CN117007940A - Self-adaptive method for detecting performance of circuit board - Google Patents
Self-adaptive method for detecting performance of circuit board Download PDFInfo
- Publication number
- CN117007940A CN117007940A CN202310966174.2A CN202310966174A CN117007940A CN 117007940 A CN117007940 A CN 117007940A CN 202310966174 A CN202310966174 A CN 202310966174A CN 117007940 A CN117007940 A CN 117007940A
- Authority
- CN
- China
- Prior art keywords
- circuit board
- detection
- printed circuit
- point
- adaptive
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 41
- 238000001514 detection method Methods 0.000 claims abstract description 102
- 239000000523 sample Substances 0.000 claims abstract description 42
- 238000012360 testing method Methods 0.000 claims abstract description 39
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 29
- 238000012545 processing Methods 0.000 claims abstract description 26
- 238000013135 deep learning Methods 0.000 claims abstract description 12
- 238000010586 diagram Methods 0.000 claims abstract description 7
- 230000003044 adaptive effect Effects 0.000 claims description 15
- 238000004458 analytical method Methods 0.000 claims description 8
- 239000013598 vector Substances 0.000 claims description 8
- 238000001914 filtration Methods 0.000 claims description 7
- 238000007637 random forest analysis Methods 0.000 claims description 7
- 238000003466 welding Methods 0.000 claims description 7
- 238000006073 displacement reaction Methods 0.000 claims description 6
- 238000012549 training Methods 0.000 claims description 5
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 claims description 4
- 238000003066 decision tree Methods 0.000 claims description 4
- 230000006870 function Effects 0.000 claims description 4
- 238000012937 correction Methods 0.000 claims description 3
- 238000012544 monitoring process Methods 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 claims description 2
- 238000006243 chemical reaction Methods 0.000 claims description 2
- 239000011159 matrix material Substances 0.000 claims description 2
- 238000010606 normalization Methods 0.000 claims description 2
- 238000013461 design Methods 0.000 abstract 1
- 238000012546 transfer Methods 0.000 abstract 1
- 238000011161 development Methods 0.000 description 3
- 230000018109 developmental process Effects 0.000 description 3
- 230000007547 defect Effects 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 230000011218 segmentation Effects 0.000 description 2
- 238000010998 test method Methods 0.000 description 2
- 241000668842 Lepidosaphes gloverii Species 0.000 description 1
- 238000005299 abrasion Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- WABPQHHGFIMREM-UHFFFAOYSA-N lead(0) Chemical compound [Pb] WABPQHHGFIMREM-UHFFFAOYSA-N 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000001360 synchronised effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/28—Testing of electronic circuits, e.g. by signal tracer
- G01R31/2801—Testing of printed circuits, backplanes, motherboards, hybrid circuits or carriers for multichip packages [MCP]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/70—Determining position or orientation of objects or cameras
- G06T7/73—Determining position or orientation of objects or cameras using feature-based methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
- G06T2207/30141—Printed circuit board [PCB]
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Data Mining & Analysis (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Computing Systems (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Quality & Reliability (AREA)
- Computer Hardware Design (AREA)
- Microelectronics & Electronic Packaging (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a self-adaptive method for detecting the performance of a circuit board, which is characterized in that a printed circuit board diagram is led into an upper computer, a to-be-detected printed circuit board is placed at the same time, based on a deep learning network and a self-adaptive image processing algorithm, the upper computer automatically identifies the optimal coordinates of test points of the printed circuit board (based on the original point and auxiliary point positions on the printed circuit board diagram), according to detection requirements, the information transfer type at the test points is operated on the upper computer, one or more detection probes are controlled to move to the pointing test point positions, the power-on and multi-point signal detection of the circuit board are completed, meanwhile, the upper computer can store test data, and according to requirements, whether signal output meets design requirements is judged. The invention improves the efficiency and accuracy of circuit board detection by using the self-adaptive identification method, and the method can meet the requirements of testing various different printed circuit boards and testing multiple test points simultaneously, and improves the testing speed, the testing precision and the applicability of the testing method.
Description
Technical Field
The invention relates to the field of circuit board performance detection, in particular to a self-adaptive method for circuit board performance detection.
