A collaborative control system and method for multi-process PCB processing

Abstract

This invention discloses a collaborative control system and method for multi-process PCB manufacturing. It collects multi-source process data from each process, establishes a process status baseline through cross-verification of redundant sensors, identifies process drift characteristics based on process capability review, and constructs a process benchmark library using dynamic threshold hierarchies to detect and generate process deviation parameters. It performs correlation analysis on adjacent processes to determine the transmission coupling degree, locates defect transmission links through bidirectional verification combining forward tracing and reverse backtracking, and accumulates process deviation parameters along the links to form a process correlation matrix. It backtracks defect propagation to determine the affected area, generates process status parameters based on the affected area, and extracts trend reversal features to construct a process feature map. It extracts defect amplification factors through sensitivity decoupling, identifies critical process nodes, constructs a control rule library, integrates the process feature map, and outputs process control commands, thus realizing the location of defect sources and the tracking of propagation paths.

Description

Technical Field

[0001] This invention relates to the field of intelligent PCB manufacturing technology, and in particular to a collaborative control system and method for multi-process PCB manufacturing. Background Technology

[0002] As the core carrier of electronic products, the manufacturing process of PCBs involves multiple steps, including inner layer pattern transfer, lamination and stacking, drilling, copper plating, and solder mask printing. These steps involve complex parameter transfer and quality correlations. When the manufacturing parameters of a certain step deviate, this deviation often does not remain confined to that step but propagates downstream along the production flow, accumulating step by step and ultimately manifesting as serious quality defects in the finished product. For example, improper temperature control in the lamination and stacking process can lead to abnormal stress distribution within the board. This problem manifests as hole misalignment in the drilling process and uneven plating in the copper plating process. This cross-process defect propagation characteristic poses significant challenges to PCB quality control.

[0003] Existing PCB process monitoring technologies mainly adopt a single-process independent monitoring approach, with each process having its own parameter thresholds set and alarms triggered when limits are exceeded. This approach has significant limitations: it lacks analysis of the impact transmission mechanism between upstream and downstream processes, making it difficult to accurately trace back to the source process when quality problems occur in downstream processes; the monitoring of process parameters is mostly based on static threshold determination, making it difficult to provide early warnings when parameters are still within the normal range but show a tendency to deteriorate; and the process control strategy fails to identify sensitive parameters and key node locations, resulting in dispersed control resources and limited intervention effects. Summary of the Invention

[0004] This invention discloses a collaborative control system and method for multi-process PCB manufacturing. By collecting multi-source process data from each process and performing cross-verification and filtering, a process status baseline is established. Based on this, process drift characteristics are identified and a hierarchical process benchmark library is constructed. Then, the transmission and coupling relationship between processes is analyzed to locate defect transmission links. The spread of defects is traced back to determine the scope of impact and trend turning features are extracted to construct a process feature map. Finally, the critical nodes of the process are identified through sensitivity decoupling and precise process control instructions are output to achieve collaborative management and control of the entire PCB production line.

[0005] The first aspect of this invention proposes a collaborative control method for multi-process PCB manufacturing, comprising the following steps:

[0006] Collect multi-source process data from each process of the PCB production line, and perform multi-source cross-verification and filtering on the multi-source process data to form a process status baseline;

[0007] Based on the process status baseline, process capability review is performed to identify process drift characteristics. Dynamic threshold hierarchical construction of process benchmark library is performed on the process drift characteristics. Offset detection is performed through the process benchmark library to generate process deviation parameters.

[0008] The process state baseline is subjected to inter-process correlation enhancement processing to determine the transmission coupling degree of adjacent processes. The transmission coupling degree is used to perform cross-process penetration tracking to locate the defect transmission link. The process deviation parameters are cascaded and accumulated along the defect transmission link to form a process correlation matrix.

[0009] Defect propagation backtracking and influence domain aggregation are performed on the process association matrix to determine the defect range. Process status parameters are generated based on the defect range. Trend turning features are extracted from the process status parameters to construct a process feature map.

