Integrated circuit board welding spot quality detection system and method based on spectral analysis

Abstract

The invention relates to the technical field of quality detection, and discloses an integrated circuit board welding spot quality detection system and method based on spectral analysis. According to the method, detection points are arranged in an integrated circuit board welding area, Internet of Things equipment is installed, different batches of welding spot samples are collected based on the set detection points, and a spectrograph is used for collecting irradiated spectral data in real time to construct a training sample set; analyzing and processing the spectral data in the constructed training sample set through a data processing method, detecting the quality of the welding spot corresponding to a spectral data curve, after detection is completed, collecting image data of the surface defect welding spot in the welding area through the installed Internet of Things equipment, and performing image processing; meanwhile, feature extraction is carried out on the processed defective welding spot image data, finally, welding spot quality detection modes and welding spot surface defect detection modes are summarized to construct a welding spot detection model, the welding spot quality of the integrated circuit board is detected in real time based on the constructed welding spot detection model, and the accuracy of welding spot quality detection is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of quality inspection, and particularly to an integrated circuit board solder joint quality inspection system and method based on spectral analysis. Background Art

[0002] At present, the mainstream solder joint quality inspections often adopt destructive inspections and unified post-welding inspections. However, this inspection method is single and inefficient, and it is easy to cause industrial waste.

[0003] In the existing publicly applied patent CN110887846A, this method irradiates a circuit board with tricolor lights RGB continuously at fixed time intervals and acquires images; at the same time, divides the solder joint part of the acquired images; and then recombines the effective sub-region images into RGB three-channel time series signals according to time by continuously vibrating in a specified area of the PCB board at a specified frequency; at the same time, adopts an independent component analysis algorithm to obtain three independent components with the smallest correlation; performs FFT transformation, and selects one independent component with the largest correlation with the effective sub-region image; and judges the solder joint quality according to the fitting degree between the independent component and the typical value. However, since the vibration monitoring method is likely to cause damage to the solder joints, the accuracy of solder joint detection is not high, and the detection effect is single by acquiring images for solder joint detection, which cannot meet the needs of solder joint quality inspection. Summary of the Invention

[0004] (I) Technical Problems to be Solved

[0005] Aiming at the deficiencies of the prior art, the present invention provides an integrated circuit board solder joint quality inspection system and method based on spectral analysis, which has the advantages of being real-time, accurate, efficient, etc., and solves the problems of single and low-efficiency of destructive inspection and unified post-welding inspection of solder joints.

[0006] (II) Technical Solutions

[0007] To solve the above technical problems of single and low-efficiency of destructive inspection and unified post-welding inspection of solder joints, the present invention provides the following technical solutions:

[0008] The present invention discloses an integrated circuit board solder joint quality inspection method based on spectral analysis, which specifically includes the following steps:

[0009] S1. Set detection points in the welding area of the integrated circuit board and install Internet of Things devices, and at the same time collect solder joint samples of different batches based on the set detection points;

[0010] S2. Irradiate the solder joint samples of different batches with a spectrometer, and collect the spectral data after irradiation in real time to construct a training sample set;

[0011] S3. Analyze and process the spectral data in the constructed training sample set through a data processing method to detect the solder joint quality corresponding to the spectral data curve;

[0012] S31. Construct the spectral data curve in the training sample set through a data processing method;

[0013] S32. Analyze based on the constructed spectral data curve in combination with the heat conduction method to detect the solder joint quality corresponding to the spectral data curve;

[0014] S4. After the detection is completed, collect the image data of the surface defect solder joints in the welding area through the installed Internet of Things device, and perform image processing to obtain the processed defect solder joint image data;

[0015] S5. Extract the features of the processed defect solder joint image data and save them;

[0016] S6. Summarize the solder joint quality detection method and the solder joint surface defect detection method to construct a solder joint detection model, and based on the constructed solder joint detection model, detect the solder joint quality of the integrated circuit board in real time.

[0017] In the present invention, detection points are set in the welding area of the integrated circuit board and Internet of Things devices are installed. At the same time, solder joint samples of different batches are collected based on the set detection points. The spectrometer is used to collect the spectral data after irradiation in real time to construct a training sample set, and the spectral data in the constructed training sample set is analyzed and processed through a data processing method to detect the solder joint quality corresponding to the spectral data curve. After the detection is completed, the image data of the surface defect solder joints in the welding area is collected through the installed Internet of Things device and image processing is performed. At the same time, the features of the processed defect solder joint image data are extracted. Finally, the solder joint quality detection method and the solder joint surface defect detection method are summarized to construct a solder joint detection model, and based on the constructed solder joint detection model, the solder joint quality of the integrated circuit board is detected in real time, improving the accuracy of solder joint quality detection.

