A defect prediction method and prediction system for solder paste printing process

Abstract

The present invention discloses a defect prediction method and system for solder paste printing process, which relates to the field of defect detection technology, including: obtaining the solder paste printing process flow of SMT products, using historical solder paste printing defect examples to trace the process, and segmenting the solder paste printing process flow based on the defect process traceability results; obtaining the types of solder paste printing defects that occur frequently in each solder paste printing process flow sub-segment, screening the printing parameters related to the solder paste printing defect type to generate key influencing factors; extracting the key influencing factors corresponding to the normal printing parameter samples to construct a normal printing behavior model; importing the real-time printing parameters into the model for estimation, and predicting the solder paste printing defects according to the deviation between the estimated printing parameters and the real-time printing parameters. The present invention mines the key influencing parameters of defects from SMT production data, constructs a normal printing behavior model to predict solder paste printing defects in advance, and improves the quality of solder paste printing.

Description

Technical Field

[0001] The present invention relates to the field of defect detection technology, and more specifically, to a defect prediction method and prediction system for a solder paste printing process. Background Art

[0002] Surface Mount Technology (SMT) is widely used in the electronic assembly industry. It mainly includes three process flows: solder paste printing, component mounting and reflow soldering. At present, the layout of printed circuit boards (PCBs) is becoming more and more complex and sophisticated, which brings great challenges to SMT solder paste printing. Solder paste printing is affected by many factors, and the printing quality is difficult to guarantee. With the intelligentization and digitization of SMT production line processing equipment, a large amount of data generated during SMT production can be automatically collected. Extracting useful information from the data to predict brush defects in advance is an effective way to improve the quality of solder paste printing.

[0003] At present, SMT production lines use a large number of high-density integrated circuit components intensively. If the printing parameters are not set properly or for other reasons, printing defects such as pull tips and insufficient solder will often occur during the production process. The appearance of printing defects requires cleaning the PCB, which will affect production efficiency and cause loss of manpower and material resources. In addition, since solder paste printing is affected by many factors, it is difficult to set printing parameters. Currently, SMT production lines basically set printing parameters through repeated experiments and adjustments, and parameter setting takes a long time. Since the solder paste printing process is complex and affected by many factors, and there is a correlation between the factors, how to screen the key influencing factors of important solder paste printing quality, build a defect prediction model based on the key influencing factors, and improve printing quality are problems that need to be solved. Summary of the invention

[0004] In order to solve the above technical problems, the present invention proposes a defect prediction method and prediction system for the solder paste printing process, with the aim of screening out the influencing factors of solder paste printing defects, establishing a defect prediction model in combination with a deep learning method, realizing early prediction of defects, and improving the production efficiency and yield rate of solder paste printing.

[0005] A first aspect of the present invention provides a defect prediction method for a solder paste printing process, comprising the following steps:

[0006] Obtain the process requirements of the SMT product, obtain the solder paste printing process flow according to the process requirements, use the solder paste printing defect example to perform process traceability, and divide the solder paste printing process flow into sections based on the defect process traceability results;

[0007] Obtain the most common solder paste printing defect types in each solder paste printing process sub-segment, read the solder paste printing parameter combination of each solder paste printing process sub-segment, and screen the key influencing factors of solder paste printing defects based on the most common solder paste printing defect types and solder paste printing parameter combinations;

[0008] Extracting corresponding normal printing parameter samples according to the key influencing factors to construct a normal printing behavior model for each solder paste printing process sub-segment;

[0009] The real-time printing parameters of the SMT product are obtained, the real-time printing parameters are imported into the normal printing behavior model of the corresponding solder paste printing process sub-segment for estimation, and the solder paste printing defects are predicted according to the deviation between the estimated printing parameters and the real-time printing parameters.