[0010] Sensitivity decoupling is performed on the process deviation parameters and the process correlation matrix to extract the defect amplification factor. Based on the defect amplification factor, the process critical node is identified. A control rule base is constructed around the process critical node. The process feature map is integrated through the control rule base to output the process control command.

[0011] A second aspect of this invention provides a collaborative control system for multi-process PCB manufacturing, comprising:

[0012] The data acquisition module is used to collect multi-source process data of each process in the PCB production line, and to perform multi-source cross-verification and filtering on the multi-source process data to form a process status baseline.

[0013] The benchmark detection module is used to review process capability and identify process drift characteristics based on the process status baseline, perform dynamic threshold hierarchical construction of process benchmark library in the process drift characteristics, and perform offset detection through the process benchmark library to generate process deviation parameters.

[0014] The correlation analysis module is used to strengthen the inter-process correlation of the process state baseline, determine the transmission coupling degree of adjacent processes, use the transmission coupling degree to perform cross-process penetration tracking to locate the defect transmission link, and cascade and accumulate the process deviation parameters along the defect transmission link to form a process correlation matrix.

[0015] The graph construction module is used to perform defect diffusion backtracking and influence domain aggregation on the process association matrix to determine the defect range, generate process status parameters based on the defect range, and extract trend reversal features from the process status parameters to construct a process feature graph.

[0016] The instruction output module is used to perform sensitivity decoupling on the process deviation parameters and the process correlation matrix to extract the defect amplification factor, identify the process critical node based on the defect amplification factor, construct a control rule base around the process critical node, and output process control instructions by integrating the process feature map through the control rule base. Attached Figure Description

[0017] The accompanying drawings illustrate specific examples of the technical solutions described in this invention and, together with the detailed embodiments, form part of the specification, serving to explain the technical solutions, principles, and effects of this invention.

[0018] Figure 1 This is a flowchart illustrating a collaborative control method for multi-process PCB manufacturing according to the present invention.

[0019] Figure 2 This is a structural block diagram of a collaborative control system for multi-process PCB manufacturing according to the present invention. Detailed Implementation

[0020] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0021] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0022] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0023] The technical solutions of the embodiments of this application will be described below.

[0024] likeFigure 1 As shown, this embodiment of the invention provides a collaborative control method for multi-process PCB manufacturing, including the following steps S110-S150:

[0025] Step S110: Collect multi-source process data for each process of the PCB production line, and perform multi-source cross-verification and filtering on the multi-source process data to form a process status baseline.

[0026] Specifically, multi-source process data is collected for each step of the PCB production line. Data acquisition devices are deployed at each key step of the PCB production line, including inner layer pattern transfer, lamination and stacking, drilling, copper plating, and solder mask printing. The data acquisition devices obtain process parameters from multiple information sources, including temperature sensors, pressure sensors, speed encoders, vision inspection cameras, and current monitoring modules. For the same process parameter, at least two different types of sensors are used for redundant acquisition to ensure data reliability. In the lamination and stacking process, lamination temperature is simultaneously acquired using thermocouples and infrared thermometers, with the measurements from the two sensors cross-validating each other. In the drilling process, hole position coordinates are simultaneously acquired using vision inspection cameras and mechanical probes for subsequent cross-verification. In the inner layer pattern transfer process, parameters such as exposure energy, developing temperature, developing time, and etching rate are collected. In the lamination and stacking process, parameters such as preheating temperature, lamination pressure, heating rate, and holding time are collected. In the drilling process, parameters such as spindle speed, feed rate, drill wear, and hole position deviation are collected. The collected raw data is timestamped to establish a unified time benchmark, aligning data from different processes and sensors in the time dimension. The parameter data collected from each process is then categorized and organized according to data type, forming a multi-source process data set.

[0027] In some embodiments, the step of performing multi-source cross-verification filtering on the multi-source process data to form a process state baseline includes: expanding the multi-source process data according to the process flow to form a process data sequence; performing multi-source cross-verification filtering on the process data sequence to locate core process anchor points; establishing process feature imprints using the core process anchor points as trajectory references; and generating a process state baseline based on the process feature imprints.