[0018] Preferably, the step of irradiating the solder joint samples of different batches with a spectrometer and collecting the spectral data after irradiation in real time to construct a training sample set includes the following steps:

[0019] Set the parameters of the spectrometer and irradiate and measure each solder joint sample;

[0020] In order to minimize the influence of the difference in the uniformity of the solder joint samples on the accuracy of the irradiation measurement, 30 measurement sites are randomly and uniformly selected on the solder joint samples for spectral collection;

[0021] In order to avoid the influence of laser pulse fluctuations, each spectrum is obtained by accumulating and superimposing 10 laser pulses;

[0022] Construct a training sample set based on the spectral data collected in real time.

[0023] Preferably, constructing the spectral data curve in the training sample set by the data processing method comprises the following steps:

[0024] S311, calculating the average value of all spectral data as an average spectrum;

[0025]

[0026] in, represents the calculated average spectrum, x Ns represents the spectral data of the sth spectral band sequence after normalization, m represents the number of spectral bands, and s represents the spectral band sequence;

[0027] S312, performing a univariate linear regression on the spectrum of each sample and the average spectrum, solving the least squares problem to obtain the baseline translation and offset of each sample;

[0028] S313, summarizing the baseline shift and offset of each sample to construct a spectral data curve in the training sample set.

[0029] Preferably, solving the least squares problem to obtain the baseline translation and offset of each sample comprises the following steps:

[0030]

[0031] Among them, x N represents the spectral data calculated by linear regression, represents the calculated average spectrum, g N Indicates the baseline shift of the sample spectral data, d N Indicates the offset of the sample spectral data.

[0032] Preferably, the step of aggregating the baseline shift and offset of each sample to construct a spectral data curve in the training sample set comprises the following steps:

[0033] The spectral data curve of each sample in the training sample set is as follows:

[0034]

[0035] Where n represents the order of Gaussian fitting, a i represents the height of curve i, b i represents the coordinates of the center position of curve i on the x-axis, c i represents the width of curve i, e represents the natural logarithm, f(x N ) represents the spectral data x after linear regression calculation N .

[0036] Preferably, the analysis is performed by combining the constructed spectral data curve with the heat conduction method. The steps for detecting the solder joint quality corresponding to the spectral data curve are as follows:

[0037] During the real-time monitoring of the integrated circuit board welding process, when the pulsed thermal excitation acts on the surface of the integrated circuit board during welding, the heat conduction process of the solder joint is as follows:

[0038]

[0039] Among them, k represents the thermal conductivity, T(z,t) represents the temperature at time t at a distance z from the surface of the integrated circuit board, ρ represents the density, l represents the specific heat capacity ratio, q represents the thermal excitation energy, represents the partial derivative;

[0040] Based on the heat conduction process of the solder joint, the temperature change curve during the integrated circuit board welding process is monitored in real time. When it is assumed that there are defects in the integrated circuit board, due to the difference in thermal properties between the defective area and the normal area, the temperature change curve of the defective integrated circuit board is fluctuating;

[0041] When the temperature change curve during the integrated circuit board welding process is a straight line, it indicates that there are no defects in the integrated circuit board welding;

[0042] Compare the spectral data curves of each sample without defects, set the spectral data fluctuation threshold. When the fluctuation amplitude of the spectral data curve of the sample is less than the set spectral data fluctuation threshold, it indicates that the quality of the sample is normal, otherwise there are defects.

[0043] In the present invention, the average value of all spectral data is calculated as the average spectrum, and the spectrum of each sample is subjected to a unary linear regression with the average spectrum. The least squares problem is solved to obtain the baseline translation amount and offset amount of each sample. At the same time, the baseline translation amount and offset amount of each sample are summarized to construct the spectral data curve in the training sample set; at the same time, the heat conduction method is used for analysis to determine the corresponding solder joint quality, improving the real-time performance of solder joint quality detection.

