A circuit board quality anomaly prediction system based on machine intelligence learning

Abstract

The present invention relates to the technical field of circuit board quality monitoring, specifically a circuit board quality anomaly prediction system based on machine intelligent learning. The system comprises an environmental parameter simulation and analysis module that simulates the impact of environmental parameters on material properties based on circuit board production data. Through iterative calculations, the system simulates the changes in parameters over time and generates dynamic environmental parameter simulation results. Through the environmental parameter simulation and analysis module, the system can simulate and track in detail the subtle changes in key environmental parameters such as temperature, humidity, and material properties over time, and their impact on circuit board quality. By utilizing a time trend prediction and key factor identification module, the system not only improves the accuracy of predicting future environmental parameter change trends, but also accurately identifies the environmental parameters that have the greatest impact on circuit board quality, thereby achieving targeted quality control and optimization.

Description

Technical Field

[0001] The present invention relates to the technical field of circuit board quality monitoring, and in particular to a circuit board quality anomaly prediction system based on machine intelligent learning. Background Art

[0002] The field of circuit board quality monitoring technology is used to identify and prevent defects and failures that may occur during the manufacturing and use of circuit boards. This field involves the use of various inspection and analysis methods, such as visual inspection, automated optical inspection, X-ray inspection, and electrical testing, to ensure that circuit boards meet expected standards in terms of performance and reliability.

[0003] The PCB quality anomaly prediction system is a key component of this technology. It leverages data analysis and machine learning to predict potential quality issues during PCB production. Early identification of potential quality issues reduces scrap and production costs, while improving product reliability. Furthermore, by analyzing production data, it identifies patterns and trends that lead to defects, enabling preventive measures to be taken before problems occur.

[0004] Although existing technologies have achieved the basic ability to identify and prevent possible defects and failures during the circuit board manufacturing process, their analysis depth and prediction accuracy are insufficient in capturing quality changes related to time sensitivity, such as slight changes in parameters such as temperature, humidity and material properties and their long-term impact on product quality. Therefore, it is difficult to fully capture subtle changes that evolve over time, which may eventually accumulate into significant quality differences. In addition, although existing technologies can use certain machine learning methods to analyze production data, they lack sensitivity to subtle abnormal features and accuracy in distinguishing different types of abnormal patterns in terms of the ability to automatically learn complex similarity relationships between data and efficiently distinguish different types of abnormal patterns, which affects the efficiency of early warning and prevention of potential quality problems. Summary of the Invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and propose a circuit board quality anomaly prediction system based on machine intelligent learning.

[0006] In order to achieve the above objectives, the present invention adopts the following technical solution: a circuit board quality abnormality prediction system based on machine intelligence learning, the system comprising:

[0007] The environmental parameter simulation and analysis module simulates the impact of environmental parameters on material properties based on circuit board production data. Through iterative calculations, it simulates the changes of parameters over time and generates dynamic simulation results of environmental parameters.

[0008] The time trend prediction module analyzes the seasonality and trend components of the time series data based on the dynamic simulation results of the environmental parameters, predicts future changes in environmental parameters, and generates time trend prediction results;

[0009] The key factor identification module simulates the impact of each environmental parameter on the quality of the circuit board based on the time trend prediction result, identifies the environmental parameters that affect the quality by recording the changes in the quality of the circuit board each time, and generates a key influencing factor identification result;

[0010] The circuit board production process modeling module defines key variables and parameters based on the key influencing factor identification results, establishes a cause-and-effect relationship diagram, simulates the feedback loop and nonlinear behavior in the production process, and obtains a circuit board production process dynamics model;

[0011] The abnormal pattern recognition module constructs multiple decision trees based on the circuit board production process dynamics model to train the data and summarize the prediction results of each tree, captures the abnormal patterns in the circuit board production process, and generates abnormal pattern recognition results;

[0012] The similarity measurement and analysis module constructs a twin network model and performs similarity measurement based on the abnormal pattern recognition results. By defining the model structure and adding layers, it extracts and fuses features, analyzes the similarity between production data and known abnormal patterns, and generates similarity measurement analysis results.

[0013] The quality prediction result integration and analysis module iteratively trains the weak learner to gradually reduce the prediction error based on the dynamic simulation results of the environmental parameters, the time trend prediction results, the key influencing factor identification results, the circuit board production process dynamics model, the abnormal pattern recognition results and the similarity measurement analysis results, and performs a detailed analysis of the circuit board quality status to generate circuit board quality prediction and analysis results.

[0014] The present invention has improvements in that the dynamic simulation results of environmental parameters include simulated temperature change data, humidity level adjustment records and circuit board material property response diagrams; the time trend prediction results include circuit board temperature trend prediction diagrams, humidity change prediction information and analysis of future material property change trends; the key influencing factor identification results include the impact of temperature on circuit board quality, humidity impact score and sensitivity analysis of material properties to quality; the circuit board production process dynamics model includes circuit board production speed variable analysis, raw material quality feedback loop diagram and processing temperature nonlinear response results; the abnormal pattern recognition results include circuit board welding temperature abnormality types, humidity control failure warnings and material adverse reaction classifications; the similarity measurement analysis results include known abnormal pattern similarity scores, potential quality problem similar pattern matching and risk level determination; the circuit board quality prediction and analysis results include quality risk level classification, key improvement area indications and potential defect prevention measures.

[0015] The present invention is improved in that the environmental parameter simulation and analysis module includes:

[0016] The parameter tracking submodule tracks and records environmental parameters in real time based on PCB production data, plots parameter changes over time, and generates a time series graph of environmental parameters.

