Wave soldering quality detection method and system based on laser fly scanning

Abstract

The invention relates to the field of wave-soldering detection, in particular to a wave-soldering quality detection method and system based on laser fly scanning, and the method comprises the steps: carrying out the laser fly scanning of a PCB before welding, and obtaining a laser fly scanning three-dimensional point cloud image before welding; performing laser fly scanning on the welded PCB to obtain a laser fly scanning three-dimensional point cloud image after welding; performing comparison processing to obtain laser fly scanning difference three-dimensional point cloud images before and after welding; selecting a threshold value; and constructing a deep belief network, wherein the deep belief network obtains a wave soldering quality detection result according to the threshold value and the difference three-dimensional image. According to the method, laser flying scanning is carried out on the PCB before and after welding, and the laser flying scanning three-dimensional point cloud image before and after welding is obtained.

Description

Technical Field

[0001] The present invention relates to the field of wave soldering detection, and more specifically, to a wave soldering quality detection method and system based on laser flying scanning. Background Art

[0002] In the field of wave soldering technology, melted soft solder is sprayed into a solder wave by a mechanical pump or electromagnetic pump, and a PCB board pre-installed with electronic components is passed through the solder wave, so that the solder end or pin of the electronic component can be mechanically and electrically connected to the PCB pad. As electronic components become smaller and smaller and electrical PCB boards become denser, the wave soldering process often produces various defects, such as insufficient filling, bridging, cold soldering or leaking soldering, which in turn lead to poor connection or short circuit between electronic components and PCB soldering boards. Therefore, it is very important to conduct post-weld inspection of product quality for possible quality problems in wave soldering. At present, 2D image detectors are often used for plane inspection of wave soldering solder joints in wave soldering quality inspection and analysis. However, 2D image inspection technology is easily affected by background color and ambient light, leading to misjudgment. In addition, 2D image inspection technology requires cumbersome lighting parameter adjustment. By introducing laser flying scanning inspection technology, it can be free from the influence of product status and environmental factors, and stable inspection can be achieved.

[0003] The prior art is a wave soldering quality detection method based on feature clustering and ensemble learning, which takes the preprocessed inverter wave soldering quality picture as the data basis; based on maximizing the average difference to measure the feature distribution, it proposes to minimize the intra-class MMD and maximize the inter-class MMD to guide the optimization direction of the loss function; in view of the class imbalance phenomenon in the quality data, the class weight is used to improve the problem of low classification accuracy of a few samples, improve the accuracy of quality detection and the applicability of the detection model; obtain the initial parameters of the feature extractor EfficientNet through transfer learning, and train the AdaBoost method-based learner based on this, and finally construct an ensemble learner to reduce classification bias. However, the invention is also based on 2D image detection technology for detection, which is easily affected by background color and ambient lighting, resulting in wrong judgment, thereby affecting the detection effect. Summary of the invention

[0004] The purpose of the present invention is to disclose a wave soldering quality detection method and system based on laser flying scanning with better detection effect.

[0005] In order to achieve the above object, the present invention provides a wave soldering quality detection method based on laser scanning, comprising:

[0006] S1: Performing laser flyby scanning on the PCB board before welding to obtain a three-dimensional point cloud image of the laser flyby scanning before welding; performing laser flyby scanning on the PCB board after welding to obtain a three-dimensional point cloud image of the laser flyby scanning after welding;

[0007] S2: Compare the three-dimensional point cloud image of the laser flyby scanning before welding and the three-dimensional point cloud image of the laser flyby scanning after welding to obtain the difference three-dimensional point cloud image of the laser flyby scanning before and after welding;

[0008] S3: Selecting a threshold value based on the laser scanning difference 3D point cloud image;

[0009] S4: Construct a deep belief network, which obtains the wave soldering quality detection result based on the threshold and the difference three-dimensional image.