[0044] Preferably, after the detection, the image data of the defective solder joints on the inner surface of the welding area are collected by the installed Internet of Things device and image processing is performed. The steps for obtaining the processed defective solder joint image data are as follows:

[0045] S41. Perform filtering and noise reduction on the collected image data of the defective solder joints on the inner surface of the welding area;

[0046] Perform filtering and noise reduction on 7×7 pixel points around the neighborhood of the target pixel;

[0047] Sort the gray values of 49 pixels in the 7×7 neighborhood and process them through the filtering and noise reduction formula;

[0048] The filtering and noise reduction formula is as follows:

[0049]

[0050] Wherein, W represents the total number of filtering times, represents the j-th guided filtering, F represents the image data of the defective solder joints after filtering, and I represents the gray value of the pixel point;

[0051] S42. Perform image segmentation on the image data of the defective solder joints after filtering.

[0052] Preferably, the performing image segmentation on the image data of the defective solder joints after filtering includes the following steps:

[0053] Perform binarization processing on the image data of the defective solder joints after filtering to obtain the image data after binarization processing;

[0054] Select the initial gray threshold r to divide all pixels in the image data into two categories C1 and C2;

[0055] Set C1 as the pixel category less than or equal to the gray threshold r, and C2 as the pixel category greater than the gray threshold r;

[0056] Set the gray mean value of the pixel category C1 as h1, the gray mean value of the pixel category C2 as h2, and the global gray mean value as h3;

[0057] Set the probability that a pixel in the image data belongs to the pixel category C1 as p1, and the probability that it belongs to the pixel category C2 as p2;

[0058] The binarization processing formula is as follows:

[0059] h3 = h1×p1 + h2×p2;

[0060]

[0061] Wherein, η represents the binarization threshold;

[0062] Set the gray value greater than the binarization threshold as 255, and the gray value less than or equal to the binarization threshold as 0;

[0063] Summarize the image data after binarization processing to obtain the image data after binarization processing;

[0064] Perform image segmentation on the image data after binarization processing to obtain the solder joint image;

[0065] Record the gray values of the 8 neighborhoods around each pixel point in the image data after binarization processing. When the gray values of the 8 neighborhoods around the pixel point are the same as the gray value of the center point, set that pixel point to be inside the solder joint image and delete that pixel point;

[0066] When the grayscale values of the 8 neighboring pixels around a pixel are different from the grayscale value of the central pixel, it is determined that the pixel is on the edge of the solder joint image, and the pixel is retained;

[0067] Traverse and summarize each pixel in the image data after binarization processing to obtain the solder joint image;

[0068] Set the obtained solder joint image as the processed defective solder joint image data.

[0069] The present invention filters and denoises the image data of defective solder joints on the inner surface of the welding area collected, and performs image segmentation on the filtered defective solder joint image data to obtain the processed defective solder joint image data, improving the reliability of the defective solder joint image data.

[0070] Preferably, the feature extraction of the processed defective solder joint image data includes the following steps:

[0071] S51. Divide the processed defective solder joint image data into 16×16 image data blocks, and input the divided image data blocks into a convolutional neural network;

[0072] S52. Extract features from the divided image data blocks through the convolutional neural network;

[0073] S521. Input the divided image data blocks into the input layer of the neural network;

[0074] S522. After receiving the divided image data blocks, the input layer transmits the received image data blocks to the convolutional layer, and the convolutional layer extracts local features in the image data blocks through convolutional operations;

[0075] The convolutional operation formula is as follows:

[0076] y = λ×ω + θ;

[0077] Where λ represents the input image data block, ω represents the weight value of the corresponding convolutional kernel, θ represents the bias value, and y represents the output feature;

[0078] S523. After the convolutional layer extracts the local features in the image data blocks, the pooling layer processes the local features extracted from the image data blocks, and downsamples the features through the pooling layer to reduce the data dimension;

[0079] S524. Through continuous convolution and pooling until the feature extraction converges and the convolution stops, the extracted features are summarized and input into the fully connected layer;

[0080] S525. The fully connected layer integrates the extracted features and outputs the final feature extraction result.

[0081] The present invention segments, convolves, and pools the processed defective solder joint image data using a convolutional neural network, and simultaneously summarizes to obtain the features of the defective solder joint image data, ensuring the accuracy of the comparison of the defective solder joint image data.

[0082] Preferably, the solder joint quality detection method and the solder joint surface defect detection method are summarized to construct a solder joint detection model, and the solder joint quality of the integrated circuit board is detected in real time based on the constructed solder joint detection model, including the following steps:

[0083] Determine whether there are defects in the solder joint through the solder joint quality detection method. After the solder joint quality detection is completed, the solder joint surface image data is collected in real time and feature extraction is performed. The features extracted in real time are compared with the saved features. When the features extracted in real time are consistent with the saved features, it indicates that there are surface defects in the current solder joint, otherwise it is normal.