[0017] The parameter dynamic simulation submodule uses the fourth-order Runge-Kutta method based on the environmental parameter time series diagram to simulate the influence of the parameters on the circuit board material properties and their changes over time, and generates parameter dynamic simulation data;

[0018] The fourth-order Runge-Kutta method is based on the improved formula I:

[0019]

[0020] Simulate the impact of parameters on circuit board material properties and their changes over time, and generate parameter dynamic simulation data;

[0021] Where k1 = f(t n ,y n ), k4=f(t n +Δt,y n +k3Δt), k env =g(temp,hum,pres,chem),y n+1 Refers to the predicted value of the environmental parameter for the next time step,

[0022] y nis the environmental parameter value of the current time step, k2 is the slope estimate after the influence of k1 at half of the time step, k3 is the slope estimate after using k2 to adjust again at half of the time step, k4 is the slope estimate after using k3 to adjust after the entire time step is completed, Δt is the time step, k env is the comprehensive slope estimation of the impact of environmental factors on the material properties of the circuit board. α is the weight coefficient determined based on historical data analysis, which is obtained by analyzing the relationship between historical data and the impact of environmental changes on the quality of the circuit board. Temp represents the ambient temperature, hum represents the ambient humidity, pres represents the ambient pressure, chem represents the concentration of chemical substances in the environment, and g is a function used to calculate the comprehensive impact of environmental factors, with k as the weight coefficient. env Form of expression;

[0023] The environmental impact assessment submodule analyzes the impact of changes in environmental parameters on the quality of the circuit board based on the parameter dynamic simulation data, determines the strength of the relationship between each environmental parameter and the quality of the circuit board, and generates environmental parameter dynamic simulation results.

[0024] The present invention is improved in that the time trend prediction module includes:

[0025] The sequence parsing submodule sets and processes the data based on the dynamic simulation results of the environmental parameters, decomposes the time series data into trend, seasonality and residual components, identifies the periodic changes and basic trends of the data, and generates seasonal decomposition and trend identification results;

[0026] The trend forecast analysis submodule performs time series forecasting based on the seasonal decomposition and trend identification results, and predicts the changing trend of future environmental parameters by setting seasonal change parameters and trend change parameters, thereby generating future environmental parameter forecast results;

[0027] The trend change prompt submodule displays the future environmental parameter trends through charts based on the future environmental parameter prediction results, identifies future risk points and opportunities, and generates time trend prediction results.

[0028] The present invention is improved in that the key factor identification module includes:

[0029] The environmental parameter influence analysis submodule performs a correlation analysis between the environmental parameters and the quality of the circuit board based on the time trend prediction results, calculates the Spearman rank correlation coefficient between the parameters and the quality, and performs a significance test of the correlation to generate an environmental parameter influence analysis result;

[0030] The environmental parameter simulation test submodule performs a simulation test on the effect of environmental parameter changes on the quality of the circuit board based on the environmental parameter influence analysis results, records the changes in each environmental parameter and its effect on the quality of the circuit board, and generates simulation test results;

[0031] The influencing parameter extraction submodule obtains the simulation test results, performs an influencing parameter extraction process, applies an F test to screen environmental parameters that have an impact on the quality of the circuit board, and generates a key influencing factor identification result.

[0032] The present invention is improved in that the circuit board production process modeling module includes:

[0033] The variable relationship mapping submodule maps the causal relationship between key variables and parameters based on the key influencing factor identification results, draws a causal loop diagram between variables, annotates variables, including the direct and indirect relationships between circuit board production speed, raw material quality and environmental conditions, and generates a causal loop diagram;

[0034] The production process simulation submodule performs dynamic simulation of feedback loops and nonlinear behaviors in the production process based on the causal loop diagram. By setting key variables and parameters, it simulates various scenarios on the PCB production line, including the impact of temperature and humidity fluctuations on PCB quality, the impact of raw material quality changes on production efficiency, and the impact of production speed adjustments on PCB stability, and generates dynamic simulation results for the production process.

[0035] The production process behavior analysis submodule obtains the dynamic simulation results of the production process, uses numerical analysis and data management technology to analyze the simulation data and identify key behavior patterns in the production process, and obtains a circuit board production process dynamics model.

[0036] The present invention is improved in that the abnormal pattern recognition module includes:

[0037] The data processing submodule performs denoising, missing value filling and normalization based on the output of the circuit board production process dynamics model to generate a standardized data set;

[0038] The pattern mining submodule constructs and trains multiple decision tree models based on the standardized data set, analyzes and learns patterns in the data indicating abnormalities in circuit board production quality, and generates mined abnormal pattern information;

[0039] The abnormal pattern classification submodule analyzes and classifies abnormal patterns based on the mined abnormal pattern information details, distinguishes each type of circuit board production abnormality, including multiple factors such as abnormal welding temperature and humidity control failure, and generates abnormal pattern recognition results.

[0040] The present invention is improved in that the similarity measurement and analysis module includes:

[0041] The feature analysis submodule analyzes the identified abnormal pattern data based on the abnormal pattern recognition results, extracts key features that represent production abnormalities, and gradually extracts detailed features in the data to generate a key feature vector set;

[0042] The model building submodule builds a twin network model based on the key feature vector set to analyze similarity, including using a multi-layer convolutional neural network for feature extraction and a fully connected layer for feature fusion and similarity analysis. It uses a ternary loss function as the training target to optimize the model's measurement of the distance between normal and abnormal patterns, and generates a twin network similarity model.

[0043] The ternary loss function is based on the improved formula II:

[0044] L(a,p,n)=max(ω1·d(a,p)-ω2·d(a,n)+margin+λ·R,0)

[0045] The optimization model measures the distance between normal and abnormal patterns and generates a twin network similarity model;

[0046] Where L represents the loss function, a represents the feature vector of the anchor sample, p represents the feature vector of the positive sample, n represents the feature vector of the negative sample, d represents the distance between the two feature vectors, margin is a hyperparameter used to control the minimum distance between the positive sample and the negative sample, ω1 and ω2 are weight coefficients for adjusting the distance between the anchor point and the positive sample and the anchor point and the negative sample, respectively. ω1 is determined by cross-validation performance evaluation on the training data, ω2 is determined by analyzing the distance distribution characteristics between the anchor sample and the negative sample and optimizing the model performance through cross-validation, λ is the weight of the regularization term, and R represents the regularization term of the model.

[0047] The similarity result analysis submodule performs similarity analysis based on the twin network similarity model, compares and analyzes the feature vectors between the circuit board production data and the known abnormal patterns, calculates the quantitative similarity score, identifies potential quality risks or abnormalities, and generates similarity measurement analysis results.

