinates of all points within the grid are calculated. The mean point is retained as the representative point of the voxel, and all the representative points of the voxels form a density point cloud dataset.

[0026] The smoothing operation employs the moving least squares method. For each target point M in the density point cloud dataset, points within its neighborhood are selected to construct a local quadratic fitting surface equation. This method models the local surface as an explicit quadratic function of coordinates x and y. It is worth noting that, All are spatial coordinate variables. These are all coefficients of the surface equation. The sum of squared distances from all points in the neighborhood to the fitted surface is minimized; this sum of squared distances is expressed by the following mathematical formula: in, The number of points in the neighborhood. Given the coordinates of the k-th point in the neighborhood, solve for the coefficients. Then, based on the coordinates of the target point M Substituting x and y into The surface equation yields a new z The coordinates are updated to the target point position. All updated points are combined to form a smooth point cloud dataset, eliminating surface jagged noise.

[0027] Shape feature extraction employs principal component analysis (PCA) algorithm, which first calculates the mean coordinates of all points in the smoothed point cloud dataset. Exemplary To smooth the total number of points in the point cloud dataset, For the first l Given the coordinates of each point, construct the covariance matrix. Solve for the covariance matrix. The eigenvalues ​​and corresponding eigenvectors are selected, and the eigenvectors corresponding to the three largest eigenvalues ​​are chosen. These eigenvectors represent the three main directions of the connector: length, width, and height. Core shape features such as axial direction, symmetry, and curvature are captured. These features are then combined with smooth point cloud data to form a smooth 3D point cloud model that accurately reflects the 3D structure of the connector.

[0028] For example, taking the inspection of a car connector as an example, a 3D scanner with a resolution of 0.3mm is used to emit a laser beam to capture its surface geometry information and obtain a point cloud dataset containing 500,000 3D coordinate points and the normal vectors and gray values ​​of each point. Due to the interference of ambient light in the workshop, there are some isolated outliers in the data. First, the distance between points is determined, with a preset distance threshold of 0.5 mm. For points exceeding this threshold, a statistical filter is used with a neighborhood radius of 2 mm to calculate the mean and standard deviation of the neighborhood distance. Outliers with distances exceeding twice the standard deviation are removed to obtain a denoised point cloud dataset. Then, the dataset is divided into voxel grids with a side length of 1 mm. The mean coordinates of points within each grid are calculated as representative points to generate a density point cloud dataset of 30,000 points. Subsequently, for each target point, a fitted surface is constructed with a neighborhood radius of 2 mm. The coefficients are solved by minimizing the sum of squared distances from neighboring points to the surface, and the point coordinates are updated to obtain a smoothed point cloud dataset. Finally, principal component analysis is used to calculate the mean coordinates, construct the covariance matrix, and solve for the eigenvalues. The eigenvectors corresponding to the top three eigenvalues ​​are selected and combined with the smoothed point cloud data to form a smoothed 3D point cloud model.

[0029] In step S12, a clustering operation is performed on the smoothed 3D point cloud model to divide the connector region into multiple sub-regions, and geometric shape features are extracted and obtained, including: Based on the smooth 3D point cloud model, the connector region is clustered into multiple sub-regions to obtain sub-region point cloud subsets; wherein, the multiple sub-regions include pin-dense regions and nested layer regions; Based on the sub-region point cloud subset, calculate the local surface curvature value, calculate the curvature standard deviation based on the local surface curvature value, and if the curvature standard deviation is less than a preset curvature threshold, obtain the curvature feature subset of the pin-dense region; Based on the curvature feature subset, the point spacing distribution of the nested layer region is obtained, and the nested layer region is refined based on the point spacing distribution to determine the geometric shape boundary subset. The integrated density distribution parameters are obtained from the geometric boundary subset to obtain the geometric features of the connector.

[0030] It is worth noting that the clustering operation is implemented using the K-means clustering algorithm. Based on the clustering objective, the number of clusters is determined to be K=2, corresponding to the target sub-regions containing pin-dense areas and nested layer areas. Two points are randomly selected from the smooth 3D point cloud model as initial cluster centers, one initially corresponding to the pin area and the other initially corresponding to the nested layer area. Then, the Euclidean distance from each point in the smooth 3D point cloud model to these two cluster centers is calculated, and each point is assigned to the cluster containing the closer cluster center. Then, based on the coordinates of all points in each cluster, the center of the cluster is recalculated, and the clustering steps are repeated until one of the two iteration termination conditions is met after two adjacent iterations. The first iteration termination condition is when the change in the Euclidean distance between the two cluster centers in 3D space is less than or equal to 0.01mm, indicating that the change in the position of the cluster center is small and the termination condition is met. The second iteration termination condition is when the number of iterations reaches the maximum number of iterations, such as 100. At this time, the two clusters correspond to the pin-dense area and the nested layer area, respectively, and all points in the clusters form their respective sub-region point cloud subsets. For each point in the point cloud subset of the pin-dense region, select its neighborhood points with a radius of 1 mm, and fit a local quadratic surface. The steps for fitting the local quadratic surface are the same as those in S11. Calculate the principal curvature using the second derivative of the surface. and , representing the maximum and minimum principal curvatures of the surface at a certain point, respectively. Both reflect the degree of curvature of the surface, and the average of the two is taken as the local surface curvature value at that point. . and The calculation is expressed by the following mathematical formula: ， ； ， ， ； ; in, and Let z represent the first-order partial derivatives of z with respect to x and y, respectively. Let z represent the second partial derivative of z with respect to x, the mixed second partial derivative of z with respect to x and y, and the second partial derivative of z with respect to y, respectively. Substitute these values ​​into the calculation of the principal curvature. and The principal curvature of a surface at a point is the maximum and minimum of the normal curvatures in all directions at that point. The average of these two values ​​is taken as the local surface curvature value at that point. The local surface curvature values ​​of all points in the point cloud subset of the pin-dense area are statistically analyzed, the mean curvature is calculated, and the standard deviation of curvature is further calculated. The preset curvature threshold is determined based on the connector pin design specifications and represents the maximum allowable curvature fluctuation value of the pin surface. This threshold can be automatically calculated using the connector's CAD model, extracting the maximum principal curvature distribution of the pin surface, taking the 95th percentile as the theoretical curvature value, and setting a deviation based on manufacturing tolerances, such as ±0.02mm. -1 Alternatively, refer to industry standards (such as IEC 60512) for the allowable fluctuation range of pin curvature; the theoretical curvature value of a pin is typically 0.08 mm. -1 The curvature deviation is typically 0.02mm. -1 The maximum allowable curvature fluctuation of the pin surface is the theoretical curvature value plus the upper limit of the deviation, i.e., 0.1mm. -1 Due to its regular structure, the pin-dense area has minimal curvature variation, and the upper limit of the allowable curvature fluctuation range is typically 0.1mm. -1 Therefore, the preset curvature threshold is usually set to 0.1 / mm. If the standard deviation of curvature is lower than the preset curvature threshold, the local surface curvature values, corresponding point coordinates, mean curvature, and standard deviation of curvature of all points in the point cloud subset of the pin-dense area are selected to form a curvature feature subset.

