A method and system for detecting defects in automotive evaporators

Abstract

This invention relates to the field of data processing technology, and in particular to a method and system for detecting defects in automotive evaporators. The method includes the following steps: acquiring point cloud data of the surface of automotive air conditioning evaporator fins; based on the distribution characteristics, orientation consistency factor, and flatness factor of the neighborhood data point set of each three-dimensional coordinate point in the point cloud data, determining the probability that the neighborhood data point set of each three-dimensional coordinate point belongs to the fin surface region, thereby obtaining the fin surface coordinate points; based on the density weight and quality influence factor of each fin surface coordinate point, obtaining the adaptive neighborhood radius of each fin surface coordinate point; based on the neighborhood radius, obtaining the normal vector of each fin surface coordinate point, obtaining the standard normal vector; judging the angle between the normal vector of each three-dimensional coordinate point in the point cloud data and the standard normal vector, obtaining abnormal coordinate points. This invention improves the accuracy of evaporator fin surface defect detection.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and system for detecting defects in automotive evaporators. Background Technology

[0002] The evaporator in an automotive air conditioning system is a core component of the heat exchange system, and its densely packed aluminum fins are crucial to heat exchange efficiency. During manufacturing and distribution, these high-density, relatively soft fins are highly susceptible to defects such as tipping, bending, large-scale collapse, and scratches due to external forces like impacts and compression, severely affecting product performance and reliability. To ensure product quality, precise defect detection of the evaporator fins is an essential production step.

[0003] Laser profilometry is widely used to acquire 3D point cloud data of evaporator fin surfaces for non-contact, high-precision inspection. A core technique is a normal vector estimation algorithm based on Local Principal Component Analysis (PCA). This algorithm calculates the normal vector of the local neighborhood of each 3D coordinate point in the point cloud data and compares it with a standard normal vector to identify defects such as attitude deviation, large-scale collapse, and scratches. Therefore, a standard normal vector direction is usually required for comparison. However, since the standard direction is unknown, we cannot directly compare the estimated normal vector with the standard direction.

[0004] Ideally, the standard normal vector should be determined solely by the point cloud of a defect-free, flat fin surface. However, the actual point cloud data inevitably contains a large number of defect points (such as points in collapsed or bent areas). Directly averaging these points to calculate the standard normal vector is inaccurate. Furthermore, when the principal component analysis (PCA) algorithm estimates the normal vector, its neighborhood radius is usually set to a fixed empirical value. Setting a small neighborhood radius makes the normal vector calculated by the algorithm highly unstable when faced with sparse point clouds caused by uneven scanning angles. Conversely, increasing the neighborhood radius to ensure computational stability may cause the neighborhood radius to easily cross multiple independent fin surface boundaries, leading to PCA calculating completely incorrect normal vectors. Therefore, both excessively large and excessively small neighborhood radii can cause errors in normal vector estimation, resulting in inaccurate standard normal vectors and subsequent incorrect defect identification. Summary of the Invention

[0005] To address the technical issues that point cloud data contains various types of defects, directly using the mean of the normal vectors of all point cloud data as the standard vector is inaccurate, and the fixed neighborhood radius of the principal component analysis algorithm also leads to inaccurate standard normal vectors, this invention provides a method and system for detecting defects in automotive evaporators.

[0006] In a first aspect, the present invention provides a method for detecting defects in automotive evaporators, employing the following technical solution:

[0007] A method for detecting defects in automotive evaporators, comprising the following steps:

[0008] Point cloud data of the fin surface of an automotive air conditioner evaporator is collected. Based on the geometric features of the neighborhood data point set of each three-dimensional coordinate point in the point cloud data, the distribution characteristics, orientation consistency factor, and flatness factor of the neighborhood data point set of each three-dimensional coordinate point in the point cloud data are obtained. The probability that the neighborhood data point set of each three-dimensional coordinate point belongs to the fin surface region is obtained, and the probability is positively correlated with the distribution characteristics, orientation consistency factor, and flatness factor. Based on the probability, the coordinate points of each fin surface are obtained.

[0009] Based on the local density of the coordinate points on the fin surface, the density weight of each fin surface coordinate point is obtained; the adaptive neighborhood radius of each fin surface coordinate point is obtained, and the adaptive neighborhood radius is positively correlated with the density weight; the adaptive neighborhood radius is input into the principal component analysis algorithm to obtain the normal vector of each fin surface coordinate point; the mean of the normal vectors of all fin surface coordinate points is used as the standard normal vector; the normal vector of each three-dimensional coordinate point in the point cloud data is compared with the standard normal vector to identify defects.