[0084] The present invention constructs a solder joint detection model by summarizing the solder joint quality detection method and the solder joint surface defect detection method, and comprehensively detects the solder joint quality and surface defects through the combination of the two methods, improving the effectiveness of the solder joint quality detection.

[0085] The present invention also discloses an integrated circuit board solder joint quality system based on spectral analysis for implementing the integrated circuit board solder joint quality method based on spectral analysis. The system includes: a data acquisition module, a spectral data processing module, a solder joint quality defect evaluation module, a solder joint feature extraction module, and a solder joint surface defect evaluation module;

[0086] The data acquisition module is used to collect solder joint spectral data and solder joint surface image data in real time;

[0087] The spectral data processing module is used to process the collected spectral data and generate a spectral data curve;

[0088] The solder joint quality defect evaluation module is used to evaluate the solder joint quality defect by combining the spectral data curve and the heat conduction method;

[0089] The solder joint feature extraction module is used to extract features from the solder joint surface image data collected in real time;

[0090] The solder joint surface defect evaluation module is used to evaluate the solder joint surface defect according to the feature extraction result.

[0091] (III) Beneficial effects

[0092] Compared with the prior art, the present invention provides an integrated circuit board solder joint quality detection system and method based on spectral analysis, having the following beneficial effects:

[0093] 1. The invention sets detection points in the welding area of the integrated circuit board and installs Internet of Things devices. At the same time, based on the set detection points, solder joint samples of different batches are collected. The spectrometer is used to collect the spectral data after irradiation in real time to construct a training sample set, and the spectral data in the constructed training sample set is analyzed and processed through a data processing method to detect the solder joint quality corresponding to the spectral data curve. After the detection is completed, the image data of the surface defective solder joints in the welding area is collected through the installed Internet of Things devices and image processing is carried out. At the same time, feature extraction is carried out on the processed defective solder joint image data. Finally, the solder joint quality detection method and the solder joint surface defect detection method are summarized to construct a solder joint detection model, and the solder joint quality of the integrated circuit board is detected in real time based on the constructed solder joint detection model, improving the accuracy of solder joint quality detection.

[0094] 2. The invention calculates the average value of all spectral data as the average spectrum, performs unary linear regression on the spectrum of each sample and the average spectrum, solves the least squares problem to obtain the baseline translation amount and offset amount of each sample, and at the same time summarizes the baseline translation amount and offset amount of each sample to construct the spectral data curve in the training sample set; at the same time, the heat conduction method is analyzed to determine the corresponding solder joint quality, improving the real-time performance of solder joint quality detection.

[0095] 3. The invention filters and denoises the image data of the surface defective solder joints in the collected welding area, and performs image segmentation on the filtered defective solder joint image data to obtain the processed defective solder joint image data, improving the reliability of the defective solder joint image data.

[0096] 4. The invention uses a convolutional neural network to segment, convolve, and pool the processed defective solder joint image data, and at the same time summarizes to obtain the features of the defective solder joint image data, ensuring the accuracy of the comparison of the defective solder joint image data.

[0097] 5. The invention summarizes the solder joint quality detection method and the solder joint surface defect detection method to construct a solder joint detection model, and comprehensively detects the solder joint quality and surface defects through a combination of the two methods, improving the effectiveness of solder joint quality detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0098] Figure 1 It is a schematic structural diagram of the solder joint quality detection process of the integrated circuit board for spectral analysis of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0099] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0100] Embodiment 1

[0101] Please refer to Figure 1 , this embodiment discloses a method for detecting the solder joint quality of an integrated circuit board based on spectral analysis, which specifically includes the following steps:

[0102] S1. Set detection points in the welding area of the integrated circuit board and install Internet of Things devices, and at the same time collect solder joint samples of different batches based on the set detection points;

[0103] S2. Irradiate the solder joint samples of different batches with a spectrometer, and collect the spectral data after irradiation in real time to construct a training sample set;

[0104] S3. Analyze and process the spectral data in the constructed training sample set through a data processing method to detect the solder joint quality corresponding to the spectral data curve;

[0105] S31. Construct a spectral data curve of the spectral data in the training sample set through a data processing method;

[0106] S32. Analyze based on the constructed spectral data curve in combination with the heat conduction method to detect the solder joint quality corresponding to the spectral data curve;

[0107] S4. After the detection is completed, collect the image data of the surface defect solder joints in the welding area through the installed Internet of Things devices, and perform image processing to obtain the processed defect solder joint image data;