[0031] To obtain the point spacing distribution of the nested layer region, it is first necessary to select a neighborhood of each point in the nested layer sub-region point cloud, with a neighborhood radius of 1 mm, and calculate the Euclidean distance between this point and all other points in the neighborhood. The mathematical formula is the same as that used to calculate the Euclidean distance from each point in the smoothed 3D point cloud model to these two cluster centers. Consistent, the average distance between neighboring points of each point is calculated using the following mathematical formula: in, The number of points in the neighborhood. Let be the distance between the i-th point and the j-th neighboring point. Collect the average neighboring point distances of all points in the nested layer region, arrange them in order of point coordinates, thus forming a point distance distribution. Refining the nested layer region requires setting a point distance threshold of 1.2mm based on this distribution. If the average neighboring point distance is less than the threshold, this region is classified as a tight nested layer region; if it is greater, it is classified as a transitional nested layer region, thus obtaining the refined nested layer region. Determining the geometric boundary subset requires processing the refined nested layer region using a voxel filter. The voxel side length is 0.5mm. Points within the voxel that are located at the region boundary are retained. These points simultaneously contain both the tight and transitional regions. Collect the coordinates of all boundary points and their corresponding point distance values ​​to form a geometric boundary subset, reflecting the contour boundary of the nested layer region. For each voxel, check if any of its internal points simultaneously contain points with both labels. If so, retain these points as boundary points.

[0032] Obtaining the comprehensive density distribution parameters requires selecting multiple sampling regions from the geometric boundary subset. Each sampling region is a cube with a side length of 2mm. The number of points within each sampling region is counted, and the point cloud density of each region is obtained by dividing the number of points by the volume of the sampling region. The point cloud densities of all sampling regions are collected, and the mean and variance are calculated. The mean and variance together constitute the comprehensive density distribution parameters. The number of sampling regions is obtained by dividing the volume of the nested layer region by the volume of the sampling region. When covering the sampling regions, adjacent sampling regions must not overlap, and the center coordinates must be arranged sequentially within the nested layer region at 2mm intervals to ensure no detection blind spots. The geometric features of the connector include the curvature feature subset of the pin-dense area, the point spacing distribution of the nested layer region, the geometric boundary subset, and the comprehensive density distribution parameters.

[0033] For example, taking a certain automotive connector as an example, K-means clustering was applied to its smooth 3D point cloud model, with K=2. After the initial cluster centers were randomly selected, the model converged after 3 iterations, resulting in a point cloud subset of the pin-dense region containing 8000 points and a point cloud subset of the nested layer region containing 5000 points. The local surface curvature value of each point in the pin-dense region was calculated, and the mean curvature was 0.08, in mm. -1 The standard deviation of curvature is 0.06 < 0.1, forming a curvature feature subset; for the nested layer region, the average point spacing is calculated with a neighborhood radius of 1 mm to obtain the point spacing distribution, which is then refined into a tight region and a transition region with a threshold of 1.2 mm. Boundary points are extracted using a 0.5 mm voxel filter to form a geometric shape boundary subset; 10 sampling regions are selected from the boundary subset to calculate the density, resulting in a density mean of 6 points per cubic millimeter and a variance of 0.8. Finally, the geometric shape features of the connector are obtained by combining all the data.

[0034] In step S13, a preliminary path set is generated by combining the geometric features. The preliminary path set is then optimized and redistributed to obtain conflict resolution paths. Finally, the conflict resolution paths are extracted and integrated to obtain a sensor deployment scheme.

[0035] In one implementation, the step of optimizing and reallocating the initial path set to obtain conflict resolution paths includes: Based on the initial path set, the path coverage rate is calculated. If the path coverage rate is lower than a preset path coverage threshold, path branches are added to obtain an adjusted path set. Based on the adjusted path set, the scan density distribution of the pin-dense region is analyzed to obtain a subset of the density distribution. The intersection points of the covered paths are obtained from the density distribution subset. The overlap of the intersection points of the covered paths is analyzed. If the overlap is higher than a preset overlap threshold, the intersection points of the covered paths are merged to obtain a simplified path set. For the simplified path set, combined with the geometric shape boundary subset, the path length is optimized to obtain the length-optimized path; Based on the optimized path of the length, the sensors are simulated and deployed. If there is a location conflict, the locations are reassigned to obtain a conflict resolution path.