[0010] In this solution, the solder paste printing process flow is obtained according to the process requirements, and the process is traced using historical solder paste printing defect examples. Based on the defect process traceability results, the solder paste printing process flow is divided into sections, specifically:

[0011] Obtain the process requirements of SMT products, retrieve the solder paste printing parameter combination that meets the process requirements according to the process requirements, use historical solder paste printing data to extract keywords, use solder paste printing keywords to build a solder paste printing process architecture, and add description information of different process steps;

[0012] Dividing the solder paste printing parameter combinations according to the solder paste printing process architecture to obtain the solder paste printing process after data filling;

[0013] Retrieve and obtain solder paste printing defect instances, analyze causal data of the solder paste printing defect instances, calculate similarity with description information of different process steps based on the causal data, and determine the process steps corresponding to the solder paste printing defect instances according to the similarity;

[0014] Obtain defect process traceability results according to the process steps and solder paste printing parameters corresponding to the solder paste printing defect instances, and use the defect process traceability results to divide the solder paste printing defect instances into different process step labels;

[0015] The data volume of solder paste printing instances under different process step labels is obtained, and process step labels with a data volume threshold less than a preset value are screened for marking. The marked process step labels are compared with the solder paste printing defect types contained in the adjacent process step labels for similarity. When the similarity meets the preset standard, they are divided into the same solder paste printing process flow sub-segment.

[0016] In this solution, the most common solder paste printing defect types in each solder paste printing process sub-segment are obtained, and the solder paste printing parameter combinations of each solder paste printing process sub-segment are read, specifically:

[0017] Obtain process step labels contained in each solder paste printing process sub-segment after segmentation, integrate corresponding solder paste printing defect instances based on the process step labels, count the occurrence frequencies of different solder paste printing defect types, and select a preset number of solder paste printing defect types as high-incidence solder paste printing defect types;

[0018] The solder paste printing parameter combination corresponding to each solder paste printing process flow sub-segment is read, wherein the solder paste printing parameter combination includes PCB characteristic parameters, printing process parameters, printing equipment parameters, printing state parameters and printing environment parameters, and the solder paste printing parameter combination is preprocessed.

[0019] In this solution, the key influencing factors of solder paste printing defects are screened based on the types of common solder paste printing defects and the combination of solder paste printing parameters, specifically:

[0020] The parameters in the solder paste printing parameter combination are used as features, and the Pearson correlation coefficient is used to calculate the correlation degree of each parameter feature. The solder paste printing parameters are pre-screened according to the correlation degree, and the pre-screened parameter features are feature encoded, and the partitions are randomly combined to generate the initial feature combination;

[0021] The improved gray wolf algorithm is used to optimize the feature combination, the gray wolf population and algorithm parameters are initialized based on the initial feature combination, an objective function is constructed with the goal of minimizing the number of feature selections, minimizing the prediction error rate and minimizing the feature redundancy, and the objective function is associated with the fitness function to calculate the fitness of the gray wolf;

[0022] The three best gray wolves are selected based on fitness, and the positions and algorithm parameters of the gray wolves are updated in the iterative calculation steps of surrounding, hunting and attacking prey. The gray wolves before and after the position update are evenly crossed to generate a new gray wolf population;

[0023] The fitness of the gray wolves in the new gray wolf population is calculated, and a mutation operator that changes with the number of iterations is introduced to mutate the first leader gray wolf with the highest fitness, and the position of the gray wolves in the wolf pack is continuously updated to select the best three gray wolves;

[0024] When the number of iterations meets the preset conditions, the position of the first leading gray wolf with the highest fitness in the last iteration is output to obtain the best feature combination.

[0025] In this solution, the corresponding normal printing parameter samples are extracted according to the key influencing factors to construct the normal printing behavior model of each solder paste printing process sub-segment, specifically:

[0026] Obtain key influencing factors of solder paste printing defects in each solder paste printing process sub-segment, obtain historical solder paste printing data of SMT products without printing defects, use the key influencing factors to extract parameters in the historical solder paste printing data, and use the extracted parameters to construct a normal printing parameter sample;

[0027] A normal printing behavior model is constructed according to a noise reduction self-encoding structure, and the normal printing parameter samples are used for model training. Gaussian noise is added to the input normal printing parameter samples in the normal printing behavior model, and a multi-layer perceptron is used for encoding;

[0028] The encoded data is imported into the LSTM network for time series feature extraction, and the time series features are decoded and dimensionally upgraded using a multi-layer perceptron to obtain reconstructed data of normal printing parameter samples, and the reconstruction error of the normal printing behavior model is obtained through iterative training.

