PCB forming detection method based on hole site scatter diagram and SPC analysis

Abstract

The invention relates to the technical field of printed circuit board manufacturing and quality detection, in particular to a PCB forming detection method based on a hole site scatter diagram and SPC analysis, and the method comprises the steps: collecting the actual hole site data of a PCB substrate through an optical scanner or a laser measurement device, generating a scatter diagram containing coordinates, aperture and tolerance, and carrying out the detection of the hole site scatter diagram; an abnormal region is identified by using an improved density clustering algorithm; performing multi-dimensional analysis on the scatter diagram based on an SPC model, extracting characteristic parameters such as hole site density deviation, constructing a dynamic control limit, and calculating a CPK to judge a processing state; and when the CPK is lower than a threshold value, optimization instructions such as drilling path correction and drill bit wear compensation are generated and fed back to the equipment through the anti-saturation fuzzy PID controller to realize closed-loop control. According to the method, the hole site machining quality can be accurately detected, the machining parameters are adaptively adjusted, and the PCB forming precision and the production stability are effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of printed circuit board manufacturing and quality inspection, and specifically provides a PCB forming detection method based on hole position scatter diagram and SPC analysis. Background Technique

[0002] In the field of electronic manufacturing, printed circuit boards (PCBs) are key basic components of electronic devices, and their forming quality directly affects the performance and reliability of electronic devices. With the development of electronic products towards miniaturization, high density, and high performance, the manufacturing precision requirements for PCBs are continuously increasing. In particular, indicators such as hole position accuracy and hole diameter consistency have become key factors determining the quality of PCBs.

[0003] During the PCB forming process, the hole position processing quality is affected by a variety of factors. On the one hand, process factors such as drill bit wear, equipment vibration, and substrate thermal expansion can cause problems such as hole position deviation and hole diameter variation. On the other hand, different types of PCBs have differences in hole position distribution density, board material, etc., further increasing the complexity of quality control. Traditional detection methods mostly rely on manual visual inspection or off-line sampling inspection. This method is not only inefficient but also difficult to comprehensively and accurately detect subtle quality problems, and cannot meet the requirements of modern PCB production for real-time and high precision.

[0004] Existing automated detection technologies mainly include statistical process control (SPC) and detection methods based on intelligent algorithms. However, traditional SPC models are usually based on fixed parameters and assumptions, such as data following a normal distribution and control limits remaining unchanged. It is difficult to adapt to the non-steady state changes caused by equipment aging, process fluctuations, etc. during the PCB processing, and it is easy to make misjudgments or missed judgments. Detection methods based on intelligent algorithms, such as convolutional neural networks (CNNs), although have certain advantages in image recognition, have insufficient analysis ability for multi-scale features and lack effective combination with process parameters, and the detection reliability under complex working conditions needs to be improved. In addition, most existing detection systems are in open-loop mode. Even if quality problems are detected, it is impossible to quickly feedback and adjust the processing parameters, and it is difficult to achieve dynamic optimization and closed-loop control of the PCB forming process. Therefore, a PCB forming detection method based on hole position scatter diagram and SPC analysis is proposed to solve the above problems. Summary of the Invention

[0005] The purpose of the present invention is to provide a PCB forming detection method based on hole position scatter diagram and SPC analysis to solve the problems raised in the above background technique.

[0006] To achieve the above purpose, the present invention provides the following technical solutions:

[0007] A PCB forming detection method based on a hole position scatter diagram and SPC analysis, comprising the following steps:

[0008] S1. Collect the actual hole position data of the PCB substrate to generate a hole position scatter diagram, which includes the coordinates, hole diameters and tolerance ranges of each hole position;

[0009] S2. Based on the statistical process control (SPC) model, perform multi-dimensional analysis on the hole position scatter diagram to extract hole position distribution characteristic parameters, including hole position density deviation, standard deviation of hole group spacing and coefficient of variation of hole diameter;

[0010] S3. Construct a dynamic control limit according to the characteristic parameters, calculate the process capability index CPK in real time, and judge whether the hole position processing process is in a statistically controlled state;

[0011] S4. When the CPK is lower than the preset threshold, generate an optimization instruction, which includes drilling path correction parameters, drill bit wear compensation values or equipment calibration schemes;

[0012] S5. Feed back the optimization instruction to the PCB forming equipment and adjust the processing parameters to achieve closed-loop control.

[0013] As a preferred solution, the generation of the hole position scatter diagram in step S1 includes:

[0014] Obtain the actual hole position data of the PCB substrate through an optical scanner or a laser measuring device;

[0015] Perform coordinate mapping on the actual hole position data and the design drawing to generate a comparison scatter diagram including the theoretical position and the actual position;

[0016] Use an improved density clustering algorithm to identify abnormal aggregation regions in the hole position distribution. The density threshold and minimum number of points of the improved density clustering algorithm are dynamically set according to the area of the PCB substrate, the reference area, the total number of holes and the preset number of clusters. Among them, the density threshold is adjusted proportionally to the square root of the substrate area, and the minimum number of points is determined by rounding up the ratio of the total number of holes to the preset number of clusters.

[0017] As a preferred solution, the SPC model in step S2 includes:

[0018] Establish an improved skewness-kurtosis joint test model for hole position density distribution based on historical data, and judge the distribution normality by calculating the joint statistic of sample skewness and kurtosis;

[0019] Adopt a multivariate T² control chart with dynamic covariance update to jointly monitor the hole group spacing and hole diameter. Among them, the covariance matrix is dynamically updated through the weighted average of historical data and real-time data, and the weight is controlled by a forgetting factor;

[0020] An adaptive ARIMA model is introduced to predict the processing trend, and its differencing order is dynamically adjusted according to the ratio of the standard deviation of the prediction error to the reference standard deviation.

[0021] As a preferred solution, the method for constructing the dynamic control limit in step S3 is as follows:

[0022] Adopt a non-linear adaptive control limit adjustment strategy. The bandwidth of the control limit is dynamically adjusted according to the mean and standard deviation of the characteristic parameters, and a smooth transition is achieved through the hyperbolic tangent function;

[0023] The length of the sliding window is dynamically optimized according to the logarithmic relationship between the total number of samples in the current processing batch and the complexity index, and the upper limit of the window length is the preset maximum value.

[0024] As a preferred solution, the generation of the optimization instruction in step S4 includes:

[0025] If the deviation of the hole position density exceeds the limit, optimize the drilling path using the spatial autocorrelation weight. The weight is calculated based on the Euclidean distance between hole positions and the difference in the local thermal expansion coefficient, and the default values of the attenuation factors are 0.1 and 0.05;

[0026] If the standard deviation of the hole group spacing exceeds the limit, the drill feed speed is corrected according to an exponential function. The sensitivity coefficient ranges from 0.1 to 0.3, and the reference spacing standard deviation is 0.02 mm;

[0027] If the coefficient of variation of the hole diameter exceeds the limit, the drill wear compensation amount is jointly determined by the reference compensation amount, the deviation of the process capability index, and the cumulative usage time of the drill. The time attenuation coefficient is 0.001 / mm.