[0010] Furthermore, in step S1, it includes:

[0011] Perform laser flyby scanning on the PCB board before welding to obtain optical signals, and perform three-dimensional reconstruction of the optical signals to obtain a three-dimensional point cloud image of the laser flyby scanning before welding;

[0012] Perform laser flying scan on the welded PCB board to obtain optical signals, and perform three-dimensional reconstruction on the optical signals to obtain a three-dimensional point cloud image of the laser flying scan after welding; horizontal coordinate, vertical coordinate and vertical coordinate;

[0013] The 3D point cloud image includes horizontal coordinates, vertical coordinates and vertical coordinates.

[0014] The three-dimensional reconstruction formula is as follows:

[0015]

[0016] θ is the vertical angle of the laser, α is the horizontal angle of the laser, S is the slant distance from the instrument to the scanning point, and x, y, and z are the horizontal, vertical, and vertical coordinates in the three-dimensional point cloud image, respectively.

[0017] Furthermore, in step S2, it includes:

[0018] Subtract the 3D point cloud image of the laser fly-by-wire scanning before welding and the 3D point cloud image of the laser fly-by-wire scanning after welding:

[0019] img=img2-img1

[0020] img1 is the 3D point cloud image of the laser flyby scanning before welding, img2 is the 3D point cloud image of the laser flyby scanning after welding, and img is the difference 3D point cloud image of the laser flyby scanning before and after welding.

[0021] Furthermore, in step S3, it includes:

[0022] S3.1: De-noising the laser scanning difference 3D point cloud image before and after welding to obtain a de-noised point cloud image;

[0023] S3.2: Perform empirical mode decomposition on the denoised point cloud image to obtain IMF components;

[0024] S3.3: Obtain the vertical coordinate z according to the IMF component;

[0025] S3.4: Obtain the segmentation threshold V according to the vertical coordinate z and the IMF component.

[0026] Furthermore, in step S3.1, it includes:

[0027] O={o1,o2,o3,...o n}

[0028]

[0029] reach_dist k (p,o)=max{d k (o),d(o,p)}

[0030] O is the set of data points with the largest n local outlier factors, LOFk(p) is the kth local outlier factor of data point p, Nk(p) is the kth distance neighborhood of point p, lrdk(p) is the local reachability density of data point p, reach_distk(p,o) is the kth reachable distance of each point in the kth distance neighborhood of data point p, dk(o) is the kth distance of the domain point, and d(o,p) is the distance from domain point o to point p.

[0031] Furthermore, in step S3.2, it includes:

[0032]

[0033] x(t) is the point cloud image signal of the welding spot, ci(t) is the intrinsic mode function IMF, and r(t) is the remainder after decomposition. First, find all the extreme points of x(t), use interpolation to form the lower envelope emint(t) for the minimum point, and the upper envelope emax(t) for the maximum point, then calculate the mean m(t) = (emint(t) + emax(t)) / 2, extract the details d(t) = x(t) - m(t), and finally repeat the above steps for the residual m(t) until the mean of d(t) is 0.

[0034] The IMF with physical significance is obtained by using the cubic spline interpolation algorithm, and each IMF must satisfy: the number of extreme points is equal to the number of zero points and the time of the upper and lower envelopes is symmetrical about the time axis.

[0035] Furthermore, in step S3.3, it includes: according to Nyquist judgment, the IMF components obtained by empirical mode decomposition are divided into high-frequency detail components and low-frequency approximate components, and the three-dimensional point cloud signal is reconstructed using the approximate low-frequency components. After reconstruction, the RKD signal of each scanning laser measurement point and the vertical coordinate z of the PCB board surface are obtained.

[0036] Furthermore, in step S3.4, it includes: analyzing the empirical mode decomposition signal of the solder point cloud and using the maximum inter-class variance method to obtain the segmentation threshold V; wherein the vertical coordinate z less than the segmentation threshold V is a defective position containing a pinhole or a cold solder joint, wherein the vertical coordinate z of the PCB layout is greater than or equal to the segmentation threshold V is a defective position without a cold solder joint or a pinhole.