[0029] In this solution, the real-time printing parameters of the SMT product are obtained, and the real-time printing parameters are imported into the normal printing behavior model of the corresponding solder paste printing process sub-segment for estimation, specifically:

[0030] Collect real-time printing parameters during the solder paste printing process of SMT products, determine the corresponding solder paste printing process sub-segment according to the real-time printing parameters, and call the key influencing factors and normal printing behavior model of the corresponding solder paste printing process sub-segment;

[0031] Model input data is extracted according to the key influencing factors, and the model input data is reconstructed using a normal printing behavior model to obtain an estimated value of the printing parameters under normal operation of the solder paste printing defect.

[0032] In this solution, solder paste printing defects are predicted based on the deviation between the estimated printing parameters and the real-time printing parameters, specifically:

[0033] Calculate the residual value between the estimated value output by the normal printing behavior model and the input data, use a sliding window to perform average filtering on the residual value, combine the filtered residual value with the time series to generate a residual change curve, and fit the reconstruction error change curve according to the reconstruction error of the normal printing behavior model;

[0034] The dynamic time warping distance between the residual change curve and the reconstructed error change curve is obtained, and when the average dynamic time warping distance is greater than a preset distance threshold, solder paste printing defect warning information is generated.

[0035] A second aspect of the present invention provides a defect prediction system for a solder paste printing process, the system comprising a solder paste printing defect analysis unit, a real-time printing parameter acquisition unit, a solder paste printing defect prediction unit and a defect early warning unit;

[0036] The solder paste printing defect analysis unit divides the solder paste printing process of the SMT product into sections, obtains the solder paste printing parameter combinations and the types of high-incidence solder paste printing defects in each solder paste printing process sub-section, and screens the key influencing factors of the solder paste printing defects;

[0037] The real-time printing parameter collection unit collects real-time printing parameters in the SMT product solder paste printing process according to key influencing factors corresponding to different solder paste printing process sub-segments;

[0038] The solder paste printing defect prediction unit extracts corresponding normal printing parameter samples using key influencing factors of each solder paste printing process sub-segment to construct a normal printing behavior model, imports the real-time printing parameters of each solder paste printing process sub-segment into the model for estimation, and predicts solder paste printing defects according to the deviation between the estimated printing parameters and the real-time printing parameters;

[0039] When predicting the occurrence of solder paste printing defects, the defect warning unit generates defect warning information and sends and displays it to the operation and maintenance personnel in a preset manner.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] The present invention uses the printing parameters collected during the solder paste printing process to screen out the influencing factors of solder paste printing defects, and combines the deep learning method to establish a defect prediction model to achieve early prediction of defects and improve the production efficiency and yield rate of solder paste printing.

[0042] Based on the traditional denoising autoencoder, the present invention introduces the LSTM network to construct a normal printing behavior model for multivariate spatiotemporal state monitoring. The addition of time series features improves the network's reconstruction ability. In the construction of the normal printing behavior model, the strategy of training data partitioning is introduced to divide the solder paste printing process into segments, and different segments are modeled separately to establish a data-specific model. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions in the embodiments or exemplary embodiments of the present invention, the drawings required for use in the embodiments or exemplary descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained according to the drawings without paying creative work.