[0010] This invention first utilizes the distribution characteristics, orientation consistency factor, and flatness factor of the neighborhood data point set of a three-dimensional coordinate point to obtain the probability that the neighborhood data point set of each three-dimensional coordinate point belongs to the fin surface region. It accurately extracts the ideal fin surface points from the raw data containing mixed defects. Then, for these points, it innovatively constructs a dual adaptive radius adjustment method based on local density and quality confidence, ensuring that the final obtained standard normal vector has extremely high representativeness and accuracy. This effectively solves the problem of defect identification errors caused by incorrect standard normal vectors in the prior art.

[0011] Preferably, the acquisition of the distribution characteristics of the neighborhood data point set of each three-dimensional coordinate point in the point cloud data includes:

[0012] Preset basic neighborhood radius Taking the i-th 3D coordinate point in the point cloud data as the center, all 3D coordinate points within its basic neighborhood radius are taken as the neighborhood data point set of the i-th 3D coordinate point; principal component analysis is performed on the neighborhood data point set of the i-th 3D coordinate point to obtain the distribution variance of the neighborhood data point set of the i-th 3D coordinate point in the three principal directions and the three feature vectors of the neighborhood data point set of the i-th 3D coordinate point;

[0013] , The distribution characteristics of the neighborhood data point set representing the i-th three-dimensional coordinate point; The variance of the distribution of the neighborhood data point set of the i-th three-dimensional coordinate point in the second principal direction; The variance of the distribution of the neighborhood data points of the i-th 3D coordinate point in the first principal direction.

[0014] By analyzing the ratio of the variance of the main direction distribution, the coordinates of the regions that conform to the elongated shape of the fins can be effectively identified, thereby initially eliminating interference points on non-fin surfaces and improving the data purity in subsequent processing.

[0015] Preferably, obtaining the directional consistency factor includes:

[0016] ;

[0017] In the formula, The orientation consistency factor represents the set of neighboring data points of the i-th three-dimensional coordinate point; The first feature vector representing the neighborhood data point set of the i-th three-dimensional coordinate point; Represents the global fin's main extension direction vector; || represents the vector dot product operation; || represents the absolute value symbol.

[0018] By introducing global directional constraints, we can quickly eliminate outliers that, although their shapes conform to the characteristics of elongated strips, have severely deviated extension directions (such as severe collapse), ensuring that the selected points are consistent with the ideal fins in terms of attitude.

[0019] Preferably, obtaining the flatness factor includes:

[0020] , Flatness factor representing the set of neighboring data points of the i-th 3D coordinate point; The variance of the distribution of the neighborhood data points of the i-th three-dimensional coordinate point in the third principal direction; represents the variance of the distribution of the neighborhood data points of the i-th 3D coordinate point in the second principal direction; exp() represents an exponential function with the natural constant as the base.

[0021] Preferably, the step of obtaining the probability that the neighborhood data point set of each three-dimensional coordinate point belongs to the fin surface region includes:

[0022] ;

[0023] In the formula, The probability that the set of neighborhood data points of the i-th three-dimensional coordinate point belongs to the fin surface region; The orientation consistency factor represents the set of neighboring data points of the i-th three-dimensional coordinate point; The distribution characteristics of the neighborhood data point set representing the i-th three-dimensional coordinate point; The flatness factor represents the set of neighboring data points of the i-th three-dimensional coordinate point.

[0024] It can accurately quantify the confidence level that each point belongs to the ideal fin surface, laying the foundation for subsequent high-precision calculations.

[0025] Preferably, obtaining the coordinates of each fin surface based on the probability includes:

[0026] Preset probability threshold If the probability that the neighborhood data point set of any three-dimensional coordinate point lies in the ideal fin surface region is greater than or equal to the probability threshold. When this occurs, the three-dimensional coordinate point is recorded as the fin surface coordinate point.

[0027] Preferably, obtaining the density weight of each fin surface coordinate point includes:

[0028] The number of points belonging to the fin surface coordinates in the neighborhood data point set of the j-th fin surface coordinates is taken as the local density of the j-th fin surface coordinates.

[0029] ;

[0030] In the formula, The density weight represents the coordinate point on the surface of the j-th fin. This represents the local density at the coordinate point on the surface of the j-th fin; represents the mean of the local density at all coordinate points on the fin surface; exp() represents an exponential function with the natural constant as the base.