[0028] Based on multi-source process data, a process data sequence is formed by unfolding the process flow. Various parameters from the multi-source process data are arranged according to the sequence of PCB production processes. The process flow begins with inner layer pattern transfer, proceeding sequentially through lamination, drilling, copper plating, outer layer pattern transfer, etching, and finally solder mask printing. Each process corresponds to a set of process parameters. Parameters belonging to the same process from the multi-source process data are grouped together to form a subset of parameters for that process. These parameter subsets are then arranged in series according to the sequence of processes, establishing a temporal relationship between adjacent processes, ensuring continuity between the output time of the previous process and the input time of the next process. Between the lamination and drilling processes, the interval between the lamination completion timestamp and the drilling start timestamp reflects the process connection efficiency. During normal production, this interval remains within a fixed range; an abnormally long interval may indicate a stagnation in material turnover. Through process flow unfolding, the scattered multi-source process data is integrated into a structured process data sequence, which clearly presents the evolution trajectory of process parameters as the process progresses.

[0029] Multi-source cross-validation filtering is used to locate core process anchor points in the process data sequence. The key parameter locations with the greatest impact on product quality are identified within the process data sequence. Sensitivity analysis is performed on parameters for each process step to assess the impact of parameter changes on subsequent processes and the final product quality. Data points corresponding to parameters with high sensitivity are marked as candidate anchor points. Multi-source cross-validation is performed on the data at candidate anchor points, comparing measurements from redundant sensors to verify data consistency and reliability. In the drilling process, hole position accuracy directly affects the subsequent copper plating effect and circuit conductivity. Hole position accuracy measurement points are marked as candidate anchor points. Hole position data obtained through both visual inspection and mechanical probes are cross-validated. When the difference between the hole position deviation values ​​measured by the two methods exceeds the allowable range, the data point is considered abnormal. After removing abnormal data, the location is determined as the core process anchor point. In the lamination process, lamination temperature is collected using both thermocouples and infrared thermometers. When the deviation between the two measurements is within the normal range, the data is confirmed as valid. The moment when the temperature reaches the set value is determined as the core process anchor point for this process. Candidate anchor points that pass multi-source cross-validation are selected, and one or two of the most critical data points for each process are retained as core process anchor points. Core process anchor points represent the most critical parameter locations for quality control in each process, and subsequent process monitoring will revolve around these anchor points.

[0030] A process feature imprint is established using core process anchor points as trajectory benchmarks. The parameter values ​​of the core process anchor points are extracted and used as feature vectors characterizing the process state of that process. For processes with multiple parameters, the parameters of each anchor point are weighted and combined according to their importance to form the comprehensive feature value of that process. Extending a certain range forward and backward in both directions from the core process anchor point, the variation characteristics of the parameter values ​​within this range are extracted, including the rate of change, fluctuation amplitude, and trend direction. In the lamination process, the moment when the temperature reaches the set value is used as the core process anchor point. The temperature value, pressure value, and the heating rate and pressure change rate within 30 seconds before and after that moment are extracted; these data together constitute the feature vector of that process. In the drilling process, the moment the first through hole is completed is used as the core process anchor point. The spindle speed, feed rate, and the speed fluctuation and feed stability within 10 seconds before and after that moment are extracted to form the feature vector of the drilling process. The feature vectors of each process are arranged in sequence to form a process feature imprint characterizing the process state of the entire production process. Process feature imprints are like the process fingerprints of a production line, which can uniquely identify the production process status within a specific time period. The process feature imprints of different batches of products can be used for traceability and comparative analysis.