[0044] Figure 1 A flow chart showing a defect prediction method for a solder paste printing process;

[0045] Figure 2 A flow chart showing the key influencing factors for screening solder paste printing defects is shown;

[0046] Figure 3 A flowchart for constructing a normal printing behavior model for a solder paste printing process sub-segment is shown;

[0047] Figure 4 A block diagram of a defect prediction system for solder paste printing process is shown. DETAILED DESCRIPTION

[0048] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0049] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0050] like Figure 1 As shown, in a first embodiment of the present invention, a defect prediction method for a solder paste printing process is provided, comprising:

[0051] S102, obtaining process requirements of SMT products, obtaining a solder paste printing process flow according to the process requirements, performing process tracing using a solder paste printing defect example, and dividing the solder paste printing process flow into sections based on defect process tracing results;

[0052] S104, obtaining the most common solder paste printing defect types in each solder paste printing process sub-segment, and reading the solder paste printing parameter combination of each solder paste printing process sub-segment, and screening the key influencing factors of the solder paste printing defects based on the most common solder paste printing defect types and the solder paste printing parameter combination;

[0053] S106, extracting corresponding normal printing parameter samples according to the key influencing factors to construct a normal printing behavior model for each solder paste printing process sub-segment;

[0054] S108, obtaining real-time printing parameters of the SMT product, importing the real-time printing parameters into the normal printing behavior model of the corresponding solder paste printing process sub-segment for estimation, and predicting solder paste printing defects according to the deviation between the estimated printing parameters and the real-time printing parameters.

[0055] It should be noted that, in order to obtain the process requirements of SMT products, the solder paste printing parameter combinations that meet the process requirements are retrieved from the historical solder paste printing data of the SMT products according to the process requirements, keywords are extracted using the historical solder paste printing data, and the solder paste printing process architecture is constructed using solder paste printing keywords and historical printing process timing, and descriptive information of different process steps is added; the solder paste printing parameter combinations are divided according to the solder paste printing process architecture to obtain the solder paste printing process flow after data filling.

[0056] Retrieve and obtain solder paste printing defect instances, perform semantic analysis after preprocessing the solder paste printing defect instances, obtain the cause data of the solder paste printing defect instances, calculate the similarity with the description information of different process steps based on the cause data, and determine the process flow steps corresponding to the solder paste printing defect instances according to the similarity; obtain defect process traceability results according to the process flow steps corresponding to the solder paste printing defect instances and the solder paste printing parameters of the process flow steps in the instances, and use the defect process traceability results to divide the solder paste printing defect instances into different process step labels; obtain the data volume of the solder paste printing instances under different process step labels, select the process step labels that are less than the preset data volume threshold for marking, and compare the marked process step labels with the solder paste printing defect types contained in the adjacent process step labels for similarity. When the similarity meets the preset standard, they are divided into the same solder paste printing process flow sub-segment. Segment the solder paste printing process into different segments and model them separately, establish a data-targeted model, and ensure that the subsequent models remain effective and accurate in the ever-changing actual environment.

[0057] Obtain the process step labels contained in each solder paste printing process sub-segment after segmentation, integrate the corresponding solder paste printing defect instances based on the process step labels, count the occurrence frequencies of different solder paste printing defect types, and select a preset number of solder paste printing defect types as high-incidence solder paste printing defect types; read the solder paste printing parameter combination corresponding to each solder paste printing process sub-segment, the solder paste printing parameter combination includes PCB characteristic parameters, printing process parameters, printing equipment parameters, printing state parameters and printing environment parameters, PCB characteristic parameters mainly include PCB size parameters, and parameters related to packaging on PCB; printing process parameters are mainly parameters set by the printer during solder paste printing, etc.; printing equipment parameters mainly include parameters of equipment and materials such as steel mesh and solder paste used during solder paste printing; printing state parameters mainly include data obtained by real-time detection during solder paste printing, such as real-time pressure of scraper, etc.; printing environment parameters mainly include ambient temperature and humidity data. Perform preprocessing such as missing value completion, windowing, outlier removal and data standardization on the solder paste printing parameter combination. Through data preprocessing, the efficiency and accuracy of solder paste printing defect prediction are improved.

[0058] Figure 2 A flow chart for screening key influencing factors of solder paste printing defects is shown.