[0028] As a preferred solution, the closed-loop control in step S5 includes:

[0029] Adopt an anti-saturation fuzzy PID controller to adjust the processing parameters. Its control law combines proportional, integral, differential terms and anti-chattering compensation terms. The integral term is limited by a threshold to prevent overshoot;

[0030] After adjusting the processing parameters, the covariance matrix is updated in real time based on the data within the current sliding window.

[0031] As a preferred solution, the method includes:

[0032] Introduce a multi-head attention convolutional neural network MA-CNN, calculate the attention weights of different dimensions through multiple query and key matrices, and enhance the detection ability for local anomalies;

[0033] The anomaly probability is determined by fusing the SPC analysis results, the output of the MA-CNN, and the mean value of the historical anomaly probability. The fusion weights are normalized through the Sigmoid function, and the historical weight ranges from 0.2 to 0.4.

[0034] As a preferred solution, the method is applicable to the detection of high-density thermoelectric separation aluminum substrates, and the adjustment coefficient of the dynamic control limit is dynamically adjusted according to the substrate thermal deformation amount, original size, temperature rise, and thermal expansion coefficient.

[0035] As a preferred solution, the calculation of the local thermal expansion coefficient includes:

[0036] Obtain the temperature distribution of the PCB substrate in real time based on the infrared thermal imaging data;

[0037] Dynamically correct the thermal expansion coefficient according to the difference between the local temperature and the reference temperature, and the reference temperature is 25 °C.

[0038] As a preferred solution, the historical weight is dynamically adjusted according to the harmonic mean of the precision rate of the SPC model and the recall rate of the MA-CNN model.

[0039] It can be seen from the technical solutions provided by the present invention above that a PCB forming detection method based on a hole position scatter diagram and SPC analysis provided by the present invention has the following beneficial effects:

[0040] High-precision data acquisition and visualization analysis: Use an optical scanner or a laser measurement device to obtain hole position data, generate a comparison scatter diagram including theoretical and actual positions through coordinate mapping, and dynamically identify abnormal aggregation areas in combination with an improved density clustering algorithm, which can accurately locate abnormal hole position distributions and provide an intuitive basis for quality analysis;

[0041] Multi-dimensional SPC analysis and dynamic monitoring: Based on an improved skewness-kurtosis joint test, a multivariate T² control chart with dynamic covariance update, and an adaptive ARIMA model, realize real-time monitoring and trend prediction of parameters such as hole position density, spacing, and aperture. Compared with a single model, the detection sensitivity is significantly improved, and small fluctuations in the processing process can be effectively captured;

[0042] Adaptive control and closed-loop optimization: Construct a non-linear adaptive control limit, dynamically adjust it according to the sample size and batch complexity, and reduce the misjudgment rate of the control limit; when the CPK is lower than the threshold, generate targeted optimization instructions based on characteristic parameters such as hole position density deviation, spacing standard deviation, and aperture coefficient of variation, such as optimizing the drilling path based on spatial autocorrelation weight, dynamic compensation for drill bit wear, etc., effectively reducing the aperture coefficient of variation and significantly improving the processing stability;

[0043] Intelligent algorithm fusion to enhance anomaly detection: Introduce a multi-head attention convolutional neural network (MA-CNN) to fuse anomaly probabilities with the SPC model and historical data, improve the detection accuracy of small hole position offsets, effectively reduce the false alarm rate compared with traditional methods, and effectively avoid unnecessary shutdowns;

[0044] Special scenario adaptation ability: For high-density thermoelectric separation aluminum substrates, dynamically control the correlation between the limit adjustment coefficient and the thermal deformation amount, combine infrared thermal imaging to calculate the local thermal expansion coefficient in real time, effectively improve the hole position processing accuracy in thermally sensitive areas, and meet the requirements of high-end PCB manufacturing;

[0045] Model dynamic evolution and reliability: In closed-loop control, an anti-saturation fuzzy PID controller is adopted to reduce the overshoot of parameter adjustment; after the processing parameters are adjusted, the covariance matrix of the SPC model is updated in real time, and the historical weight is dynamically adjusted according to the model confidence level, effectively shortening the adaptation period of the system to process changes and continuously maintaining the detection accuracy. Brief Description of the Drawings

[0046] Figure 1 It is a schematic flow chart of a PCB forming detection method based on a hole position scatter diagram and SPC analysis according to the present invention. Detailed Embodiments

[0047] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, 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 used to limit the present invention.

[0048] In order to better understand the above technical solutions, the above technical solutions will be described in detail below in conjunction with the specification drawings and specific embodiments.

[0049] As Figure 1 shown, an embodiment of the present invention provides a PCB forming detection method based on a hole position scatter diagram and SPC analysis, including the following steps:

[0050] S1. Collect the actual hole position data of the PCB substrate to generate a hole position scatter diagram, which includes the coordinates, aperture and tolerance range of each hole position;

[0051] S2. Based on the statistical process control SPC model, perform multi-dimensional analysis on the hole position scatter diagram to extract hole position distribution characteristic parameters, including hole position density deviation, hole group spacing standard deviation and aperture variation coefficient;

[0052] S3. Construct a dynamic control limit according to the characteristic parameters, calculate the process capability index CPK in real time, and judge whether the hole position processing process is in a statistically controlled state;

[0053] S4. When the CPK is lower than the preset threshold, generate an optimization instruction, and the optimization instruction includes drilling path correction parameters, drill bit wear compensation values or equipment calibration schemes;

[0054] S5. Feed the optimization instruction back to the PCB forming equipment to adjust the processing parameters to achieve closed-loop control.

[0055] In this embodiment, the generation of the hole position distribution diagram in step S1 includes:

[0056] Obtain the actual hole position data of the PCB substrate through an optical scanner or laser measuring device;

[0057] Map the actual hole position data to the design drawings to generate a scatter plot comparing the theoretical position and the actual position;

[0058] An improved density clustering algorithm is used to identify abnormal clustering areas in hole position distribution. The density threshold and minimum number of points of the improved density clustering algorithm are dynamically set according to the area of the PCB substrate, the reference area, the total number of holes and the preset number of clusters. The density threshold is adjusted proportionally with the square root of the substrate area, and the minimum number of points is determined by rounding up the ratio of the total number of holes to the preset number of clusters.