[0031] Preferably, obtaining the adaptive neighborhood radius of each fin surface coordinate point includes:

[0032] , The mass influence factor represents the coordinate point on the surface of the j-th fin. The probability that the set of neighborhood data points of the j-th fin surface coordinate point belongs to the fin surface region; Represents a preset probability threshold; This represents the preset adjustment coefficient;

[0033] , The adaptive neighborhood radius represents the coordinate point on the surface of the j-th fin. Represents the preset basic neighborhood radius; The density weight represents the coordinate point on the surface of the j-th fin. The mass influence factor represents the coordinate point on the surface of the j-th fin.

[0034] For fin points with just a passing confidence level, their neighborhood radius can be appropriately expanded by using a quality influence factor to introduce more neighborhood information to compensate for the uncertainty of their own characteristics, which significantly improves the robustness of the overall normal vector estimation.

[0035] Preferably, the step of comparing the normal vector of each three-dimensional coordinate point in the point cloud data with the standard normal vector to identify defects includes:

[0036] Preset angle deviation threshold The normal vector of each three-dimensional coordinate point in the point cloud data is obtained. The angle between the normal vector of each three-dimensional coordinate point and the standard normal vector is obtained as the abnormal angle of each three-dimensional coordinate point in the point cloud data. If the abnormal angle of any three-dimensional coordinate point in the point cloud data is greater than the angle deviation threshold, the three-dimensional coordinate point is regarded as the abnormal coordinate point. The area formed by all abnormal coordinate points is regarded as the defect area.

[0037] Secondly, this invention provides an automotive evaporator defect detection system, which adopts the following technical solution:

[0038] A vehicle evaporator defect detection system includes a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the aforementioned vehicle evaporator defect detection method is implemented.

[0039] By adopting the above technical solution, a computer program for detecting defects in automotive evaporators is generated and stored in a memory, so that it can be loaded and executed by a processor. A terminal device can then be made based on the memory and the processor for convenient use.

[0040] The present invention has the following technical effects: The present invention integrates the distribution, orientation, and flatness multi-dimensional features of the neighborhood data point set of three-dimensional coordinate points, accurately extracts the ideal fin surface points from the original data of mixed defects, and then innovatively constructs a dual adaptive neighborhood radius acquisition method based on local density and quality confidence for the ideal fin surface points, ensuring that the final obtained standard normal vector has extremely high representativeness and accuracy, thereby effectively solving the problem of defect identification errors caused by incorrect standard normal vectors in the prior art. Attached Figure Description

[0041] Figure 1 This is a flowchart of a method for detecting defects in an automotive evaporator according to an embodiment of the present invention. Detailed Implementation

[0042] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0043] This invention discloses a method for detecting defects in automotive evaporators, referring to... Figure 1 This includes steps S1-S4:

[0044] S1: Collect point cloud data of the surface of the evaporator fins of the car air conditioner.

[0045] In this embodiment of the invention, a line laser profile measurement sensor scans the surface of the automotive air conditioning evaporator. Specifically, the automotive air conditioning evaporator moves at a constant speed on a conveyor belt, and the sensor gradually collects the three-dimensional coordinate points of the evaporator fin surface, ultimately forming point cloud data. ,in, as well as This represents the first and second 3D coordinate points in the point cloud data. This represents the i-th 3D coordinate point in the point cloud data.

[0046] S2: Obtain the neighborhood data point set of each 3D coordinate point in the point cloud data. Perform principal component analysis on the neighborhood data point set of each 3D coordinate point to obtain three eigenvalues ​​and three eigenvectors. Obtain the distribution characteristics, orientation consistency factor, and flatness factor of the neighborhood data point set of each 3D coordinate point. Based on the distribution characteristics, orientation consistency factor, and flatness factor, obtain the probability that the neighborhood data point set of each 3D coordinate point belongs to the fin surface region. Based on the probability, obtain the fin surface coordinate point.

[0047] It should be noted that the purpose of this invention is to identify defective fins that have tilted, bent, collapsed, or scratched by calculating the normal vector of each three-dimensional coordinate point in the point cloud data and comparing it with the standard normal vector. Since there may be different types of defects on the fins of automotive air conditioning evaporators, the collected point cloud data contains defect points. Because directly using the mean of the normal vectors of all point cloud data as the standard vector is inaccurate, this invention combines the geometric features of the ideal fin to analyze the neighborhood data point set of each three-dimensional coordinate point, obtain the probability that the neighborhood data point set of each three-dimensional coordinate point belongs to the fin surface area, and then obtain the three-dimensional coordinate points that belong to the ideal fin area.