[0037] Furthermore, in step S4, the deep belief network includes: an input layer, an output layer for outputting recognition results, and two hidden layers for extracting internal features and learning the mapping relationship between input and output. The activation functions of the hidden layer and the output layer are the rectified linear unit function and the LogSig function respectively.

[0038] In addition, the present invention provides a wave soldering quality detection system based on laser scanning, comprising:

[0039] Scanning module: perform laser flyback scanning on the PCB board before welding to obtain a three-dimensional point cloud image of the laser flyback scanning before welding; perform laser flyback scanning on the PCB board after welding to obtain a three-dimensional point cloud image of the laser flyback scanning after welding;

[0040] Difference module: compares the 3D point cloud image of laser flyby scanning before welding and the 3D point cloud image of laser flyby scanning after welding to obtain the difference 3D point cloud image of laser flyby scanning before and after welding;

[0041] Threshold module: select threshold value according to the difference 3D point cloud image of laser scanning;

[0042] Detection module: construct a deep belief network, which obtains the wave soldering quality detection results based on the threshold and the difference three-dimensional image.

[0043] Compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0044] The present invention performs laser fly scanning on the PCB board before welding to obtain the laser fly scanning three-dimensional point cloud image before welding; performs laser fly scanning on the PCB board after welding to obtain the laser fly scanning three-dimensional point cloud image after welding; then compares the laser fly scanning three-dimensional point cloud image before welding and the laser fly scanning three-dimensional point cloud image after welding to obtain the laser fly scanning difference three-dimensional point cloud image before and after welding and selects the threshold value according to the laser fly scanning difference three-dimensional point cloud image; finally, a deep confidence network is constructed, and the deep confidence network obtains the wave soldering quality detection result according to the threshold value and the difference three-dimensional image. The influence of background color and ambient light in 2D images can be effectively avoided, thereby improving the detection effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 This is a diagram of a wave soldering quality detection method based on laser scanning according to the first embodiment;

[0046] Figure 2 This is a block diagram of a wave soldering quality detection system based on laser scanning as described in Example 4;

[0047] Figure 3 This is a flow chart of wave soldering quality detection based on laser scanning as described in Example 5; DETAILED DESCRIPTION

[0048] The drawings are for illustrative purposes only and should not be construed as limiting the present patent;

[0049] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0050] Embodiment 1:

[0051] This embodiment provides Figure 1 The wave soldering quality inspection method based on laser flying scanning includes:

[0052] S1: Performing laser flyby scanning on the PCB board before welding to obtain a three-dimensional point cloud image of the laser flyby scanning before welding; performing laser flyby scanning on the PCB board after welding to obtain a three-dimensional point cloud image of the laser flyby scanning after welding;

[0053] S2: Compare the three-dimensional point cloud image of the laser flyby scanning before welding and the three-dimensional point cloud image of the laser flyby scanning after welding to obtain the difference three-dimensional point cloud image of the laser flyby scanning before and after welding;

[0054] S3: Selecting a threshold value based on the laser scanning difference 3D point cloud image;

[0055] S4: Construct a deep belief network, which obtains the wave soldering quality detection result based on the threshold and the difference three-dimensional image.

[0056] In this embodiment, the PCB board before welding is subjected to laser fly scanning to obtain a three-dimensional point cloud image of the laser fly scanning before welding; the PCB board after welding is subjected to laser fly scanning to obtain a three-dimensional point cloud image of the laser fly scanning after welding; then the three-dimensional point cloud image of the laser fly scanning before welding and the three-dimensional point cloud image of the laser fly scanning after welding are compared and processed to obtain a three-dimensional point cloud image of the laser fly scanning difference before and after welding and select a threshold value according to the three-dimensional point cloud image of the laser fly scanning difference; finally, a deep confidence network is constructed, and the deep confidence network obtains the wave soldering quality detection result according to the threshold value and the difference three-dimensional image. The influence of background color and ambient light in 2D images can be effectively avoided, thereby improving the detection effect.