[0059] According to an embodiment of the present invention, key influencing factors of solder paste printing defects are screened based on the types of common solder paste printing defects and solder paste printing parameter combinations, specifically:

[0060] S202, taking the parameters in the solder paste printing parameter combination as features, using the Pearson correlation coefficient to calculate the correlation degree of each parameter feature, pre-screening the solder paste printing parameters according to the correlation degree, feature encoding the pre-screened parameter features, and randomly combining the partitions to generate an initial feature combination;

[0061] S204, optimizing the feature combination using an improved gray wolf algorithm, initializing the gray wolf population and algorithm parameters based on the initial feature combination, constructing an objective function with the goal of minimizing the number of feature selections, minimizing the prediction error rate, and minimizing feature redundancy, and associating the objective function with a fitness function to calculate the fitness of the gray wolf;

[0062] S206, selecting the best three gray wolves according to fitness, updating the positions and algorithm parameters of the gray wolves in the steps of encircling, hunting and attacking prey in iterative calculation, performing uniform crossover of the gray wolves before and after the position update, and generating a new gray wolf population;

[0063] S208, calculating the fitness of the gray wolves in the new gray wolf population, and introducing a mutation operator that changes with the number of iterations to mutate the first leader gray wolf with the highest fitness, continuing to update the positions of the gray wolves in the wolf pack, and selecting the best three gray wolves;

[0064] S210, when the number of iterations meets the preset conditions, the position of the first leading gray wolf with the highest fitness in the last iteration is output to obtain the best feature combination.

[0065] It should be noted that the gray wolf algorithm divides gray wolf individuals into four social levels: alpha (𝛼), beta (𝛽), delta (𝛿), and omega (𝜔). Based on the hierarchy, 𝛼, 𝛽, and 𝛿 are regarded as the best leaders, and the remaining gray wolf individuals are named 𝜔 wolves. 𝜔 wolves will follow the leaders towards promising areas to complete the predation process. The hunting process of gray wolves mainly includes three steps: encirclement, hunting, and attacking prey. In each iteration, 𝜔 wolf completes the hunt based on the location information of the three best wolves 𝛼, 𝛽, and 𝛿.

[0066] Initialize feature combinations to minimize the number of feature selections Minimize prediction error rate and minimize feature redundancy For the purpose, represents the number of selected features, represents the total number of features, represents false negative examples, false positive examples, true negative examples, and true positive examples, Indicates selected features The Pearson correlation coefficient of the three objectives is integrated using linear weighting to construct the objective function, and the gray wolf population and algorithm parameters are initialized. The algorithm parameters include the coordination coefficient and convergence factor in encirclement, hunting and attacking prey.

[0067] The gray wolves before and after the position update are uniformly crossed, and the fitness of the gray wolf population individuals is sorted by the objective function. The gray wolf individuals with the largest fitness are uniformly crossed with the gray wolf individuals with the smallest fitness, and the order is executed in sequence, which improves the diversity of the gray wolf population. In addition, a mutation operator that is updated with the number of iterations is introduced. As the number of iterations increases, the linear growth of the mutation operator promotes The increased probability of variation in wolf feature dimensions increases the possibility of new feature combinations, which can avoid the occurrence of local optimal situations to a certain extent.

[0068] Figure 3 A flowchart for constructing a normal printing behavior model for a sub-segment of a solder paste printing process flow is shown.

[0069] According to an embodiment of the present invention, a normal printing behavior model of each solder paste printing process sub-segment is constructed by extracting corresponding normal printing parameter samples according to the key influencing factors, specifically:

[0070] S302, obtaining key influencing factors of solder paste printing defects in each solder paste printing process sub-segment, obtaining historical solder paste printing data of SMT products without printing defects, extracting parameters from the historical solder paste printing data using the key influencing factors, and constructing a normal printing parameter sample using the extracted parameters;

[0071] S304, constructing a normal printing behavior model according to the noise reduction self-encoding structure, using the normal printing parameter samples to perform model training, adding Gaussian noise to the input normal printing parameter samples in the normal printing behavior model, and encoding them using a multi-layer perceptron;

[0072] S306, importing the encoded data into the LSTM network to extract time series features, using a multi-layer perceptron to decode and upgrade the time series features, obtaining reconstructed data of normal printing parameter samples, and obtaining the reconstruction error of the normal printing behavior model through iterative training.