[0059] Furthermore, the function of step S1 is to construct a visual map including hole coordinates, hole diameters, tolerances and abnormal distributions through high-precision data collection and intelligent analysis, so as to provide multi-dimensional basic data for subsequent SPC analysis. The following is a detailed description of step S1:

[0060] Step S1-1: Multi-device collaborative data collection:

[0061] Dual collection of hole position data using optical scanners and laser measuring devices:

[0062] Optical scanner: Use a linear array CCD camera to scan the PCB substrate line by line, identify the hole contour through image edge detection algorithm (such as Canny operator), combine perspective transformation to correct optical distortion, and output the two-dimensional coordinates (X, Y) of the hole with an accuracy of up to ±5μm;

[0063] Laser measurement device: Use semiconductor laser emitter to emit a thin light beam. When the light beam scans the edge of the hole, the hole center coordinates and aperture size are calculated by the triangular reflection principle. The specific process is: the laser spot is diffusely reflected by the edge of the hole and then received by the CMOS sensor. The actual hole position (error ≤ ±2μm) and aperture (resolution 0.1μm) are calculated based on the spot offset.

[0064] Data fusion: Perform consistency check on the coordinate and aperture data collected by the two types of equipment, remove abnormal points whose deviation exceeds the tolerance range (such as ±10μm), and generate the initial hole position data set;

[0065] Step S1-2: Coordinate mapping and comparison scatter plot construction:

[0066] Spatial registration and visual mapping of actual hole position data and design drawings:

[0067] Coordinate system alignment:

[0068] Extract the reference marks (such as positioning holes, board edge contours) in the design drawings, and calculate the transformation parameters (translation vector (ΔX, ΔY), rotation angle θ) between the actual coordinate system and the design coordinate system through the Iterative Closest Point (ICP) algorithm to achieve rigid registration between the two;

[0069] Perform scaling correction on the registered actual hole position coordinates to compensate for the thermal expansion or mechanical deformation during the processing of the PCB substrate (the deformation coefficient is calculated through the CTE parameter of the substrate material and the processing temperature difference);

[0070] Comparison graph generation:

[0071] Use professional drawing tools (such as MATLAB, Matplotlib in Python) to draw a scatter plot, where:

[0072] Theoretical hole positions: Represented by hollow circles (diameter 2mm), colored blue, and the center coordinates correspond to the design drawing data;

[0073] Actual hole positions: Represented by solid circles, colored red, the center coordinates are the measured data after registration, and the size of the circles is scaled according to the aperture ratio (e.g., a 1mm aperture corresponds to a 1mm diameter circle);

[0074] Deviation connection lines: Connect the theoretical and actual centers of the same hole positions with black dashed lines. The line length intuitively reflects the position offset, and the connection lines exceeding the tolerance range (such as ±50μm) are thickened for display;

[0075] Legend annotation: Add the aperture tolerance band (such as ±10μm), coordinate scale (unit: mm), and scale (1:1) to the graph for quantitative analysis;

[0076] Step S1-3: Identification of abnormal regions based on dynamic density clustering:

[0077] Adopt an improved DBSCAN algorithm to detect abnormal aggregations or discrete points in the hole position distribution, and the algorithm parameters are dynamically adapted to the characteristics of the PCB layout:

[0078] Dynamic parameter calculation: , (where is the density threshold of the current PCB (unit: mm), reflecting the tightness of the hole position aggregation, increasing non-linearly with the increase of the substrate area to avoid misjudging large-area substrates with sparse hole positions as abnormal; is the reference density threshold, default 0.1mm, applicable to standard PCBs (area ≤ 100 cm²); is the measured area of the PCB substrate (unit: cm²), calculated by the image edge detection algorithm for the contour area; is the reference area, fixed at 100 cm² for area normalization, is the minimum number of points; is the total number of holes on the substrate, obtained from the data statistics in step S1-1; is the preset number of clusters, with a value range of 3-10, manually set according to the complexity of the hole distribution (e.g., 8-10 for high-density boards and 3-5 for low-density boards); is the ceiling function to ensure that MinPts is a positive integer and avoid clustering failure due to insufficient number of holes);

[0079] Clustering execution:

[0080] Taking each hole as the core point, calculate its the number of holes in the neighborhood (denoted as ):

[0081] If , it is determined as a core point, and all holes in the neighborhood are marked as the same cluster (such as cluster 1, cluster 2,...);

[0082] If and does not belong to the neighborhood of any core point, it is determined as a noise point (abnormal aggregation area);

[0083] Visual annotation:

[0084] In the hole scatter plot:

[0085] Normal clustering holes: represented by green diamonds, and the holes within the same cluster are connected by gray solid lines to highlight the distribution pattern;

[0086] Noise points: marked with red pentagrams, and their coordinates, aperture diameters, and distances from the cluster center are listed beside the graph (such as "Noise point A: coordinates (10.2, 15.5) mm, 2.3 mm away from the nearest cluster center");

[0087] Add analysis notes: such as "The abnormal aggregation area may be caused by drill bit vibration and substrate clamping offset, and needs to be monitored closely";

[0088] Step S1-4: Tolerance range and attribute annotation:

[0089] Overlay the tolerance information and physical attributes of the holes in the scatter plot:

[0090] Aperture tolerance: Mark the allowable deviation range of each hole in the form of concentric circles. For example, if the theoretical aperture is 2 mm and the tolerance is ±0.05 mm, draw gray dashed circles with radii of 1.95 mm (inner circle) and 2.05 mm (outer circle);

[0091] Material properties: Different areas of the substrate (such as those with a metal heat dissipation layer and FR-4 substrate) are distinguished by shading, and the shading color corresponds to the coefficient of thermal expansion (CTE) value (for example, a red shading indicates CTE = 18×10 -6 / °C, and a blue shading indicates CTE = 6×10 -6 / °C), providing a visual reference for subsequent thermal deformation compensation;

[0092] Output results of step S1:

[0093] Hole position scatter plot file: An interactive PDF or vector graph containing coordinates, hole diameters, tolerances, and clustering results, supporting zooming to view details;

[0094] Data file: A hole position data set in CSV format, with fields including hole position ID, theoretical coordinates, actual coordinates, measured hole diameter value, clustering label, noise marker, etc., for the SPC model in step S2 to call;

[0095] Technical advantages: Through multi-source data fusion, dynamic clustering, and visual annotation, a "data-graph" two-dimensional representation of the hole position processing quality is achieved, laying a precise foundation for subsequent process capability analysis based on SPC.