Background
Along with the continuous development of science and technology, the application of integrated circuits is wider, and because of the development and use of circuit units such as a singlechip, an FPGA and the like, the functions of the circuits are diversified and complicated, the pins of the components of the circuits are extremely dense, and new requirements are put forward for circuit testing. The original test lead wire test mode consumes a great deal of manpower and material resources, has lower precision and has very low adaptability to different circuit boards.
Aiming at the problems of the original testing method, the international method generally adopts a boundary scan test (JTAG test), and the method can realize digital, digital-analog hybrid circuit and digital network test, is mainly suitable for chip test and is suitable for application of large-scale and ultra-large-scale integrated circuits. The JTAG test is characterized in that a boundary scanning unit formed by a shift register is added between a chip pin and an internal logic circuit, so that the setting and reading of the state of the chip pin are realized, and the pin has controllability and observability. The test method mainly uses a boundary scan test algorithm to identify the pins, namely a test vector set generation algorithm, wherein the selection of the test vector set directly influences the output of a boundary scan test result, and the capability of the test vector set for detecting faults is called fault resolution. The main generation algorithm is based on a test vector generation algorithm of Serial Test Vectors (STVs), a binary counting algorithm (CSA algorithm), an adaptive test algorithm and the like. According to different algorithms, the performance index of the circuit board can be rapidly judged. The self-adaptive algorithm is used for deducing possible faults according to the prior experience, and adding the test vector to accurately judge the type and the position of the faults, so that the self-adaptive algorithm can improve the stability and the recognition speed of the system, and has high adaptability compared with STV and CSA algorithms.
However, the JTAG test method can only identify and detect the chip, but cannot detect various performances of the circuit board, such as on-off condition of a certain circuit, voltage boosting capability of a certain circuit, and the like. Aiming at the problems at home and abroad, various recognition algorithms are provided, and mainly comprise a circuit board recognition mode based on deep learning, a segmentation detection algorithm, a boundary scanning algorithm, a self-adaptive recognition algorithm and the like, wherein the circuit recognition mode based on the deep learning has better development due to the advantages of wide analysis range, strong capability and the like, but has the characteristics of weak self-adaptive performance, poor robustness, poor recognition and avoidance performance on defects such as dirt, damage and the like on the surface of the circuit board and the like. Although the self-adaptive algorithm solves the defect of deep learning to a certain extent, the existing self-adaptive algorithm is not applied to the performance detection direction of the circuit board in an overall good way, the self-adaptive algorithm for accurately identifying the to-be-detected point is not available, and the precision of self-adaptive identification is greatly reduced if the breaking degree of the break point is too large when the boundary of the circuit board is connected. In addition, the identification algorithm is not matched with a test platform for use, and has a gap from actual detection, so that the identification and detection of various and complex circuit boards can not be realized.
Disclosure of Invention
The invention provides a self-adaptive method for detecting the performance of a circuit board, which improves the efficiency and the accuracy of the detection of the circuit board.
The technical scheme for realizing the invention is as follows: an adaptive method for detecting the performance of a circuit board comprises the following steps:
step 1, leading a printed circuit board diagram into an upper computer, placing a printed circuit board to be tested on a test platform, detecting an origin point and an auxiliary point of the printed circuit board to be tested by the test platform, judging whether the printed circuit board is placed accurately, and obtaining that the origin point coordinate is (x) if the origin point coordinate is accurate t0 ,y t0 ) And (2) switching to step 2.
And step 2, utilizing an industrial lens with a light source to collect a photo of the printed circuit board to be tested, and turning to step 3.
And step 3, after gray processing is carried out on the picture of the printed circuit board to be tested, the image is segmented based on the deep learning network, so that a picture only containing the information of the printed circuit board is obtained, and the step 4 is carried out.
And 4, performing on-line detection and identification on the picture only containing the information of the printed circuit board by utilizing a self-adaptive image identification algorithm to obtain the position coordinate (x) of the to-be-detected point p1 ,y p1 ),(x p2 ,y p2 ),...,(x pn ,y pn ) N is the number of points to be measured; go to step 5.