[0073] It should be noted that the traditional denoising autoencoder network structure is improved to obtain the dynamic time series characteristics of different printing parameters in the time series data. The normal printing behavior model includes a noise addition module, a multi-layer perceptron module, and an LSTM autoencoder network module. Gaussian noise is added to the input data so that the autoencoder network can learn more valuable data for data encoding. The input data after adding noise is reduced in dimension using a multi-layer perceptron, and the nonlinear characteristics between different printing parameters are obtained. The LSTM autoencoder network is used to obtain the time series characteristics after dimensionality reduction, and the time series characteristics are used to optimize the encoding reconstruction and reduce the reconstruction error. Finally, the multi-layer perceptron is used for decoding to realize the data reconstruction of the input data. The root mean square error, mean absolute error and goodness of fit are selected as evaluation indicators to obtain the reconstruction error of the normal printing behavior model.

[0074] It should be noted that the real-time printing parameters in the solder paste printing process of SMT products are collected, the corresponding solder paste printing process sub-segment is determined according to the real-time printing parameters, and the key influencing factors and normal printing behavior model of the corresponding solder paste printing process sub-segment are called; the model input data is extracted according to the key influencing factors, and the model input data is reconstructed using the normal printing behavior model to obtain the estimated value of the printing parameters under normal operation without solder paste printing defects. Preferably, a probability distribution is constructed based on the obtained estimated value, and it is determined whether the estimated value probability distribution obeys the probability distribution composed of the reconstructed data of the normal printing parameter sample, and it is inferred whether solder paste printing defects will occur.

[0075] Calculate the residual value between the estimated value output by the normal printing behavior model and the input data, use a sliding window to filter the residual value by the average value, solve the problem of frequent alarms caused by frequent fluctuations in residual data, combine the filtered residual value with the time series to generate a residual change curve, and reconstruct the error change curve according to the reconstruction error of the normal printing behavior model; obtain the dynamic time warping distance between the residual change curve and the reconstructed error change curve, and generate solder paste printing defect warning information when the average dynamic time warping distance is greater than the preset distance threshold. The dynamic time warping algorithm can find the best alignment between the corresponding data points from the two curves, thereby generating a minimum Euclidean distance alignment sequence between the two curves, so that the time scale can be nonlinearly warped by stretching, and the obtained Euclidean distance is used to measure the similarity of the two curves. Use the dynamic time warping distance to characterize the degree of deviation between the printing parameters of the real-time solder paste printing of SMT products and the normal printing parameters. When the degree of deviation is too large, it indicates that there is an abnormality.

[0076] like Figure 4As shown, the second embodiment of the present invention provides a defect prediction system for a solder paste printing process, the system comprising a solder paste printing defect analysis unit 401, a real-time printing parameter acquisition unit 402, a solder paste printing defect prediction unit 403 and a defect warning unit 404;

[0077] The solder paste printing defect analysis unit divides the solder paste printing process of the SMT product into sections, obtains the solder paste printing parameter combinations and the types of high-incidence solder paste printing defects in each solder paste printing process sub-section, and screens the key influencing factors of the solder paste printing defects;

[0078] The real-time printing parameter collection unit collects real-time printing parameters in the SMT product solder paste printing process according to key influencing factors corresponding to different solder paste printing process sub-segments;

[0079] The solder paste printing defect prediction unit extracts corresponding normal printing parameter samples using key influencing factors of each solder paste printing process sub-segment to construct a normal printing behavior model, imports the real-time printing parameters of each solder paste printing process sub-segment into the model for estimation, and predicts solder paste printing defects according to the deviation between the estimated printing parameters and the real-time printing parameters;

[0080] When predicting the occurrence of solder paste printing defects, the defect warning unit generates defect warning information and sends and displays it to the operation and maintenance personnel in a preset manner.

[0081] The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored, or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms. In addition, the functional units in the embodiments of the present invention can be all integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the above-mentioned integrated units can be implemented in the form of hardware or in the form of hardware plus software functional units.

[0082] Those skilled in the art can understand that all or part of the steps of implementing the above method embodiments can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiments; and the aforementioned storage medium includes: mobile storage devices, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), disks or optical disks, and other media that can store program codes.