[0096] In this embodiment, the SPC model in step S2 includes:

[0097] An improved skewness-kurtosis joint test model for establishing the hole position density distribution is established, and the normality of the distribution is judged by calculating the joint statistic of the sample skewness and kurtosis;

[0098] A multivariate T² control chart with dynamic covariance update is used to jointly monitor the hole group spacing and hole diameter, where the covariance matrix is dynamically updated through the weighted average of historical data and real-time data, and the weight is controlled by a forgetting factor;

[0099] An adaptive ARIMA model is introduced to predict the processing trend, and its differencing order is dynamically adjusted according to the ratio of the standard deviation of the prediction error to the reference standard deviation;

[0100] Furthermore, the role of step S2 is to perform multi-dimensional in-depth analysis on the data in the hole position scatter plot based on the statistical process control (SPC) model, extract key characteristic parameters that can reflect the stability and quality characteristics of the hole position processing process, and provide a quantitative basis for subsequent judgment of whether the processing process is under control; the specific steps are as follows:

[0101] Step S2-1: Establish a normality test model for the hole position density distribution:

[0102] Data preprocessing: From the hole position scatter plot dataset generated in step S1, extract the coordinate information of each hole position, calculate the number of hole positions per unit area, and use this as the hole position density data; to ensure the validity of the data, eliminate extreme values caused by measurement errors or abnormal factors;

[0103] Calculate the sample skewness and kurtosis: For the hole position density sample data 、 、 、 , the sample skewness and the sample kurtosis are calculated as follows: , (where, represents the th sample value, represents the index of the sample ( ), is the number of samples, is the sample mean, is the sample standard deviation);

[0104] Normality test: Use the improved combined skewness-kurtosis test formula: (where, is the test statistic, is the sample skewness, is the sample kurtosis); Compare the calculated value with the preset critical value. If the value is less than the critical value, it is considered that the hole position density distribution conforms to the normal distribution; otherwise, it indicates that the distribution is abnormal and there may be problems such as unstable processing;

[0105] Step S2-2: Construct a multivariate T² control chart with dynamic covariance update:

[0106] Determine the monitoring variables: Select the hole position coordinates ( 、 direction), and the hole diameter size as the monitoring variables to form a multivariate data vector (where, is the hole diameter);

[0107] Initialize the covariance matrix: Based on the hole position data during the historical stable production process, calculate the initial covariance matrix ;

[0108] Dynamic covariance update: As new hole position data is continuously collected, use the following formula to update the covariance matrix: (where, is the updated covariance matrix, is the historical covariance matrix, is the real-time covariance matrix calculated based on the current new sample data, is the forgetting factor, with a value range of 0.7 - 0.9, used to adjust the weights of historical data and new data in covariance update);

[0109] Calculate the T² statistic: According to the updated covariance matrix, calculate the statistic of the multivariate T² control chart: (where, is the statistic, is the sample mean vector, is the covariance matrix 's inverse matrix), by comparing the value with the control limit, determine whether the current hole position processing process is in a controlled state;

[0110] Step S2 - 3: Introduce an adaptive ARIMA model to predict the processing trend:

[0111] Data stationary processing: Conduct time series analysis on the hole position characteristic data (such as hole position density, hole diameter size, etc.) to judge the stationarity of the data; if the data is not stationary, make it stationary through differencing operations;

[0112] Determine the differencing order: The differencing order d is dynamically adjusted according to the processing fluctuations, and the calculation formula is: (where, is the differencing order, is the standard deviation of the prediction error, reflecting the deviation degree between the model prediction value and the actual value; is the reference standard deviation threshold, with a value of ; is the floor function);

[0113] Model parameter estimation and prediction: According to the determined differencing order , combined with the autocorrelation function (ACF) and partial autocorrelation function (PACF) to determine the other parameters of the ARIMA model ( , ), use historical data to estimate the model parameters; through the trained adaptive ARIMA model, predict the future hole position processing data, discover the changes in the processing trend in advance, and provide a basis for timely adjustment of processing parameters;

[0114] Step S2 - 4: Extract the hole position distribution characteristic parameters:

[0115] Calculate the hole position density deviation: Compare the actual hole position density obtained in step S2-1 with the standard hole position density required by the design, and calculate the difference between the two, which is the hole position density deviation; this deviation value reflects the degree of deviation between the actual hole position distribution and the design requirements.

[0116] Calculate the standard deviation of the hole group spacing: For each hole position in the hole group, calculate the spacing between every two hole positions to obtain a hole group spacing data set; calculate the standard deviation based on this data set. The standard deviation of the hole group spacing can reflect the dispersion degree of the hole positions within the hole group. The larger the standard deviation, the greater the fluctuation of the hole position spacing and the lower the processing accuracy.

[0117] Calculate the coefficient of variation of the hole diameter: According to the hole diameter data collected in step S1, calculate the average value and standard deviation of the hole diameter. The formula for the coefficient of variation of the hole diameter is: (where is the coefficient of variation of the hole diameter, is the standard deviation of the hole diameter, is the average value of the hole diameter); the coefficient of variation of the hole diameter is used to measure the dispersion degree of the hole diameter size relative to the average hole diameter and is an important indicator for evaluating the consistency of hole diameter processing.

[0118] Through the above steps, comprehensively and deeply analyze the data of the hole position scatter diagram, and extract key characteristic parameters such as hole position density deviation, standard deviation of hole group spacing, and coefficient of variation of hole diameter, which provides important data support for subsequent construction of dynamic control limits, calculation of process capability indices, and judgment of the stability of the hole position processing process.

[0119] In this embodiment, the method for constructing the dynamic control limit in step S3 is as follows:

[0120] Adopt a non-linear adaptive control limit adjustment strategy, where the control limit bandwidth is dynamically adjusted according to the mean and standard deviation of the characteristic parameters, and a smooth transition is achieved through the hyperbolic tangent function.

[0121] The sliding window length is dynamically optimized according to the logarithmic relationship between the total sample number of the current processing batch and the complexity index, and the upper limit of the window length is a preset maximum value.