And 5, selecting at least one point to be detected as a detection point according to requirements, wherein each detection point corresponds to one detection probe, respectively transmitting network and coordinate position information at the detection point by using an upper computer, controlling the detection probes to move to the positions of the appointed detection points, and switching to the step 6.
And 6, judging whether the detection probe is at the optimal position again, designing an iterative compensation algorithm to finely adjust the position of the detection probe until the position reaches the optimal position, and turning to the step 7.
And 7, supplying power to the circuit board, monitoring the current detection point signal, simultaneously storing the detection point signal by using the upper computer, and judging whether the output signal meets the designed performance requirement according to the normal signal characteristics set in advance in the upper computer.
Compared with the prior art, the invention has the remarkable advantages that:
(1) The method for identifying the origin and the auxiliary points of the printed circuit board is adopted to judge the placement position of the circuit board, so that the speed of position determination is improved, the method is suitable for various printed circuit boards, and the adaptability of the overall detection mode is improved.
(2) The original image is segmented through the deep learning network, the network is optimized, the cross loss is reduced, the boundary recognition capability is improved, and the problem that the target welding spot cannot be recognized with high precision when the breakpoint is encountered in the self-adaptive recognition is avoided.
(3) The method for identifying the to-be-detected point by the self-adaptive identification algorithm is provided, the optimal position of the actual probe is analyzed in an iterative mode, the accuracy of the probe is improved, the robustness of the overall detection mode is improved, and the measurement error rate is reduced; by means of the image processing method, the detection probe is automatically controlled to reach the optimal detection position, the accuracy of the probe is improved, the measurement error rate is reduced, the detection flow is greatly reduced, and the detection efficiency is improved.
(4) The multi-detection-point synchronous detection is realized, the operation is simple, the detection of various multifunctional circuit boards can be satisfied, and the method is suitable for the performance detection of the circuit boards in each period.
(5) The method realizes batch detection of the circuit boards, signal analysis and fault display, greatly improves the speed and precision of detection of the circuit boards, and reduces the detection cost.
Drawings
FIG. 1 is a schematic diagram of a test platform.
Fig. 2 is a top view of the placement platform of fig. 1 with a printed circuit board to be tested placed thereon.
Fig. 3 is a flow chart of segmentation of an image by a deep learning network.
Fig. 4 is a flow chart of random forest method data processing.
Fig. 5 is a flow chart of an adaptive method for circuit board performance detection.
FIG. 6 is a flow chart for adaptive position fine tuning.
Detailed Description
The invention is described in further detail below with reference to the accompanying drawings.
Referring to fig. 1 to 6, the self-adaptive method for detecting performance of a circuit board according to the present invention comprises the following steps:
step 1, leading a printed circuit board diagram into an upper computer 1, placing a printed circuit board to be tested on a test platform 3, fixing the printed circuit board by a circuit board fixture 4, wherein the detection probes 2 are all positioned at the edge of equipment when not started, the test platform 3 detects an origin 5 and an auxiliary point 6 of the printed circuit board to be tested, judging whether the printed circuit board is accurately placed, and if so, obtaining that the origin coordinate is (x) t0 ,y t0 )。
And 2, acquiring a photo of the printed circuit board to be tested by using the industrial lens with the light source, wherein the photo comprises the circuit board to be tested and a background.
Step 3, after gray processing is performed on the photo of the printed circuit board to be tested, the image is segmented based on a deep learning network, as shown in fig. 3, to obtain a picture only containing information of the printed circuit board, wherein the deep learning network comprises a deep convolution network with residual blocks and a feature network, and specifically comprises the following steps:
step 3-1, a deep convolution network with residual blocks is built, the photo of the printed circuit board to be tested obtained in the step 2 is sent into the network, convolution kernels are divided into three groups of 1×1 and 2×2, 2×2 and 5×5, 5×5 and 7×7 according to the sizes, the method improves the difference recognition capability, the 1×1 and 2×2 can recognize local space and small scale features, the 2×2 and 5×5 can better recognize middle scale and long scale dependency relationships, the 5×5 and 7×7 improve wide recognition and dependency relationships, and the classification enhances the overall recognition capability. Obtaining feature graphs with different scales by using a residual network and a feature pyramid mode:
wherein f i For the characteristic differences of different groups, kappa and beta are training parameters, and Deltax i For each set of differences Δσ i For each group of variance differences ΔE i [x]Epsilon and gamma are both constant for each group of expected differences. And optimizing training parameters, increasing variance difference and expected difference, and improving scale feature precision.