[0048] Specifically, this invention performs principal component analysis on the neighborhood data point set of each three-dimensional coordinate point in the point cloud data to obtain three eigenvalues ​​and three eigenvectors. It is known that the fins of an air conditioner evaporator are distributed in a long strip shape. Therefore, if the neighborhood data point set of any three-dimensional coordinate point satisfies the long strip shape distribution, then the distribution variance (first eigenvalue) of the neighborhood data point set of that three-dimensional coordinate point in the first principal direction is greater than the distribution variance (second eigenvalue) in the second principal direction. That is to say, the length of the fin surface area is greater than the width. Therefore, this invention obtains the distribution characteristics of the neighborhood data point set of that three-dimensional coordinate point based on the difference between the first eigenvalue and the second eigenvalue. This index can exclude non-fin areas that do not have a significant long strip shape distribution. For example, the collapse defect area is approximately circular. In the analysis, the first eigenvalue and the second eigenvalue of the neighborhood data point set of the three-dimensional coordinate point in this area are approximately the same.

[0049] In this embodiment of the invention, a preset basic neighborhood radius is defined. Taking the i-th 3D coordinate point in the point cloud data as the center, all 3D coordinate points within its basic neighborhood radius are considered as the neighborhood data set of the i-th 3D coordinate point. Principal component analysis is performed on the neighborhood data set of the i-th 3D coordinate point to obtain three eigenvalues ​​and three eigenvectors.

[0050] It should be noted that the three feature values ​​represent the distribution variance of the neighborhood data point set of the i-th three-dimensional coordinate point in the three principal directions, and , The variance of the distribution of the neighborhood data points of the i-th 3D coordinate point in the first principal direction; Let represent the variance of the distribution of the neighborhood data points of the i-th 3D coordinate point along the second principal direction. The variance of the distribution of the neighborhood data points of the i-th three-dimensional coordinate point in the third principal direction;

[0051] The three feature vectors represent the three main directions of the neighborhood data point set, reflecting the three main directions of the neighborhood data point set distribution.

[0052] In this embodiment of the invention, the distribution characteristics of the neighborhood data point set of the i-th three-dimensional coordinate point are obtained:

[0053]

[0054] In the formula, The distribution characteristics of the neighborhood data point set representing the i-th three-dimensional coordinate point; The variance of the distribution of the neighborhood data point set of the i-th three-dimensional coordinate point in the second principal direction; The variance of the distribution of the neighborhood data points of the i-th 3D coordinate point in the first principal direction;

[0055] It is known that the fins of an air conditioner evaporator are elongated and strip-shaped. Therefore, the distribution of the fins varies more significantly along the length direction, and less so along the height direction. This difference is reflected as eigenvalues ​​in principal component analysis. and The size difference, therefore if The smaller the ratio, the better. When the value approaches 1, it indicates that the distribution of the neighborhood data point set of the i-th three-dimensional coordinate point conforms to the characteristic of the elongated strip distribution of the fins of the air conditioner evaporator.

[0056] like The ratio approaches 1. When the value approaches 0, it indicates that the distribution of the neighborhood data point set of the i-th three-dimensional coordinate point does not conform to the characteristic of the elongated strip distribution of the fins of the air conditioner evaporator. The distribution of the neighborhood data point set of the i-th three-dimensional coordinate point may belong to the characteristics of circular collapse defects, etc.

[0057] It should be noted that for any three-dimensional coordinate point located on the surface of a normal fin, the main extension direction of its neighborhood data point set must be highly consistent with the global fin extension direction. Furthermore, the first eigenvector obtained by principal component analysis (i.e., the principal eigenvector of the neighborhood data point set) represents the main extension direction vector of the neighborhood data point set. Therefore, if the absolute value of the dot product between the first eigenvector of the neighborhood data point set of any three-dimensional coordinate point and the global fin main extension direction vector is close to 1, then the main extension direction is highly consistent with the global fin extension direction. Thus, based on the main extension direction vector and the global fin extension direction vector, the direction consistency factor of the neighborhood data point set of each three-dimensional coordinate point is obtained.