[0031] A process status baseline is generated based on process feature imprints. Process feature imprint data from multiple normal production batches are collected to establish a statistical sample library of process features. Statistical analysis is performed on the process feature imprint data in the sample library to calculate the mean and distribution range of each process feature value. The statistical mean of each process feature value is connected according to the process sequence to form the center curve of the process status baseline. The normal fluctuation range of each process feature value is calculated, and upper and lower boundaries are established on both sides of the center curve to form the normal interval of the process status baseline. In the copper plating process, the statistical mean of the current density feature value is the baseline center. During normal production, the current density fluctuates within a certain range above and below the baseline center; fluctuations exceeding this range are considered abnormal. In the inner layer pattern transfer process, the normal fluctuation range of the exposure energy feature value is narrower; even slight deviations can lead to a decrease in linewidth accuracy. Therefore, the baseline interval for this process is set more strictly. Once the process status baseline is established, it can be used for real-time process monitoring. The process feature imprints of the current production process are compared with the baseline to identify abnormal processes that deviate from the baseline. As production data accumulates, the process status baseline is updated periodically to reflect parameter drift caused by changes in equipment status and process optimization, maintaining the timeliness and accuracy of the baseline.

[0032] Step S120: Based on the process status baseline, process capability review is performed to identify process drift characteristics. Dynamic threshold layering is performed on the process drift characteristics to construct a process benchmark library. Offset detection is performed through the process benchmark library to generate process deviation parameters.

[0033] Specifically, process capability review is conducted based on the process status baseline to identify process drift characteristics. The process characteristic imprints of the current production batch are compared step-by-step with the process status baseline, and the deviation of each process characteristic value from the baseline center curve is calculated. The deviation is continuously tracked to identify the cumulative change pattern over time; a sustained unidirectional increase in deviation indicates that process parameters are drifting. The process capability index Cpk for each process characteristic value is calculated. The process capability index Cpk is determined by the relationship between the characteristic value distribution and the normal range of the baseline: Cpk = min[(USL-μ) / 3σ, (μ-LSL) / 3σ], where USL and LSL are the upper and lower boundaries of the baseline, μ is the mean of the characteristic value, and σ is the standard deviation of the characteristic value. When the process capability index is below a set threshold, the process is considered to have a risk of process drift. In the drilling process, when the hole position accuracy characteristic value continuously deviates from the upper boundary of the baseline for multiple consecutive batches, and the Cpk value gradually decreases from 1.33 to below 1.0, the hole position accuracy parameter of this process is considered to be drifting upwards. In the lamination process, although the lamination temperature characteristic value was within the baseline range, it showed a slow upward trend for five consecutive batches, with an increase of approximately 0.3℃ per batch, indicating a drift tendency for this parameter. The drift direction, drift amplitude, and rate of change for each process were extracted, and this drift information was integrated to form process drift characteristics. These characteristics include six dimensions: process identifier, drift parameter type, drift direction, drift amplitude, rate of change, and process capability index, comprehensively describing the deviation of each process from the process baseline.

[0034] In some embodiments, the step of performing dynamic threshold stratification to construct a process benchmark library in the process drift features includes: obtaining the drift amplitude of the process drift features and comparing it with standard process specifications to determine the process standard value; setting an adaptive threshold boundary based on the process standard value and the rate of change of the process drift features; dividing the process drift features into layers according to the adaptive threshold boundary to form a drift level classification; and performing hazard assessment processing on the drift level classification to form a process benchmark library.

[0035] The drift amplitude of the process drift feature is compared with the standard process specification to determine the process standard value. Drift amplitude data for each process step is extracted from the process drift feature; the drift amplitude represents the absolute distance of the current process parameter from the baseline center. The drift amplitude in the process drift feature is compared with the parameter tolerance range specified in the industry standard process specification, which is derived from IPC standards or internal control standard documents. The proportion of the drift amplitude to the tolerance range is calculated; this proportion reflects the severity of the current drift relative to the allowable deviation. In the copper plating process, the process drift feature shows a current density drift amplitude of 0.5 A / dm², while the standard process specification stipulates a current density tolerance range of ±1.0 A / dm², meaning the drift amplitude accounts for 50% of the tolerance. Based on the comparison results of the drift amplitude and tolerance, combined with the process capability index in the process drift feature, the process standard value of each process parameter is determined. The process standard value is defined as the target value of the process parameter that can be stably achieved under the current equipment and material conditions; this value is within the standard process specification tolerance range and matches the current actual production capacity. In the lamination process, the standard process specification stipulates a lamination temperature of 180±5℃. Based on process drift characteristics, the temperature during stable equipment operation is slightly higher than normal. Therefore, the standard process value for this process is set to 181℃. The determined standard process values ​​for each process are then compiled to form a complete dataset of standard process values.