[0108] S5. Extract the features of the processed defect solder joint image data and save them;

[0109] S6. Summarize the solder joint quality detection method and the solder joint surface defect detection method to construct a solder joint detection model, and based on the constructed solder joint detection model, detect the solder joint quality of the integrated circuit board in real time;

[0110] Further, please refer to Figure 1 , irradiating the solder joint samples of different batches with a spectrometer and collecting the spectral data after irradiation in real time to construct a training sample set includes the following steps:

[0111] Set the parameters of the spectrometer and irradiate and measure each solder joint sample;

[0112] In order to minimize the impact of different uniformity of solder joint samples on the accuracy of irradiation measurement, 30 measurement sites were randomly and uniformly selected on the solder joint samples for spectrum acquisition;

[0113] To avoid the influence of laser pulse fluctuation, each spectrum was obtained by accumulating and superimposing 10 laser pulses;

[0114] Construct a training sample set based on the spectral data collected in real time;

[0115] For further information, see Figure 1 ,Constructing the spectral data curve in the training sample set by the data processing method includes the following steps:

[0116] S311, calculating the average value of all spectral data as an average spectrum;

[0117]

[0118] in, represents the calculated average spectrum, x Ns represents the spectral data of the sth spectral band sequence after normalization, m represents the number of spectral bands, and s represents the spectral band sequence;

[0119] S312, performing a univariate linear regression on the spectrum of each sample and the average spectrum, solving the least squares problem to obtain the baseline translation and offset of each sample;

[0120]

[0121] Among them, x N represents the spectral data calculated by linear regression, represents the calculated average spectrum, g N Indicates the baseline shift of the sample spectral data, d N Indicates the offset of the sample spectrum data;

[0122] S313, summarizing the baseline translation and offset of each sample to construct a spectral data curve in the training sample set;

[0123] The spectral data curve of each sample in the training sample set is as follows:

[0124]

[0125] Where n represents the order of Gaussian fitting, a i represents the height of curve i, b i represents the coordinates of the center position of curve i on the x-axis, c i represents the width of curve i, e represents the natural logarithm, f(x N) represents the spectral data x after linear regression calculation N ;

[0126] Further, please refer to Figure 1 , and based on the constructed spectral data curve, analyze it in combination with the heat conduction method. Detecting the solder joint quality corresponding to the spectral data curve includes the following steps:

[0127] During the real-time monitoring of the integrated circuit board welding process, when the pulsed thermal excitation acts on the surface of the integrated circuit board during welding, the heat conduction process of the solder joint is as follows:

[0128]

[0129] Among them, k represents the thermal conductivity, T(z,t) represents the temperature at time t at a distance z from the surface of the integrated circuit board, ρ represents the density, l represents the heat capacity ratio, q represents the thermal excitation energy, represents the partial derivative;

[0130] Further, based on the heat conduction process of the solder joint, the temperature change curve during the integrated circuit board welding process is monitored in real time. When it is set that there are defects in the integrated circuit board, due to the difference in thermal properties between the defective area and the normal area, the temperature change curve of the defective integrated circuit board is fluctuating;

[0131] When the temperature change curve during the integrated circuit board welding process is a straight line, it indicates that there are no defects in the integrated circuit board welding;

[0132] Further, compare the spectral data curves of each sample without defects, set the spectral data fluctuation threshold. When the fluctuation amplitude of the spectral data curve of the sample is less than the set spectral data fluctuation threshold, it indicates that the quality of the sample is normal, otherwise there are defects;

[0133] Further, please refer to Figure 1 , after the detection is completed, collect the image data of the defective solder joints on the inner surface of the welding area through the installed Internet of Things device, and perform image processing. The steps to obtain the processed defective solder joint image data include the following:

[0134] S41. Filter and denoise the image data of the defective solder joints on the inner surface of the welding area collected;

[0135] Filter and denoise the 7×7 pixel points around the target pixel neighborhood;

[0136] Sort the gray values of the 49 pixels in the 7×7 neighborhood and process them through the filter denoising formula;

[0137] The filter denoising formula is as follows:

[0138]

[0139] Among them, W represents the total number of filtering times, represents the j-th guided filtering, F represents the image data of defective solder joints after filtering, and I represents the gray value of pixel points;

[0140] S42. Perform image segmentation on the image data of defective solder joints after filtering;

[0141] Perform binarization processing on the image data of defective solder joints after filtering to obtain the image data after binarization processing;