[0058] In this embodiment of the invention, the orientation consistency factor of the neighborhood data point set of the i-th three-dimensional coordinate point is obtained:

[0059]

[0060] In the formula, The orientation consistency factor represents the set of neighboring data points of the i-th three-dimensional coordinate point; The first feature vector representing the neighborhood data point set of the i-th three-dimensional coordinate point; Represents the global fin's main extension direction vector; || represents the vector dot product operation; || represents the absolute value symbol.

[0061] It should be noted that the first eigenvector of the neighborhood data point set of the i-th 3D coordinate point reflects the first principal direction of the neighborhood data point set of the i-th 3D coordinate point, i.e., the principal extension direction of the neighborhood data point set. Therefore, for any 3D coordinate point located on the surface of a normal fin, the principal extension direction of its neighborhood data point set must be parallel or antiparallel to the global fin extension direction. Thus, the absolute value of the dot product between the first eigenvector of the neighborhood data point set of the i-th 3D coordinate point and the global fin principal extension direction vector approaches 1. If there is a significant angle between the principal extension direction of the neighborhood data point set of the 3D coordinate point and the global fin principal extension direction, resulting in... When the value is significantly less than 1, it indicates that the set of neighboring data points at the three-dimensional coordinate point may belong to a tilted and curved defect region.

[0062] It should be noted that the variance of the distribution of the neighborhood data point set of each three-dimensional coordinate point in the third principal direction can reflect the thickness of the neighborhood data point set, and the variance in the distribution in the second principal direction can reflect the width of the neighborhood data point set. Therefore, if the variance of the distribution of the neighborhood data point set of any three-dimensional coordinate point in the third principal direction is smaller and the variance in the distribution in the second principal direction is larger, it indicates that the three-dimensional coordinate point is located on a plane and has a large width. The larger the flatness factor of the neighborhood data point set of the three-dimensional coordinate point, the more likely it is to belong to the fin surface area.

[0063] In this embodiment of the invention, the flatness factor of the neighborhood data point set of the i-th three-dimensional coordinate point is obtained:

[0064]

[0065] In the formula, Flatness factor representing the set of neighboring data points of the i-th 3D coordinate point; The variance of the distribution of the neighborhood data points of the i-th three-dimensional coordinate point in the third principal direction; represents the variance of the distribution of the neighborhood data points of the i-th 3D coordinate point in the second principal direction; exp() represents an exponential function with the natural constant as the base.

[0066] The value approaches 0. The larger the value of , the smaller the variance of the distribution of the neighborhood data points of the i-th 3D coordinate point in the third principal direction and the larger the variance in the distribution of the distribution in the second principal direction. Therefore, the neighborhood data points of the i-th 3D coordinate point are more likely to belong to the fin surface region; conversely, When the value is large, The smaller the value, the smaller the distribution variance of the neighborhood data point set of the three-dimensional coordinate point in the third principal direction and the smaller the distribution variance in the second principal direction. This indicates that the neighborhood data point set of the three-dimensional coordinate point may have linear defect characteristics. Therefore, the flatness factor can eliminate some linear defects.

[0067] The probability that the neighborhood data point set of the i-th 3D coordinate point belongs to the fin surface region:

[0068]

[0069] In the formula, The probability that the set of neighborhood data points of the i-th three-dimensional coordinate point belongs to the fin surface region; The orientation consistency factor represents the set of neighboring data points of the i-th three-dimensional coordinate point; The distribution characteristics of the neighborhood data point set representing the i-th three-dimensional coordinate point; Let represent the flatness factor of the neighborhood data point set of the i-th 3D coordinate point; if the neighborhood data point set of any 3D coordinate point simultaneously satisfies the following conditions: having a long strip shape, high consistency with the global fin extension direction vector, and located in a flat planar region, then... The larger the value, the greater the probability that the neighborhood data point set of the i-th three-dimensional coordinate point belongs to the fin surface region.

[0070] Preset probability threshold In other embodiments, implementers can preset a probability threshold according to specific implementation methods. If the probability that the neighborhood data point set of any three-dimensional coordinate point is located in the ideal fin surface area is greater than or equal to the probability threshold, then... When this point is reached, the three-dimensional coordinate point is recorded as the fin surface coordinate point.

[0071] S3: Based on the local density distribution of coordinate points on the fin surface, obtain the density weight of each coordinate point on the fin surface; based on the difference between the probability that each coordinate point on the fin surface belongs to the fin surface region and the probability threshold, obtain the quality influence factor of each coordinate point on the fin surface; based on the density weight and the quality influence factor, obtain the adaptive neighborhood radius of each coordinate point on the fin surface.