[0048] The present invention is improved in that the quality prediction result integration and analysis module includes:

[0049] The information integration submodule performs data fusion operations based on the environmental parameter dynamic simulation results, time trend prediction results, key influencing factor identification results, circuit board production process dynamics model, abnormal pattern recognition results and similarity measurement analysis results to generate a comprehensive feature data set;

[0050] The training and optimization submodule performs iterative training and model optimization operations based on the comprehensive feature data set, and generates an optimized prediction model by adjusting parameters and performing cross-validation of model performance;

[0051] The quality index analysis submodule performs detailed analysis of the quality index of the circuit board based on the optimized prediction model, analyzes the impact of each key factor on the quality of the circuit board, identifies the key factors in quality prediction, and obtains the circuit board quality prediction and analysis results.

[0052] Compared with the prior art, the advantages and positive effects of the present invention are:

[0053] In the present invention, through the environmental parameter simulation and analysis module, the system can simulate and track in detail the slight changes in key environmental parameters such as temperature, humidity and material properties over time and their impact on the quality of the circuit board. By using the time trend prediction and key factor identification module, not only the system's prediction accuracy of future environmental parameter change trends is improved, but also the environmental parameters that have the greatest impact on the quality of the circuit board can be identified, thereby achieving targeted quality control and optimization. By constructing a dynamic model of the circuit board production process and abnormal pattern recognition, the system can deeply analyze the feedback loops and nonlinear behaviors in the production process, as well as accurately capture and classify abnormal patterns in the production process. The ability to identify potential quality risks in the production process at an early stage is improved. Through the similarity measurement and analysis module, the system can efficiently learn the complex similarity relationships between data, improve sensitivity to subtle abnormal features, and accurately distinguish different types of abnormal patterns. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 The present invention proposes a module diagram of a circuit board quality anomaly prediction system based on machine intelligence learning;

[0055] Figure 2 The present invention proposes a system framework diagram of a circuit board quality anomaly prediction system based on machine intelligence learning;

[0056] Figure 3 The present invention proposes a schematic diagram of an environmental parameter simulation and analysis module in a circuit board quality anomaly prediction system based on machine intelligence learning;

[0057] Figure 4 A schematic diagram of a time trend prediction module in a circuit board quality anomaly prediction system based on machine intelligence learning is proposed in the present invention;

[0058] Figure 5 A schematic diagram of a key factor identification module in a circuit board quality anomaly prediction system based on machine intelligence learning is proposed in the present invention;

[0059] Figure 6The present invention proposes a schematic diagram of a circuit board production process modeling module in a circuit board quality anomaly prediction system based on machine intelligence learning;

[0060] Figure 7 A schematic diagram of an abnormal pattern recognition module in a circuit board quality abnormality prediction system based on machine intelligence learning is proposed in the present invention;

[0061] Figure 8 The present invention proposes a schematic diagram of a similarity measurement and analysis module in a circuit board quality anomaly prediction system based on machine intelligence learning;

[0062] Figure 9 The present invention proposes a schematic diagram of a quality prediction result integration and analysis module in a circuit board quality anomaly prediction system based on machine intelligent learning. DETAILED DESCRIPTION

[0063] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0064] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings and are only for the convenience of describing the present invention and simplifying the description. They do not indicate or imply that the devices or elements referred to must have a specific direction, be constructed and operate in a specific direction, and therefore should not be understood as limiting the present invention. In addition, in the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0065] Example

[0066] See also Figure 1 The present invention provides a technical solution: a circuit board quality abnormality prediction system based on machine intelligence learning, the system comprising:

[0067] The environmental parameter simulation and analysis module simulates the impact of environmental parameters on material properties based on circuit board production data. Through iterative calculations, it simulates the changes of parameters over time and generates dynamic simulation results of environmental parameters.

[0068] The time trend prediction module analyzes the seasonality and trend components of time series data based on the dynamic simulation results of environmental parameters, predicts future changes in environmental parameters, and generates time trend prediction results;

[0069] The key factor identification module simulates the impact of each environmental parameter on PCB quality based on the time trend prediction results. By recording the changes in PCB quality each time, it identifies the environmental parameters that affect the quality and generates key influencing factor identification results.

[0070] The PCB production process modeling module defines key variables and parameters based on the results of key influencing factor identification, establishes a cause-and-effect diagram, simulates the feedback loop and nonlinear behavior in the production process, and obtains a PCB production process dynamics model;

[0071] The abnormal pattern recognition module builds multiple decision trees based on the PCB production process dynamics model to train the data and summarize the prediction results of each tree, capturing abnormal patterns in the PCB production process and generating abnormal pattern recognition results;

[0072] The similarity measurement and analysis module builds a twin network model and performs similarity measurement based on the abnormal pattern recognition results. By defining the model structure and adding layers, it extracts and fuses features, analyzes the similarity between production data and known abnormal patterns, and generates similarity measurement analysis results.

[0073] The quality prediction result integration and analysis module is based on the dynamic simulation results of environmental parameters, time trend prediction results, key influencing factor identification results, circuit board production process dynamics model, abnormal pattern recognition results and similarity measurement analysis results. It iteratively trains weak learners to gradually reduce prediction errors, conducts detailed analysis of circuit board quality status, and generates circuit board quality prediction and analysis results.

[0074] The dynamic simulation results of environmental parameters include simulated temperature change data, humidity level adjustment records and circuit board material property response diagrams. The time trend prediction results include circuit board temperature trend prediction diagrams, humidity change prediction information and analysis of future material property change trends. The key influencing factor identification results include the impact of temperature on circuit board quality, humidity impact score and sensitivity analysis of material properties to quality. The circuit board production process dynamics model includes circuit board production speed variable analysis, raw material quality feedback loop diagram and processing temperature nonlinear response results. The abnormal pattern recognition results include circuit board welding temperature abnormality types, humidity control failure warnings and material adverse reaction classifications. The similarity measurement analysis results include known abnormal pattern similarity scores, potential quality problem similar pattern matching and risk level determination. The circuit board quality prediction and analysis results include quality risk level classification, key improvement area indications and potential defect prevention measures.