[0142] Select the initial gray threshold r to divide all pixels in the image data into two categories C1 and C2;

[0143] Set C1 as the pixel category less than or equal to the gray threshold r, and C2 as the pixel category greater than the gray threshold r;

[0144] Set the gray mean value of pixel category C1 as h1, the gray mean value of pixel category C2 as h2, and the global gray mean value as h3;

[0145] Set the probability that a pixel in the image data belongs to pixel category C1 as p1, and the probability of belonging to pixel category C2 as p2;

[0146] The binarization processing formula is as follows:

[0147] h3 = h1×p1 + h2×p2;

[0148]

[0149] Among them, η represents the binarization threshold;

[0150] Set the gray value greater than the binarization threshold as 255, and the gray value less than or equal to the binarization threshold as 0;

[0151] Aggregate the image data after binarization processing to obtain the image data after binarization processing;

[0152] Perform image segmentation on the image data after binarization processing to obtain the solder joint image;

[0153] Record the gray values of the 8 neighborhoods around each pixel point in the image data after binarization processing. When the gray values of the 8 neighborhoods around the pixel point are the same as the gray value of the center point, set the pixel point to be inside the solder joint image and delete the pixel point;

[0154] When the gray values of the 8 neighborhoods around the pixel point are different from the gray value of the center point, set the pixel point to be on the edge of the solder joint image and retain the pixel point;

[0155] Traverse and summarize each pixel point in the binarized image data to obtain a solder joint image;

[0156] Set the obtained solder joint image as the processed defective solder joint image data;

[0157] Further, please refer to Figure 1 , and the feature extraction of the processed defective solder joint image data includes the following steps:

[0158] S51. Divide the processed defective solder joint image data into 16×16 image data blocks, and input the divided image data blocks into a convolutional neural network;

[0159] S52. Extract features from the divided image data blocks through the convolutional neural network;

[0160] S521. Input the divided image data blocks into the input layer of the neural network;

[0161] S522. After the input layer receives the divided image data blocks, it transmits the received image data blocks to the convolutional layer, and the convolutional layer extracts local features in the image data blocks through convolutional operations;

[0162] The convolutional operation formula is as follows:

[0163] y = λ×ω + θ;

[0164] where λ represents the input image data block, ω represents the weight of the corresponding convolutional kernel, θ represents the bias value, and y represents the output feature;

[0165] S523. After the convolutional layer extracts the local features in the image data blocks, the pooling layer processes the local features extracted from the image data blocks, and downsamples the features through the pooling layer to reduce the data dimension;

[0166] S524. Through continuous convolution and pooling until the extracted features converge, the convolution stops, and the extracted features are summarized and input into the fully connected layer;

[0167] S525. The fully connected layer integrates the extracted features, outputs the final feature extraction result, and saves it;

[0168] Further, please refer to Figure 1 , summarize the solder joint quality detection method and the solder joint surface defect detection method to build a solder joint detection model, and based on the built solder joint detection model, real-time detect the quality of the solder joints on the integrated circuit board, including the following steps:

[0169] Determine whether there are defects in the solder joints within the welding area of the integrated circuit board through solder joint quality detection. After the solder joint quality detection is completed, collect the surface image data of the solder joints in real time and perform feature extraction. Compare the features extracted in real time with the saved features. When the features extracted in real time are consistent with the saved features, it indicates that there are surface defects in the current solder joint; otherwise, it is normal.

[0170] Embodiment 2

[0171] Please refer to Figure 1 , this embodiment also discloses an integrated circuit board solder joint quality system based on spectral analysis, which is used to implement the integrated circuit board solder joint quality method based on spectral analysis. The system includes: a data acquisition module, a spectral data processing module, a solder joint quality defect evaluation module, a solder joint feature extraction module, and a solder joint surface defect evaluation module;

[0172] The data acquisition module is used to collect solder joint spectral data and solder joint surface image data in real time;

[0173] The spectral data processing module is used to process the collected spectral data and generate a spectral data curve;

[0174] The solder joint quality defect evaluation module is used to evaluate the solder joint quality defects by combining the spectral data curve and the heat conduction method;

[0175] The solder joint feature extraction module is used to extract features from the solder joint surface image data collected in real time;

[0176] The solder joint surface defect evaluation module is used to evaluate the solder joint surface defects according to the feature extraction results.