[0072] It should be noted that coordinate points on the fin surface are more likely to belong to the fin surface, so it is necessary to estimate their normal vectors to obtain standard normal vectors. However, when performing normal vector estimation, the neighborhood radius of the principal component analysis algorithm is usually set to a fixed empirical value. If a small neighborhood radius is set, the normal vector calculated by the algorithm will be extremely unstable when faced with sparse point clouds caused by uneven scanning angles. Conversely, if the neighborhood radius is increased to ensure computational stability, the neighborhood radius may easily cross multiple independent fin surface boundaries, causing the principal component analysis to calculate completely incorrect normal vectors. Therefore, both excessively large and excessively small neighborhood radii will cause errors in normal vector estimation, resulting in inaccurate standard normal vectors and subsequent incorrect defect identification.

[0073] Therefore, point cloud density affects the stability of normal vectors. In sparse point cloud regions, a larger neighborhood radius is needed to include enough 3D coordinate points to ensure the effectiveness of normal vector calculation. In dense point cloud regions, the basic neighborhood radius is appropriately reduced to reduce interference from distant points, ensuring that the points within the neighborhood are more closely connected and more relevant, and the subsequent normal vector calculation is more accurate. Therefore, this step constructs a density weight based on the local point cloud density information of the 3D coordinate points in the point cloud concentration on the fin surface.

[0074] In this embodiment of the invention, the number of fin surface coordinate points in the neighborhood data point set of the j-th fin surface coordinate point is taken as the local density of the j-th fin surface coordinate point.

[0075] Obtain the density weight of the j-th fin surface coordinate point:

[0076]

[0077] In the formula, The density weight represents the coordinate point on the surface of the j-th fin. This represents the local density at the coordinate point on the surface of the j-th fin; The mean of the local density at all coordinate points on the fin surface; exp() represents an exponential function with the natural constant as the base.

[0078] When the value approaches 1, A value of 1 indicates that the density around the j-th fin surface coordinate point is appropriate and has no effect on adjusting the basic neighborhood radius.

[0079] like When the local density of the j-th fin surface coordinate point is less than the average local density of all fin surface coordinate points, it indicates that the number of fin surface coordinate points surrounding the j-th fin surface coordinate point is relatively small. If the value is positive and greater than 1, the radius of the basic neighborhood needs to be increased.

[0080] like When the density of the j-th fin surface coordinate point is greater than the average of the local densities of all fin surface coordinate points, it indicates that the number of fin surface coordinate points surrounding the j-th fin surface coordinate point is relatively large. If the value is negative and less than 1, the radius of the basic neighborhood needs to be reduced to decrease interference from points that are too far away, ensuring that the points in the neighborhood are closer and more related, and that the normal vectors calculated later are more accurate.

[0081] It should be noted that when the probability that the neighborhood data point set of any fin surface coordinate point belongs to the fin surface region just exceeds the preset probability threshold, the fin surface coordinate point may belong to the fin surface region, but its quality is not high. Therefore, its neighborhood radius can be appropriately expanded to include more point information. This helps to estimate the normal vector more stably. Especially in this uncertain case, the uncertainty can be compensated by increasing the density of neighborhood points to ensure that the calculation results are more reliable.

[0082] In this embodiment of the invention, the quality influence factor of the j-th fin surface coordinate point is obtained:

[0083]

[0084] In the formula, The mass influence factor represents the coordinate point on the surface of the j-th fin. The probability that the set of neighborhood data points of the j-th fin surface coordinate point belongs to the fin surface region; Represents a preset probability threshold; Representing a preset adjustment coefficient, in this embodiment of the invention, the preset... In other embodiments, implementers may pre-determine specific implementation methods. The value;

[0085] When it approaches 0, at this time The value of is close to 0, indicating that the j-th fin surface coordinate point is more likely to belong to the fin surface region, therefore its quality influence factor is... The value approaches 1, indicating that when adjusting the neighborhood radius of the j-th fin surface coordinate point, only the local density around the j-th fin surface coordinate point is considered.

[0086] like The value just exceeds the preset probability threshold. At this time, the coordinate point on the surface of the j-th fin may belong to the fin surface region, but its quality is not high. The value approaches 1. The value tends to This allows for a more appropriate increase in the neighborhood radius of the j-th fin surface coordinate point to include more three-dimensional coordinate points and thus estimate a more stable normal vector.