[0075] See also Figure 2 and Figure 3 , environmental parameter simulation and analysis modules include:

[0076] The parameter tracking submodule tracks and records environmental parameters in real time based on circuit board production data, plots parameter changes over time, and generates a time series graph of environmental parameters. The specific process is as follows:

[0077] The parameter tracking submodule is based on circuit board production data and adopts real-time data acquisition technology. It uses the SQL database to execute the INS ERT command to insert real-time environmental parameter data, uses the Pandas library to time-series the data, sets the timestamp as the index, and uses the Matplotlib library to draw a chart showing the changes in parameters over time to generate a time series graph of environmental parameters.

[0078] The parameter dynamic simulation submodule uses the fourth-order Runge-Kutta method based on the environmental parameter time series diagram to simulate the influence of parameters on the material properties of the circuit board and its changes over time. The specific process of generating parameter dynamic simulation data is as follows:

[0079] The parameter dynamic simulation submodule is based on the environmental parameter time series diagram and adopts the fourth-order Runge-Kutta method to define the initial conditions and step size. The iterative formula is written in Python programming language, including the calculation of the intermediate slopes k1, k2, k3, and k4. The parameter value of each step is updated according to the Runge-Kutta formula, and the NumPy library is used for numerical calculations to generate parameter dynamic simulation data.

[0080] The fourth-order Runge-Kutta method is based on the improved formula I:

[0081]

[0082] Simulate the impact of parameters on circuit board material properties and their changes over time, and generate parameter dynamic simulation data;

[0083] Where k1 = f(t n ,y n ), k4=f(t n +Δt,y n +k3Δt), k env =g(temp,hum,pres,chem),y n+1 is the predicted value of the environmental parameter for the next time step, y n is the environmental parameter value of the current time step, k2 is the slope estimate after considering the influence of k1 at half of the time step, k3 is the slope estimate after adjusting with k2 again at half of the time step, k4 is the slope estimate after adjusting with k3 after the entire time step is completed, Δt is the time step, that is, the length of time spanned by each step prediction, k envis the comprehensive slope estimation of the impact of environmental factors on the material properties of the circuit board. α is the weight coefficient determined based on historical data analysis. It is obtained by analyzing the relationship between historical data and the impact of environmental changes on the quality of the circuit board. It is used to adjust the impact of environmental factors on the simulation results. Temp represents the ambient temperature, hum represents the ambient humidity, pres represents the ambient pressure, and chem represents the concentration of chemical substances in the environment. g is a function used to calculate the comprehensive impact of environmental factors such as temperature, humidity, pressure, and chemical concentration. k env Form of expression;

[0084] The execution process is as follows:

[0085] S1: Set the initial condition y0 (initial state or characteristic value of the circuit board material), time step Δt, and environmental parameters (temperature temp, humidity hum, pressure pres, chemical concentration chem).

[0086] S2: Use the initial condition or the state value y of the current step n and time t n Calculate k1=f(t n ,y n ), f represents the function of the relationship between environmental parameters and circuit board material properties.

[0087] S3: Based on the result of k1, calculate

[0088] S4: Calculate based on the result of k2

[0089] S5: Finally, calculate k4=f(t n +Δt,y n +k3Δt).

[0090] S6: Calculate the comprehensive slope estimate k of environmental factors through the function g(temp,hum,pres,chem) env .

[0091] S7: Based on the relationship between historical data and the change of circuit board quality, the weight coefficient α is determined through data analysis methods to balance the impact of environmental factors on the simulation results.

[0092] S8: Calculate the predicted value of the environmental parameter for the next time step according to formula I.

[0093] S9: Update time step t n+1 =t n +Δt and environmental parameter value y n =y n+1 , repeat steps S2 to S8 until the calculation of the entire simulation period is completed.

[0094] The environmental impact assessment submodule analyzes the impact of changes in environmental parameters on circuit board quality based on parameter dynamic simulation data, determines the strength of the relationship between each environmental parameter and circuit board quality, and generates the specific process of environmental parameter dynamic simulation results as follows:

[0095] The environmental impact assessment submodule is based on parameter dynamic simulation data, adopts regression analysis method, uses SciPy library to execute linear regression function, sets environmental parameters as independent variables, and circuit board quality indicators as dependent variables, and calculates the determination coefficient R 2 The model fit was evaluated, and the matplotlib library was used to draw a scatter plot of the relationship strength between environmental parameters and quality indicators to generate dynamic simulation results of environmental parameters.

[0096] See also Figure 2 and Figure 4 , the time trend forecast module includes:

[0097] The sequence parsing submodule sets and processes the data based on the dynamic simulation results of environmental parameters, decomposes the time series data into trend, seasonality, and residual components, identifies the cyclical changes and basic trends of the data, and generates seasonal decomposition and trend identification results. The specific process is as follows:

[0098] The sequence parsing submodule uses time series analysis technology based on the dynamic simulation results of environmental parameters. It uses the Python statsmodels library to call the seasonal decomposition function seasonal_decompose to index and process the data, decompose the time series data into trend, seasonal, and residual components, and generate seasonal decomposition and trend identification results by calculating and analyzing the cyclical changes and basic trends of the data.

[0099] The trend forecast analysis submodule performs time series forecasting based on seasonal decomposition and trend identification results. By setting seasonal change parameters and trend change parameters, it predicts the changing trend of future environmental parameters. The specific process of generating future environmental parameter forecast results is as follows:

[0100] The trend forecast analysis submodule adopts the ARIMA model based on the seasonal decomposition and trend identification results, uses Python's statsmodels library to execute the ARIMA function, specifies the autoregressive term, the number of differences, and the moving average term parameters, sets the seasonal change parameters and the trend change parameters, and applies the model to predict the changing trends of future environmental parameters and generate future environmental parameter prediction results.