Step 3-2, feature images with different scales are transmitted into candidate areas to generate a feature network, and the specific process is as follows: scanning on the feature map by means of a sliding window, outputting a fixed number of feature networks for each location. And matching the candidate frames in each feature network with known positive and negative samples, and carrying out candidate frame regression prediction to obtain frame position and size information:
L' 1 ＝ω 1 l 1 +λ 1 L 1
L' 2 ＝ω 2 l 2 +λ 2 L 2
wherein L' 1 、L' 2 For the length and width of the adjusted bounding box, l 1 、l 2 Omega is the original length and width 1 、ω 2 As weight, lambda 1 、λ 2 L is the crossover loss 1 、L 2 Is the length and width of the bounding box; weight and cross loss are added, the feature degree of the boundary frame is increased, boundary distinguishing is facilitated, and the method is suitable for complex boundary recognition; and transmitting the updated frame position and size information back to the feature network.
And 3-3, classifying and dividing targets in the feature network to obtain pictures only containing the information of the printed circuit board.
Step 4, processing the picture only containing the information of the printed circuit board by utilizing an adaptive image processing algorithm to obtain the position coordinates (x p1 ,y p1 ),(x p2 ,y p2 ),...,(x pn ,y pn ) The method is characterized by comprising the following steps:
and 4-1, performing texture processing on the picture only containing the information of the printed circuit board by adopting a gray level co-occurrence matrix, wherein the corresponding frequency accumulated value of the point to be detected is GLCM (p (x) 1n ,y 1n ),p(x 1n +dx 1n ,y 1n +dy 1n ) To obtain the position coordinates (x) of the point to be measured after the first analysis 11 ,y 11 ),(x 12 ,y 12 ),...,(x 1n ,y 1n ) N represents the number of points to be measured, p (x) 1n ,y 1n )、p(x 1n +dx 1n ,y 1n +dy 1n ) Is a pixel value, (dx) 1n ,dy 1n ) Representing the amount of offset in the relative position between two pixels in a two-dimensional image.
Step 4-2, recognizing welding spots by using a random forest, and normalizing the existing data by combining the processing of the welding spots in step 4-1 to obtain a normalized feature vector x' n 、y' n So that the features are at the same scale, namely:
(x' 1n ,y' 1n )＝((x 1n -mean(x 1n ))/std(x 1n ),y 1n -mean(y 1n ))/std(y 1n ))
wherein, by (x) 1n ,y 1n ) Processing is exemplified by mean (x 1n ) Is the mean (y) 1n ) Std (x) 1n ) Std (y 1n ) The coordinate precision can be improved by using the normalized coordinates as the ordinate standard deviation.
As shown in fig. 4, a random forest model is established, and feature subsets are randomly selected for training when each node is split by constructing a plurality of decision trees, so that the generalization capability and the robustness of the algorithm are improved; according to the voting results and average results of all decision trees, obtaining the position coordinates (x 21 ,y 21 ),(x 22 ,y 22 ),...,(x 2n ,y 2n )。
Step 4-3, extracting the existing template with the characteristic quantity calculated by the random forest, carrying out Gaussian processing, carrying out convolution matching identification G (x, y) with the template which is not subjected to Gaussian processing, and carrying out convolution operation on a Gaussian kernel and the template to obtain a convolution result G ij Wherein G is ij The (i, j) th element being G (x, y); and combining non-average filtering to obtain a corresponding Gaussian kernel function:
wherein, (p-k) 2 +(q-l) 2 H is the square of the Euclidean distance, xi is the bandwidth parameter, Z (x, y) is the normalization factor, P (P, q), P (x, y) is the local block pixel value. The non-average filtering is added on the original Gaussian kernel function, the overall adaptability of the algorithm is improved, the detail characteristics of the image are improved while the noise is reduced, the smooth parameters are multiplied by the standard deviation, and the robustness of the smooth parameters is improved.
