IGBT bonding wire 3D point cloud adaptive filtering and arc height surface extraction method

Abstract

Disclosed is an IGBT bonding wire 3D point cloud adaptive filtering and arc height surface extraction method, comprising: acquiring an image to be detected; performing local discrete denoising processing on the image to be detected, and acquiring a processed image to be detected; calculating a grayscale histogram of the processed image to be detected, and acquiring an IGBT bonding wire grayscale histogram smooth curve; performing second-order derivation on the IGBT bonding wire grayscale histogram smooth curve, and acquiring a dense noise region and a grayscale distribution range of an arc height surface; and on the basis of the dense noise region and the grayscale distribution range of the arc height surface, removing the dense noise region, and acquiring a bonding wire arc height region. In the present invention, acquired 3D point cloud information of the IGBT bonding wire can be adaptively filtered, effectively replacing a large amount of work for manually setting parameters. After filtering, key information of a target bonding wire can be successfully extracted, detection work on the bonding wire is accurately and efficiently completed, and the overall working efficiency and reliability of a system are effectively improved.

Description

Adaptive filtering and arc height surface extraction method for IGBT wire 3D point cloud TECHNICAL FIELD

[0001] The application belongs to the field of IGBT detection, and particularly relates to an adaptive filtering and arc height surface extraction method for IGBT wire 3D point cloud. BACKGROUND

[0002] IGBT (Insulated Gate Bipolar Transistor) is a key component widely used in power electronic devices, and its performance and quality directly affect the reliability, efficiency, safety and service life of the entire system. AOI (Automated Optical Inspection) uses high-resolution cameras and image processing algorithms to quickly scan the surface of the IGBT module, accurately locate the wire, solder joints and other components, automatically identify and classify defects, and significantly improve production efficiency. Among them, the current and signal-dependent wire transmits between different components, and the wire also ensures the physical stability of the components, so it is necessary to use AOI to detect the wire.

[0003] When processing the 3D point cloud data of the wire based on AOI, in order to optimize the quality and accuracy of the data and improve the processing efficiency and stability of the algorithm, point cloud filtering is needed to eliminate noise and outliers. Common point cloud filtering methods include voxel filter, radius filter, statistical filter, straight-through filter, bilateral filter, Gaussian filter, median filter, etc. Generally, due to the sparsity and irregularity of point cloud data and the complexity of noise, point cloud filtering generally needs to manually set a large number of parameters. Therefore, it is very important to develop an adaptive point cloud filtering method that can dynamically adjust the filtering parameters according to the local characteristics of the point cloud data and achieve accurate denoising.

[0004] Defects in the prior art: using AOI to detect the wire, it is necessary to identify and classify defects such as offset and fracture, and accurately calculate data such as arc height and line width. Traditional 2D image algorithms cannot obtain height information, and are easily disturbed by the background when detecting defects, making it difficult to effectively identify the wire. Through 3D point cloud information, the height can be calculated, but noise and data redundancy problems will be caused, which requires a large amount of parameter setting work to reduce the amount of calculation, making it difficult to realize high-speed automatic detection, and the overall generality and reliability of the system will be further reduced. SUMMARY

[0005] To solve the above technical problems, the application provides an adaptive filtering and arc height surface extraction method for IGBT wire 3D point cloud, which can perform adaptive filtering on the acquired 3D point cloud information of the IGBT wire, effectively replacing the large amount of work of manually setting parameters.

[0006] To achieve the above object, the application provides an IGBT wire 3D point cloud adaptive filtering and arc height surface extraction method, comprising:

[0007] An image of an IGBT module to be detected is acquired.

[0008] The image of the IGBT module to be detected is subjected to local discrete noise denoising processing to obtain a processed image to be detected.

[0009] A gray histogram of the processed image to be detected is calculated to obtain an IGBT module wire gray histogram smooth curve.