[0101] The trend change prompt submodule displays the future environmental parameter trends through charts based on the future environmental parameter prediction results, and identifies future risk points and opportunities. The specific process of generating time trend prediction results is as follows:

[0102] The trend change prompt submodule uses data visualization technology and Python's matplotlib library to draw charts based on the prediction results of future environmental parameters. It displays the trends of future environmental parameters through charts, analyzes the chart data to identify future risk points and opportunities, and generates time trend prediction results.

[0103] See also Figure 2 and Figure 5 , the key factor identification module includes:

[0104] The environmental parameter influence analysis submodule performs a correlation analysis between environmental parameters and PCB quality based on the time trend prediction results, calculates the Spearman rank correlation coefficient between the parameters and quality, and performs a significance test for the correlation. The specific process for generating the environmental parameter influence analysis results is as follows:

[0105] The environmental parameter influence analysis submodule adopts statistical analysis methods based on the time trend prediction results, uses Python's SciPy library to execute the spearmanr function, calculates the Spearman rank correlation coefficient between environmental parameters and circuit board quality, and performs a p-value test to determine the significance of the correlation by setting the significance level alpha to 0.05, thereby generating the environmental parameter influence analysis results.

[0106] The environmental parameter simulation test submodule performs a simulation test on the impact of environmental parameter changes on the quality of the circuit board based on the results of the environmental parameter influence analysis. It records the changes of each environmental parameter and its impact on the quality of the circuit board. The specific process of generating the simulation test results is as follows:

[0107] The environmental parameter simulation test submodule adopts simulation test technology based on the results of environmental parameter influence analysis. It uses Python scripts to change the environmental parameter values ​​one by one through a loop structure, records the changes of each environmental parameter and its impact on the quality of the circuit board, and generates simulation test results by using the Pandas library to record and organize data.

[0108] The influencing parameter extraction submodule obtains the simulation test results, performs the influencing parameter extraction process, applies the F test to screen the environmental parameters that have an impact on the quality of the circuit board, and generates the key influencing factor identification results. The specific process is as follows;

[0109] The influencing parameter extraction submodule obtains the simulation test results, adopts the variance analysis method, uses Python's SciPy library to execute the f_oneway function, and applies the F test to perform statistical analysis on the influence of each environmental parameter in the data set. By comparing the F value and the corresponding p value, the environmental parameters that have a significant impact on the quality of the circuit board are screened, and the key influencing factor identification results are generated.

[0110] See also Figure 2 and Figure 6 , the circuit board production process modeling module includes:

[0111] The variable relationship mapping submodule maps the causal relationships between key variables and parameters based on the key influencing factor identification results, draws a causal loop diagram between variables, and labels the variables, including the direct and indirect relationships between circuit board production speed, raw material quality, and environmental conditions. The specific process of generating a causal loop diagram is as follows;

[0112] The variable relationship mapping submodule adopts the system dynamics method based on the key influencing factor identification results and uses Vensim software to construct a causal relationship diagram. By defining circuit board production speed, raw material quality and environmental conditions as key variables and marking the direct and indirect connections between variables, a causal loop diagram is generated.

[0113] The production process simulation submodule uses a causal loop diagram to perform dynamic simulation of feedback loops and nonlinear behaviors in the production process. By setting key variables and parameters, it simulates various scenarios on the PCB production line, including the impact of temperature and humidity fluctuations on PCB quality, the impact of raw material quality changes on production efficiency, and the impact of production speed adjustments on PCB stability. The specific process for generating dynamic simulation results of the production process is as follows:

[0114] The production process simulation submodule is based on the causal loop diagram and adopts system dynamics simulation technology. It uses Simulink to implement simulation in the MATLAB environment. By setting key variables and parameters such as production speed control parameters, raw material quality standards and environmental control parameters, it performs dynamic simulation of various future situations on the circuit board production line, considers the influence of feedback loops and nonlinear behavior, and generates dynamic simulation results of the production process.

[0115] The production process behavior analysis submodule obtains the dynamic simulation results of the production process, uses numerical analysis and data management technology to analyze the simulation data and identify the key behavior patterns in the production process, and obtains the specific process of the circuit board production process dynamics model;

[0116] The production process behavior analysis submodule obtains the dynamic simulation results of the production process, adopts numerical analysis methods, and uses Python's NumPy and Pandas libraries for data processing and analysis. By analyzing the simulated data sets, it identifies key behavioral patterns in the production process, such as the impact of production delays, raw material shortages or quality fluctuations on the quality of the final product, and obtains a dynamic model of the circuit board production process.

[0117] See also Figure 2 and Figure 7 , the abnormal pattern recognition module includes:

[0118] The data processing submodule performs denoising, missing value filling, and normalization based on the output of the circuit board production process dynamics model to generate a standardized data set. The specific process is as follows:

[0119] The data processing submodule is based on the output of the circuit board production process dynamics model. It adopts data preprocessing technology, uses Python's Pandas library for data cleaning, removes missing values ​​through the dropna function, and applies the MinMaxScaler of the Scikit-learn library for data normalization. The data range is adjusted to between 0 and 1 to eliminate dimensionality effects and reduce the impact of noise data on model performance, thereby generating a standardized data set.

[0120] The pattern mining submodule builds and trains multiple decision tree models based on standardized data sets, analyzes and learns patterns in the data that indicate abnormalities in PCB production quality, and generates the mined abnormal pattern information in the following specific process:

[0121] The pattern mining submodule is based on a standardized data set and adopts a machine learning algorithm. It uses Python's Scikit-learn library to build and train multiple decision tree models, builds a decision tree through the DecisionTreeClassifier function, sets the max_depth parameter to control the maximum depth of the tree, and uses the fit method to learn the data, analyze and learn the patterns in the data that indicate abnormal circuit board production quality, and generate mined abnormal pattern information.

[0122] The abnormal pattern classification submodule analyzes and classifies abnormal patterns based on the mined abnormal pattern information details, distinguishing each type of circuit board production abnormality, including various factors such as abnormal soldering temperature and humidity control failure. The specific process of generating abnormal pattern recognition results is as follows;

[0123] The abnormal pattern classification submodule uses classification technology based on the mined abnormal pattern information. It uses Python's Scikit-learn library to execute RandomForestClassifier to build a random forest model, integrates multiple decision trees to improve classification accuracy and stability, and determines the number of trees by setting the n_estimators parameter. The model is used to analyze and classify abnormal patterns, distinguish different types of circuit board production abnormalities, such as abnormal soldering temperature and humidity control failure, and generate abnormal pattern recognition results.