[0057] Embodiment 2:

[0058] This embodiment further discloses on the basis of the first embodiment:

[0059] In step S1, it includes:

[0060] Perform laser flyby scanning on the PCB board before welding to obtain optical signals, and perform three-dimensional reconstruction of the optical signals to obtain a three-dimensional point cloud image of the laser flyby scanning before welding;

[0061] Perform laser flying scan on the welded PCB board to obtain optical signals, and perform three-dimensional reconstruction on the optical signals to obtain a three-dimensional point cloud image of the laser flying scan after welding; horizontal coordinate, vertical coordinate and vertical coordinate;

[0062] The 3D point cloud image includes horizontal coordinates, vertical coordinates and vertical coordinates.

[0063] The three-dimensional reconstruction formula is as follows:

[0064]

[0065] θ is the vertical angle of the laser, α is the horizontal angle of the laser, S is the slant distance from the instrument to the scanning point, and x, y, and z are the horizontal, vertical, and vertical coordinates in the three-dimensional point cloud image, respectively.

[0066] In step S2, it includes:

[0067] Subtract the 3D point cloud image of the laser fly-by-wire scanning before welding and the 3D point cloud image of the laser fly-by-wire scanning after welding:

[0068] img=img2-img1

[0069] img1 is the 3D point cloud image of the laser flyby scanning before welding, img2 is the 3D point cloud image of the laser flyby scanning after welding, and img is the difference 3D point cloud image of the laser flyby scanning before and after welding.

[0070] In step S3, it includes:

[0071] S3.1: De-noising the laser scanning difference 3D point cloud image before and after welding to obtain a de-noised point cloud image;

[0072] S3.2: Perform empirical mode decomposition on the denoised point cloud image to obtain IMF components;

[0073] S3.3: Obtain the vertical coordinate z according to the IMF component;

[0074] S3.4: Obtain the segmentation threshold V according to the vertical coordinate z and the IMF component.

[0075] In this embodiment, the PCB board before welding is subjected to laser flying scanning to obtain the laser flying scanning 3D point cloud image before welding; the PCB board after welding is subjected to laser flying scanning to obtain the laser flying scanning 3D point cloud image after welding; then the laser flying scanning 3D point cloud image before welding and the laser flying scanning 3D point cloud image after welding are compared and processed to obtain the laser flying scanning difference 3D point cloud image before and after welding and select the threshold value according to the laser flying scanning difference 3D point cloud image; finally, a deep confidence network is constructed, and the deep confidence network obtains the wave soldering quality detection result according to the threshold value and the difference 3D image. The influence of background color and ambient light in 2D images can be effectively avoided, thereby improving the detection effect.

[0076] Embodiment three:

[0077] This embodiment further discloses on the basis of the second embodiment:

[0078] In step S3.1, it includes:

[0079] O={o1,o2,o3,...o n}

[0080]

[0081]

[0082] reach_dist k (p,o)=max{d k (o),d(o,p)}

[0083] O is the set of data points with the largest n local outlier factors, LOFk(p) is the kth local outlier factor of data point p, Nk(p) is the kth distance neighborhood of point p, lrdk(p) is the local reachability density of data point p, reach_distk(p,o) is the kth reachable distance of each point in the kth distance neighborhood of data point p, dk(o) is the kth distance of the domain point, and d(o,p) is the distance from domain point o to point p.

[0084] In step S3.2, it includes:

[0085]

[0086] x(t) is the point cloud image signal of the welding spot, ci(t) is the intrinsic mode function IMF, and r(t) is the remainder after decomposition. First, find all the extreme points of x(t), use interpolation to form the lower envelope emint(t) for the minimum point, and the upper envelope emax(t) for the maximum point, then calculate the mean m(t) = (emint(t) + emax(t)) / 2, extract the details d(t) = x(t) - m(t), and finally repeat the above steps for the residual m(t) until the mean of d(t) is 0.