[0177] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for detecting the quality of solder joints of integrated circuit boards based on spectral analysis, characterized in that: The following steps are involved: S1. Set up inspection points in the integrated circuit board welding area and install IoT devices, and collect solder joint samples from different batches based on the set inspection points; S2, irradiating different batches of solder joint samples using a spectrometer and collecting the spectral data after irradiation in real time to build a training sample set; S3, analyzing and processing the spectral data in the constructed training sample set by a data processing method, and detecting the quality of the solder joint corresponding to the spectral data curve; S31, constructing a spectral data curve in a training sample set by a data processing method; S32, analyzing the constructed spectral data curve in combination with the heat conduction method to detect the quality of the solder joint corresponding to the spectral data curve; S4. After the detection is completed, the image data of the surface defective solder joints in the welding area is collected through the installed Internet of Things device, and the image is processed to obtain the processed image data of the defective solder joints; S5, extracting features from the processed defective solder joint image data and saving the features; S6. Summarize the solder joint quality detection methods and solder joint surface defect detection methods to construct a solder joint detection model, and detect the quality of the solder joints of the integrated circuit board in real time based on the constructed solder joint detection model.

2. The method for detecting the quality of solder joints of integrated circuit boards based on spectral analysis according to claim 1, characterized in that: The method of constructing the spectral data curve in the training sample set by the data processing method comprises the following steps: S311, calculating the average value of all spectral data as an average spectrum; in, represents the calculated average spectrum, x Ns represents the spectral data of the sth spectral band sequence after normalization, m represents the number of spectral bands, and s represents the spectral band sequence; S312, performing a univariate linear regression on the spectrum of each sample and the average spectrum, solving the least squares problem to obtain the baseline translation and offset of each sample; S313, summarizing the baseline shift and offset of each sample to construct a spectral data curve in the training sample set.

3. The method for detecting the quality of solder joints of integrated circuit boards based on spectral analysis according to claim 2, characterized in that: Solving the least squares problem to obtain the baseline translation and offset of each sample includes the following steps: Among them, x N represents the spectral data calculated by linear regression, represents the calculated average spectrum, g N Indicates the baseline shift of the sample spectral data, d N Indicates the offset of the sample spectral data.

4. The method for detecting the quality of solder joints of integrated circuit boards based on spectral analysis according to claim 2, characterized in that: The method of summarizing the baseline shift and offset of each sample to construct a spectral data curve in the training sample set includes the following steps: The spectral data curve of each sample in the training sample set is as follows: Where n represents the order of Gaussian fitting, a i represents the height of curve i, b i represents the coordinates of the center position of curve i on the x-axis, c i represents the width of curve i, e represents the natural logarithm, f(x N ) represents the spectral data x after linear regression calculation N .

5. The method for detecting the quality of solder joints of integrated circuit boards based on spectral analysis according to claim 1, characterized in that: The analysis based on the constructed spectral data curve combined with the heat conduction method and detecting the quality of the solder joint corresponding to the spectral data curve include the following steps: During the real-time monitoring of the integrated circuit board welding process, when the pulse heat excitation acts on the surface of the integrated circuit board during welding, the heat conduction process of the solder joint is as follows: Where k represents the thermal conductivity, T(z,t) represents the temperature at time t at a distance z from the surface of the integrated circuit board, ρ represents the density, l represents the heat capacity ratio, and q represents the thermal excitation energy. represents partial differential; Based on the heat conduction process of the solder joint, the temperature change curve of the integrated circuit board during the soldering process is monitored in real time. When the integrated circuit board is defective, the temperature change curve of the defective integrated circuit board fluctuates due to the difference in thermal properties between the defective area and the normal area. When the temperature change curve during the integrated circuit board welding process is a straight line, it means that there is no defect in the integrated circuit board welding; The spectral data curve of each sample without defects is compared, and a spectral data fluctuation threshold is set. When the fluctuation amplitude of the spectral data curve of the sample is less than the set spectral data fluctuation threshold, it indicates that the quality of the sample is normal, otherwise there is a defect.

6. The method for detecting the quality of solder joints of integrated circuit boards based on spectral analysis according to claim 1, characterized in that: After the detection is completed, the image data of the surface defective weld spots in the welding area are collected by the installed Internet of Things device, and the image processing is performed to obtain the processed image data of the defective weld spots, which includes the following steps: S41, filtering and denoising the collected image data of surface defective weld spots in the welding area; Filter and reduce noise on 7×7 pixels around the target pixel neighborhood; Sort the grayscale values ​​of 49 pixels in the 7×7 neighborhood and process them using the filtering and noise reduction formula; The filtering noise reduction formula is as follows: Where W represents the total number of filtering times, represents the jth guided filtering, F represents the defective solder joint image data after filtering, and I represents the gray value of the pixel; S42, performing image segmentation on the filtered defective solder joint image data.