[0124] See also Figure 2 and Figure 8 ,Similarity measurement and analysis modules include:

[0125] Based on the abnormal pattern recognition results, the feature analysis submodule analyzes the identified abnormal pattern data, extracts the key features that represent production abnormalities, and gradually extracts detailed features in the data to generate a key feature vector set. The specific process is as follows:

[0126] Based on the abnormal pattern recognition results, the feature analysis submodule adopts feature extraction technology and uses Python's Scikit-learn library to perform principal component analysis. The fit_transform method is used to reduce the dimension of the identified abnormal pattern data to extract the key features that represent production anomalies, and further refine the detailed features in the data to generate a set of key feature vectors.

[0127] The model building submodule builds a twin network model based on the key feature vector set to analyze similarity. This includes using a multi-layer convolutional neural network for feature extraction and a fully connected layer for feature fusion and similarity analysis. It uses a ternary loss function as the training objective to optimize the model's measurement of the distance between normal and abnormal patterns. The specific process for generating a twin network similarity model is as follows:

[0128] The model building submodule is based on a set of key feature vectors and adopts a deep learning method. It uses the TensorFlow and Keras libraries to build a twin network model. It adds multiple convolutional layers Conv2D through the Sequential model for feature extraction, and uses the fully connected layer Dense for feature fusion and similarity analysis. It sets the ternary loss function TripletLoss as the training target, optimizes the model's measurement of the distance between normal and abnormal patterns, and generates a twin network similarity model.

[0129] The ternary loss function follows the improved formula II:

[0130] L(a,p,n)=max(ω1·d(a,p)-ω2·d(a,n)+margin+λ·R,0)

[0131] The optimization model measures the distance between normal and abnormal patterns and generates a twin network similarity model;

[0132] Among them, L represents the loss function, which is used to calculate the loss value of each training sample and guide the parameter update during the model training process.

[0133] a represents the feature vector of the anchor sample, which is a reference point for comparison with positive and negative samples.

[0134] p represents the feature vector of the positive sample, which belongs to the same category as the anchor sample, so that the distance between the anchor sample and the positive sample is as small as possible.

[0135] n represents the feature vector of the negative sample, which does not belong to the same category as the anchor sample, so that the distance between the anchor sample and the negative sample is as large as possible. d represents the distance between the two feature vectors, using Euclidean distance as the metric.

[0136] Margin is a hyperparameter used to control the minimum distance between positive samples and negative samples, ensuring that the model can effectively distinguish samples of different categories.

[0137] ω1 and ω2 are weight coefficients for adjusting the distance between the anchor point and the positive sample and the anchor point and the negative sample, respectively. By adjusting these two parameters, the contribution of the distance between the positive and negative samples to the loss function can be controlled.

[0138] ω1 is determined by cross-validation performance evaluation on the training data,

[0139] ω2 is determined by analyzing the distance distribution characteristics between anchor samples and negative samples and optimizing model performance through cross-validation. λ is the weight of the regularization term, which is used to balance the distance metric and model complexity in the loss function to avoid overfitting.

[0140] R represents the regularization term of the model, such as L2 regularization, which is used to penalize the complexity of the model and encourage the model to learn more generalized feature representations;

[0141] The execution process is as follows:

[0142] S1: Set the initial parameters in the Siamese network model, including the weights and biases of the convolutional layer and the fully connected layer.

[0143] S2: Select a set of training samples, including anchor sample a, positive sample p and negative sample n, to ensure that the positive sample and the anchor sample belong to the same category, while the negative sample and the anchor sample do not belong to the same category.

[0144] S3: Feature extraction is performed on anchor samples, positive samples, and negative samples through the two identical convolutional network architectures of the twin network to obtain their respective feature vectors.

[0145] S4: Calculate the distance d(a,p) between the anchor point sample feature vector a and the positive sample feature vector p, and the distance d(a,n) between the anchor point sample feature vector a and the negative sample feature vector n.

[0146] S5: Determine the weight coefficients ω1 and ω2 through cross-validation and analysis of model performance. In addition, determine the weight λ of the regularization term through grid search method to balance the complexity and generalization ability of the model.

[0147] S6: Calculate the loss value under the current model parameters according to Formula II.

[0148] S7: Use gradient descent or other optimization algorithms to update the model parameters according to the results of the loss function to reduce the distance between the anchor sample and the positive sample, while increasing the distance between the anchor sample and the negative sample.

[0149] S8: Repeat steps S3 to S7 until the performance of the model on the training set reaches a preset threshold or the training round is completed.

[0150] S9: Evaluate model performance on an independent validation set to ensure that the model has good generalization ability.

[0151] The similarity result analysis submodule performs similarity analysis based on the twin network similarity model, compares and analyzes the feature vectors between the circuit board production data and the known abnormal patterns, calculates the quantitative similarity score, identifies potential quality risks or abnormalities, and generates the similarity measurement analysis results. The specific process is as follows:

[0152] The similarity result analysis submodule is based on the twin network similarity model and adopts similarity calculation technology. It uses Python's Scikit-learn library to execute the cosine_similarity function to compare and analyze the feature vectors between circuit board production data and known abnormal patterns. The quantitative similarity score is obtained by calculation to identify potential quality risks or abnormal situations and generate similarity measurement analysis results.

[0153] See also Figure 2 and Figure 9 , the quality prediction result integration and analysis module includes:

[0154] The information integration submodule performs data fusion operations based on the dynamic simulation results of environmental parameters, time trend prediction results, key influencing factor identification results, circuit board production process dynamics model, abnormal pattern recognition results and similarity measurement analysis results to generate a comprehensive feature data set. The specific process is as follows:

[0155] The information integration submodule adopts data fusion technology based on the dynamic simulation results of environmental parameters, time trend prediction results, key influencing factor identification results, circuit board production process dynamics model, abnormal pattern recognition results and similarity measurement analysis results. It uses Python's Pandas library to merge data and integrates data sets from different sources according to the corresponding key values ​​through the merge function to create a comprehensive feature data set containing all relevant information, generating a comprehensive feature data set.