[0087] The IMF with physical significance is obtained by using the cubic spline interpolation algorithm, and each IMF must satisfy: the number of extreme points is equal to the number of zero points and the time of the upper and lower envelopes is symmetrical about the time axis.

[0088] In step S3.3, it includes: according to Nyquist judgment, the IMF components obtained by empirical mode decomposition are divided into high-frequency detail components and low-frequency approximate components, and the three-dimensional point cloud signal is reconstructed using the approximate low-frequency components. After reconstruction, the RKD signal of each scanning laser measurement point and the vertical coordinate z of the PCB board surface are obtained.

[0089] Furthermore, in step S3.4, it includes: analyzing the empirical mode decomposition signal of the solder point cloud and using the maximum inter-class variance method to obtain the segmentation threshold V; wherein the vertical coordinate z less than the segmentation threshold V is a defective position containing a pinhole or a cold solder joint, wherein the vertical coordinate z of the PCB layout is greater than or equal to the segmentation threshold V is a defective position without a cold solder joint or a pinhole.

[0090] In step S4, the deep belief network includes: an input layer, an output layer for outputting recognition results, and two hidden layers for extracting internal features and learning the mapping relationship between input and output. The activation functions of the hidden layer and the output layer are the rectified linear unit function and the LogSig function respectively.

[0091] In this embodiment, the PCB board before welding is subjected to laser fly scanning to obtain a three-dimensional point cloud image of the laser fly scanning before welding; the PCB board after welding is subjected to laser fly scanning to obtain a three-dimensional point cloud image of the laser fly scanning after welding; then the three-dimensional point cloud image of the laser fly scanning before welding and the three-dimensional point cloud image of the laser fly scanning after welding are compared and processed to obtain a three-dimensional point cloud image of the laser fly scanning difference before and after welding and select a threshold value according to the three-dimensional point cloud image of the laser fly scanning difference; finally, a deep confidence network is constructed, and the deep confidence network obtains the wave soldering quality detection result according to the threshold value and the difference three-dimensional image. The influence of background color and ambient light in 2D images can be effectively avoided, thereby improving the detection effect.

[0092] Embodiment 4:

[0093] In addition, the present invention provides Figure 2 A wave soldering quality inspection system based on laser flying scanning is shown, comprising:

[0094] Scanning module: perform laser flyback scanning on the PCB board before welding to obtain a three-dimensional point cloud image of the laser flyback scanning before welding; perform laser flyback scanning on the PCB board after welding to obtain a three-dimensional point cloud image of the laser flyback scanning after welding;

[0095] Difference module: compares the 3D point cloud image of laser flyby scanning before welding and the 3D point cloud image of laser flyby scanning after welding to obtain the difference 3D point cloud image of laser flyby scanning before and after welding;

[0096] Threshold module: select threshold value according to the difference 3D point cloud image of laser scanning;

[0097] Detection module: construct a deep belief network, which obtains the wave soldering quality detection results based on the threshold and the difference three-dimensional image.

[0098] In this embodiment, the PCB board before welding is subjected to laser fly scanning to obtain a three-dimensional point cloud image of the laser fly scanning before welding; the PCB board after welding is subjected to laser fly scanning to obtain a three-dimensional point cloud image of the laser fly scanning after welding; then the three-dimensional point cloud image of the laser fly scanning before welding and the three-dimensional point cloud image of the laser fly scanning after welding are compared and processed to obtain a three-dimensional point cloud image of the laser fly scanning difference before and after welding and select a threshold value according to the three-dimensional point cloud image of the laser fly scanning difference; finally, a deep confidence network is constructed, and the deep confidence network obtains the wave soldering quality detection result according to the threshold value and the difference three-dimensional image. The influence of background color and ambient light in 2D images can be effectively avoided, thereby improving the detection effect.