[0036] An adaptive threshold boundary is set based on the rate of change of process standard values ​​and process drift characteristics. The process standard value serves as the central reference for the threshold boundary, with allowable deviation ranges set on both sides of the standard value. The rate of change data for each process parameter is extracted from the process drift characteristics; the rate of change reflects the speed of drift development. The boundary width on both sides of the standard value is dynamically adjusted according to the magnitude of the rate of change. Parameters with a fast rate of change have a narrower boundary for early warning, while parameters with a slow rate of change have a wider boundary to avoid frequent false alarms. The formula for calculating the width of the adaptive threshold boundary is W = W0 × (1 - k × v), where W is the adaptive boundary width, W0 is the basic boundary width, k is the rate adjustment coefficient, and v is the normalized rate of change; all dimensions are dimensionless or consistent with the parameters. In the solder resist printing process, the process standard value is an ink thickness of 25 μm, and the basic boundary width is ±3 μm. When the rate of change is fast, the boundary narrows to ±2 μm, and when the rate of change is slow, the boundary remains ±3 μm. In the drilling process, the rate of change in hole position accuracy is rapid. The basic boundary width is narrowed from ±20μm to ±15μm to ensure timely detection of hole position deviations. Two layers of adaptive threshold boundaries are set, centered on the process standard value: an early warning boundary and a control boundary. The early warning boundary indicates that process parameters are beginning to deviate, while the control boundary triggers mandatory intervention. The adaptive threshold boundaries for each process parameter are summarized, resulting in a boundary dataset for each parameter containing a center value, early warning upper and lower limits, and control upper and lower limits.

[0037] A stratified drift level classification is formed by hierarchically dividing process drift characteristics based on adaptive threshold boundaries. The current drift amplitude of each process step is compared with the adaptive threshold boundary to determine the boundary range into which the drift amplitude falls. Based on the location of the drift amplitude within this range, the drift state is divided into different levels. A drift amplitude fluctuating near the center value of the threshold boundary is classified as normal, indicating that the process parameters are in a stable and controlled state. A drift amplitude exceeding the warning boundary but not reaching the control boundary is classified as a warning level, indicating that the process parameters are beginning to deviate and require attention. A drift amplitude exceeding the control boundary is classified as out of control, indicating that the process parameters are severely deviating and require immediate intervention. In the drilling process, the adaptive threshold boundary warning limit for hole position accuracy is ±15μm, and the control limit is ±25μm. The current drift amplitude is 18μm, exceeding the warning boundary but not reaching the control boundary; therefore, the hole position accuracy of this process is classified as a warning level. In the copper plating process, the current density drift is currently 1.2 A / dm², exceeding the control boundary of ±1.0 A / dm², classifying this parameter as out of control. Boundary comparisons and level classifications were performed on each parameter in each process, resulting in a drift level classification covering all processes and parameters. The criticality weights of each process parameter in the drift level classification were assigned based on their impact on the product's electrical performance, mechanical strength, and reliability. Parameters related to circuit conduction had the highest weight, while parameters related to appearance had lower weights. The drift level classification is presented in tabular form, with each row corresponding to a process parameter. Column fields include process name, parameter type, current drift amplitude, level, and criticality weight.