According to the operation result and Gaussian kernel weight, obtaining the position coordinates (x 31 ,y 31 ),(x 32 ,y 32 ),...,(x 3n ,y 3n )。
Step 4-4, after filtering and denoising the existing template, weighting the center of the denoising template, wherein the center weighting coefficient W (i, j) is as follows:
W(i,j)＝exp{-(i 2 +j 2 ) 1/2 }/(2τ 2 )
wherein, (i, j) is the center position, and τ is the standard deviation parameter of the weight distribution. And carrying out standard deviation parameter processing on the basis of the original center weighting coefficient, and adjusting the whole weight range and the intensity.
The noise filtering image is self-adaptively processed, and the processing correction method is as follows:
t (i, j, T) is the template currently used, O (i, j, T) is the best matching position sub-image of the current frame,and (3) for adaptively correcting values, improving the adaptive capacity of image processing, wherein T (i, j, t+1) is information of a predicted next frame template, and alpha is a weighting coefficient.
The change rate of the correlation tracking confidence coefficient along with analysis and matching degree is obtained in the correlation tracking process, and in the identification, the pixel value distribution of a neighborhood block of the welding spot position to be tested is obtained through calculation to determine the threshold value eta of the test point 0 Thereafter, a correlation tracking confidence C is calculated 0 The solving process is as follows:
R max the coordinate position (x) of the point to be detected after the p-th analysis is obtained by self-adaptive processing of the image for the optimal matching degree number of the current frame p1 ,y p1 ),(x p2 ,y p2 ),...,(x pn ,y pn ) I.e. the optimal position of the point to be measured.
Step 5, selecting at least one point to be detected as a detection point according to the requirement, wherein each detection point corresponds to one detection probe 2, and utilizing the upper positionThe machine 1 transmits network and coordinate position information at the detection point respectively, and controls the detection probe 2 to move to the position of the appointed detection point in the following moving mode: before the detection probe 2 moves, the upper computer 1 needs to compare the x-axis movement displacement between a plurality of detection points, and the detection probe 2 is moved to the corresponding (x pm ,y 10 ) Wherein, the angle mark of the optimal coordinate position of the point to be measured after the p-th self-adaptive processing is represented, m is the m-th detection point, y 10 The origin y-axis coordinate of the printed circuit board is; the detection probe 2 is moved along the y-axis, and finally, the z-axis direction is moved until the detection point position is reached. The method avoids the mutual collision of the detection probes 2 in the moving process, and finally moves in the z-axis direction, thereby effectively avoiding the abrasion and damage of the detection probes 2 to the circuit board.
Step 6, judging whether the detection probe 2 is at the optimal position again: when t detection probes 2 reach the specified detection point (x p1 ,y p1 ),(x p2 ,y p2 ),...,(x pt ,y pt ) Then, the detection probe 2 further identifies and determines the current position, the image of the current position is transmitted back to the upper computer 1, the upper computer 1 compares the current position data with the original position data, and the residual delta of the current position and the original position data is calculated n For residual delta n Normal conversion is carried out, and an adaptive threshold value eta 'is calculated' n And judging whether to update the data coordinates, obtaining a movement displacement difference value, transmitting the movement displacement difference value back to the detection probe 2, and performing fine adjustment on the existing coordinates until the optimal position is reached.
And 7, supplying power to the circuit board, monitoring the current detection point signal, simultaneously storing the detection point signal by using the upper computer 1, and judging whether the output signal meets the designed performance requirement according to the normal signal characteristics set in advance in the upper computer 1.