[0036] It should be noted that the determination of the initial path set needs to be combined with the geometric shape characteristics of the connector sub-region. A hybrid path planning algorithm combining the grid method and the boundary following method is used to generate the initial coverage path. The specific implementation logic of the hybrid path planning algorithm combining the grid method and the boundary following method is as follows: First, the three-dimensional space of the connector is divided into grids with a side length of 0.5mm. Based on the geometric shape characteristics of the sub-region, the grids containing pin-dense areas and nested layer areas are marked as valid detection grids. For pin-dense areas, a parallel coverage path is generated within the valid grids according to the configured path step size of 1mm and the scan line spacing of 0.8mm. For nested layer areas, a circumferential path is generated by following the boundary according to the parameters that the path extends along the boundary tangent direction and the turning radius is 2mm. Finally, the two types of paths are integrated to form an initial path set that does not cross invalid areas and is adapted to the sub-region characteristics.

[0037] For example, taking the pin-dense region as an example, the point cloud density is 60 points per cubic millimeter, the basic path step size is 1 mm, parallel paths are generated along the pin arrangement direction, and for the nested layer region, a surrounding path is generated along the boundary contour. All paths together form a preliminary path set.

[0038] The path coverage rate is the ratio of the surface area of ​​the connector region covered by the initial path set to the total surface area of ​​the connector region to be inspected. The preset path coverage threshold is determined based on the tolerance of industrial inspection for blind spots, and is usually required to cover no obvious blind spots; the preset threshold is set to 80%. If the path coverage rate is lower than the preset path coverage threshold, path branches are added to obtain an adjusted path set.

[0039] To analyze the scanning density distribution in the pin-dense area, the scanning area corresponding to the adjusted path set needs to be divided into grid cells of equal size, such as a grid with a side length of 2mm, covering the pin-dense area. Then, the number of scanning paths in each grid cell is counted. The scanning density of each grid is calculated based on the number of scanning paths and the grid area. That is, the scanning density is the ratio of the number of paths in the grid to the grid area. The scanning density values ​​of all grids are sorted in order of grid coordinates to form a density distribution subset, which intuitively reflects the scanning density at different positions in the pin-dense area.

[0040] Analyzing the overlap of intersections of coverage paths requires extracting the coordinates of all intersections of coverage paths from a subset of the density distribution. For each intersection, the number of paths passing through that intersection is counted, and the overlap is calculated using the following mathematical formula: in, The degree of overlap at the intersection points. For each intersection point, this represents the number of paths that pass through that intersection point. This represents the total number of paths within the local area where the intersection point is located, where the local area is a circle with a radius of 1 mm centered at the intersection point. The preset overlap threshold is determined based on the need to avoid path redundancy and improve detection efficiency, and is usually set to 20%. For example, in the detection of a certain automotive connector, to ensure efficiency, the threshold is set to 20%. When the overlap of an intersection point reaches 25%, the intersection point is merged and redundant path segments are deleted. If the overlap is 15%, the original path is retained. After the above processing, a simplified path set is obtained.

[0041] Path length optimization first considers the geometric boundary subset, and based on the simplified path set, aims to minimize the total path length. Algorithm planning and optimization path, In the algorithm, the connector sub-region is divided into a grid map, with each grid being a cube with a side length of 1mm. The cost function for each grid is set as the sum of the distance from the grid to the target point and the penalty value for whether the grid crosses the boundary. If the grid crosses the geometric boundary, the penalty value increases, thereby preventing the path from exceeding the detection area. The penalty value for whether the grid crosses the boundary depends on whether the grid crosses the boundary. If it does not cross the boundary, the penalty value is 0; otherwise, it is 100, making the penalty value much larger than the normal grid cost, thus forcibly avoiding invalid boundaries. To determine whether a grid crosses the boundary, the boundary grid of the detection area is first determined based on the geometric boundary subset, such as the boundary grid between the nested layer area and the non-detection area. Then, it is determined whether the grids traversed by the path include the non-detection area grids outside the boundary grids. If they do, it is determined that the boundary has been crossed; otherwise, it has not been crossed. Next, the path nodes in the simplified path set are traversed, and the total length of different path combinations is calculated. The path with the smallest cost function value is selected first, that is, the path with the shortest length and that does not cross invalid boundaries. Finally, redundant detour segments in the original path are deleted, that is, paths with non-minimum cost functions, and the optimized path nodes are retained to form the length-optimized path. For example, if the total length of a simplified path set is 15cm, after optimization, the redundant boundary segments of the nested layer area are avoided, and the total length is shortened to 12cm, resulting in the length-optimized path.

[0042] Sensor simulation deployment requires using key nodes on the length-optimized path as candidate sensor deployment locations. Each node corresponds to one sensor, and the three-dimensional coordinates of each sensor are recorded. If the Euclidean distance between any two candidate sensor locations is less than 2mm, it is determined to be a location conflict. The calculation of the Euclidean distance is the same as that in step S12. For sensors with location conflicts, the coordinates are gradually fine-tuned in steps of 0.5mm along the vertical direction of the length-optimized path until the Euclidean distance is calculated to be greater than or equal to 2mm. The path segments corresponding to all the adjusted sensor locations are then integrated to obtain the conflict resolution path.

[0043] The extraction of the conflict resolution path requires extracting its core information, including the final deployment coordinates of each sensor and the scanning parameters of the corresponding path segments. The scanning parameters for each path segment include a 1mm path step size and a 0.8mm scan line spacing in the pin-dense area, and a 2mm turning radius in the nested layer area. The integration of the conflict resolution path requires categorizing the extracted information by sub-region. The extracted information is categorized by pin-dense area and nested layer area, and the number of sensors in each sub-region is counted. The categorized and integrated information is then organized into a table format: sub-region – sensor coordinates – scanning parameters – number, forming a sensor deployment plan.