[0122] Further, the function of step S3 is: Based on the characteristic parameters extracted in step S2, establish an adaptive control boundary and quantify the processing stability to achieve real-time statistical monitoring of the hole position processing process. The specific operation steps are as follows:

[0123] Step S3-1: Construction of dynamic control limit:

[0124] Adopt a non-linear adaptive control limit model, and adjust the control limit range in combination with the change in sample size and the characteristics of the processing batch. The formula is as follows:

[0125] Non-linear control limit adjustment formula: (where are the upper / lower control limits of the characteristic parameters; is the sample mean of the hole position characteristic parameters (such as density deviation, coefficient of variation of hole diameter), reflecting the central tendency of the processing process; is the sample standard deviation, measuring the fluctuation range of the characteristic parameters; is the adjusted control limit coefficient, with a default value of 3 (corresponding to a 99.73% confidence interval), and is dynamically adjusted according to the thermal deformation of the substrate in the detection scenario of high-density thermoelectric separation aluminum substrates ( , where is the reference control limit coefficient (default value of 3, corresponding to a 99.73% confidence interval), is the basic thermal deformation amount; is the basic original size; is the thermal expansion coefficient of the aluminum substrate; is the temperature rise, that is, the difference between the current temperature of the substrate and the normal temperature (reference temperature 25°C), and is obtained in real time through infrared thermal imaging); is updated in real time to the number of holes already collected; is the reference sample size (50), used to smooth the transition of the control limits and avoid misjudgment caused by a small sample size; is the hyperbolic tangent function, which approaches 1 when , and the control limits converge to the traditional 3σ boundary; when , it approaches -1, and the control limits are temporarily expanded to adapt to data fluctuations);

[0126] Implementation process:

[0127] For each characteristic parameter (such as hole position density deviation), calculate the current mean and the standard deviation in real time;

[0128] Dynamically adjust the control limit width according to the current sample size n, and draw the control limit curve changing with time (such as the UCL is a red dashed line, the LCL is a blue dashed line, and the mean is a black solid line);

[0129] Overlay the control limits in the hole position scatter plot or SPC control chart, and mark whether the current sample point exceeds the boundary (if it exceeds, it is marked with a red diamond);

[0130] Dynamic optimization of the sliding window length: (where is the maximum window length (200), to avoid lag in response to sudden anomalies caused by too long a window; is the cumulative processing time of the current batch (unit: minute), reflecting the processing progress; is the batch complexity index (0 - 1), which is comprehensively evaluated by factors such as the number of substrate layers, hole density, material type, etc. (such as high-density interconnect boards , single-sided boards ); is the natural logarithm function, which is used to suppress the linear influence of the complexity index on the window length and ensure the smooth change of the window length;

[0131] Application logic:

[0132] Complex batches (such as multi-layer boards) use a longer window ( ) to improve the stability of trend analysis;

[0133] Simple batches (such as single-sided boards) use a shorter window ( ) to speed up the response to anomalies;

[0134] Step S3-2: Real-time calculation of the process capability index (CPK):

[0135] For the key quality characteristics of hole position processing (such as hole diameter, hole position coordinate deviation), the following formula is used to calculate CPK: (where CPK is the process capability index, which reflects the ability of the processing process to meet the specification requirements. The larger the value, the stronger the process capability; LSL is the lower specification limit, that is, the minimum allowable value of the hole position characteristic parameter (such as the minimum tolerance value of the hole diameter); USL is the upper specification limit, that is, the maximum allowable value of the hole position characteristic parameter (such as the maximum tolerance value of the hole diameter); is the current sample mean, is the current sample standard deviation, both calculated based on the latest data within the sliding window);

[0136] Step S3-3: Judgment of statistical control status:

[0137] Control limit judgment rule:

[0138] If the real-time values of the hole position characteristic parameters (such as density deviation, spacing standard deviation, hole diameter coefficient of variation) all fall within the UCL / LCL range, and there are no abnormal patterns such as 7 consecutive points increasing / decreasing, 9 consecutive points on the same side of the mean (following the SPC control chart out-of-control criteria), then it is determined that the process is in a statistically controlled state;

[0139] If any characteristic parameter breaks through the control limit or an abnormal pattern appears, a warning signal is triggered, indicating that the cause needs to be further analyzed;

[0140] CPK threshold judgment:

[0141] The preset CPK threshold is usually 1.33 (corresponding to the "sufficient ability" level of the ISO9001 standard):

[0142] When CPK ≥ 1.33, it is determined that the process capability is sufficient and the processing quality is stable;

[0143] When CPK < 1.33, it is determined that the process capability is insufficient, and the optimization mechanism of step S4 needs to be started;

[0144] Step S3-4: Visual monitoring of control limits and CPK:

[0145] Integrate the dynamic control limits and CPK values into the real-time monitoring system interface:

[0146] Control chart display: With time as the horizontal axis and hole position characteristic parameters as the vertical axis, draw the UCL, LCL, mean line (CL) and real-time data points. Abnormal points are automatically marked in red and a pop-up window prompts (such as "The coefficient of variation of the hole diameter breaks through the UCL, the current value 0.08 > 0.05");

[0147] CPK trend curve: Real-time draw the curve of the CPK value changing with time, set the warning intervals of green (CPK ≥ 1.33), yellow (1.0 ≤ CPK < 1.33), and red (CPK < 1.0), and intuitively reflect the fluctuation of the process capability;

[0148] Data storage and traceability: Save the control limit parameters, CPK values and abnormal records according to the time stamp, support historical data query and trend analysis, and provide data support for process improvement;

[0149] Key technical points of step S3:

[0150] Dynamic adaptability: Through the hyperbolic tangent function and the sliding window mechanism, the control limits can be automatically adjusted according to the sample size and batch complexity, solving the limitations of traditional fixed control limits in small samples or complex processes;

[0151] Multi-dimensional monitoring: Combine control limit outlier judgment and CPK quantitative evaluation, judge the processing status from the two dimensions of "process stability" and "capability sufficiency", and improve the detection reliability;

[0152] Real-time response: Update the control limits and CPK based on the latest data (within the window length W), ensure that the monitoring system responds to changes in the processing process within seconds, and promptly capture signs of quality fluctuations.

[0153] In this embodiment, the generation of the optimization instructions in step S4 includes:

[0154] If the deviation of the hole position density exceeds the limit, optimize the drilling path using spatial autocorrelation weights. The weights are calculated based on the Euclidean distance between hole positions and the difference in local thermal expansion coefficients, and the default values of the attenuation factors are 0.1 and 0.05;

[0155] If the standard deviation of the hole group spacing exceeds the limit, the drill feed speed is corrected according to the exponential function, the sensitivity coefficient ranges from 0.1 to 0.3, and the reference standard deviation of the spacing is 0.02 mm;

[0156] If the coefficient of variation of the hole diameter exceeds the limit, the drill wear compensation amount is jointly determined by the reference compensation amount, the deviation of the process capability index, and the cumulative usage time of the drill. The time decay coefficient is 0.001 / mm;

[0157] Furthermore, the function of step S4 is as follows: for the insufficient machining process capability (CPK < preset threshold, such as 1.33) determined in step S3, by analyzing the reasons for the overlimit of the hole position distribution characteristic parameters in multiple dimensions, generating accurate process optimization instructions, and realizing the intelligent adjustment of machining parameters. The specific operation steps are as follows:

[0158] Step S4-1: Optimization strategy for the overlimit of hole position density deviation:

[0159] Trigger condition: The hole position density deviation > the design allowable threshold (such as ±5%), indicating that the hole position distribution deviates from the theoretical value, which may be caused by an unreasonable drilling path or substrate thermal deformation;