[0156] The training and optimization submodule performs iterative training and model optimization operations based on the comprehensive feature dataset. By adjusting parameters and cross-validating model performance, the specific process of generating an optimized prediction model is as follows:

[0157] The training and optimization submodule uses a machine learning algorithm based on a comprehensive feature dataset and Python's Scikit-learn library for model training. It builds a random forest model through the RandomForestClassifier, sets the n_estimators parameter to define the number of trees, and applies the cross_val_score function for cross-validation to evaluate the performance of the model on different data subsets. It then adjusts the model parameters to optimize the prediction performance and generate an optimized prediction model.

[0158] The quality index analysis submodule performs detailed analysis of PCB quality indicators based on the optimized prediction model, analyzes the impact of each key factor on PCB quality, identifies key factors in quality prediction, and obtains PCB quality prediction and analysis results. The specific process is as follows:

[0159] The quality index analysis submodule is based on the optimized prediction model and adopts statistical analysis methods. It uses Python's SciPy library to analyze the impact of key factors. By executing the anova function, it conducts variance analysis on the impact of each key factor on the quality of the circuit board, identifies the factors that have the most significant impact on the quality of the circuit board, and deeply understands the specific impact of each factor on the quality of the final product, and obtains the circuit board quality prediction and analysis results.

[0160] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A circuit board quality anomaly prediction system based on machine intelligence learning, characterized in that: The system comprises: The environmental parameter simulation and analysis module simulates the impact of environmental parameters on material properties based on circuit board production data. Through iterative calculations, it simulates the changes of parameters over time and generates dynamic simulation results of environmental parameters. The time trend prediction module analyzes the seasonality and trend components of the time series data based on the dynamic simulation results of the environmental parameters, predicts future changes in environmental parameters, and generates time trend prediction results; The key factor identification module simulates the impact of each environmental parameter on the quality of the circuit board based on the time trend prediction result, identifies the environmental parameters that affect the quality by recording the changes in the quality of the circuit board each time, and generates a key influencing factor identification result; The circuit board production process modeling module defines key variables and parameters based on the key influencing factor identification results, establishes a cause-and-effect relationship diagram, simulates the feedback loop and nonlinear behavior in the production process, and obtains a circuit board production process dynamics model; The abnormal pattern recognition module constructs multiple decision trees based on the circuit board production process dynamics model to train the data and summarize the prediction results of each tree, captures the abnormal patterns in the circuit board production process, and generates abnormal pattern recognition results; The similarity measurement and analysis module constructs a twin network model and performs similarity measurement based on the abnormal pattern recognition results. By defining the model structure and adding layers, it extracts and fuses features, analyzes the similarity between production data and known abnormal patterns, and generates similarity measurement analysis results. The quality prediction result integration and analysis module iteratively trains the weak learner to gradually reduce the prediction error based on the dynamic simulation results of the environmental parameters, the time trend prediction results, the key influencing factor identification results, the circuit board production process dynamics model, the abnormal pattern recognition results and the similarity measurement analysis results, and performs a detailed analysis of the circuit board quality status to generate circuit board quality prediction and analysis results.

2. The circuit board quality anomaly prediction system based on machine intelligence learning according to claim 1 is characterized in that: The dynamic simulation results of environmental parameters include simulated temperature change data, humidity level adjustment records and circuit board material property response diagrams; the time trend prediction results include circuit board temperature trend prediction diagrams, humidity change prediction information and analysis of future material property change trends; the key influencing factor identification results include the impact of temperature on circuit board quality, humidity impact score and sensitivity analysis of material properties to quality; the circuit board production process dynamics model includes circuit board production speed variable analysis, raw material quality feedback loop diagram and processing temperature nonlinear response results; the abnormal pattern recognition results include circuit board welding temperature abnormality types, humidity control failure warnings and material adverse reaction classifications; the similarity measurement analysis results include known abnormal pattern similarity scores, potential quality problem similar pattern matching and risk level determination; the circuit board quality prediction and analysis results include quality risk level classification, key improvement area indications and potential defect prevention measures.

3. The circuit board quality anomaly prediction system based on machine intelligence learning according to claim 1 is characterized in that: The environmental parameter simulation and analysis module includes: The parameter tracking submodule tracks and records environmental parameters in real time based on PCB production data, plots parameter changes over time, and generates a time series graph of environmental parameters. The parameter dynamic simulation submodule uses the fourth-order Runge-Kutta method based on the environmental parameter time series diagram to simulate the influence of the parameters on the circuit board material properties and their changes over time, and generates parameter dynamic simulation data; The fourth-order Runge-Kutta method is based on the improved formula I: Simulate the impact of parameters on circuit board material properties and their changes over time, and generate parameter dynamic simulation data; Where k1 = f(t n ,y n ), k4=f(t n +Δt,y n +k3Δt), k env =g(temp,hum,pres,chem),y n+1 Refers to the predicted value of the environmental parameter for the next time step, y n is the environmental parameter value of the current time step, k2 is the slope estimate after the influence of k1 at half of the time step, k3 is the slope estimate after using k2 to adjust again at half of the time step, k4 is the slope estimate after using k3 to adjust after the entire time step is completed, Δt is the time step, k env is the comprehensive slope estimation of the impact of environmental factors on the material properties of the circuit board. α is the weight coefficient determined based on historical data analysis, which is obtained by analyzing the relationship between historical data and the impact of environmental changes on the quality of the circuit board. Temp represents the ambient temperature, hum represents the ambient humidity, pres represents the ambient pressure, chem represents the concentration of chemical substances in the environment, and g is a function used to calculate the comprehensive impact of environmental factors, with k as the weight coefficient. env Form of expression; The environmental impact assessment submodule analyzes the impact of changes in environmental parameters on circuit board quality based on the parameter dynamic simulation data, determines the strength of the relationship between each environmental parameter and circuit board quality, and generates environmental parameter dynamic simulation results.