[0099] Embodiment five:

[0100] In a specific embodiment, Figure 3As shown in the figure, the wave soldering process of common PCB is mainly divided into four steps: spraying flux, preheating, wave soldering, and forced air cooling. In order to detect the PCB wave soldering before and after, a laser fly scanning pre-soldering detection system is added before the PCB board is sprayed with flux, and a laser fly scanning post-soldering quality detection system is added after the PCB board completes forced air cooling. The pre-soldering laser scanning detection system scans the pre-soldering state of the PCB to obtain a laser fly scanning 3D point cloud image before welding; the post-soldering laser scanning detection system scans the post-soldering state of the PCB to obtain a laser fly scanning 3D point cloud image after welding; the laser fly scanning 3D point cloud image before welding and the laser fly scanning 3D point cloud image after welding are compared and processed to obtain the laser fly scanning difference 3D point cloud image before and after welding; the threshold is selected according to the laser fly scanning difference 3D point cloud image; a deep confidence network is constructed, and the deep confidence network obtains the wave soldering quality detection result according to the threshold and the difference 3D image.

[0101] In this embodiment, the PCB board before welding is subjected to laser fly scanning to obtain a three-dimensional point cloud image of the laser fly scanning before welding; the PCB board after welding is subjected to laser fly scanning to obtain a three-dimensional point cloud image of the laser fly scanning after welding; then the three-dimensional point cloud image of the laser fly scanning before welding and the three-dimensional point cloud image of the laser fly scanning after welding are compared and processed to obtain a three-dimensional point cloud image of the laser fly scanning difference before and after welding and select a threshold value according to the three-dimensional point cloud image of the laser fly scanning difference; finally, a deep confidence network is constructed, and the deep confidence network obtains the wave soldering quality detection result according to the threshold value and the difference three-dimensional image. The influence of background color and ambient light in 2D images can be effectively avoided, thereby improving the detection effect.

[0102] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. A wave soldering quality inspection method based on laser scanning, characterized in that: include: S1: Performing laser flyby scanning on the PCB board before welding to obtain a three-dimensional point cloud image of the laser flyby scanning before welding; performing laser flyby scanning on the PCB board after welding to obtain a three-dimensional point cloud image of the laser flyby scanning after welding; S2: Compare and process the three-dimensional point cloud image of the laser flyby scanning before welding and the three-dimensional point cloud image of the laser flyby scanning after welding to obtain the difference three-dimensional point cloud image of the laser flyby scanning before and after welding; S3: selecting a threshold value based on the laser scanning difference 3D point cloud image; S4: Construct a deep belief network, which obtains the wave soldering quality detection result based on the threshold and the difference three-dimensional image.

2. The wave soldering quality detection method based on laser scanning according to claim 1 is characterized in that: In step S1, it includes: Perform laser flyby scanning on the PCB board before welding to obtain optical signals, and perform three-dimensional reconstruction of the optical signals to obtain a three-dimensional point cloud image of the laser flyby scanning before welding; Perform laser flying scan on the welded PCB board to obtain optical signals, and perform three-dimensional reconstruction on the optical signals to obtain a three-dimensional point cloud image of the laser flying scan after welding; horizontal coordinate, vertical coordinate and vertical coordinate; The 3D point cloud image includes horizontal coordinates, vertical coordinates and vertical coordinates. The three-dimensional reconstruction formula is as follows: θ is the vertical angle of the laser, α is the horizontal angle of the laser, S is the slant distance from the instrument to the scanning point, and x, y, and z are the horizontal, vertical, and vertical coordinates in the three-dimensional point cloud image, respectively.

3. The wave soldering quality detection method based on laser scanning according to claim 1 is characterized in that: In step S2, it includes: Subtract the 3D point cloud image of the laser fly-by-wire scanning before welding and the 3D point cloud image of the laser fly-by-wire scanning after welding: img=img2-img1 img1 is the 3D point cloud image of the laser flyby scanning before welding, img2 is the 3D point cloud image of the laser flyby scanning after welding, and img is the difference 3D point cloud image of the laser flyby scanning before and after welding.