[0160] Optimize the drilling path with spatial autocorrelation weight: (where is the spatial weight between hole positions and ), which is used to measure the influence of the drilling sequence on thermal deformation. The smaller the weight, the higher the priority; is the Euclidean distance (unit: mm) between hole positions and . The closer the distance, the more significant the thermal accumulation effect; , are the thermal expansion coefficients (unit: , in the regions where hole positions C) are located, reflecting the thermal deformation characteristics of the material; (default 0.1 mm -2 ), (default 0.05 C / mm) are attenuation factors, which adjust the influence ratio of the distance and CTE difference on the weight);

[0161] Path planning implementation:

[0162] Construct a priority queue for the drilling sequence according to the weight matrix , and give priority to machining the holes with low weights (such as the holes at the edge of the dense area) to reduce the local temperature rise caused by continuous drilling;

[0163] For regions with high CTE differences (such as the junction between metal inserts and FR-4 substrates), insert cooling intervals (such as pausing for 2 seconds every 5 holes drilled) to reduce the accumulation of thermal deformation;

[0164] Step S4-2: Optimization strategy for the standard deviation of hole group spacing exceeding the limit:

[0165] Trigger condition: The standard deviation of hole group spacing > reference value (such as 0.02 mm), indicating that the hole position spacing fluctuates too much, which may be caused by unstable drill feed speed or equipment vibration;

[0166] Drill feed speed correction: (where, is the corrected feed speed (unit: mm / s), is the original speed; is the sensitivity coefficient (0.1 - 0.3), and the value decreases as the drill diameter increases (such as 0.3 for a φ1 mm drill and 0.1 for a φ3 mm drill); is the deviation between the current standard deviation and the target value (0.02 mm); is the reference spacing standard deviation, fixed at 0.02 mm);

[0167] Vibration suppression measures:

[0168] If the corrected speed is lower than the minimum process speed (such as 5 mm / s), automatically start the equipment vibration monitoring module, and detect the spindle vibration frequency through an acceleration sensor;

[0169] When the vibration frequency is coupled with the drill's natural frequency (such as detecting a 100 Hz resonance peak), adjust the spindle speed (such as from 20000 rpm to 18000 rpm) to avoid the resonance range;

[0170] Step S4-3: Optimization strategy for the coefficient of variation of hole diameter exceeding the limit:

[0171] Trigger condition: The coefficient of variation of hole diameter (CV) > 5%, indicating poor hole diameter consistency, mainly caused by drill wear or spindle runout;

[0172] Drill wear compensation calculation: (where, is the drill wear compensation value (unit: mm), used to adjust the drilling depth or hole diameter compensation parameters; is the reference compensation amount (default 0.01 mm), preset based on the drill material and the hardness of the processed board; is the current process capability index (only reflecting the hole diameter size), is the target value (1.33); is the time decay coefficient (0.001 mm / min), reflecting the cumulative effect of drill wear over time; is the cumulative drill usage time (unit: minutes), which is collected in real time by the equipment sensor);

[0173] Spindle accuracy calibration:

[0174] If the CV is still > 5% after compensation, trigger the spindle radial runout detection: Use a dial indicator to measure the radial displacement when the spindle rotates. If the runout amount > 10μm, automatically generate an equipment calibration instruction, including:

[0175] Adjust the preload force of the spindle bearing (e.g., re-tighten with a torque wrench according to the standard value of 15 N·m);

[0176] Clean and align the drill holder (use a drill holder cleaner to remove oil stains, and calibrate the tool yaw < 5μm through a tool setter);

[0177] Step S4-4: Multi-modal generation of optimization instructions:

[0178] Instruction type and format:

[0179] Digital signal: The drilling path correction parameters (such as the hole position processing sequence list) and the feed speed value (such as v = 12 mm / s) are transmitted to the machine tool controller in real time through the RS-232 interface;

[0180] Text work order: The equipment calibration plan (such as "The spindle runout exceeds the tolerance, and the accuracy calibration process needs to be executed") is pushed to the mobile device of the equipment maintenance personnel in PDF format;

[0181] Visual prompt: Mark the optimization items with icons on the machine tool HMI interface (such as a blue arrow indicating path adjustment and a yellow wrench indicating calibration requirements);

[0182] Instruction priority management:

[0183] Emergency instructions (such as the drill bit wear causing the hole diameter to exceed the tolerance) have the highest priority, immediately interrupt the current machining and execute compensation;

[0184] Optimization instructions (such as path adjustment) are automatically loaded after the current batch of machining is completed to avoid interrupting continuous production;

[0185] Calibration instructions are forced to stop after being triggered 3 times to ensure the restoration of equipment accuracy;

[0186] Technical advantages of step S4:

[0187] Accurate positioning of the root cause of the fault: Correlate specific process problems through the over-limit types of characteristic parameters (density deviation / spacing fluctuation / hole diameter variation), avoiding the blindness of empirical adjustment;

[0188] Multi-parameter collaborative optimization: Integrate geometric parameters (hole position coordinates), physical parameters (CTE, vibration frequency), and processing parameters (speed, compensation amount) to achieve multi-field coupling optimization of heat-machine-control;

[0189] Closed-loop control pre-adaptability: The optimization instructions include both immediate adjustments (such as speed correction) and long-term maintenance (such as equipment calibration), taking into account both production efficiency and quality stability.

[0190] In this embodiment, the closed-loop control in step S5 includes:

[0191] Adopt an anti-saturation fuzzy PID controller to adjust the processing parameters. Its control law combines proportional, integral, differential terms, and anti-chattering compensation terms. The integral term is limited by a threshold to prevent overshoot;

[0192] After the processing parameters are adjusted, the covariance matrix is updated in real time based on the data within the current sliding window;

[0193] Furthermore, the function of step S5 is: Through a reliable data transmission link and intelligent control algorithm, the optimization instructions generated in step S4 are accurately mapped to the actuator of the PCB forming equipment, forming a complete closed loop of "detection - analysis - adjustment" to ensure that the processing parameters are dynamically adapted to the quality requirements. The specific operation steps are as follows:

[0194] Step S5-1: Protocol conversion and transmission of optimization instructions:

[0195] Multi-device interface adaptation:

[0196] For equipment of different brands (such as KohYoung in Germany, Han's Laser in China), develop a dedicated communication protocol adapter to convert the optimization instructions (such as drilling path, compensation value) into control codes recognizable by the equipment:

[0197] Drilling path: Convert the hole position priority queue into a processing sequence instruction in G code (such as G00X10Y15; G01Z-2F100);