[0010] The IGBT module wire gray histogram smooth curve is subjected to second-order derivation, and a gray distribution range of a dense noise region and an arc height surface is obtained.

[0011] Based on the gray distribution range of the dense noise region and the arc height surface, dense noise point clouds are removed, and a wire arc height region is obtained.

[0012] Optionally, the image of the IGBT module to be detected is acquired by:

[0013] A two-dimensional height image of an IGBT module is acquired, and the two-dimensional height image is preprocessed to obtain the image of the IGBT module to be detected.

[0014] Optionally, the two-dimensional height image of the IGBT module is acquired by:

[0015] Point cloud data of the IGBT module under a single field of view is acquired.

[0016] The point cloud data under the single field of view is spliced to obtain complete point cloud data of the IGBT module.

[0017] The complete point cloud data of the IGBT module is converted into a two-dimensional height image.

[0018] Optionally, the two-dimensional height image is preprocessed to obtain the image of the IGBT module to be detected by:

[0019] A gray value of the two-dimensional height image is acquired, and the gray value is subtracted by a minimum gray value to obtain a first processed image.

[0020] The first processed image is subjected to linear stretching to obtain a second processed image, i.e., the image of the IGBT module to be detected.

[0021] Optionally, the gray histogram of the processed image to be detected is calculated to obtain the IGBT module wire gray histogram smooth curve by:

[0022] The gray histogram of the processed image to be detected is calculated.

[0023] Based on the gray histogram, an IGBT module wire bonding height histogram curve is obtained;

[0024] The IGBT module wire bonding height histogram curve is filtered by using a Gaussian function to obtain an IGBT module wire bonding gray histogram smooth curve.

[0025] Optionally, the IGBT module wire bonding gray histogram smooth curve is second-order differentiated to obtain a gray distribution range of the dense noise region and the arc height surface, including:

[0026] The IGBT module wire bonding gray histogram smooth curve is second-order differentiated to obtain a second-order derivative curve;

[0027] The zero-crossing points of the second-order derivative curve are calculated, and the range corresponding to the zero-crossing points is the gray distribution range of the dense noise region and the arc height surface.

[0028] Optionally, based on the gray distribution range of the dense noise region and the arc height surface, the method for obtaining the wire bonding arc height region is:

[0029] Wherein, p1 and p2 are zero-crossing points corresponding to the arc height region, I(x) represents a height map image, and g(x) is a wire bonding arc height region.

[0030] Compared with the prior art, the present application has the following advantages and technical effects:

[0031] 1. The present application accurately extracts the arc height surface of the IGBT wire by introducing 3D point cloud information, accurately calculates the relevant height data, and breaks through the limitations of 2D detection lacking height information. Compared with 2D detection, the present application is not easily disturbed by the background, and can better and more accurately detect defects such as broken wire and offset.

[0032] 2. The present application designs an automatic filtering process for 3D point cloud, which has many advantages compared with manual parameter setting filtering: (1) The adaptive filtering has a high degree of automation, can automatically adjust the filtering parameters according to the input data, reduces the need for manual parameter adjustment, makes the processing process more simple and efficient, and is more suitable for high-speed automatic detection. (2) Adaptive filtering can dynamically adjust the filtering parameters according to the local features of the point cloud data, more accurately retain valuable information and key features, while eliminating noise and errors, improving the overall quality and precision of the point cloud data. (3) Adaptive filtering is more efficient than fixed parameter filtering when processing large-scale point cloud data, and dynamic adjustment of parameters makes the robustness better, so that consistent filtering effect can be obtained in various complex environments.