Claims (8)
1. An adaptive method for detecting the performance of a circuit board is characterized by comprising the following steps:
step 1, leading the printed circuit board diagram into an upper computer (1), placing the printed circuit board to be tested on a test platform, and detecting the original printed circuit board to be tested by the test platformThe point (5) and the auxiliary point (6) are used for judging whether the printed circuit board is accurately placed, and if so, the origin coordinate is (x) t0 ,y t0 ) Turning to step 2;
step 2, utilizing an industrial lens with a light source to collect a photo of the printed circuit board to be tested, and turning to step 3;
step 3, after gray processing is carried out on the picture of the printed circuit board to be tested, the image is segmented based on the deep learning network, so that a picture only containing the information of the printed circuit board is obtained, and the step 4 is carried out;
and 4, performing on-line detection and identification on the picture only containing the information of the printed circuit board by utilizing a self-adaptive image identification algorithm to obtain the position coordinate (x) of the to-be-detected point p1 ,y p1 ),(x p2 ,y p2 ),...,(x pn ,y pn ) N is the number of points to be measured; turning to step 5;
step 5, selecting at least one point to be detected as a detection point according to requirements, wherein each detection point corresponds to one detection probe (2), respectively transmitting network and coordinate position information at the detection point by using an upper computer (1), controlling the detection probes (2) to move to the position of the appointed detection point, and turning to step 6;
step 6, judging whether the detection probe (2) is at the optimal position again, designing an iterative compensation algorithm to finely adjust the position of the detection probe (2) until the optimal position is reached, and turning to step 7;
and 7, supplying power to the circuit board, monitoring the current detection point signal, simultaneously storing the detection point signal by using the upper computer (1), and judging whether the output signal meets the designed performance requirement according to the normal signal characteristics set in advance in the upper computer (1).
2. The self-adaptive method for detecting the performance of the circuit board according to claim 1, wherein in the step 1, a detection probe device is added on a test platform, the detection probe device adopts a plurality of detection probes (2), and each detection probe (2) can independently carry out position adjustment;
the positions of an origin (5) and an auxiliary point (6) on the printed circuit board diagram are processed by an image recognition methodDetecting, judging whether the printed circuit board is correctly placed, and if so, obtaining the origin coordinate of the printed circuit board as (x) t0 ,y t0 )。
3. The adaptive method for circuit board performance detection of claim 1, wherein: in step 3, after gray processing is performed on the photo of the printed circuit board to be tested, the image is segmented based on a deep learning network to obtain a picture only containing the information of the printed circuit board, and the deep learning network comprises a deep convolution network with residual blocks and a feature network, and specifically comprises the following steps:
step 3-1, a deep convolution network with residual blocks is built, the photo of the printed circuit board to be tested is sent into the network, and convolution kernels are divided into three groups according to the size and are respectively: 1×1 and 2×2, 2×2 and 5×5, 5×5 and 7×7; obtaining feature graphs with different scales by using a residual network and a feature pyramid mode:
wherein f i For the characteristic differences of different groups, kappa and beta are training parameters, and Deltax i For each set of differences Δσ i For each group of variance differences ΔE i [x]Epsilon and gamma are constant for each group of expected differences;
step 3-2, feature images with different scales are transmitted into candidate areas to generate a feature network, and the specific process is as follows:
scanning on the feature map by utilizing a sliding window mode, outputting a fixed number of feature networks for each position, and carrying out candidate frame regression prediction on candidate frames in each feature network by matching with known positive and negative samples to obtain frame position and size information:
L' 1 ＝ω 1 l 1 +λ 1 L 1
L' 2 ＝ω 2 l 2 +λ 2 L 2
wherein L' 1 、L' 2 To adjustLength and width of the subsequent bounding box, l 1 、l 2 Omega is the original length and width 1 、ω 2 As weight, lambda 1 、λ 2 L is the crossover loss 1 、L 2 Is the length and width of the bounding box; transmitting the updated frame position and size information back to the feature network;
and 3-3, classifying and dividing targets in the feature network to obtain pictures only containing the information of the printed circuit board.