4. The wave soldering quality detection method based on laser scanning according to claim 1 is characterized in that: In step S3, it includes: S3.1: De-noising the laser scanning difference 3D point cloud image before and after welding to obtain a de-noised point cloud image; S3.2: Perform empirical mode decomposition on the denoised point cloud image to obtain IMF components; S3.3: Obtain the vertical coordinate z according to the IMF component; S3.4: Obtain the segmentation threshold V according to the vertical coordinate z and the IMF component.

5. The wave soldering quality detection method based on laser scanning according to claim 4 is characterized in that: In step S3.1, it includes: O={o1,o2,o3,……o n } reach_dist k (p,o)=max{d k (o),d(o,p)} O is the set of data points with the largest n local outlier factors, LOFk(p) is the kth local outlier factor of data point p, Nk(p) is the kth distance neighborhood of point p, lrdk(p) is the local reachability density of data point p, reach_distk(p,o) is the kth reachable distance of each point in the kth distance neighborhood of data point p, dk(o) is the kth distance of the domain point, and d(o,p) is the distance from domain point o to point p.

6. The wave soldering quality detection method based on laser scanning according to claim 4 is characterized in that: In step S3.2, it includes: x(t) is the point cloud image signal of the welding spot, ci(t) is the intrinsic mode function IMF, and r(t) is the remainder after decomposition; first find all the extreme points of x(t), use interpolation to form the lower envelope emint(t) for the minimum point, and the upper envelope emax(t) for the maximum point, then calculate the mean m(t) = (emint(t) + emax(t)) / 2, extract the details d(t) = x(t) - m(t), and finally repeat the above steps for the residual m(t) until the mean of d(t) is 0; The IMF with physical significance is obtained by using the cubic spline interpolation algorithm, and each IMF must satisfy: the number of extreme points is equal to the number of zero points and the time of the upper and lower envelopes is symmetrical about the time axis.

7. The wave soldering quality detection method based on laser scanning according to claim 4 is characterized in that: In step S3.3, it includes: according to Nyquist judgment, the IMF components obtained by empirical mode decomposition are divided into high-frequency detail components and low-frequency approximate components, and the three-dimensional point cloud signal is reconstructed using the approximate low-frequency components. After reconstruction, the RKD signal of each scanning laser measurement point and the vertical coordinate z of the PCB board surface are obtained.

8. The wave soldering quality detection method based on laser scanning according to claim 4 is characterized in that: In step S3.4, it includes: analyzing the empirical mode decomposition signal of the solder point cloud and using the maximum inter-class variance method to obtain the segmentation threshold V; wherein the vertical coordinate z less than the segmentation threshold V is a defective position containing a pinhole or a cold solder joint, wherein the vertical coordinate z of the PCB layout is greater than or equal to the segmentation threshold V is a defective position without a cold solder joint or a pinhole.

9. The wave soldering quality detection method based on laser scanning according to claim 1 is characterized in that: In step S4, the deep belief network includes: an input layer, an output layer for outputting recognition results, and two hidden layers for extracting internal features and learning the mapping relationship between input and output; the activation functions of the hidden layer and the output layer are the rectified linear unit function and the LogSig function respectively.

10. A wave soldering quality inspection system based on laser scanning, comprising: Scanning module: perform laser flyback scanning on the PCB board before welding to obtain a three-dimensional point cloud image of the laser flyback scanning before welding; perform laser flyback scanning on the PCB board after welding to obtain a three-dimensional point cloud image of the laser flyback scanning after welding; Difference module: compares the 3D point cloud image of laser flyby scanning before welding and the 3D point cloud image of laser flyby scanning after welding to obtain the difference 3D point cloud image of laser flyby scanning before and after welding; Threshold module: select threshold value according to the difference 3D point cloud image of laser scanning; Detection module: construct a deep belief network, which obtains the wave soldering quality detection results based on the threshold and the difference three-dimensional image.

Images