4. The circuit board quality anomaly prediction system based on machine intelligence learning according to claim 1 is characterized in that: The time trend prediction module includes: The sequence parsing submodule sets and processes the data based on the dynamic simulation results of the environmental parameters, decomposes the time series data into trend, seasonality and residual components, identifies the periodic changes and basic trends of the data, and generates seasonal decomposition and trend identification results; The trend forecast analysis submodule performs time series forecasting based on the seasonal decomposition and trend identification results, and predicts the changing trend of future environmental parameters by setting seasonal change parameters and trend change parameters, thereby generating future environmental parameter forecast results; The trend change prompt submodule displays the future environmental parameter trends through charts based on the future environmental parameter prediction results, identifies future risk points and opportunities, and generates time trend prediction results.

5. The circuit board quality anomaly prediction system based on machine intelligence learning according to claim 1 is characterized in that: The key factor identification module includes: The environmental parameter influence analysis submodule performs a correlation analysis between the environmental parameters and the quality of the circuit board based on the time trend prediction results, calculates the Spearman rank correlation coefficient between the parameters and the quality, and performs a significance test of the correlation to generate an environmental parameter influence analysis result; The environmental parameter simulation test submodule performs a simulation test on the effect of environmental parameter changes on the quality of the circuit board based on the environmental parameter influence analysis results, records the changes in each environmental parameter and its effect on the quality of the circuit board, and generates simulation test results; The influencing parameter extraction submodule obtains the simulation test results, performs an influencing parameter extraction process, applies an F test to screen environmental parameters that have an impact on the quality of the circuit board, and generates a key influencing factor identification result.

6. The circuit board quality anomaly prediction system based on machine intelligence learning according to claim 1 is characterized in that: The circuit board production process modeling module includes: The variable relationship mapping submodule maps the causal relationship between key variables and parameters based on the key influencing factor identification results, draws a causal loop diagram between variables, annotates variables, including the direct and indirect relationships between circuit board production speed, raw material quality and environmental conditions, and generates a causal loop diagram; The production process simulation submodule performs dynamic simulation of feedback loops and nonlinear behaviors in the production process based on the causal loop diagram. By setting key variables and parameters, it simulates various scenarios on the PCB production line, including the impact of temperature and humidity fluctuations on PCB quality, the impact of raw material quality changes on production efficiency, and the impact of production speed adjustments on PCB stability, and generates dynamic simulation results for the production process. The production process behavior analysis submodule obtains the dynamic simulation results of the production process, uses numerical analysis and data management technology to analyze the simulation data and identify key behavior patterns in the production process, and obtains a circuit board production process dynamics model.

7. The circuit board quality anomaly prediction system based on machine intelligence learning according to claim 1 is characterized in that: The abnormal pattern recognition module includes: The data processing submodule performs denoising, missing value filling and normalization based on the output of the circuit board production process dynamics model to generate a standardized data set; The pattern mining submodule constructs and trains multiple decision tree models based on the standardized data set, analyzes and learns patterns in the data indicating abnormalities in circuit board production quality, and generates mined abnormal pattern information; The abnormal pattern classification submodule analyzes and classifies abnormal patterns based on the mined abnormal pattern information details, distinguishes each type of circuit board production abnormality, including multiple factors such as abnormal welding temperature and humidity control failure, and generates abnormal pattern recognition results.

8. The circuit board quality anomaly prediction system based on machine intelligence learning according to claim 1 is characterized in that: The similarity measurement and analysis module includes: The feature analysis submodule analyzes the identified abnormal pattern data based on the abnormal pattern recognition results, extracts key features that represent production abnormalities, and gradually extracts detailed features in the data to generate a key feature vector set; The model building submodule builds a twin network model based on the key feature vector set to analyze similarity, including using a multi-layer convolutional neural network for feature extraction and a fully connected layer for feature fusion and similarity analysis. It uses a ternary loss function as the training target to optimize the model's measurement of the distance between normal and abnormal patterns, and generates a twin network similarity model. The ternary loss function is based on the improved formula II: L(a,p,n)=max(ω1·d(a,p)-ω2·d(a,n)+margin+λ·R,0) The optimization model measures the distance between normal and abnormal patterns and generates a twin network similarity model; Where L represents the loss function, a represents the feature vector of the anchor sample, p represents the feature vector of the positive sample, n represents the feature vector of the negative sample, d represents the distance between the two feature vectors, margin is a hyperparameter used to control the minimum distance between the positive sample and the negative sample, ω1 and ω2 are weight coefficients for adjusting the distance between the anchor point and the positive sample and the anchor point and the negative sample, respectively. ω1 is determined by cross-validation performance evaluation on the training data, ω2 is determined by analyzing the distance distribution characteristics between the anchor sample and the negative sample and optimizing the model performance through cross-validation, λ is the weight of the regularization term, and R represents the regularization term of the model. The similarity result analysis submodule performs similarity analysis based on the twin network similarity model, compares and analyzes the feature vectors between the circuit board production data and the known abnormal patterns, calculates the quantitative similarity score, identifies potential quality risks or abnormalities, and generates similarity measurement analysis results.

9. The circuit board quality anomaly prediction system based on machine intelligence learning according to claim 1 is characterized in that: The quality prediction result integration and analysis module includes: The information integration submodule performs data fusion operations based on the environmental parameter dynamic simulation results, time trend prediction results, key influencing factor identification results, circuit board production process dynamics model, abnormal pattern recognition results and similarity measurement analysis results to generate a comprehensive feature data set; The training and optimization submodule performs iterative training and model optimization operations based on the comprehensive feature data set, and generates an optimized prediction model by adjusting parameters and performing cross-validation of model performance; The quality index analysis submodule performs detailed analysis of the quality index of the circuit board based on the optimized prediction model, analyzes the impact of each key factor on the quality of the circuit board, identifies the key factors in quality prediction, and obtains the circuit board quality prediction and analysis results.

Images