[0198] Bit compensation: Convert the wear compensation value into the tool radius compensation parameter of the machine tool (such as , assuming the drill bit diameter is 1mm, is the register number for tool radius compensation in the machine tool control system, used to store the compensation value of the drill bit radius);

[0199] Equipment calibration: Convert the spindle runout calibration instruction into a PLC control signal (such as the M60 code triggers the calibration program);

[0200] Adopt the OPCUA industrial communication standard to achieve real-time transmission of instruction data through Ethernet, with a communication delay ≤ 50ms to ensure the synchronization of instructions and processing progress;

[0201] Transmission reliability guarantee:

[0202] Append a CRC check code to the instruction data packet. After successful verification at the receiving end, execute it. If the verification fails, automatically retransmit (up to 3 times);

[0203] Establish an instruction execution log system to record the sending time of each instruction, the device receiving status, and the execution result (success / failure) for subsequent traceability and analysis;

[0204] Step S5-2: Parameter adjustment based on anti-windup fuzzy PID:

[0205] Use an anti-windup fuzzy PID controller to achieve dynamic adjustment of machining parameters. The control law formula is: (where, is the controller output (such as the feed speed adjustment amount, spindle speed correction value); , , are the proportional coefficient, integral coefficient, and differential coefficient respectively, and are adjusted online through fuzzy inference (such as when the error m, automatically increase to speed up the response); Target value - actual value, such as hole position coordinate deviation, hole diameter size error; is the saturation function: ( is the integral anti-windup threshold, default , to prevent the integral term from accumulating excessively during large errors and causing system overshoot); is the sign function, used to judge the error direction; 1 is the anti-chattering coefficient( ), to suppress the control quantity fluctuation during small errors);

[0206] Application scenario:

[0207] Feed speed adjustment: When there is a deviation between the calculated in step S4-2 and the current speed, the controller dynamically adjusts the output of the frequency converter according to the real-time error , and the adjustment accuracy is ;

[0208] Drill bit compensation execution: Drive the servo motor through the controller to adjust the Z-axis feed depth to compensate for the hole diameter deviation caused by drill bit wear, and the positioning accuracy is ;

[0209] Step S5-3: Real-time iterative update of the SPC model:

[0210] After adjusting the processing parameters, the SPC model parameters in step S2 are updated synchronously to ensure that the monitoring system reflects the latest process status:

[0211] Covariance matrix update: (where reflects the dispersion degree of the hole position data after adjustment; is the current sliding window length (dynamically determined by step S3-1); is the th hole position data vector (including coordinates and aperture) within the window; is the data mean vector within the window);

[0212] Adaptive correction of model parameters:

[0213] If the CPK values of 5 consecutive samples after adjustment are stable above 1.33, it is determined that the optimization is effective, and the current covariance matrix is stored in the model library as historical data;

[0214] If the CPK does not meet the standard (such as <1.0), secondary optimization is automatically triggered: step S2-(4) is re-executed, and the optimization times are recorded in the system (manual intervention prompt is triggered when the cumulative number is ≥3 times);

[0215] Step S5-4: Closed-loop control effect verification and visualization:

[0216] Real-time data feedback:

[0217] Embed a closed-loop control monitoring dashboard in the device HMI interface and display in real time:

[0218] Execution status of optimization instructions (such as "Path adjustment has been loaded", "Drill bit compensation value +0.005mm");

[0219] Trends of key quality parameters (such as mean aperture, standard deviation of hole position deviation), use a solid green line to represent the data after adjustment, and a dashed red line to represent the data before adjustment for an intuitive comparison of the optimization effect;

[0220] Statistical process verification:

[0221] After each processing batch is completed (such as 50 substrates), a closed-loop control effect report is automatically generated, including:

[0222] Increase in CPK value (such as from 0.8 to 1.5);

[0223] Reduction in feature parameter fluctuation (such as standard deviation of hole group spacing from 0.03mm to 0.015mm);

[0224] Execution efficiency of optimization instructions (such as average instruction transmission delay of 23ms, execution success rate of 99.7%);

[0225] Exception handling mechanism:

[0226] If the closed-loop control fails to meet the CPK standard twice in a row, the system automatically generates a "Process Abnormality Analysis Report", summarizes the hole position data, equipment status logs, and optimization records, and pushes them to the process engineer's account to prompt for manual intervention (such as changing the drill bit model and adjusting the substrate fixing fixture);

[0227] Core technical points of step S5:

[0228] Heterogeneous system integration ability: Through protocol conversion and standardized interfaces, seamless docking of the detection system with equipment of different brands is achieved, and mainstream PCB processing equipment (such as drill presses and routing machines) is compatible;

[0229] Robustness of control algorithm: Anti-saturation fuzzy PID combined with feedforward compensation effectively suppresses the inertial delay of the mechanical system and external interference (such as workshop temperature fluctuations) to ensure the smoothness of parameter adjustment;

[0230] Model dynamic evolution: Continuously iterate the SPC model based on the latest processing data, gradually optimize the monitoring standard with the improvement of the process, and form an intelligent closed loop of "detection - adjustment - learning".

[0231] In this embodiment, the method includes:

[0232] Introduce the multi-head attention convolutional neural network MA-CNN, calculate the attention weights of different dimensions through multiple groups of query and key matrices, and enhance the detection ability of local anomalies. The calculation of its attention weights is: (where, is the attention weight between position and in the th attention head, , are the query and key matrices of the th attention head, is the dimensional scaling factor, is the position index in the sequence with a value range of 1 to , is the sequence length);

[0233] The anomaly probability is determined by fusing the SPC analysis results, the output of MA-CNN, and the historical anomaly probability mean. The fusion weights are normalized through the Sigmoid function. The value range of the historical weights is 0.2 - 0.4, and the anomaly probability fusion formula is improved to: (where, is the final anomaly probability, is the Sigmoid function, is the abnormal probability predicted by the SPC model, is the abnormal probability predicted by the MA-CNN model, is the mean value of historical abnormal probabilities, is the SPC model weight (0.6 - 0.8), is the historical weight (0.2 - 0.4)).

[0234] In this embodiment, the method is applicable to the detection of high-density thermoelectrically separated aluminum substrates, and the adjustment coefficient of the dynamic control limit is dynamically adjusted according to the substrate thermal deformation amount, original size, temperature rise, and thermal expansion coefficient. The formula is: (where, is the adjusted control limit coefficient, is the reference control limit coefficient, is the substrate thermal deformation amount, is the original size of the substrate, is the thermal expansion coefficient of the aluminum substrate ( ), is the temperature rise).