4. The adaptive method for circuit board performance detection of claim 1, wherein: in step 4, the image only containing the information of the printed circuit board is processed by using the self-adaptive image processing algorithm to obtain the position coordinates (x p1 ,y p1 ),(x p2 ,y p2 ),...,(x pn ,y pn ) The method is characterized by comprising the following steps:
step 4-1, texture processing is carried out on the picture only containing the information of the printed circuit board by adopting the gray level co-occurrence matrix, and the corresponding frequency accumulated value of the point to be detected is GLCM (p (x) 1n ,y 1n ),p(x 1n +dx 1n ,y 1n +dy 1n ) To obtain the position coordinates (x) of the point to be measured after the first analysis 11 ,y 11 ),(x 12 ,y 12 ),...,(x 1n ,y 1n ) N represents the number of points to be measured, p (x) 1n ,y 1n )、p(x 1n +dx 1n ,y 1n +dy 1n ) Are pixel values, (dx) 1n ,dy 1n ) An offset representing a relative position between two pixels in the two-dimensional image;
step 4-2, recognizing welding spots by using a random forest, and normalizing the existing data by combining the processing of the welding spots in step 4-1 to obtain a normalized feature vector x' n 、y′ n So that the features are at the same scale, namely:
(x' 1n ,y' 1n )＝((x 1n -mean(x 1n ))/std(x 1n ),y 1n -mean(y 1n ))/std(y 1n ))
wherein the method comprises the steps of，(x' 1n ,y' 1n ) To normalize the coordinates, mean (x 1n ) Is the mean (y) 1n ) Std (x) 1n ) Std (y 1n ) Is the ordinate standard deviation;
establishing a random forest model, and randomly selecting a feature subset for training by constructing a plurality of decision trees when each node is split; according to the voting results and average results of all decision trees, obtaining the position coordinates (x 21 ,y 21 ),(x 22 ,y 22 ),...,(x 2n ,y 2n )；
Step 4-3, extracting the existing template with the characteristic quantity calculated by the random forest, carrying out Gaussian processing, carrying out convolution matching identification G (x, y) with the template which is not subjected to Gaussian processing, and carrying out convolution operation on a Gaussian kernel and the template to obtain a convolution result G ij Wherein G is ij The (i, j) th element being G (x, y); and combining non-average filtering to obtain a corresponding Gaussian kernel function:
wherein, (p-k) 2 +(q-l) 2 H is the square of Euclidean distance, xi is the bandwidth parameter, Z (x, y) is the normalization factor, P (P, q) and P (x, y) are local block pixel values;
according to the operation result and Gaussian kernel weight, obtaining the position coordinates (x 31 ,y 31 ),(x 32 ,y 32 ),...,(x 3n ,y 3n )；
Step 4-4, after filtering and denoising the existing template, weighting the center of the denoising template, wherein the center weighting coefficient W (i, j) is as follows:
W(i,j)＝exp{-(i 2 +j 2 ) 1/2 }/(2τ 2 )
wherein, (i, j) is a central position coordinate, and τ is a standard deviation parameter of weight distribution;
the noise filtering image is self-adaptively processed, and the processing correction method is as follows:
t (i, j, T) is the template currently used, O (i, j, T) is the best matching position sub-image of the current frame,for the self-adaptive correction value, T (i, j, t+1) is the information of the next frame template obtained by prediction, and alpha is a weighting coefficient;
the change rate of the correlation tracking confidence coefficient along with analysis and matching degree is obtained in the correlation tracking process, and in the identification, the pixel value distribution of a neighborhood block of the welding spot position to be tested is obtained through calculation to determine the threshold value eta of the test point 0 Thereafter, a correlation tracking confidence C is calculated 0 The solving process is as follows:
R max the coordinate position (x) of the point to be detected after the p-th analysis is obtained by self-adaptive processing of the image for the optimal matching degree number of the current frame p1 ,y p1 ),(x p2 ,y p2 ),...,(x pn ,y pn ) I.e. the optimal position of the point to be measured.
5. The adaptive method for circuit board performance detection of claim 1, wherein: in step 5, at least one point to be detected is selected as a detection point according to requirements, information at the detection point is respectively transmitted by using the upper computer (1), each detection point corresponds to one detection probe (2), and the detection probes (2) are controlled to move to the position of the appointed detection point, specifically as follows:
after selecting detection points according to requirements, controlling the detection probes (2) to move to the positions corresponding to the detection points in the mode of x-axis first, y-axis second and z-axis direction last until the detection points are reached.
6. The adaptive method for circuit board performance detection of claim 5, wherein: before the detection probes (2) move, the upper computer (1) needs to compare the displacement of the x-axis movement among a plurality of detection points, and the detection probes (2) move to the corresponding (x) according to the sequence from small to large according to the x-axis coordinates pm ,y 10 ) Wherein, the angle mark of the optimal coordinate position of the point to be measured after the p-th self-adaptive processing is represented, m is the m-th detection point, y 10 The origin y-axis coordinate of the printed circuit board is; and then the detection probe (2) is moved along the y axis, so that the movement efficiency is improved, and the collision of the detection probe (2) is avoided.