[0038] A hazard assessment was conducted to create a process benchmark library based on drift level classifications. Hazard assessments were performed on each process parameter within the drift level classifications to evaluate the impact of parameter drift on the final product quality. The hazard assessment considered two factors: the severity of the parameter drift and the parameter's criticality to product function. Level information for each parameter was extracted from the drift level classifications, with the out-of-control level having the highest hazard, followed by the warning level, and the normal level having the lowest hazard. Combining the functional criticality weights of each parameter, a comprehensive hazard score was calculated: Comprehensive Hazard H = Level Coefficient × Criticality Weight, where the level coefficient is 1 for normal, 3 for warning, and 5 for out-of-control. The criticality weights were determined based on the parameter's impact on the product's electrical performance, mechanical strength, and reliability, with parameters related to circuit continuity having the highest weight and those related to appearance having lower weights. In the inner layer pattern transfer process, linewidth accuracy is a critical parameter for circuit function, directly affecting circuit impedance and signal transmission quality, with a criticality weight of 0.9. The current level is warning, and the comprehensive hazard score is 3 × 0.9 = 2.7. In the solder resist printing process, ink thickness primarily affects appearance and protective performance, with a criticality weight of 0.5. Even at the warning level, the overall hazard score is only 3 × 0.5 = 1.5. Based on the overall hazard score, the parameters in the drift level classification are ranked to identify the critical drift parameters with the highest hazard. The hazard assessment results are integrated with the drift level classification data to establish a process benchmark library. The process benchmark library contains three types of information: process standard values ​​and adaptive threshold boundaries for each process parameter, the current drift level classification result, and a hazard assessment ranking list. The process benchmark library is stored in database form and supports multiple search methods, including querying by process, filtering by level, and sorting by hazard.

[0039] Process deviation parameters are generated through offset detection using a process benchmark library. The process standard values ​​and adaptive threshold boundary data for each process parameter are retrieved from the process benchmark library as the basis for offset detection. Real-time process characteristic data of the current production process are collected and compared with the corresponding process standard values ​​in the process benchmark library to calculate the offset. The offset calculation formula is D=X-X0, where D is the offset, X is the current measured value, and X0 is the process standard value in the process benchmark library. The unit of the offset is consistent with the unit of the process parameter. The offset is compared with the adaptive threshold boundary in the process benchmark library to determine whether the offset exceeds the allowable range. In the etching forming process, the etching rate process standard value retrieved from the process benchmark library is 3.5μm / s, the adaptive threshold boundary is ±0.3μm / s, the current measured value is 3.9μm / s, and the offset is +0.4μm / s, exceeding the threshold boundary and is determined to be an abnormal offset. In the drilling process, the standard value for hole position accuracy is retrieved from the process reference library as 0 μm, the adaptive threshold boundary warning limit is ±15 μm, and the current measured hole position deviation is +12 μm. Since the offset does not exceed the warning boundary, it is considered normal. For parameters with detected abnormal offsets, the hazard assessment information of the parameter is extracted from the process reference library. Combined with the offset direction and offset magnitude, process deviation parameters are generated. Process deviation parameters include four fields: deviation process, deviation parameter name, offset direction, and offset magnitude. The abnormal offset parameters detected in each process are summarized and sorted from highest to lowest hazard level to form a complete list of process deviation parameters.

[0040] Step S130: Strengthen the inter-process correlation of the process status baseline to determine the transmission coupling degree of adjacent processes, use the transmission coupling degree to perform cross-process penetration tracking to locate the defect transmission link, and cascade and accumulate process deviation parameters along the defect transmission link to form a process correlation matrix.

[0041] Specifically, the process baseline is processed to enhance inter-process correlation and determine the transmission coupling degree between adjacent processes. Time-series data of characteristic values ​​between adjacent processes are extracted from the process baseline to analyze the correlation between the output characteristics of the preceding process and the input characteristics of the following process. The correlation coefficient of the characteristic values ​​of adjacent processes is calculated; the correlation coefficient reflects the degree of influence of changes in the parameters of the preceding process on the parameters of the following process. A sliding window analysis is performed on historical data in the process baseline to statistically analyze the probability and magnitude of the shift in characteristic values ​​of the following process after a shift in the characteristic values ​​of the preceding process. Between the lamination and drilling processes, historical data extracted from the process baseline is analyzed. It is found that whenever the lamination temperature is too high in a batch, the hole position deviation in the subsequent drilling process also increases accordingly. This is because uneven heating of the sheet metal leads to changes in the internal stress distribution, and micro-deformation of the sheet metal during drilling affects the hole position accuracy. The transmission coupling degree between adjacent processes is calculated by combining the correlation coefficient and the transmission probability: Transmission coupling degree C = |ρ| × P, where ρ is the correlation coefficient, |ρ| is taken as its absolute value to represent the correlation strength, and P is the transmission probability. The coupling degree ranges from 0 to 1. For each pair of adjacent processes covered in the process baseline, a correlation analysis is performed to calculate the transitive coupling degree between each pair of adjacent processes. The transitive coupling degree is stored in matrix form, with the rows and columns of the matrix corresponding to the preceding and following processes, and the element values ​​being the coupling degree values ​​of the corresponding process pair. The higher the coupling degree, the more significant the influence of the preceding process on the following process.