[0235] In this embodiment, the calculation of the local thermal expansion coefficient includes:

[0236] Obtaining the temperature distribution of the PCB substrate in real time based on infrared thermal imaging data;

[0237] Dynamically correcting the thermal expansion coefficient according to the difference between the local temperature and the reference temperature, and the reference temperature is 25°C;

[0238] Specifically:

[0239] The thermal expansion coefficient difference term The calculation of includes:

[0240] Obtaining the local area temperature distribution in real time based on infrared thermal imaging data ;

[0241] Dynamically calculating the local thermal expansion coefficient: (where, is the coordinate The local thermal expansion coefficient at, is the normal temperature thermal expansion coefficient, is the real-time temperature at this position, is the reference temperature ( ))).

[0242] In this embodiment, the historical weight is dynamically adjusted according to the harmonic mean of the precision rate of the SPC model and the recall rate of the MA-CNN model;

[0243] The historical weight is dynamically adjusted according to the model confidence: (wherein, is the historical weight after dynamic adjustment, is the precision rate of the SPC model, is the recall rate of the MA-CNN model).

[0244] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principle and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A PCB forming detection method based on a hole position scatter diagram and SPC analysis, characterized in that: It includes the following steps: S1. Collect the actual hole position data of the PCB substrate to generate a hole position scatter diagram, and the hole position scatter diagram includes the coordinates, hole diameters, and tolerance ranges of each hole position; S2. Based on the statistical process control (SPC) model, perform multi-dimensional analysis on the hole position scatter diagram, and extract hole position distribution characteristic parameters, including hole position density deviation, standard deviation of hole group spacing, and coefficient of variation of hole diameter; S3. Construct a dynamic control limit according to the characteristic parameters, calculate the process capability index CPK in real time, and determine whether the hole position processing process is in a statistically controlled state; S4. When the CPK is lower than the preset threshold, generate an optimization instruction, and the optimization instruction includes drilling path correction parameters, drill bit wear compensation values, or equipment calibration schemes; S5. Feed back the optimization instruction to the PCB forming equipment and adjust the processing parameters to achieve closed-loop control.

2. A PCB forming detection method based on a hole position scatter diagram and SPC analysis according to claim 1, characterized in that: The generation of the hole position scatter diagram in step S1 includes: Obtain the actual hole position data of the PCB substrate through an optical scanner or a laser measuring device; Perform coordinate mapping on the actual hole position data and the design drawing to generate a comparison scatter diagram including the theoretical position and the actual position; Use an improved density clustering algorithm to identify abnormal aggregation areas in the hole position distribution. The density threshold and the minimum number of points of the improved density clustering algorithm are dynamically set according to the area of the PCB substrate, the reference area, the total number of holes, and the preset number of clusters. Among them, the density threshold is adjusted proportionally to the square root of the substrate area, and the minimum number of points is determined by rounding up the ratio of the total number of holes to the preset number of clusters.

3. A PCB forming detection method based on a hole position scatter diagram and SPC analysis according to claim 1, characterized in that: The SPC model in step S2 includes: Establish an improved skewness-kurtosis joint test model for hole position density distribution based on historical data, and judge the distribution normality by calculating the joint statistic of sample skewness and kurtosis; Adopt a multivariate T² control chart with dynamic covariance update to jointly monitor the hole group spacing and hole diameter, and the covariance matrix is dynamically updated through the weighted average of historical data and real-time data, and the weight is controlled by a forgetting factor; Introduce an adaptive ARIMA model to predict the processing trend, and its differencing order is dynamically adjusted according to the ratio of the prediction error standard deviation to the reference standard deviation.

4. A PCB forming detection method based on a hole position scatter diagram and SPC analysis according to claim 1, characterized in that: The method for constructing the dynamic control limit in step S3 is: Adopt a non-linear adaptive control limit adjustment strategy, and the control limit bandwidth is dynamically adjusted according to the mean and standard deviation of the characteristic parameters, and a smooth transition is achieved through the hyperbolic tangent function; The sliding window length is dynamically optimized according to the logarithmic relationship between the total number of samples and the complexity index of the current processing batch, and the upper limit of the window length is the preset maximum value.

5. A PCB forming detection method based on a hole position scatter diagram and SPC analysis according to claim 1, characterized in that: The generation of the optimization instruction in step S4 includes: If the hole position density deviation exceeds the limit, optimize the drilling path by using spatial autocorrelation weights, and the weights are calculated according to the Euclidean distance between hole positions and the difference in local thermal expansion coefficients, and the default values of the attenuation factors are 0.1 and 0.05; If the standard deviation of the hole group spacing exceeds the limit, the drill feed speed is corrected according to an exponential function, and the sensitivity coefficient ranges from 0.1 to 0.3, and the reference spacing standard deviation is 0.02 mm; If the coefficient of variation of the hole diameter exceeds the limit, the drill bit wear compensation amount is jointly determined by the reference compensation amount, the process capability index deviation, and the cumulative usage time of the drill bit, and the time attenuation coefficient is 0.001 / mm.

6. A PCB forming detection method based on a hole position scatter diagram and SPC analysis according to claim 1, characterized in that: The closed-loop control in step S5 includes: An anti-saturation fuzzy PID controller is used to adjust the processing parameters. Its control law combines proportional, integral, differential terms and anti-chattering compensation terms. The integral term is limited by a threshold to prevent overshoot; After the processing parameters are adjusted, the covariance matrix is updated in real time based on the data within the current sliding window.

7. A PCB forming detection method based on a hole position scatter diagram and SPC analysis according to claim 1, characterized in that: The method includes: Introduce a multi-head attention convolutional neural network MA-CNN, calculate the attention weights of different dimensions through multiple groups of query and key matrices, and enhance the detection ability for local anomalies; The anomaly probability is determined by fusing the SPC analysis results, the output of MA-CNN and the mean value of historical anomaly probabilities. The fusion weights are normalized by the Sigmoid function, and the value range of the historical weights is 0.2 - 0.

4.

8. A PCB forming detection method based on a hole position scatter diagram and SPC analysis according to claim 1, characterized in that: The method is applicable to the detection of high-density thermoelectric separation aluminum substrates, and the adjustment coefficient of its dynamic control limit is dynamically adjusted according to the substrate thermal deformation amount, original size, temperature rise and thermal expansion coefficient.

9. A PCB forming detection method based on a hole position scatter diagram and SPC analysis according to claim 5, characterized in that: The calculation of the local thermal expansion coefficient includes: Based on the infrared thermal imaging data, the temperature distribution of the PCB substrate is obtained in real time; The thermal expansion coefficient is dynamically corrected according to the difference between the local temperature and the reference temperature, and the reference temperature is 25°C.

10. A PCB forming detection method based on a hole position scatter diagram and SPC analysis according to claim 7, characterized in that: The historical weight is dynamically adjusted according to the harmonic mean of the precision rate of the SPC model and the recall rate of the MA-CNN model.

Images