[0033] 3.The application can perform adaptive filtering on the acquired 3D point cloud information of the IGBT bonding wire, effectively replacing the large amount of work of manually setting parameters. After filtering, the key information of the target bonding wire can be successfully extracted, the detection of the bonding wire can be accurately and efficiently completed, and the overall work efficiency and reliability of the system can be effectively improved. BRIEF DESCRIPTION OF DRAWINGS

[0034] The accompanying drawings, which form a part of this application, are intended to provide further understanding of the application and are incorporated herein in their entirety. The schematic embodiments of the application and their descriptions are used to explain the application and do not constitute an improper limitation on the application. In the drawings:

[0035] Fig. 1 is an IGBT bonding wire height map according to an embodiment of the application;

[0036] Fig. 2 is a local maximum discrete point filtering diagram of an IGBT bonding wire according to an embodiment of the application;

[0037] Fig. 3 is a local minimum discrete point filtering diagram of an IGBT bonding wire according to an embodiment of the application;

[0038] Fig. 4 is a local maximum and minimum value filtering diagram of an IGBT bonding wire according to an embodiment of the application;

[0039] Fig. 5 is an IGBT bonding wire gray scale histogram statistical diagram according to an embodiment of the application;

[0040] Fig. 6 is a Gaussian smoothing diagram of an IGBT bonding wire gray scale histogram according to an embodiment of the application;

[0041] Fig. 7 is a second derivative curve diagram of an IGBT bonding wire gray scale histogram according to an embodiment of the application;

[0042] Fig. 8 is an IGBT bonding wire arc height surface extraction result diagram according to an embodiment of the application;

[0043] Fig. 9 is a flowchart of an IGBT bonding wire 3D point cloud adaptive filtering and arc height surface extraction method according to an embodiment of the application. DETAILED DESCRIPTION

[0044] It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0045] It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a group of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described herein can be executed in an order different from that shown herein.

[0046] The present application proposes a knowledge tracking method based on multi-task enhancement, as shown in Fig. 9, which specifically includes the following steps:

[0047] acquire an image of the IGBT module to be detected;

[0048] perform local discrete noise denoising processing on the image of the IGBT module to be detected, and acquire a processed image to be detected;

[0049] calculate a gray histogram of the processed image to be detected, and acquire a smooth curve of an IGBT module wire gray histogram;

[0050] perform second-order derivation on the smooth curve of the IGBT module wire gray histogram, and acquire a dense noise region and an arc height surface gray distribution range;

[0051] remove the dense noise region based on the dense noise region and the arc height surface gray distribution range, and acquire a wire arc height region.

[0052] Specifically, the IGBT wire 3D point cloud information acquired by the application can be adaptively filtered, and a large amount of work of manually setting parameters can be effectively replaced. After filtering, the key information of the target wire can be successfully extracted, the detection of the wire can be accurately and efficiently completed, and the overall work efficiency and reliability of the system can be effectively improved.

[0053] Further, acquiring the image of the IGBT module to be detected comprises:

[0054] acquiring a two-dimensional height image of the IGBT module, and pre-processing the two-dimensional height image to acquire the image of the IGBT module to be detected.

[0055] Further, acquiring the two-dimensional height image of the IGBT module comprises:

[0056] using a 3D camera to shoot the IGBT module to be detected to acquire point cloud data;

[0057] splicing the point cloud data under a single field of view to acquire complete point cloud data of the IGBT module;

[0058] converting the complete point cloud data of the IGBT module into a two-dimensional height image.

[0059] Specifically, the 3D camera is used to shoot the IGBT module to be detected to acquire point cloud data, the point cloud data under a single field of view is spliced to obtain complete point cloud data of the IGBT module, and then the point cloud data is converted into a two-dimensional height image, as shown in FIG. 1.

[0060] The single field of view means that the camera has a single shooting field of view. Since the target size is large, the camera cannot be shot at one time, and needs to be moved and shot multiple times.

[0061] The method for converting 3D point cloud data into a 2D height map is: taking the height Z information of the 3D point cloud as a gray value to assign to the pixel position of the corresponding 2D image, as the brightness value of the 2D image.

[0062] Further, the 2D height image is preprocessed, and the image to be detected is obtained, including:

[0063] The gray value of the 2D height image is obtained, and the gray value is subtracted from the minimum gray value to obtain a first processed image.