[0087] It should be noted that the basic neighborhood radius is adjusted based on the density weight of the coordinate points on the fin surface and the quality influence factor. The larger the density weight, the larger the neighborhood radius is adjusted, and the smaller the density weight, the smaller the neighborhood radius is adjusted. In addition, the neighborhood radius of the three-dimensional coordinate points with a quality influence factor greater than 1 is appropriately increased.

[0088] In this embodiment of the invention, the adaptive neighborhood radius of the j-th fin surface coordinate point is obtained:

[0089]

[0090] In the formula, The adaptive neighborhood radius represents the coordinate point on the surface of the j-th fin. Represents the preset basic neighborhood radius; The density weight represents the coordinate point on the surface of the j-th fin. The mass influence factor represents the coordinate point on the surface of the j-th fin.

[0091] When the value is greater than 1, it indicates that the density distribution around the coordinate point on the surface of the j-th fin is sparse. Increasing the value of makes the neighborhood radius contain more three-dimensional coordinate points, resulting in higher accuracy of the subsequently calculated normal vector;

[0092] When the value is less than 1, it indicates that the density distribution around the coordinate point on the surface of the j-th fin is dense, which is beneficial to... By reducing the value of , while ensuring the stability of the calculated normal vector, it is possible to capture the subtle deformation of the fins, making it easier to detect subtle deformation defects in the future.

[0093] Fine-tuning is performed based on the quality influence factor of the j-th fin surface coordinate point. For three-dimensional coordinate points with a quality influence factor greater than 1, their neighborhood radius is adjusted to be larger, so as to include more and farther three-dimensional coordinate points and improve the accuracy of the calculated normal vector.

[0094] S4: Based on the adaptive neighborhood radius of each fin surface coordinate point, obtain the normal vector of each fin surface coordinate point, and then obtain the standard normal vector; determine the angle between the normal vector of each three-dimensional coordinate point in the point cloud data and the standard normal vector to obtain abnormal coordinate points.

[0095] In this embodiment of the invention, the adaptive neighborhood radius of each fin surface coordinate point is input into the principal component analysis algorithm to calculate the normal vector of each fin surface coordinate point, and the mean of the normal vectors of all fin surface coordinate points is used as the standard normal vector.

[0096] Angle deviation threshold In other embodiments of the invention, the implementer can preset an angle deviation threshold according to the actual implementation situation; obtain the normal vector of each three-dimensional coordinate point in the point cloud data, obtain the angle between the normal vector of each three-dimensional coordinate point and the standard normal vector, and use it as the abnormal angle of each three-dimensional coordinate point in the point cloud data. If the abnormal angle of any three-dimensional coordinate point in the point cloud data is greater than the angle deviation threshold, the three-dimensional coordinate point is regarded as an abnormal coordinate point, and the area formed by all abnormal coordinate points is regarded as a defect area.

[0097] This invention also discloses an automotive air conditioning evaporator defect detection system, including a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement the automotive air conditioning evaporator defect detection method provided by this invention.

[0098] The system also includes other components well-known to those skilled in the art, such as communication buses and communication interfaces, the setup and functions of which are known in the art and will not be described in detail here. In this invention, the aforementioned memory can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0099] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method of detecting a defect of an evaporator of an automobile, characterized by, The method comprises: acquiring point cloud data on the surface of the fins of an automobile air conditioner evaporator; based on the geometric features of the neighborhood data point set of each three-dimensional coordinate point in the point cloud data, acquiring the distribution feature, the direction consistency factor and the planeness factor of the neighborhood data point set of each three-dimensional coordinate point in the point cloud data; acquiring the possibility that each three-dimensional coordinate point belongs to the fin surface region, which is positively correlated with the distribution feature, the direction consistency factor and the planeness factor; based on the possibility, acquiring each fin surface coordinate point; based on the local density of the fin surface coordinate point, acquiring the density weight of each fin surface coordinate point; acquiring the adaptive neighborhood radius of each fin surface coordinate point, which is positively correlated with the density weight; inputting the adaptive neighborhood radius into a principal component analysis algorithm to acquire the normal vector of each fin surface coordinate point; taking the mean value of the normal vectors of all fin surface coordinate points as a standard normal vector; comparing the normal vector of each three-dimensional coordinate point in the point cloud data with the standard normal vector to identify defects.