7. The method for detecting the quality of solder joints of integrated circuit boards based on spectral analysis according to claim 6, characterized in that: The image segmentation of the filtered defective solder joint image data comprises the following steps: Binarization is performed on the filtered defective solder joint image data to obtain binarized image data; Select the initial grayscale threshold r to divide all pixels in the image data into two categories C1 and C2; Set C1 to be the pixel category that is less than or equal to the grayscale threshold r, and C2 to be the pixel category that is greater than the grayscale threshold r; Set the grayscale mean of pixel category C1 to h1, the grayscale mean of pixel category C2 to h2, and the global grayscale mean to h3; Assume that the probability that a pixel in the image data belongs to pixel category C1 is p1, and the probability that a pixel belongs to pixel category C2 is p2; The binarization formula is as follows: h3=h1×p1+h2×p2; Wherein, η represents the binarization threshold; The grayscale values ​​greater than the binarization threshold are set to 255, and the grayscale values ​​less than or equal to the binarization threshold are set to 0; Summarize the image data after binarization processing to obtain the image data after binarization processing; Perform image segmentation on the binary processed image data to obtain a welding spot image; Record the grayscale values ​​of the eight neighborhoods around each pixel in the binary processed image data. When the grayscale values ​​of the eight neighborhoods around the pixel are the same as the grayscale value of the center point, the pixel is set to be inside the welding spot image and the pixel is deleted. When the grayscale values ​​of the eight neighborhoods around a pixel point are different from the grayscale value of the center point, the pixel point is set at the edge of the welding point image and the pixel point is retained; Traverse and summarize each pixel point in the binary processed image data to obtain a welding point image; The obtained solder joint image is set as the processed defective solder joint image data.

8. The method for detecting the quality of solder joints of integrated circuit boards based on spectral analysis according to claim 1, characterized in that: The feature extraction of the processed defective solder joint image data comprises the following steps: S51, dividing the processed defective solder joint image data into 16×16 image data blocks, and inputting the divided image data blocks into a convolutional neural network; S52, extracting features from the divided image data blocks through a convolutional neural network; S521, inputting the divided image data blocks into the input layer of the neural network; S522, after receiving the divided image data blocks, the input layer transmits the received image data blocks to the convolution layer, and the convolution layer extracts local features in the image data blocks through convolution operations; The convolution operation formula is as follows: y = λ × ω + θ; Among them, λ represents the input image data block, ω represents the weight of the corresponding convolution kernel, θ represents the bias value, and y represents the output feature; S523, after the convolution layer extracts the local features in the image data block, the local features in the extracted image data block are processed by the pooling layer, and the features are downsampled by the pooling layer to reduce the data dimension; S524, through continuous convolution and pooling until the extracted features converge, the convolution stops, and the extracted features are summarized and input into the fully connected layer; S525, the fully connected layer integrates the extracted features and outputs the final feature extraction results.

9. The method for detecting the quality of solder joints of integrated circuit boards based on spectral analysis according to claim 1, characterized in that: The method of summarizing the solder joint quality detection method and the solder joint surface defect detection method to construct a solder joint detection model, and detecting the quality of the solder joints of the integrated circuit board in real time based on the constructed solder joint detection model includes the following steps: The solder joint quality inspection method is used to determine whether the solder joint has defects. When the solder joint quality inspection is completed, the solder joint surface image data is collected in real time and feature extraction is performed. The real-time extracted features are compared with the saved features. When the real-time extracted features are consistent with the saved features, it means that the current solder joint has surface defects, otherwise it is normal.

10. A system for implementing the integrated circuit board solder joint quality detection method based on spectral analysis as described in any one of claims 1 to 9, characterized in that: include: Data acquisition module, spectral data processing module, solder joint quality defect assessment module, solder joint feature extraction module and solder joint surface defect assessment module; The data acquisition module is used to collect solder spot spectrum data and solder spot surface image data in real time; The spectral data processing module is used to process the collected spectral data and generate a spectral data curve; The solder joint quality defect assessment module is used to assess solder joint quality defects by combining the spectral data curve and the heat conduction method; The solder joint feature extraction module is used to extract features from solder joint surface image data collected in real time; The solder joint surface defect evaluation module is used to evaluate the solder joint surface defects according to the feature extraction result.

Images