[0083] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A defect prediction method for solder paste printing process, characterized in that: The following steps are involved: Obtain the process requirements of the SMT product, obtain the solder paste printing process flow according to the process requirements, use the solder paste printing defect example to perform process traceability, and divide the solder paste printing process flow into sections based on the defect process traceability results; Obtain the most common solder paste printing defect types in each solder paste printing process sub-segment, read the solder paste printing parameter combination of each solder paste printing process sub-segment, and screen the key influencing factors of solder paste printing defects based on the most common solder paste printing defect types and solder paste printing parameter combinations; Extracting corresponding normal printing parameter samples according to the key influencing factors to construct a normal printing behavior model for each solder paste printing process sub-segment; Acquire real-time printing parameters of SMT products, import the real-time printing parameters into a normal printing behavior model of a corresponding solder paste printing process sub-segment for estimation, and predict solder paste printing defects according to deviations between the estimated printing parameters and the real-time printing parameters; The solder paste printing process flow is obtained according to the process requirements, and the process is traced using historical solder paste printing defect examples. The solder paste printing process flow is divided into sections based on the defect process traceability results, specifically: Obtain the process requirements of SMT products, retrieve the solder paste printing parameter combination that meets the process requirements according to the process requirements, use historical solder paste printing data to extract keywords, use solder paste printing keywords to build a solder paste printing process architecture, and add description information of different process steps; Dividing the solder paste printing parameter combinations according to the solder paste printing process architecture to obtain the solder paste printing process after data filling; Retrieve and obtain solder paste printing defect instances, analyze causal data of the solder paste printing defect instances, calculate similarity with description information of different process steps based on the causal data, and determine the process steps corresponding to the solder paste printing defect instances according to the similarity; Obtain defect process traceability results according to the process steps and solder paste printing parameters corresponding to the solder paste printing defect instances, and use the defect process traceability results to divide the solder paste printing defect instances into different process step labels; The data volume of solder paste printing instances under different process step labels is obtained, and process step labels with a data volume threshold less than a preset value are screened for marking. The marked process step labels are compared with the solder paste printing defect types contained in the adjacent process step labels for similarity. When the similarity meets the preset standard, they are divided into the same solder paste printing process flow sub-segment.

2. A defect prediction method for solder paste printing process according to claim 1, characterized in that: Obtain the most common solder paste printing defect types in each solder paste printing process sub-segment, and read the solder paste printing parameter combination of each solder paste printing process sub-segment, specifically: Obtain process step labels contained in each solder paste printing process sub-segment after segmentation, integrate corresponding solder paste printing defect instances based on the process step labels, count the occurrence frequencies of different solder paste printing defect types, and select a preset number of solder paste printing defect types as high-incidence solder paste printing defect types; The solder paste printing parameter combination corresponding to each solder paste printing process flow sub-segment is read, wherein the solder paste printing parameter combination includes PCB characteristic parameters, printing process parameters, printing equipment parameters, printing state parameters and printing environment parameters, and the solder paste printing parameter combination is preprocessed.

3. The defect prediction method for solder paste printing process according to claim 1, characterized in that: Based on the types of common solder paste printing defects and the combination of solder paste printing parameters, the key influencing factors of solder paste printing defects are screened, specifically: The parameters in the solder paste printing parameter combination are used as features, and the Pearson correlation coefficient is used to calculate the correlation degree of each parameter feature. The solder paste printing parameters are pre-screened according to the correlation degree, and the pre-screened parameter features are feature encoded, and the partitions are randomly combined to generate the initial feature combination; The improved gray wolf algorithm is used to optimize the feature combination, the gray wolf population and algorithm parameters are initialized based on the initial feature combination, an objective function is constructed with the goal of minimizing the number of feature selections, minimizing the prediction error rate and minimizing the feature redundancy, and the objective function is associated with the fitness function to calculate the fitness of the gray wolf; The three best gray wolves are selected based on fitness, and the positions and algorithm parameters of the gray wolves are updated in the iterative calculation steps of surrounding, hunting and attacking prey. The gray wolves before and after the position update are evenly crossed to generate a new gray wolf population; The fitness of the gray wolves in the new gray wolf population is calculated, and a mutation operator that changes with the number of iterations is introduced to mutate the first leader gray wolf with the highest fitness, and the position of the gray wolves in the wolf pack is continuously updated to select the best three gray wolves; When the number of iterations meets the preset conditions, the position of the first leading gray wolf with the highest fitness in the last iteration is output to obtain the best feature combination.