[0044] In step S14, coverage path data is extracted from the sensor deployment scheme, sensor parameters are adjusted according to the coverage path data, and adaptive parameter configuration is determined.

[0045] It should be noted that, in order to extract coverage path data from the sensor deployment scheme, it is necessary to first classify the data by sub-region, and then extract the complete list of path nodes that the sensor is responsible for covering from the sensor deployment information of each sub-region. The path node list includes the three-dimensional coordinates of each node and the node connection order.

[0046] In one implementation, adjusting sensor parameters based on the coverage path data to determine the adaptive parameter configuration includes: The curvature value distribution is calculated using the coverage path data, and the curvature mapping set is determined based on the curvature value distribution. If the curvature value of the curvature mapping set is higher than the preset curvature threshold, the sensor gain parameter is adjusted to determine the enhanced signal strength configuration. The parameter correlation is obtained from the enhanced signal strength configuration, and an adaptive parameter configuration that adapts to the shape characteristics of the connector is determined based on the parameter correlation.

[0047] The calculation of curvature value distribution and determination of curvature mapping set using the covered path data requires first calculating the curvature value distribution. The covered path data contains the three-dimensional coordinates of all nodes on the path. Three consecutive adjacent nodes are selected, and an arc is fitted using these three points to obtain the radius of the arc. The specific process of fitting the arc is as follows: first, let the coordinates of the three consecutive adjacent nodes be... Take the coordinates projected onto the XOY plane. , Substitute into the general equation of a circle This yields a system of equations, which can be represented by the following mathematical expressions: in, All are coefficients. Solving for the coefficients. Then, using the formula for the radius of a circle, it can be expressed by the following mathematical expression: The radius of the arc is then calculated. The reciprocal of the arc radius is the curvature value of that path segment. Curvature values ​​are calculated by traversing all consecutive nodes, and all curvature values ​​are arranged in the order of the path nodes to obtain the curvature value distribution. Next, a curvature mapping set is determined. The connector detection area is divided into equal-sized grids with a side length of 1mm. Each calculated curvature value is associated with the grid coordinates of its corresponding path node, forming a grid coordinate-curvature value mapping relationship. These mapping relationships are then integrated to form the curvature mapping set.

[0048] The preset curvature threshold is determined based on the connector pin design specifications, and is typically set to 0.1 / mm. Adjusting the sensor gain parameters using a linear correspondence between curvature difference and gain compensation results in the adjusted gain expressed by the following mathematical formula: in, Based on the base gain, such as 5dB, The gain adjustment factor is set to 10 dB / mm. -1 , This represents the curvature value of the current region. The preset curvature threshold is 0.1 / mm. The enhanced signal strength configuration involves collecting adjusted gain parameters from all high-curvature regions. If the curvature value of a region exceeds the preset curvature threshold, that region is considered a high-curvature region. This is combined with the sensor sampling frequency and signal acquisition duration. The sensor sampling frequency is typically set to 100Hz, and the signal acquisition duration is set based on detection efficiency, typically 2 seconds. The region location, gain value, sampling frequency, and acquisition duration together constitute the enhanced signal strength configuration.

[0049] Obtaining parameter correlations requires extracting curvature values ​​from the enhanced signal strength configuration. and adjusted gain For the corresponding data pairs, the least squares method is used to fit the data pairs to obtain a linear correlation formula between the two, which is the parameter correlation relationship. Determining the adaptive parameter configuration requires combining the parameter correlation relationship with the geometric features of the connector sub-regions. For the pin-dense area and the nested layer area, the corresponding gain range is calculated by substituting into the linear correlation formula. Then, the gain range and sampling frequency of each sub-region are integrated to form an adaptive parameter configuration that adapts to different shape features, so that the parameters can be automatically adjusted according to the geometric features of the region.

[0050] For example, taking the detection of a certain automotive connector as an example, coverage path data is extracted from the sensor deployment scheme, and the curvature value is calculated by fitting a circular arc at three consecutive points to obtain the curvature value distribution, ranging from 0.08 to 0.20 mm. -1 Filter out those with a curvature > 0.1mm -1 The mapping relationships form a curvature mapping set, with a preset curvature threshold of 0.1mm. -1 For the mapped set with a curvature of 0.18 mm -1 In the region, the gain was adjusted to G=5 + 10×(0.18 - 0.1)=5.8dB, and combined with a 100Hz sampling frequency to form an enhanced signal strength configuration. 15 sets of curvature-gain data were extracted and fitted to obtain G=12. +3.2, Substituting into the pin-dense area, the curvature range is 0.12-0.20mm. -1 The resulting gain range is 4.64-5.6 dB, and the curvature range of the nested layer region is 0.08-0.15 mm. -1 The gain range is 3.96-5.0dB. The gain range and sampling frequency of each sub-region are integrated to form an adaptive parameter configuration that adapts to different shape characteristics.

[0051] In step S15, the adaptive parameter configuration is performed, real-time sensor signal data is collected, and the real-time sensor signal data is denoised to obtain a denoised signal sequence.

[0052] It should be noted that the sensor sampling frequency must satisfy the Nyquist sampling theorem, i.e., the sampling frequency must be greater than or equal to twice the highest frequency of the signal. This is determined in conjunction with the connector's detection accuracy and efficiency requirements, and is typically set to 100Hz. Noise reduction is achieved using a Kalman filter algorithm. First, initial filter parameters are set, including the state transition matrix A, process noise covariance Q, measurement noise covariance R, initial optimal signal value h, and initial error covariance. Since the signal collected by the sensor changes smoothly within a short period of time, A=1 is set to represent that the current signal state is approximately the same as the previous moment. Considering the electromagnetic interference intensity in the workshop, Q=0.01 is set to [unit missing]. 2 The measurement noise covariance R is determined by the sensor accuracy. Let R = 0.02. The initial optimal signal value h is taken as the measured value of the first data point of the original signal sequence. The initial error covariance... Set to 0.03, unit is 2 The Kalman filter algorithm for denoising consists of a prediction phase and an update phase. The state prediction formula in the prediction phase is expressed by the following mathematical formula: in, Let v be the predicted signal value at time v. Given the predicted signal value from the previous moment, when A=1, That is, based on the optimal signal value at the previous moment, the theoretical signal value at the current moment is directly estimated. The error covariance prediction formula is expressed by the following mathematical expression: in, Let be the covariance of the prediction error at time k. for transpose, Let $\mathbf{v}$ be the prediction error covariance from the previous time step. This formula estimates the uncertainty of the current prediction value by combining the error covariance from the previous time step with process noise. The Kalman gain is calculated during the update phase and is expressed by the following mathematical formula: in, Kalman gain is used to balance the weights of predicted and measured values. Small, meaning accurate prediction. A smaller value indicates a greater reliance on predicted values. Small, meaning precise measurement. A value that is too large indicates a greater reliance on measured values. The optimal signal value update is represented by the following mathematical formula: in, Let be the optimal signal value after denoising at time k. Let be the measurement value at time k. The error covariance update is expressed by the following mathematical formula: in, The updated error covariance at time k is used for iterative calculation at the next time step. By iterating through all data points of the original signal sequence and repeating the prediction and update stages, the optimal signal values ​​at each time step are arranged in chronological order to obtain the denoised signal sequence. The denoised signal sequence is presented as a two-dimensional list of "timestamp - optimal denoised signal value," where the timestamp is determined by the sensor sampling frequency of 100Hz, and each timestamp corresponds to an optimal voltage value after Kalman filtering, arranged chronologically to visually reflect the signal's changing trend during the detection process.

[0053] For example, taking the testing of a certain automotive connector as an example, the voltage of the original signal sequence fluctuates between 2.5-3.2V due to electromagnetic interference in the workshop. At timestamp 0.01s, the original voltage is 2.5V, and the optimal voltage after denoising is 2.5V; at timestamp 0.02s, the original voltage is 2.7V, the predicted value is 2.5V, the Kalman gain is 0.33, and the optimal voltage after denoising is 2.57V; at timestamp 0.03s, the original voltage is 2.9V, the predicted value is 2.57V, the Kalman gain is 0.33, and the optimal voltage after denoising is 2.64V; at timestamp 0.04s, the original voltage is 2.8V, the predicted value is 2.64V, the Kalman gain is 0.33, and the optimal voltage after denoising is 2.67V; at timestamp 0.05s, the original voltage is 3.0V, the predicted value is 2.67V, the Kalman gain is 0.33, and the optimal voltage after denoising is 2.73V. After traversing all the data, the fluctuation of the denoised sequence was reduced to within 0.1V.

[0054] In step S16, signal attenuation values ​​are obtained from the denoised signal sequence to obtain an attenuation distribution sequence. If the attenuation distribution sequence exceeds a preset attenuation threshold, the signal segments in the attenuation distribution sequence that exceed the preset attenuation threshold are locally amplified to determine the location of internal defects and generate a detection result sequence, including: The signal attenuation value is obtained from the denoised signal sequence, and the signal is segmented according to the signal characteristics to obtain a denoised signal segment. The attenuation degree of the denoised signal segment is calculated based on the denoised signal segment. The attenuation levels are integrated according to the signal segmentation order to obtain an attenuation distribution sequence; If the attenuation distribution sequence exceeds a preset attenuation threshold, the denoised signal segment in the attenuation distribution sequence that exceeds the preset attenuation threshold is enhanced to obtain an amplified signal sequence. Local feature values ​​are extracted from the amplified signal sequence, the local feature values ​​are classified, the internal defect locations are determined, the internal defect locations are integrated, and a detection result sequence is generated.

[0055] It should be noted that the signal attenuation value refers to the voltage difference between two adjacent sampling times. All signal attenuation values ​​can be obtained by traversing the denoised signal sequence. Segmentation based on signal characteristics uses signal stationarity as the segmentation feature, and segmentation is achieved by calculating the voltage variance within a sliding window, with the window size set to 10 sampling points. The voltage variance is calculated using the following mathematical formula: in, The voltage variance within the sliding window. This refers to the window size, which is 10 sampling points. Let i be the voltage value of the i-th sampling point within the window. This represents the average voltage across all sampling points within the window. If the variance difference between two consecutive windows is less than 0.01V... 2 If the signals are in the same segment, they are grouped into the same signal segment; otherwise, they are divided into new segments, resulting in multiple denoised signal segments. For each denoised signal segment, calculating its attenuation level first requires statistically analyzing the attenuation values ​​of all signals within the segment, taking the absolute values, and then calculating the average value as the attenuation level of that segment.

[0056] It is worth noting that the process of integrating the attenuation level into the attenuation distribution sequence is as follows: first, each denoised signal segment is assigned a number according to the time sequence of the signal segmentation; then, the attenuation level of each signal segment is associated with the corresponding number to form a key-value pair of number-attenuation level; finally, all attenuation levels are arranged in ascending order of segment number to obtain the attenuation distribution sequence.

[0057] It should be noted that the preset attenuation threshold is determined based on the signal attenuation pattern during normal connector operation. Signal attenuation data from defect-free connectors is collected through extensive experiments, and the upper limit of the 95% confidence interval is used as the threshold, typically set to 0.2V. The enhancement processing is implemented as follows: for denoised signal segments exceeding the preset attenuation threshold, a gain compensation coefficient is first calculated, which is the ratio of the actual attenuation level to the preset attenuation threshold. Then, each voltage value within the signal segment is multiplied by the gain compensation coefficient to stretch the signal amplitude, thereby obtaining an amplified signal sequence.

[0058] In one implementation, local feature values ​​are extracted from the amplified signal sequence. These local feature values ​​include voltage peak value, voltage variance, and the number of signal abrupt changes. The voltage peak value is the highest voltage within the segment, the number of signal abrupt changes is the number of times the voltage change rate exceeds 5V / s, and the voltage change rate is the ratio of the voltage difference between two adjacent sampling times within the denoised signal segment to the time interval between the two sampling times. The classification of the local feature values ​​is performed using a support vector machine (SVM) classification algorithm. This algorithm's model uses a linear kernel function, which is the dot product of the new input signal segment feature vector and the signal segment feature vector of a sample in the training set. This adapts to the linear separability of connector signal features. The hyperparameters include a penalty coefficient and a relaxation variable. The penalty coefficient is set to 1 to balance classification accuracy and generalization ability, while the relaxation variable is set to 0.1 to allow for a small number of misclassified samples. 1000 sets of signal segment feature data from defect-free and defective connectors are collected as training data. Each set contains three-dimensional features: peak voltage, voltage variance, and number of signal mutations. 600 sets are used as the training set, and 400 sets are used as the test set. The training process involves first standardizing the training set features using Min-Max standardization, where Min is the minimum value of the feature in the training set, and Max is the maximum value. After standardizing the training set features, they are input into the model to minimize the loss function, resulting in a classification hyperplane. The model is then used to classify new feature values ​​after the accuracy is verified to be greater than or equal to 95% on a test machine. The minimized loss function is expressed by the following mathematical formula: in, For the classification hyperplane weights, For the sample size, This is the penalty coefficient, which is 1. For example, a classification model is first trained using historical data, specifically when the voltage variance is less than 0.02V. 2 Signal segments with a mutation count of 1 or less are labeled as normal, while those with more than 1 mutation count are labeled as abnormal. This labeling rule is only used to generate supervisory signals for the training set; the final classification result is output by the trained SVM model. During classification, newly extracted local feature values ​​are input into the model, and the model outputs the classification result. The location of an internal defect is the connector region where the sensor path node corresponding to the signal segment classified as abnormal is located. When generating the detection result sequence, the determined internal defect location is first associated with the corresponding signal segment number and sub-region identifier, forming a "number-sub-region-defect location" association. Then, all associated data are arranged in order of signal segment number to clarify the connector region to which each defect location belongs. The final output detection result sequence includes the defect location coordinates, the sub-region to which it belongs, and the corresponding signal feature parameters.

[0059] In step S17, based on the detection result sequence, the omission rate is calculated. If the omission rate is lower than a preset omission threshold, the final optimized result sequence is determined, including: Based on the detection result sequence, extract the feature value sequence; The feature value sequence is classified to obtain the missing region sequence; If the sequence of missing regions is lower than a preset missing threshold, the detection result sequence is weighted and fused to generate a full-coverage report dataset. Using the full coverage report dataset, the omission rate distribution is calculated to determine the final optimized result sequence.

[0060] It is worth noting that extracting the feature value sequence first requires locating the detection data items corresponding to each sub-region from the detection result sequence, and then extracting the signal attenuation degree, the local curvature value corresponding to the internal defect location, and the point cloud density parameters from the data items.

[0061] It should be noted that the classification of the feature value sequence uses a decision tree classification algorithm. The core function of this algorithm uses the "Gini coefficient" as the node splitting criterion, and is expressed by the following mathematical formula: in, Gini coefficient, The total number of sample categories, The Gini coefficient represents the percentage of potential missing regions / no missing regions for class v samples in a given node. A smaller Gini coefficient indicates higher node purity. A decision tree is constructed by traversing all possible split points of the feature values ​​and selecting the split point with the smallest Gini coefficient. Key parameters of the decision tree classification algorithm include the maximum tree depth, minimum number of sample splits, and minimum number of leaf node samples. The maximum tree depth is set to 3, the minimum number of sample splits to 10, and the minimum number of leaf node samples to 5. First, classification rules are constructed based on historical detection data. If the signal attenuation of a sub-region is less than 0.1V (signal missing), or the point cloud density parameter deviates from the standard density of that sub-region by more than 10% (meaning the region's geometric features have not been fully collected), the sub-region is determined as a "potential missing region" and marked as 1; otherwise, it is determined as a "no missing region" and marked as 0. The parameters of all sub-regions in the feature value sequence are traversed, and the marking results are recorded in sub-region order to obtain the missing region sequence.

[0062] In one implementation, the preset omission threshold is determined based on the industrial inspection's tolerance for omission rates and connector inspection standards. It is typically set to the maximum allowable proportion of the number of potential omission regions obtained from the decision tree classification in S17 to the total number of sub-regions. For example, if a connector inspection is divided into 8 sub-regions, and a maximum of one potential omission is allowed, the preset omission threshold is set to 0.125. Technicians can set the threshold based on the actual number of sub-regions. The inspection result sequence is weighted and fused to generate a full-coverage report dataset. This requires first weighting the inspection data of each sub-region according to their weights, then integrating them into a unified data format. The final full-coverage report dataset includes the inspection data of each sub-region, which contains defect locations and signal characteristics. The weights for the weighted fusion are set according to the inspection priority of the sub-regions; the weights for pin-dense areas and nested layer areas (critical areas) are set to 0.6, and the weights for ordinary areas are set to 0.4.

[0063] For example, the full coverage report dataset includes feature values ​​of each sub-region, classification labels of each sub-region, and an overall detection overview of the connector. The overall detection overview of the connector includes the total number of sub-regions and the number of potential missed regions. Calculating the omission rate distribution requires first grouping the sub-regions in the full coverage report dataset into pin-dense areas and nested layer areas, calculating the proportion of potential missed regions in each group to the total number of regions in that group, which is the omission rate of each group, and then calculating the overall omission rate, which is the proportion of all potential missed regions to the total number of sub-regions. The results are then organized in the format of "region type - omission rate" to obtain the omission rate distribution. The sub-region type is obtained from step S12 to ensure that the grouping is consistent with the geometric features.

[0064] It should be noted that determining the final optimization result sequence requires first extracting the sub-region type, omission rate, and feature value from the full coverage report dataset. For sub-regions with an omission rate greater than 0, optimization measures are formulated based on their feature values. For example, the nested layer region is omitted due to "point cloud density deviation > 10%", and the optimization measure is "increase the voxel filter sampling density". Finally, the optimization content of all sub-regions is arranged in order according to the format of "sub-region number - optimization measure - expected improvement target" to form the final optimization result sequence.

[0065] In summary, this invention solves the problems of difficulty in adapting to complex geometric shapes of connectors and insufficient ability to identify internal defects by using multi-sensor collaboration and three-dimensional point cloud processing.

[0066] Reference Figure 2 The second embodiment of the present invention provides an automatic connector inspection system based on multiple sensors, comprising: The point cloud model construction module acquires a point cloud dataset containing the original point cloud data of the connector and surface contour information, and performs filtering and smoothing operations on the point cloud dataset to obtain a smoothed 3D point cloud model. The sub-region feature extraction module performs clustering operations on the smooth 3D point cloud model, dividing the connector region into multiple sub-regions and extracting and obtaining geometric shape features. The sensor deployment optimization module, in conjunction with the geometric features, generates a preliminary path set, optimizes and reallocates the preliminary path set to obtain conflict resolution paths, and extracts and integrates the conflict resolution paths to obtain a sensor deployment scheme. The sensor parameter configuration module extracts coverage path data from the sensor deployment scheme, adjusts sensor parameters according to the coverage path data, and determines adaptive parameter configuration. The signal acquisition and denoising module executes the adaptive parameter configuration, acquires real-time sensor signal data, performs denoising processing on the real-time sensor signal data, and obtains a denoised signal sequence. The sub-region detection module obtains the signal attenuation value from the denoised signal sequence to obtain the attenuation distribution sequence. If the attenuation distribution sequence exceeds a preset attenuation threshold, the signal segment in the attenuation distribution sequence that exceeds the preset attenuation threshold is locally magnified to determine the location of the internal defect and generate a detection result sequence. The result integration and optimization module calculates the omission rate based on the detection result sequence. If the omission rate is lower than a preset omission threshold, the final optimized result sequence is determined.

[0067] It should be noted that the multi-sensor-based automatic connector detection system provided in this embodiment of the invention is used to execute all the process steps of the multi-sensor-based automatic connector detection method in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0068] This invention also provides an electronic device. The electronic device includes a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a multi-sensor-based automatic connector detection program. When the processor executes the computer program, it implements the steps in the various multi-sensor-based automatic connector detection method embodiments described above, for example... Figure 1 The step S11 shown. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the above-described device embodiments, such as the point cloud model construction module.

[0069] For example, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the electronic device.

[0070] The electronic device may be a desktop computer, laptop, handheld computer, or smart tablet, etc. The electronic device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the above components are merely examples of electronic devices and do not constitute a limitation on the electronic device. It may include more or fewer components than described above, or combine certain components, or different components. For example, the electronic device may also include input / output devices, network access devices, buses, etc.

[0071] The processor referred to may be a central processing unit. Central Processing Unit (CPU) It can also be other general-purpose processors or digital signal processors (DSPs). Application Specific Integrated Circuit (ASIC) The processor can be an off-the-shelf field-programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc. The processor is the control center of the electronic device, connecting various parts of the entire electronic device via various interfaces and lines.

[0072] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0073] Wherein, if the modules / units integrated in the electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0074] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0075] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. An automatic connector detection method based on multiple sensors, characterized in that, include: A point cloud dataset containing the original point cloud data of the connector and surface contour information is obtained, and the point cloud dataset is filtered and smoothed to obtain a smoothed 3D point cloud model. Clustering is performed on the smooth 3D point cloud model to divide the connector region into multiple sub-regions, and geometric shape features are extracted and obtained. Based on the geometric features, a preliminary path set is generated. The preliminary path set is then optimized and redistributed to obtain conflict resolution paths. Finally, the conflict resolution paths are extracted and integrated to obtain a sensor deployment scheme. Extract coverage path data from the sensor deployment scheme, adjust sensor parameters based on the coverage path data, and determine adaptive parameter configuration; The adaptive parameter configuration is executed, real-time sensor signal data is collected, and the real-time sensor signal data is denoised to obtain a denoised signal sequence. The signal attenuation value is obtained from the denoised signal sequence to obtain the attenuation distribution sequence. If the attenuation distribution sequence exceeds the preset attenuation threshold, the signal segment in the attenuation distribution sequence that exceeds the preset attenuation threshold is locally amplified to determine the location of the internal defect and generate a detection result sequence. Based on the detection result sequence, the omission rate is calculated. If the omission rate is lower than a preset omission threshold, the final optimized result sequence is determined.

2. The automatic connector detection method based on multiple sensors according to claim 1, characterized in that, The step of filtering and smoothing the point cloud dataset to obtain a smoothed 3D point cloud model includes: Based on the point cloud dataset, determine whether the distance between points exceeds a preset distance threshold. If it does, perform a filtering and denoising operation on the point cloud dataset to obtain a denoised point cloud dataset. If it does not exceed the threshold, use the point cloud dataset as the denoised point cloud dataset. The denoised point cloud dataset is downsampled to obtain a density point cloud dataset; Smoothing operation is performed on the density point cloud dataset to obtain a smoothed point cloud dataset; Shape feature extraction is performed on the smoothed point cloud dataset to obtain a smoothed 3D point cloud model.