2. The method of claim 1, wherein, The acquisition of the distribution feature of the neighborhood data point set of each three-dimensional coordinate point in the point cloud data comprises: Pre-set base neighborhood radius all three-dimensional coordinate points within the base neighborhood radius of the i-th three-dimensional coordinate point in the point cloud data as the neighborhood data point set of the i-th three-dimensional coordinate point; performing principal component analysis on the neighborhood data point set of the i-th three-dimensional coordinate point to obtain distribution variances of the neighborhood data point set of the i-th three-dimensional coordinate point in three principal directions and three feature vectors of the neighborhood data point set of the i-th three-dimensional coordinate point; , distribution feature of the neighbor data point set representing the i-th three-dimensional coordinate point; distribution variance of the neighbor data point set representing the i-th three-dimensional coordinate point in the second principal direction; distribution variance of the neighbor data point set representing the i-th three-dimensional coordinate point in the first principal direction.

3. The method of claim 1 or 2, wherein The acquisition of the direction consistency factor comprises: ； wherein, a direction consistency factor of the neighborhood data point set representing the i-th three-dimensional coordinate point; a first eigenvector of the neighborhood data point set representing the i-th three-dimensional coordinate point; a global fin main extension direction vector; represents a vector dot product operation; || represents an absolute value symbol.

4. The method of claim 1 or 2, wherein The acquisition of the planeness factor comprises: a planarity factor of the neighborhood data point set representing the i-th three-dimensional coordinate point; a distribution variance of the neighborhood data point set representing the i-th three-dimensional coordinate point in the third principal direction; a distribution variance of the neighborhood data point set representing the i-th three-dimensional coordinate point in the second principal direction; exp() represents an exponential function with a natural constant as a base number.​ 5. The method of claim 1, wherein the method further comprises: The acquisition of the possibility that each three-dimensional coordinate point belongs to the fin surface region comprises: ； wherein a likelihood that the neighborhood data point set representing the i-th three-dimensional coordinate point belongs to the fin surface region; a direction consistency factor of the neighborhood data point set representing the i-th three-dimensional coordinate point; a distribution feature of the neighborhood data point set representing the i-th three-dimensional coordinate point; a planarity factor of the neighborhood data point set representing the i-th three-dimensional coordinate point.

6. The method of claim 1, wherein, The acquisition of each fin surface coordinate point based on the possibility comprises: preset possibility threshold , if the possibility that the neighborhood data point set of any three-dimensional coordinate point is located in the ideal fin surface region is greater than or equal to the possibility threshold , the three-dimensional coordinate point is recorded as a fin surface coordinate point.

7. The method of claim 1, wherein the method further comprises: The acquisition of the density weight of each fin surface coordinate point comprises: taking the number of fin surface coordinate points in the neighborhood data point set of the jth fin surface coordinate point as the local density of the jth fin surface coordinate point; ； wherein a density weight representing the jth fin surface coordinate point; a local density representing the jth fin surface coordinate point; a mean value of the local densities representing all fin surface coordinate points; exp() represents an exponential function with the natural constant as the base number.

8. The method of claim 1, wherein, The acquisition of the adaptive neighborhood radius of each fin surface coordinate point comprises: , a mass influence factor representing a jth fin surface coordinate point; a likelihood that a neighborhood data point set representing the jth fin surface coordinate point belongs to a fin surface region; a preset likelihood threshold value; a preset adjustment coefficient; , adaptive neighborhood radius representing the jth fin surface coordinate point; preset base neighborhood radius representing; density weight representing the jth fin surface coordinate point; quality influence factor representing the jth fin surface coordinate point.

9. The method of claim 1, wherein, The comparison of the normal vector of each three-dimensional coordinate point in the point cloud data with the standard normal vector to identify defects comprises: Pre-set angle deviation threshold , a normal vector of each three-dimensional coordinate point in the point cloud data is obtained, an included angle between the normal vector of each three-dimensional coordinate point and a standard normal vector is obtained as an abnormal angle of each three-dimensional coordinate point in the point cloud data, if the abnormal angle of any three-dimensional coordinate point in the point cloud data is greater than the angle deviation threshold, the three-dimensional coordinate point is taken as an abnormal coordinate point, and a region formed by all the abnormal coordinate points is taken as a defect region.

10. An automotive evaporator defect detection system, comprising: The method comprises: a processor and a memory, wherein the memory stores computer program instructions which, when executed by the processor, implement the automobile evaporator defect detection method according to any one of claims 1-9.

Images