7. The adaptive method for circuit board performance detection of claim 1, wherein: in step 6, it is again determined whether the detection probe (2) is at the optimal position, if not, the coordinate information is updated, and the coordinates of the detection probe (2) are finely adjusted until the optimal position is reached, specifically as follows:
when t detection probes (2) reach a specified detection point (x) p1 ,y p1 ),(x p2 ,y p2 ),...,(x pt ,y pt ) Then, the detection probe (2) further identifies and determines the current position, the image of the current position is transmitted back to the upper computer (1), the upper computer (1) compares the current position data with the original position data, and the residual delta of the current position and the original position data is calculated n For residual delta n Normal conversion is carried out, and an adaptive threshold value eta 'is calculated' n And judging whether to update the data coordinates, obtaining a movement displacement difference value, transmitting the movement displacement difference value back to the detection probe (2), and performing fine adjustment on the existing coordinates.
8. The adaptive method for circuit board performance detection of claim 7, wherein: after the coordinates of the x axis and the y axis are adjusted, the detection probe (2) moves in the z axis and contacts with the detection point.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310966174.2A CN117007940A (en) | 2023-08-02 | 2023-08-02 | Self-adaptive method for detecting performance of circuit board |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310966174.2A CN117007940A (en) | 2023-08-02 | 2023-08-02 | Self-adaptive method for detecting performance of circuit board |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117007940A true CN117007940A (en) | 2023-11-07 |
Family
ID=88572218
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310966174.2A Pending CN117007940A (en) | 2023-08-02 | 2023-08-02 | Self-adaptive method for detecting performance of circuit board |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117007940A (en) |
-
2023
- 2023-08-02 CN CN202310966174.2A patent/CN117007940A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109977808B (en) | Wafer surface defect mode detection and analysis method | |
| US9075026B2 (en) | Defect inspection device and defect inspection method | |
| CN107992898B (en) | A kind of electronic component welding point defect detection method | |
| CN113763312B (en) | Detection of defects in semiconductor samples using weak labels | |
| CN109919908B (en) | Method and device for detecting defects of light-emitting diode chip | |
| KR20200116926A (en) | In-depth learning based inspection method of semiconductor specimen and its system | |
| CN112053318A (en) | Two-dimensional PCB defect real-time automatic detection and classification device based on deep learning | |
| CN114155181B (en) | Automatic optimization of inspection schemes | |
| CN108369915B (en) | Reduce the noise examined in the bare die caused near registration and design | |
| CN111815564B (en) | Method and device for detecting silk ingots and silk ingot sorting system | |
| CN113627457A (en) | Method and system for classifying defects in wafer using wafer defect image based on deep learning | |
| JP2021515885A (en) | Methods, devices, systems and programs for setting lighting conditions and storage media | |
| KR20220012217A (en) | Machine Learning-Based Classification of Defects in Semiconductor Specimens | |
| CN115719338A (en) | PCB (printed circuit board) surface defect detection method based on improved YOLOv5 | |
| JPH1074812A (en) | Inspecting method of pattern to be inspected, diagnosing method of manufacturing process and manufacturing method of semiconductor wafer | |
| CN114910480A (en) | Wafer surface defect detection method based on machine vision | |
| KR20220014805A (en) | Generating training data usable for examination of a semiconductor specimen | |
| KR20240004240A (en) | Deep generative model-based alignment for semiconductor applications | |
| CN109919215B (en) | Target detection method for improving characteristic pyramid network based on clustering algorithm | |
| Wang et al. | Local defect detection and print quality assessment | |
| CN111582332B (en) | Picture identification method for high-speed railway contact net dropper component | |
| CN117455917A (en) | Establishment of false alarm library of etched lead frame and false alarm on-line judging and screening method | |
| CN117274258A (en) | Method, system, equipment and storage medium for detecting defects of main board image | |
| CN117007940A (en) | Self-adaptive method for detecting performance of circuit board | |
| CN113012121B (en) | Method and device for processing bare chip scanning result, electronic equipment and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |