PCBA welding temperature control method based on intelligent compensation mechanism

Abstract

This invention discloses a PCBA welding temperature control method based on an intelligent compensation mechanism, belonging to the field of PCBA welding control technology. It involves collecting and preprocessing historical welding data and PCBA characteristic data, marking abnormal data and adding anomaly type labels; based on the historical welding data and PCBA characteristic data, a PCBA welding temperature prediction model is trained using a random forest regression algorithm, and the predicted temperature curve is output. This invention, by collecting historical welding data and PCBA characteristic data, and using a random forest regression algorithm to train a PCBA welding temperature prediction model, outputs a predicted temperature curve, achieving advance prediction and dynamic adjustment of welding temperature. This significantly improves the accuracy of welding temperature control, effectively avoids the lag adjustment problem caused by changes in the external environment in traditional control methods, ensures temperature stability during the welding process, reduces welding defects caused by temperature fluctuations, and improves the overall welding quality of the PCBA.

Description

Technical Field

[0001] This invention relates to the field of PCBA welding control technology, and specifically to a PCBA welding temperature control method based on an intelligent compensation mechanism. Background Technology

[0002] PCBA (Printed Circuit Board Assembly) soldering temperature control is a crucial step in the electronics manufacturing process. Modern electronic devices are developing towards miniaturization, lightweighting, and high performance. The component density on PCBAs is constantly increasing, and the component size is constantly shrinking. Under such high-density assembly conditions, soldering temperature control is particularly important. Excessive soldering temperature may cause thermal damage to adjacent components or even lead to component failure, while excessively low soldering temperature may result in weak solder joints, defects such as cold solder joints, etc. Therefore, precise control of soldering temperature is required to ensure the quality of each solder joint. Especially in SMT (Surface Mount Technology) and wave soldering processes, the correctness of soldering temperature control directly affects the soldering quality, reliability, and circuit board performance of electronic components.

[0003] In existing technologies, when the furnace temperature changes, the adjustment of the PCBA soldering process is somewhat delayed. Furthermore, different PCBA board types and solder types require specific adjustments to the temperature profile, resulting in a slow response speed in the PCBA soldering process. Therefore, the problem to be solved by this invention is how to predict the PCBA soldering temperature by combining historical soldering data, adjust the temperature control strategy in advance, and iteratively optimize the temperature profile to form a closed-loop mechanism of "detection-analysis-compensation". To this end, a PCBA soldering temperature control method based on an intelligent compensation mechanism is proposed. Summary of the Invention

[0004] The purpose of this invention is to provide a PCBA soldering temperature control method based on an intelligent compensation mechanism to solve the problems mentioned in the background art.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0006] A PCBA soldering temperature control method based on an intelligent compensation mechanism includes the following steps:

[0007] S1: Collect and preprocess historical welding data and PCBA characteristic data, mark abnormal data and add abnormality type labels;

[0008] S2, based on historical welding data and PCBA characteristic data, combines random forest regression algorithm to train PCBA welding temperature prediction model and outputs predicted temperature curve;

[0009] S3. Collect current welding data and PCBA characteristic data, compare with the predicted temperature curve, detect any abnormal deviations, and activate the temperature control compensation mechanism.

[0010] S4. Based on PCBA soldering process knowledge, establish a compensation rule knowledge base, combine the predicted temperature curve with the compensation rules, and generate and execute corresponding compensation strategies according to abnormal deviations.

[0011] S5. Inspect the quality of the weld and correlate the inspection results with the temperature curve. If defects are found, locate the cause of the defects.

[0012] S6. Based on the new defect cases, generate corresponding compensation strategies, store them in and optimize the compensation rule knowledge base.

[0013] A further improvement to the technical solution of the present invention is that: S1 specifically includes:

[0014] Data from various sensors in the PCBA soldering area and historical databases are collected, including sensor data from the soldering equipment, PCBA characteristic data, and quality inspection data. The collected data is categorized and stored by batch, process type, and equipment model. A unique identifier is assigned to each PCBA sample, and the collected data is matched with the identifier of the PCBA sample.

[0015] The collected data was cleaned, and the data in the data warehouse was standardized using Z-Score standardization to eliminate differences in units. Noise was smoothed by sliding window filtering, and samples with more than 30% missing values ​​were removed.

[0016] Abnormal data points are identified using statistical methods. Abnormal types are defined according to welding process standards, including sudden temperature rise / fall, curve deviation from preset threshold, and component overheating / underheating. Abnormal type labels are marked, and the time, location, and characteristic parameters of the abnormality are recorded.

[0017] Structured tags are added to abnormal data, which include the abnormality type, scope of impact, and defect association. The preprocessed data, abnormal tags, and PCBA sample identifiers are then integrated and stored in the database.

[0018] A further improvement to the technical solution of this invention is that the process of labeling abnormal types and storing them in the database includes:

[0019] The dynamic threshold method and sliding window statistics are used to analyze the time series data of welding temperature and airflow. By calculating the moving average and standard deviation, the temperature change rate threshold and airflow fluctuation range are set, and data points exceeding the threshold are marked as anomalies. At the same time, cluster analysis is performed on the component temperature to identify overheating / underheating hot spots that deviate from the normal operating range.

[0020] Based on the welding process specification (IPC-A-610 standard), anomalies are classified into three categories: sudden temperature rise / fall, curve deviation from preset threshold, and component thermal anomaly. Then, the anomaly type is automatically labeled through the rule engine.

[0021] A structured label is generated for each anomaly, including type, timestamp, spatial coordinates, and feature parameters. The feature parameters are temperature peak and duration. An anomaly event log table is designed to record the time, spatial coordinates, feature parameters, and associated PCBA sample identifier of the anomaly. A time series database (InfluxDB) is used to store the time series anomaly data, and a relational database (PostgreSQL) is used to store the structured labels.

[0022] A further improvement to the technical solution of the present invention is that: S2 specifically includes:

[0023] By integrating and preprocessing historical welding data and PCBA characteristic data, an input feature matrix is ​​constructed. The temperature curve is discretized into time series points, and statistical features including peak temperature, slope, and duration are extracted. Discrete attributes are encoded, and board type and solder type are converted into numerical representations to form a structured dataset. The training set and test set are divided by stratified sampling to ensure data distribution consistency and preserve the continuity of the time series.

[0024] A PCBA soldering temperature prediction model is trained using a training set combined with a random forest regression algorithm. The input feature matrix and target variables are determined, and hyperparameters including the number of trees, depth, and feature selection method are set. The model is optimized through cross-validation, and the prediction error of the model on the test set is verified to ensure the model's generalization ability, thereby obtaining the trained PCBA soldering temperature prediction model.

[0025] Deploy a PCBA soldering temperature prediction model, input a feature matrix containing the characteristic data and process parameters of the new PCBA, and output the predicted temperature curve.

[0026] A further improvement to the technical solution of this invention lies in the fact that the specific training process of the PCBA soldering temperature prediction model includes:

[0027] Determine the input feature matrix and target variable for training the PCBA soldering temperature prediction model. The input feature matrix includes soldering process parameters (soldering power, soldering time, preheating temperature), PCBA characteristics (material thermal conductivity, number of layers, solder joint density), and statistical characteristics of historical temperature curves (peak temperature, heating rate, cooling time). The target variable is the temperature value at each time point during the soldering process.

[0028] The hyperparameters for training the PCBA soldering temperature prediction model are set, including the number of trees, maximum tree depth, feature selection method, minimum number of samples required for internal node splitting, and minimum number of samples required for leaf nodes. The number of trees is initially set to 100 and adjusted later through cross-validation. The maximum tree depth is initially set to None (fully grown) and will be limited later to prevent overfitting. The feature selection method is randomly selected. The minimum number of samples required for internal node splitting is initially set to 2, and the minimum number of samples required for leaf node splitting is initially set to 1.

[0029] Using the training set as input feature matrix and target variable, train a random forest regression model. Further divide the training set into K subsets, use K-1 subsets to train the model each time, and validate the model with the remaining 1 subset. Repeat this process K times, take the average error as the model performance index, and adjust the hyperparameters based on the cross-validation results to find the optimal parameter combination.

[0030] The random forest incorporates a feature importance scoring mechanism. By calculating the Gini index reduction of each feature across all decision trees, its contribution to the prediction results is evaluated. The features are then sorted by importance score, retaining those that significantly affect temperature prediction and removing those with low contribution, thus optimizing model performance.

[0031] Using the test set as input feature matrix, predict temperature values, and calculate error indices such as mean square error, mean absolute error, and coefficient of determination. If the test set error is large, further adjust the hyperparameters (increase the number of trees, adjust the feature selection method) or increase the amount of training data, and then save the trained PCBA soldering temperature prediction model.

[0032] A further improvement to the technical solution of this invention lies in the fact that the specific process of outputting the predicted temperature curve includes:

[0033] Load the trained PCBA soldering temperature prediction model, build an API interface to receive the input feature matrix containing the characteristic data and process parameters of the new PCBA, and return the predicted temperature curve.

[0034] The input feature matrix is ​​fed into the PCBA soldering temperature prediction model, and the output of the PCBA soldering temperature prediction model is a time series of temperature prediction values, covering the entire soldering process (preheating, reflow, and cooling stages). The time series of temperature prediction values ​​are presented in the form of discrete temperature points.

[0035] The output discrete temperature points are processed by spline interpolation to generate a continuous temperature curve. Key parameters (heating / cooling rate, peak temperature and duration, cooling slope) are marked according to the welding process stage. The predicted temperature curve is compared with the historical standard temperature curve (IPC standard) to check whether it meets the process requirements. If it does not meet the standard, the cause of the abnormality is marked and a risk warning is returned.

[0036] A further improvement to the technical solution of the present invention is that: S3 specifically includes:

[0037] Welding data and PCBA characteristic data during the welding process are collected in real time by sensors to ensure the time synchronization and integrity of the data. The collected raw data is filtered, denoised and normalized to eliminate environmental interference and measurement errors, and transformed into a standardized input feature matrix. At the same time, timestamps are recorded to match the time dimension of the prediction curve.

[0038] The real-time temperature data is compared point by point with the predicted temperature curve. The temperature deviation value at each time point is calculated. The distribution characteristics of the deviation are analyzed by using a sliding window method. The average value and standard deviation of the deviation within the window are calculated. The statistical method of mean difference is used to calculate the statistical quantity of mean difference, quantify the degree of deviation, and identify whether there are systematic or sudden abnormal deviations.

[0039] Based on the deviation analysis results, anomaly detection rules are established to determine whether the current deviation exceeds the allowable range of the process. Anomalies are classified by setting deviation thresholds, and the long-term drift or sudden change of the temperature curve is analyzed. Then, combined with PCBA characteristic data and process parameters, the potential causes of the anomalies are analyzed.

[0040] Based on the detected abnormal deviation, the temperature control compensation mechanism is automatically triggered, and it is matched and analyzed with a pre-established compensation rule knowledge base to output a compensation strategy.

[0041] A further improvement to the technical solution of the present invention is that: S4 specifically includes:

[0042] A structured compensation rule knowledge base is established based on PCBA welding process knowledge (IPC standards, historical cases). According to the characteristics of PCBA welding process, the compensation rule knowledge base is divided into a multi-level structure, including material properties, process parameters, equipment status and historical abnormal cases. When an abnormal deviation is identified, the compensation rule in the compensation rule knowledge base that is closest to the current process conditions is automatically matched, and the corresponding compensation strategy framework is extracted.

[0043] By combining the predicted temperature curve with real-time deviation data, the deviation trend is analyzed, and the priority of the compensation strategy is dynamically adjusted based on the degree of influence of PCBA characteristics on the compensation effect.

[0044] Based on the matching compensation rules, a compensation strategy containing specific parameter adjustments is generated. After the compensation strategy is generated, its feasibility is ensured through multi-dimensional verification, including process constraint verification, thermodynamic simulation verification, and risk assessment verification, and the optimal compensation strategy is selected.

[0045] The selected optimal compensation strategy is implemented, and the temperature profile and process parameters are continuously monitored.

[0046] A further improvement to the technical solution of the present invention is that: S5 specifically includes:

[0047] The solder joints are evaluated using automated optical inspection (AOI) to obtain post-soldering quality inspection results, and the quality inspection results are correlated with temperature profiles using timestamp alignment technology.

[0048] A comprehensive analysis of the correlated post-weld quality inspection results and temperature curves was conducted to screen out weld areas with defects and to focus on abnormal fluctuation points in the temperature curves. Based on the defect characteristics shown in the post-weld quality inspection results, the correlation between defects and abnormal temperature curves was preliminarily determined.

[0049] Based on the process knowledge and analysis results of PCBA soldering, the root cause of defects is located, including temperature parameter attribution, process parameter correlation and equipment / material factors;

[0050] To identify the cause of the defect, a detailed verification plan was developed. Through simulated welding tests, relevant parameters were adjusted, and the changes in temperature curves and weld quality were observed to verify the accuracy of the identified cause. Based on the verification results, targeted improvement measures were developed.

[0051] A further improvement to the technical solution of the present invention is that: S6 specifically includes:

[0052] Collect data on new defect cases, conduct systematic analysis, identify defect types and their spatial distribution characteristics, extract key process parameters related to defects, including abnormal temperature curves, process parameter deviations and equipment status data, and construct defect feature vectors through multi-dimensional data correlation analysis;

[0053] Based on the defect feature vector, similar cases in the compensation rule knowledge base are called, and an initial compensation strategy is generated through parameter mapping and logical reasoning. The content of the initial compensation strategy covers temperature curve adjustment, process parameter optimization and equipment calibration. The generated initial compensation strategy is verified by simulation. The temperature curve and solder joint quality after compensation are predicted by thermodynamic model, the effectiveness of the strategy is evaluated, and candidate solutions that meet the process requirements are selected.

[0054] The verified compensation strategies are stored in the compensation rule knowledge base and classified and stored according to multi-dimensional tags such as defect type, process parameters, and equipment status. By analyzing the matching relationship between defect features and compensation strategies, the retrieval logic of the knowledge base is optimized and the strategy matching efficiency is improved. At the same time, historical cases are analyzed again to extract general compensation rules and further enhance the generalization ability of the knowledge base.

[0055] Establish a closed-loop feedback mechanism for the compensation strategy, continuously monitor the execution effect of the compensation strategy on the production site, collect temperature curves, process parameters and solder joint quality data in real time, evaluate the effectiveness of the compensation strategy by comparing the quality inspection results before and after compensation, and if the compensation effect is found to be unsatisfactory, trigger the strategy iteration process, optimize the compensation parameters in combination with on-site data, and form a dynamic optimization closed loop of "defect identification - strategy generation - effect verification - knowledge base update".

[0056] Due to the adoption of the above technical solution, the technical progress achieved by this invention compared to the prior art is as follows:

[0057] 1. This invention provides a PCBA soldering temperature control method based on an intelligent compensation mechanism. By collecting historical soldering data and PCBA characteristic data, a PCBA soldering temperature prediction model is trained using a random forest regression algorithm, and a temperature curve prediction is output. This enables advance prediction and dynamic adjustment of the soldering temperature, significantly improving the accuracy of soldering temperature control. It effectively avoids the problem of delayed adjustment caused by changes in the external environment in traditional control methods, ensures the stability of temperature during the soldering process, reduces soldering defects caused by temperature fluctuations, and improves the overall soldering quality of PCBA.

[0058] 2. This invention provides a PCBA welding temperature control method based on an intelligent compensation mechanism. By establishing a multi-level compensation rule knowledge base, it can automatically match the compensation rule closest to the current process conditions and dynamically generate and execute the corresponding compensation strategy. This enables the welding process to flexibly cope with various complex scenarios and ensure that the optimal welding effect can be achieved under different conditions. Furthermore, by continuously accumulating new defect cases and compensation strategies during the control process and storing them in the compensation rule knowledge base, the effective accumulation of process knowledge is achieved. Attached Figure Description

[0059] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0060] Figure 1 This is a flowchart of the present invention;

[0061] Figure 2 This is a schematic diagram of the method flow of the present invention. Detailed Implementation

[0062] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0063] Example 1, as Figure 1 , Figure 2 As shown, this invention provides a PCBA soldering temperature control method based on an intelligent compensation mechanism, comprising the following steps:

[0064] S1. Collect and preprocess historical welding data and PCBA characteristic data, mark abnormal data and add abnormality type labels. Collect welding equipment sensor data, PCBA characteristic data and quality inspection data from various sensors in the PCBA welding area and historical databases. Store the collected data by batch, process type, and equipment model, and assign a unique identifier to each PCBA sample. Match the collected data with the PCBA sample identifiers. Welding equipment sensor data includes temperature, airflow, and time; PCBA characteristic data includes board type, number of layers, and solder type; and quality inspection data is the AOI result. Perform data cleaning on the collected data using Z-factor authentication. - The Score standardization method standardizes the data in the data warehouse to eliminate differences in units, smooths noise through sliding window filtering, removes samples with more than 30% missing values ​​to ensure data continuity and integrity, identifies abnormal data points through statistical methods, defines anomaly types according to welding process standards, including sudden temperature rise / fall, curve deviation from preset threshold, and component overheating / underheating, and labels the anomaly type. The time, location and characteristic parameters of the anomaly are recorded, and structured labels are added to the abnormal data. The labels include anomaly type, scope of impact and defect association. Finally, the preprocessed data, anomaly labels and PCBA sample identifiers are integrated and stored in the database.

[0065] The process of labeling exception types and storing them in the database includes:

[0066] Dynamic thresholding and sliding window statistics are used to analyze the time-series data of welding temperature and airflow. By calculating the moving average and standard deviation, a threshold for the rate of temperature change and the range of airflow fluctuation are set. Data points exceeding the threshold are marked as anomalies. At the same time, cluster analysis is performed on component temperatures to identify overheating / underheating hot spots that deviate from the normal operating range. According to the welding process specification (IPC-A-610 standard), anomalies are divided into three categories: sudden temperature rise / fall, curve deviation from the preset threshold, and component thermal anomaly. Among them, sudden temperature rise / fall is the temperature change rate exceeding the preset threshold, curve deviation from the preset threshold is the temperature curve exceeding the limit, and component thermal anomaly is the component temperature exceeding the process window. Then, the rule engine is used to automatically label the anomaly type, generating a structured label for each anomaly point, including type, timestamp, spatial coordinates, and feature parameters. The feature parameters are temperature peak and duration. An anomaly event log table is designed to record the time, spatial coordinates, feature parameters, and associated PCBA sample identifiers of the anomaly. A time-series database (InfluxDB) is used to store the time-series anomaly data, and a relational database (PostgreSQL) is used to store the structured labels.

[0067] S2, based on historical welding data and PCBA characteristic data, uses a random forest regression algorithm to train a PCBA welding temperature prediction model, outputting a predicted temperature curve. It integrates collected and preprocessed historical welding data and PCBA characteristic data to construct an input feature matrix, discretizes the temperature curve into time series points, extracts statistical features including peak temperature, slope, and duration, encodes discrete attributes, and converts board type and solder type into numerical representations to form a structured dataset. A stratified sampling method is used to divide the dataset into training and test sets to ensure data distribution consistency and preserve the continuity of the time series. The training set is used in conjunction with the random forest regression algorithm to train the PCBA welding temperature prediction model, determining the input feature matrix and target variables, and setting hyperparameters including the number of trees, depth, and feature selection method. The model is optimized through cross-validation. During training, the importance of features is analyzed, and factors significantly affecting temperature prediction are screened. The prediction error of the model on the test set is verified to ensure the model's generalization ability, thus obtaining the trained PCBA welding temperature prediction model. The PCBA welding temperature prediction model is then deployed, with an input feature matrix containing the characteristic data and process parameters of the new PCBA as input, outputting the predicted temperature curve.

[0068] The specific training process of the PCBA welding temperature prediction model includes:

[0069] The input feature matrix and target variable for training the PCBA soldering temperature prediction model are determined. The input feature matrix includes soldering process parameters (soldering power, soldering time, preheating temperature), PCBA characteristics (material thermal conductivity, number of layers, solder joint density), and statistical characteristics of historical temperature curves (peak temperature, heating rate, cooling time). The target variable is the temperature value at each time point during the soldering process. Hyperparameters for training the PCBA soldering temperature prediction model are set, including the number of trees, maximum tree depth, feature selection method, minimum number of samples required for internal node splitting, and minimum number of samples required for leaf nodes. The number of trees is initially set to 100 and adjusted later through cross-validation. The maximum tree depth is initially set to None (fully grown) and subsequently limited to prevent overfitting. The feature selection method is randomly selected. The initial minimum number of samples required for internal node splitting is set to 2, and the initial minimum number of samples for leaf nodes is set to 1. Using the training set as input for the feature matrix and target variable, a random forest regression model is trained. The training set is further divided into K subsets. Each time, K-1 subsets are used to train the model, and the remaining subset is used for validation. This process is repeated K times, and the average error is used as the model performance metric. Based on the cross-validation results, hyperparameters are adjusted to find the optimal parameter combination. A feature importance scoring mechanism is built into the random forest. By calculating the Gini index reduction of each feature across all decision trees, its contribution to the prediction results is evaluated. Features are ranked by importance score, and those with a significant impact on temperature prediction are retained. Features were identified, low-contribution features were removed, and model performance was optimized. The feature matrix was input using the test set to predict temperature values. Error indices such as mean squared error, mean absolute error, and coefficient of determination were calculated. The mean squared error measures the average squared difference between the predicted and true values, the mean absolute error measures the average absolute difference between the predicted and true values, and the coefficient of determination measures the model's explanatory power for the target variable. The distribution of the prediction results on the test set was analyzed to ensure the model's generalization ability under different PCBA types and process parameters. If the error on the test set was large, the hyperparameters were further adjusted (by increasing the number of trees or adjusting the feature selection method) or the amount of training data was increased. Finally, the trained PCBA soldering temperature prediction model was saved.

[0070] The specific process of outputting the predicted temperature curve includes:

[0071] The trained PCBA soldering temperature prediction model is loaded, and an API interface is built to receive an input feature matrix containing the characteristic data and process parameters of the new PCBA. The model returns the predicted temperature curve. The input feature matrix is ​​then fed into the PCBA soldering temperature prediction model, which outputs time-series temperature prediction values ​​covering the entire soldering process (preheating, reflow, and cooling stages). These time-series temperature prediction values ​​are presented as discrete temperature points. Spline interpolation is used to process these discrete temperature points, generating a continuous temperature curve. Key parameters (heating / cooling) are labeled according to the soldering process stages. Compare the predicted temperature curve with historical standard temperature curves (IPC standard) to check whether the predicted temperature curve meets the process requirements (including peak temperature range, heating / cooling rate, and curve shape; for peak temperature range, check whether the reflux peak temperature is within ±5℃ of the standard value; for heating / cooling rate, ensure that the heating rate in the preheating stage is ≤3℃ / second, and the slope in the cooling stage meets the equipment limits; for curve shape, verify whether the duration of each stage is within the standard range). If it does not meet the standard, mark the cause of the abnormality ("peak temperature exceeds the limit" or "cooling slope is too steep") and return a risk warning.

[0072] S3. Collect current welding data and PCBA characteristic data, compare with the predicted temperature curve, detect any abnormal deviations, and activate the temperature control compensation mechanism.

[0073] S4. Based on PCBA soldering process knowledge, establish a compensation rule knowledge base, combine the predicted temperature curve with the compensation rules, and generate and execute corresponding compensation strategies according to abnormal deviations.

[0074] S5. Inspect the quality of the weld and correlate the inspection results with the temperature curve. If defects are found, locate the cause of the defects.

[0075] S6. Based on the new defect cases, generate corresponding compensation strategies, store them in and optimize the compensation rule knowledge base.

[0076] Example 2, as Figure 1 , Figure 2 As shown, based on Embodiment 1, the present invention provides a technical solution: preferably, S3 specifically includes:

[0077] Welding data and PCBA characteristic data during the welding process are collected in real time by sensors to ensure the time synchronization and integrity of the data. The collected raw data is filtered, denoised, and normalized to eliminate environmental interference and measurement errors, and transformed into a standardized input feature matrix. At the same time, timestamps are recorded to match the time dimension of the predicted curve. The real-time temperature data is compared with the predicted temperature curve point by point, and the temperature deviation value at each time point is calculated. The distribution characteristics of the deviation are analyzed by using a sliding window method, and the average value and standard deviation of the deviation within the window are calculated. The statistical method of mean difference is used to calculate the statistical quantity of mean difference to quantify the degree of deviation and identify whether there are systematic or sudden abnormal deviations. Based on the deviation analysis results, anomaly detection rules are established to determine whether the current deviation exceeds the process allowable range. Anomalies are classified by setting deviation thresholds, and the long-term drift or sudden change of the temperature curve is analyzed. Then, combined with PCBA characteristic data and process parameters, the potential causes of anomalies are analyzed. According to the detected abnormal deviation, the temperature control compensation mechanism is automatically triggered and matched with the pre-established compensation rule knowledge base to output the compensation strategy.

[0078] The expression for the temperature deviation value at each time point is:

[0079] ;

[0080] In the formula, For the first Temperature deviation values ​​at each time point For the first Temperature data collected in real time at each point in time. For the first Temperature data predicted at a given time point; when the real-time collected temperature data is higher than the predicted temperature data. When the real-time temperature data is lower than the predicted temperature data, When the real-time temperature data equals the predicted temperature data, ;

[0081] The deviation distribution characteristics are analyzed by using a sliding window. The average value and standard deviation of the deviations within the window are calculated. Let the size of the sliding window be... Then for the first windows ( =1,2,...,n- +1), the temperature deviation value within the window is... ( =0,1,..., -1);

[0082] The expressions for the mean and standard deviation of the deviation within the window are:

[0083] ;

[0084] ;

[0085] In the formula, For the first The average temperature deviation within each window reflects the overall level of temperature deviation within that window. For the first The standard deviation of the temperature deviation within a window reflects the degree of dispersion of the temperature deviation within that window. The larger the value, the more dispersed the temperature deviation within the window.

[0086] The statistical method for mean difference is used to calculate the statistic of mean difference, quantifying the degree of deviation. Let the set of average temperature deviations within all windows be denoted as . Its mean is The standard deviation is ;

[0087] The expression for the statistic of mean difference is:

[0088] ;

[0089] ;

[0090] ;

[0091] In the formula, This is a statistic of mean difference, used to determine whether the average temperature deviation within all windows significantly deviates from the zero mean. Under the null hypothesis (that the mean of the average temperature deviation within all windows is zero), It follows a standard normal distribution N(0,1). This is the mean of the temperature deviations within all windows. The standard deviation of the average temperature deviation within all windows;

[0092] Identify systematic or sudden abnormal deviations, where systematic abnormal deviations are identified by observing the average temperature deviation within a sliding window. and standard deviation To determine whether there is a systematic abnormal deviation, we need to look at the long-term trend. For a prolonged period of time, it remains greater than zero or less than zero, and If it is relatively stable, then there is a systematic bias. The extended duration of this event indicates that the overall predicted temperature is lower than expected. The prolonged duration of this event indicates that the overall predicted temperature is higher than expected.

[0093] Sudden abnormal deviations are addressed by setting a threshold. To determine whether there is a sudden abnormal deviation, the standard deviation of the temperature deviation within a certain window is used. Greater than a certain set threshold When a sudden abnormal deviation is detected within the window, the threshold is set. Settings are based on historical data and actual needs;

[0094] S4 specifically includes:

[0095] A structured compensation rule knowledge base is established based on PCBA soldering process knowledge (IPC standards, historical cases). According to the characteristics of PCBA soldering processes, the knowledge base is divided into a multi-level structure, including material properties, process parameters, equipment status, and historical anomaly cases. When an abnormal deviation is identified, the compensation rule closest to the current process conditions in the knowledge base is automatically matched, and the corresponding compensation strategy framework is extracted. Combining the predicted temperature curve with real-time deviation data, the deviation trend is analyzed, and the priority of the compensation strategy is dynamically adjusted based on the degree of influence of PCBA characteristics on the compensation effect. If the deviation originates from an excessively fast heating rate causing the peak temperature to exceed the limit, then priority is given to adjusting the compensation strategy. If the deviation in heating power is caused by insufficient cooling slope, the cooling parameters are optimized first. Based on the matching compensation rules, a compensation strategy containing specific parameter adjustments is generated. After the compensation strategy is generated, its feasibility is ensured through multi-dimensional verification, including process constraint verification, thermodynamic simulation verification, and risk assessment verification. The optimal compensation strategy is selected. Specifically, the process constraint verification ensures that the adjusted parameters do not exceed the equipment's capacity range. The thermodynamic simulation verification predicts whether the temperature curve after compensation meets the process requirements. The risk assessment verification analyzes the potential impact of the compensation strategy on PCBA quality. The selected optimal compensation strategy is executed, and the temperature curve and process parameters are continuously monitored.

[0096] S5 specifically includes:

[0097] Solder joints are evaluated using automated optical inspection (AOI) to obtain post-soldering quality inspection results. Timestamp alignment technology is used to correlate these results with temperature profiles. A comprehensive analysis of the correlated results and profiles identifies defective weld areas and focuses on abnormal fluctuations in the temperature profile. Based on the defect characteristics shown in the post-soldering quality inspection results, a preliminary assessment of the correlation between defects and temperature profile anomalies is made. Combining PCBA soldering process knowledge and analysis results, the root cause of the defects is located, including temperature parameter attribution, process parameter correlation, and equipment / material factors. Specifically, if the cold solder joints are concentrated in areas with insufficient reflow peak, it is determined to be due to "insufficient temperature leading to solder failure." "Melting" If the edge bridging of the multilayer board is related to local overheating, it is attributed to "uneven temperature distribution". If the same defect recurs, it is traced back to uneven heating of the equipment, deviation of solder paste printing thickness or insufficient heat resistance of components. The main cause is determined by confidence ranking and a structured cause description is generated. For the located defect cause, a detailed verification plan is developed. Through simulated welding tests, relevant parameters are adjusted, and the changes in temperature curves and weld quality are observed to verify the accuracy of the located cause. Based on the verification results, targeted improvement measures are developed. If it is a process parameter problem, the welding process specification is optimized. If it is a material factor, the material selection is adjusted or the material pretreatment is carried out. If it is an environmental factor, corresponding protective measures are taken.

[0098] S6 specifically includes:

[0099] Data from new defect cases is collected and systematically analyzed to identify defect types and their spatial distribution characteristics. Key process parameters related to the defects are extracted, including abnormal temperature curves, process parameter deviations, and equipment status data. Through multi-dimensional data correlation analysis, defect feature vectors are constructed. Based on these feature vectors, similar cases from the compensation rule knowledge base are invoked. Initial compensation strategies are generated through parameter mapping and logical reasoning. These strategies cover temperature curve adjustment, process parameter optimization, and equipment calibration. The generated initial compensation strategies are then simulated and verified. A thermodynamic model is used to predict the temperature curve and solder joint quality after compensation, evaluating the effectiveness of the strategies. Candidate solutions that meet process requirements are selected, and verified compensation strategies are stored in the compensation rule knowledge base. The system categorizes and stores data using multi-dimensional tags representing defect types, process parameters, and equipment status. By analyzing the correlation between defect features and compensation strategies, the system optimizes the retrieval logic of the knowledge base and improves strategy matching efficiency. Simultaneously, it performs secondary analysis on historical cases to extract generalized compensation rules, further enhancing the generalization ability of the knowledge base. A closed-loop feedback mechanism for compensation strategies is established, continuously monitoring the execution effect of compensation strategies on the production floor. Real-time data collection includes temperature curves, process parameters, and weld quality data. By comparing quality inspection results before and after compensation, the effectiveness of the compensation strategy is evaluated. If the compensation effect is found to be unsatisfactory, a strategy iteration process is triggered, combining on-site data to optimize compensation parameters, forming a dynamic optimization closed loop of "defect identification - strategy generation - effect verification - knowledge base update".

[0100] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A PCBA soldering temperature control method based on an intelligent compensation mechanism, characterized in that, Includes the following steps: S1: Collect and preprocess historical welding data and PCBA characteristic data, mark abnormal data and add abnormality type labels; S2, based on historical welding data and PCBA characteristic data, combines random forest regression algorithm to train PCBA welding temperature prediction model and outputs predicted temperature curve; S3. Collect current welding data and PCBA characteristic data, compare with the predicted temperature curve, detect any abnormal deviations, and activate the temperature control compensation mechanism. S4. Based on PCBA soldering process knowledge, establish a compensation rule knowledge base, combine the predicted temperature curve with the compensation rules, and generate and execute corresponding compensation strategies according to abnormal deviations. S5. Inspect the quality of the weld and correlate the inspection results with the temperature curve. If defects are found, locate the cause of the defects. S6. Based on the new defect cases, generate corresponding compensation strategies, store them in and optimize the compensation rule knowledge base.

2. The PCBA soldering temperature control method based on an intelligent compensation mechanism according to claim 1, characterized in that: S1 specifically includes: Data from various sensors in the PCBA soldering area and historical databases are collected, including sensor data from the soldering equipment, PCBA characteristic data, and quality inspection data. The collected data is categorized and stored by batch, process type, and equipment model. A unique identifier is assigned to each PCBA sample, and the collected data is matched with the identifier of the PCBA sample. The collected data was cleaned, and the data in the data warehouse was standardized using Z-Score standardization. Noise was smoothed by sliding window filtering, and samples with more than 30% missing values ​​were removed. Abnormal data points are identified using statistical methods, abnormal types are defined according to welding process standards, and abnormal type labels are marked. The time, location, and characteristic parameters of the abnormality are recorded. Structured tags are added to abnormal data, which include the abnormality type, scope of impact, and defect association. The preprocessed data, abnormal tags, and PCBA sample identifiers are then integrated and stored in the database.

3. The PCBA soldering temperature control method based on an intelligent compensation mechanism according to claim 2, characterized in that: The process of labeling and storing anomaly type tags in the database includes: The dynamic threshold method and sliding window statistics are used to analyze the time series data of welding temperature and airflow. By calculating the moving average and standard deviation, the temperature change rate threshold and airflow fluctuation range are set, and data points exceeding the threshold are marked as anomalies. At the same time, cluster analysis is performed on the component temperature to identify overheating / underheating hot spots that deviate from the normal operating range. Based on the welding process specifications, anomalies are classified into three categories: sudden temperature rise / fall, curve deviation from preset threshold, and component thermal anomalies. Then, the anomaly type is automatically labeled through the rule engine. A structured label is generated for each anomaly, including type, timestamp, spatial coordinates, and feature parameters. The feature parameters are temperature peak and duration. An anomaly event log table is designed to record the time, spatial coordinates, feature parameters, and associated PCBA sample identifiers of the anomaly. A time-series database is used to store the time-series anomaly data, and a relational database is used to store the structured labels.

4. The PCBA soldering temperature control method based on intelligent compensation mechanism according to claim 1, characterized in that: S2 specifically includes: Integrate and preprocess historical welding data and PCBA characteristic data to construct an input feature matrix. Discretize the temperature curve into time series points, extract statistical features including peak temperature, slope, and duration, encode discrete attributes, convert board type and solder type into numerical representations, form a structured dataset, and use stratified sampling to divide the training set and test set. The PCBA soldering temperature prediction model is trained using a training set combined with a random forest regression algorithm. The input feature matrix and target variables are determined, and hyperparameters are set. The model is optimized through cross-validation, and the prediction error of the model on the test set is verified, thus obtaining the trained PCBA soldering temperature prediction model. Deploy a PCBA soldering temperature prediction model, input a feature matrix containing the characteristic data and process parameters of the new PCBA, and output the predicted temperature curve.

5. The PCBA soldering temperature control method based on an intelligent compensation mechanism according to claim 4, characterized in that: The specific training process of the PCBA soldering temperature prediction model includes: Determine the input feature matrix and target variable for training the PCBA soldering temperature prediction model. The input feature matrix includes the statistical features of soldering process parameters, PCBA characteristics, and historical temperature curves. The target variable is the temperature value at each time point during the soldering process. Set the hyperparameters for training the PCBA soldering temperature prediction model, including the number of trees, maximum tree depth, feature selection method, minimum number of samples required for internal node splits, and minimum number of samples required for leaf nodes. The number of trees is initially set to 100, the maximum tree depth is initially set to None, and the feature selection method is randomly selected. The minimum number of samples required for internal node splitting is initially set to 2, and the minimum number of samples required for leaf node splitting is initially set to 1. Using the training set as input feature matrix and target variable, train a random forest regression model. Further divide the training set into K subsets, use K-1 subsets to train the model each time, and validate the model with the remaining 1 subset. Repeat this process K times, take the average error as the model performance index, and adjust the hyperparameters based on the cross-validation results to find the optimal parameter combination. The random forest incorporates a feature importance scoring mechanism. By calculating the Gini index reduction of each feature across all decision trees, its contribution to the prediction results is evaluated. The features are then sorted by importance score, retaining those that significantly affect temperature prediction and removing those with low contribution, thus optimizing model performance. The test set is used as the input feature matrix to predict temperature values. Error indices such as mean square error, mean absolute error, and coefficient of determination are calculated. If the test set error is large, the hyperparameters are further adjusted or the amount of training data is increased, and then the trained PCBA soldering temperature prediction model is saved.

6. The PCBA soldering temperature control method based on an intelligent compensation mechanism according to claim 5, characterized in that: The specific process of outputting the predicted temperature curve includes: Load the trained PCBA soldering temperature prediction model, build an API interface to receive the input feature matrix containing the characteristic data and process parameters of the new PCBA, and return the predicted temperature curve. The input feature matrix is ​​fed into the PCBA welding temperature prediction model, and the output of the PCBA welding temperature prediction model is a time series of temperature prediction values, covering the entire welding process. The time series of temperature prediction values ​​are presented in the form of discrete temperature points. The output discrete temperature points are processed by spline interpolation to generate a continuous temperature curve. Key parameters are marked according to the welding process stage. The predicted temperature curve is compared with historical standard temperature curves to check whether it meets the process requirements. If it does not meet the standard, the cause of the abnormality is marked and a risk warning is returned.

7. The PCBA soldering temperature control method based on intelligent compensation mechanism according to claim 1, characterized in that: S3 specifically includes: Welding data and PCBA characteristic data during the welding process are collected in real time by sensors. The collected raw data is filtered, denoised and normalized to be transformed into a standardized input feature matrix. At the same time, timestamps are recorded to match the temporal dimension of the prediction curve. The real-time temperature data is compared point by point with the predicted temperature curve. The temperature deviation value at each time point is calculated. The distribution characteristics of the deviation are analyzed by using a sliding window method. The average value and standard deviation of the deviation within the window are calculated. The statistical method of mean difference is used to calculate the statistical quantity of mean difference, quantify the degree of deviation, and identify whether there are systematic or sudden abnormal deviations. Based on the deviation analysis results, anomaly detection rules are established to determine whether the current deviation exceeds the allowable range of the process. Anomalies are classified by setting deviation thresholds, and the long-term drift or sudden change of the temperature curve is analyzed. Then, combined with PCBA characteristic data and process parameters, the potential causes of the anomalies are analyzed. Based on the detected abnormal deviation, the temperature control compensation mechanism is automatically triggered, and it is matched and analyzed with a pre-established compensation rule knowledge base to output a compensation strategy.

8. The PCBA soldering temperature control method based on intelligent compensation mechanism according to claim 1, characterized in that: S4 specifically includes: A structured compensation rule knowledge base is established based on PCBA welding process knowledge. According to the characteristics of PCBA welding process, the compensation rule knowledge base is divided into a multi-level structure, including material properties, process parameters, equipment status and historical abnormal cases. When an abnormal deviation is identified, the compensation rule in the compensation rule knowledge base that is closest to the current process conditions is automatically matched, and the corresponding compensation strategy framework is extracted. By combining the predicted temperature curve with real-time deviation data, the deviation trend is analyzed, and the priority of the compensation strategy is dynamically adjusted based on the degree of influence of PCBA characteristics on the compensation effect. Based on the matching compensation rules, a compensation strategy containing specific parameter adjustments is generated. After the compensation strategy is generated, its feasibility is ensured through multi-dimensional verification, including process constraint verification, thermodynamic simulation verification, and risk assessment verification, and the optimal compensation strategy is selected. The selected optimal compensation strategy is implemented, and the temperature profile and process parameters are continuously monitored.

9. The PCBA soldering temperature control method based on intelligent compensation mechanism according to claim 1, characterized in that: S5 specifically includes: The solder joints are evaluated by automated optical inspection to obtain post-soldering quality inspection results, and the quality inspection results are correlated with temperature profiles using timestamp alignment technology. A comprehensive analysis of the correlated post-weld quality inspection results and temperature curves was conducted to screen out weld areas with defects and to focus on abnormal fluctuation points in the temperature curves. Based on the defect characteristics shown in the post-weld quality inspection results, the correlation between defects and abnormal temperature curves was preliminarily determined. Based on the process knowledge and analysis results of PCBA soldering, the root cause of defects is located, including temperature parameter attribution, process parameter correlation and equipment / material factors; To identify the cause of the defect, a detailed verification plan was developed. Through simulated welding tests, relevant parameters were adjusted, and the changes in temperature curves and weld quality were observed to verify the accuracy of the identified cause. Based on the verification results, targeted improvement measures were developed.

10. A PCBA soldering temperature control method based on an intelligent compensation mechanism according to claim 9, characterized in that: S6 specifically includes: Collect data on new defect cases, conduct systematic analysis, identify defect types and their spatial distribution characteristics, extract key process parameters related to defects, including abnormal temperature curves, process parameter deviations and equipment status data, and construct defect feature vectors through multi-dimensional data correlation analysis; Based on the defect feature vector, similar cases in the compensation rule knowledge base are called, and an initial compensation strategy is generated through parameter mapping and logical reasoning. The content of the initial compensation strategy covers temperature curve adjustment, process parameter optimization and equipment calibration. The generated initial compensation strategy is verified by simulation. The temperature curve and solder joint quality after compensation are predicted by thermodynamic model, the effectiveness of the strategy is evaluated, and candidate solutions that meet the process requirements are selected. The verified compensation strategies are stored in the compensation rule knowledge base and classified and stored according to multi-dimensional tags such as defect type, process parameters, and equipment status. The retrieval logic of the knowledge base is optimized by analyzing the matching relationship between defect features and compensation strategies. At the same time, historical cases are analyzed again to extract general compensation rules. Establish a closed-loop feedback mechanism for the compensation strategy, collect temperature curves, process parameters and solder joint quality data in real time, evaluate the effectiveness of the compensation strategy by comparing the quality inspection results before and after compensation, and if the compensation effect is found to be unsatisfactory, trigger the strategy iteration process, optimize the compensation parameters by combining on-site data, and form a dynamic optimization closed loop of "defect identification - strategy generation - effect verification - knowledge base update".

Images