4. A defect prediction method for solder paste printing process according to claim 1, characterized in that: According to the key influencing factors, the corresponding normal printing parameter samples are extracted to construct the normal printing behavior model of each solder paste printing process sub-segment, specifically: Obtain key influencing factors of solder paste printing defects in each solder paste printing process sub-segment, obtain historical solder paste printing data of SMT products without printing defects, use the key influencing factors to extract parameters in the historical solder paste printing data, and use the extracted parameters to construct a normal printing parameter sample; A normal printing behavior model is constructed according to a noise reduction self-encoding structure, and the normal printing parameter samples are used for model training. Gaussian noise is added to the input normal printing parameter samples in the normal printing behavior model, and a multi-layer perceptron is used for encoding; The encoded data is imported into the LSTM network for time series feature extraction, and the time series features are decoded and dimensionally upgraded using a multi-layer perceptron to obtain reconstructed data of normal printing parameter samples, and the reconstruction error of the normal printing behavior model is obtained through iterative training.

5. The defect prediction method for solder paste printing process according to claim 1, characterized in that: The real-time printing parameters of the SMT product are obtained, and the real-time printing parameters are imported into the normal printing behavior model of the corresponding solder paste printing process sub-segment for estimation, specifically: Collect real-time printing parameters during the solder paste printing process of SMT products, determine the corresponding solder paste printing process sub-segment according to the real-time printing parameters, and call the key influencing factors and normal printing behavior model of the corresponding solder paste printing process sub-segment; Model input data is extracted according to the key influencing factors, and the model input data is reconstructed using a normal printing behavior model to obtain an estimated value of the printing parameters under normal operation of the solder paste printing defect.

6. A defect prediction method for solder paste printing process according to claim 1, characterized in that: Predict solder paste printing defects based on the deviation between the estimated printing parameters and the real-time printing parameters, specifically: Calculate the residual value between the estimated value output by the normal printing behavior model and the input data, use a sliding window to perform average filtering on the residual value, combine the filtered residual value with the time series to generate a residual change curve, and fit the reconstruction error change curve according to the reconstruction error of the normal printing behavior model; The dynamic time warping distance between the residual change curve and the reconstructed error change curve is obtained, and when the average dynamic time warping distance is greater than a preset distance threshold, solder paste printing defect warning information is generated.

7. A defect prediction system for solder paste printing process, characterized in that: Implementing the defect prediction method for solder paste printing process as described in any one of claims 1 to 6, the system comprises a solder paste printing defect analysis unit, a real-time printing parameter acquisition unit, a solder paste printing defect prediction unit and a defect warning unit; The solder paste printing defect analysis unit divides the solder paste printing process of the SMT product into sections, obtains the solder paste printing parameter combinations and the types of high-incidence solder paste printing defects in each solder paste printing process sub-section, and screens the key influencing factors of the solder paste printing defects; The real-time printing parameter collection unit collects real-time printing parameters in the SMT product solder paste printing process according to key influencing factors corresponding to different solder paste printing process sub-segments; The solder paste printing defect prediction unit extracts corresponding normal printing parameter samples using key influencing factors of each solder paste printing process sub-segment to construct a normal printing behavior model, imports the real-time printing parameters of each solder paste printing process sub-segment into the model for estimation, and predicts solder paste printing defects according to the deviation between the estimated printing parameters and the real-time printing parameters; When predicting the occurrence of solder paste printing defects, the defect warning unit generates defect warning information and sends and displays it to the operation and maintenance personnel in a preset manner.

Images