3. The automatic connector detection method based on multiple sensors according to claim 1, characterized in that, The clustering operation on the smooth 3D point cloud model, dividing the connector region into multiple sub-regions, and extracting and obtaining geometric shape features includes: Based on the smooth 3D point cloud model, the connector region is clustered into multiple sub-regions to obtain sub-region point cloud subsets; wherein, the multiple sub-regions include pin-dense regions and nested layer regions; Based on the sub-region point cloud subset, calculate the local surface curvature value, calculate the curvature standard deviation based on the local surface curvature value, and if the curvature standard deviation is less than a preset curvature threshold, obtain the curvature feature subset of the pin-dense region; Based on the curvature feature subset, the point spacing distribution of the nested layer region is obtained, and the nested layer region is refined based on the point spacing distribution to determine the geometric shape boundary subset. The integrated density distribution parameters are obtained from the geometric boundary subset to obtain the geometric features of the connector.

4. The automatic connector detection method based on multiple sensors according to claim 3, characterized in that, The step of optimizing and reallocating the initial path set to obtain conflict resolution paths includes: Based on the initial path set, the path coverage rate is calculated. If the path coverage rate is lower than a preset path coverage threshold, path branches are added to obtain an adjusted path set. Based on the adjusted path set, the scan density distribution of the pin-dense region is analyzed to obtain a subset of the density distribution. The intersection points of the covered paths are obtained from the density distribution subset. The overlap of the intersection points of the covered paths is analyzed. If the overlap is higher than a preset overlap threshold, the intersection points of the covered paths are merged to obtain a simplified path set. For the simplified path set, combined with the geometric shape boundary subset, the path length is optimized to obtain the length-optimized path; Based on the optimized path of the length, the sensors are simulated and deployed. If there is a location conflict, the locations are reassigned to obtain a conflict resolution path.

5. The automatic connector detection method based on multiple sensors according to claim 1, characterized in that, The step of adjusting sensor parameters based on the coverage path data to determine the adaptive parameter configuration includes: The curvature value distribution is calculated using the coverage path data, and the curvature mapping set is determined based on the curvature value distribution. If the curvature value of the curvature mapping set is higher than the preset curvature threshold, the sensor gain parameter is adjusted to determine the enhanced signal strength configuration. The parameter correlation is obtained from the enhanced signal strength configuration, and an adaptive parameter configuration that adapts to the shape characteristics of the connector is determined based on the parameter correlation.

6. The automatic connector detection method based on multiple sensors according to claim 1, characterized in that, The step involves obtaining signal attenuation values ​​from the denoised signal sequence to obtain an attenuation distribution sequence. If the attenuation distribution sequence exceeds a preset attenuation threshold, the signal segments in the attenuation distribution sequence that exceed the preset attenuation threshold are locally amplified to determine the location of internal defects and generate a detection result sequence, including: The signal attenuation value is obtained from the denoised signal sequence, and the signal is segmented according to the signal characteristics to obtain a denoised signal segment. The attenuation degree of the denoised signal segment is calculated based on the denoised signal segment. The attenuation levels are integrated according to the signal segmentation order to obtain an attenuation distribution sequence; If the attenuation distribution sequence exceeds a preset attenuation threshold, the denoised signal segment in the attenuation distribution sequence that exceeds the preset attenuation threshold is enhanced to obtain an amplified signal sequence. Local feature values ​​are extracted from the amplified signal sequence, the local feature values ​​are classified, the internal defect locations are determined, the internal defect locations are integrated, and a detection result sequence is generated.

7. The automatic connector detection method based on multiple sensors according to claim 6, characterized in that, The step of calculating the omission rate based on the detection result sequence, and determining the final optimized result sequence if the omission rate is lower than a preset omission threshold, includes: Based on the detection result sequence, a feature value sequence is extracted; wherein, the feature value sequence includes the signal attenuation degree detected in each sub-region, the local curvature value corresponding to the internal defect location, and the point cloud density parameter; The feature value sequence is classified to obtain the missing region sequence; If the sequence of missing regions is lower than a preset missing threshold, the detection result sequence is weighted and fused to generate a full-coverage report dataset. Using the full coverage report dataset, the omission rate distribution is calculated to determine the final optimized result sequence.

8. An automatic connector inspection system based on multiple sensors, characterized in that, include: The point cloud model construction module acquires a point cloud dataset containing the original point cloud data of the connector and surface contour information, and performs filtering and smoothing operations on the point cloud dataset to obtain a smoothed 3D point cloud model. The sub-region feature extraction module performs clustering operations on the smooth 3D point cloud model, dividing the connector region into multiple sub-regions and extracting and obtaining geometric shape features. The sensor deployment optimization module, in conjunction with the geometric features, generates a preliminary path set, optimizes and reallocates the preliminary path set to obtain conflict resolution paths, and extracts and integrates the conflict resolution paths to obtain a sensor deployment scheme. The sensor parameter configuration module extracts coverage path data from the sensor deployment scheme, adjusts sensor parameters according to the coverage path data, and determines adaptive parameter configuration. The signal acquisition and denoising module executes the adaptive parameter configuration, acquires real-time sensor signal data, performs denoising processing on the real-time sensor signal data, and obtains a denoised signal sequence. The sub-region detection module obtains the signal attenuation value from the denoised signal sequence to obtain the attenuation distribution sequence. If the attenuation distribution sequence exceeds a preset attenuation threshold, the signal segment in the attenuation distribution sequence that exceeds the preset attenuation threshold is locally magnified to determine the location of the internal defect and generate a detection result sequence. The result integration and optimization module calculates the omission rate based on the detection result sequence. If the omission rate is lower than a preset omission threshold, the final optimized result sequence is determined.

Images