[0064] The first processed image is linearly stretched to obtain a second processed image.

[0065] The second processed image is the image to be detected.

[0066] Specifically, the minimum gray value of the complete height image is obtained, and the gray value of all pixel points is subtracted from the minimum gray value, and the minimum gray value of the updated image is 0.

[0067] Next, the image of the previous step is multiplied by 1000 to convert um, and at the same time, the height image is linearly stretched once to make the gray histogram of the height image contain more detailed information.

[0068] Specifically, the local discrete noise is filtered out, the image is windowed and slid, the local maximum and minimum points in the neighborhood are removed, and the neighborhood window can be selected as 4-neighborhood and 8-neighborhood to remove local discrete points. The filtering effect is shown in Figures 2-4, where the black color is the filtered discrete points.

[0069] Further, the gray histogram of the processed image to be detected is calculated, and the IGBT wire height gray histogram smooth curve is obtained, including:

[0070] The gray histogram of the processed image to be detected is calculated.

[0071] Based on the gray histogram, the IGBT wire height gray histogram curve is obtained.

[0072] The IGBT wire height gray histogram curve is filtered using a Gaussian function to obtain the IGBT wire gray histogram smooth curve.

[0073] Specifically, the maximum and minimum gray values of the image to be detected are obtained, the minimum gray value is taken as the starting value, and the maximum gray value is taken as the end value, and the mathematical expression is shown in formula (1), wherein I(x) The height map image is represented. Then a one-dimensional array is constructed with a step size of 10 as a lookup table LUT for looking up the corresponding height value at the corresponding position. And the length of the lookup table is used as the number of cells of the histogram.

[0074] The look-up table LUT can be expressed as formula 2, where step represents a step length, and is set as 10 by default: LUT = [MinGray:step:MaxGray] (2)

[0075] Based on the above, the gray scale histogram is constructed to obtain the data distribution of the IGBT wire bonding height map, and the gray scale histogram is shown in FIG. 5. Since the histogram curve has some fluctuation burr areas, a Gaussian function is used to perform sliding filtering on the curve to obtain a smooth curve, which is convenient for function differentiation in the following, as shown in FIG. 6.

[0076] Further, the IGBT wire bonding gray scale histogram smooth curve is second-order differentiated to obtain the gray scale distribution range of the dense noise area and the arc height surface, including:

[0077] The IGBT wire bonding gray scale histogram smooth curve is second-order differentiated to obtain the second-order derivative curve.

[0078] The zero-crossing points of the second-order derivative curve are calculated, and the range corresponding to the zero-crossing points is the gray scale distribution range of the dense noise area and the arc height surface.

[0079] Specifically, based on the obtained IGBT wire bonding gray scale histogram smooth curve, the second-order derivative curve thereof is obtained by second-order differentiation, the zero-crossing points thereof are calculated, and the range corresponding to the zero-crossing points is the obtained gray scale distribution range of the dense noise area and the wire bonding arc height. Since the numerical distribution of the height gray scale value of the arc height is concentrated on the right side of the histogram, the rightmost group of zero-crossing points is selected as the wire bonding arc height extraction range. The remaining zero-crossing points are removed as dense noise. The second-order derivative curve of the IGBT wire bonding gray scale histogram is shown in FIG. 7.

[0080] The calculation method of the zero-crossing point: for the second-order derivative, the calculation method of the zero-crossing point is to first find the point on the x coordinate axis whose function value is 0, and then calculate the positive and negative of the function values on the left and right sides of the coordinate point. If one is positive and the other is negative on the left and right sides of the point, the point is a zero-crossing point.

[0081] After obtaining the rightmost group of zero-crossing points [p1, p2], the corresponding gray scale value of the height map is obtained as the final extraction range of the wire bonding arc height [LUT[p1 , LUT[p2]] through the look-up table established by formula 2.

[0082] Further, based on the arc height surface distribution range, the method for obtaining the wire bonding arc height region is:

[0083] Wherein, p1 and p2 are both zero-crossing points, I(x) represents a height map image, and g(x) is a wire bonding arc height region.

[0084] Specifically, the arc height surface extraction result image is shown in FIG. 8.

[0085] The above merely provides the preferred embodiments of the application, and the protection scope of the application is not limited thereto, and any modification or replacement within the technical scope disclosed by the application should be covered within the protection scope of the application. Therefore, the protection scope of the application should be subject to the protection scope of the claims.

Claims

1. An IGBT wire bonding 3D point cloud adaptive filtering and camber surface extraction method, characterized in that, The method comprises the following steps: acquiring an image of an IGBT module to be detected; performing local discrete noise denoising processing on the image of the IGBT module to be detected to acquire a processed image of the IGBT module to be detected; calculating a gray histogram of the processed image of the IGBT module to be detected to acquire a smooth curve of a gray histogram of a solder wire of the IGBT module; performing second-order derivation on the smooth curve of the gray histogram of the solder wire of the IGBT module to acquire a dense noise region and a gray distribution range of an arc height surface; based on the dense noise region and the gray distribution range of the arc height surface, removing the dense noise region and acquiring a solder wire arc height region.

2. The IGBT wire bonding 3D point cloud adaptive filtering and camber surface extraction method according to claim 1, characterized in that, The acquiring of the image of the IGBT module to be detected comprises the following steps: acquiring a two-dimensional height image of the IGBT module, and performing preprocessing on the two-dimensional height image to acquire the image of the IGBT module to be detected.

3. The IGBT wire bonding 3D point cloud adaptive filtering and camber surface extraction method according to claim 2, characterized in that, The acquiring of the two-dimensional height image of the IGBT module comprises the following steps: acquiring point cloud data of the IGBT module under a single field of view; splicing the point cloud data under the single field of view to acquire complete point cloud data of the IGBT module; converting the complete point cloud data of the IGBT module into a two-dimensional height image.

4. The IGBT wire bonding 3D point cloud adaptive filtering and camber surface extraction method according to claim 2, characterized in that, The preprocessing of the two-dimensional height image to acquire the image of the IGBT module to be detected comprises the following steps: acquiring a gray value of the two-dimensional height image, subtracting a minimum gray value from the gray value to acquire a first processed image; performing linear stretching on the first processed image to acquire a second processed image, i.e., the image of the IGBT module to be detected.

5. The IGBT wire bonding 3D point cloud adaptive filtering and camber surface extraction method according to claim 1, characterized in that, The calculation of the gray histogram of the processed image of the IGBT module to be detected to acquire the smooth curve of the gray histogram of the solder wire of the IGBT module comprises the following steps: calculating the gray histogram of the processed image of the IGBT module to be detected; based on the gray histogram, acquiring a solder wire height histogram curve of the IGBT module; performing sliding filtering on the solder wire height histogram curve of the IGBT module by using a Gaussian function to acquire the smooth curve of the gray histogram of the solder wire of the IGBT module.

6. The IGBT wire bonding 3D point cloud adaptive filtering and camber surface extraction method of claim 1, wherein, The second-order derivation on the smooth curve of the gray histogram of the solder wire of the IGBT module to acquire the gray distribution range of the dense noise region and the arc height surface comprises the following steps: performing second-order derivation on the smooth curve of the gray histogram of the solder wire of the IGBT module to acquire a second-order derivative curve; calculating zero-crossing points of the second-order derivative curve, and the range corresponding to the zero-crossing points is the gray distribution range of the dense noise region and the arc height surface.

7. The IGBT wire bonding 3D point cloud adaptive filtering and camber surface extraction method according to claim 1, characterized in that, Based on the gray scale distribution range of the arc height surface, the method for obtaining the welding wire arc height region is: wherein p1 and p2 are both zero-crossing points, I(x) represents a height map image, and g(x) is a solder wire arc height region.

Images