[0042] In some embodiments, the method of using the transmission coupling degree to perform cross-process penetration tracking and locating of defect transmission links includes: generating a process topology map by mapping the transmission coupling degree to the process association space; obtaining the influence depth level by parsing the transmission level from the process topology map; extracting cross-process transmission channels by performing bidirectional association verification through the influence depth level; and forming a defect transmission link by superimposing the transmission intensity of the cross-process transmission channels.

[0043] A process topology graph is generated by mapping the transitive coupling degree to the process association space. Each process in the transitive coupling degree matrix is ​​treated as a node, and the coupling relationships between processes are treated as connecting edges, constructing a directed graph structure. In the process association space, the positions of each process node are arranged according to the production flow sequence, and the direction of the connecting edges between nodes indicates the direction of process parameter transmission. Coupling degree values ​​are extracted from the transitive coupling degree matrix and mapped to the weights of connecting edges; the higher the coupling degree, the greater the edge weight. A coupling degree threshold is set; process pairs with coupling degrees below the threshold are not connected, filtering out weak relationships to simplify the topology. In the PCB production line, the exposure parameters of the inner layer pattern transfer process have a significant impact on the interlayer alignment of the lamination stack, and strong connecting edges are established between the two processes; however, the solder mask printing process is far from the preceding inner layer pattern transfer process and has weak parameter correlation, so no connecting edges are established between them. For non-adjacent process pairs that have cross-process influence, if the transitive coupling degree data shows that their coupling degree exceeds the threshold value, cross-process connecting edges are also established. Nodes, connecting edges, and edge weights are integrated to form a complete process topology graph. The process topology diagram visually presents the strength of the relationship and the direction of transmission between each process, providing a structured basis for subsequent transmission path analysis.

[0044] The influence depth is obtained by analyzing the transmission hierarchy in the process topology diagram. A target process node is selected in the process topology diagram, and the depth of influence of upstream processes on this process is analyzed. Starting from the target process node, trace back along the connecting edges to count the number of upstream processes and the path length reaching the target node. Upstream processes directly connected to the target process are classified as the first level, processes further upstream connected to the first level are classified as the second level, and so on. When analyzing the influence sources of the copper plating process in the process topology diagram, the drilling process, directly connected to it, belongs to the first level because the hole wall quality directly affects the copper plating effect; the lamination process, indirectly connected through drilling, belongs to the second level because the board lamination quality affects the hole wall smoothness during drilling; the inner layer pattern transfer process belongs to the third level, where the integrity of the circuit pattern ultimately affects the plating uniformity through multi-level transmission. The number of processes and the corresponding connecting edge weights in each level are counted. The deeper the level, the longer the transmission path, the more indirect the influence, but there may be a cumulative effect. The process list and connecting edge weight data of each level are integrated to form the influence depth hierarchy of the target process. The influence depth hierarchy describes the hierarchical structure of how a target process is affected by upstream processes. This hierarchical information is used to identify the source process for forward tracing. Hierarchical analysis is performed on each process node in the process topology diagram to form a complete influence depth hierarchy dataset.

[0045] For example, the step of extracting cross-process transmission channels by performing bidirectional correlation verification through the influence depth level includes: performing forward transmission tracing to form a forward correlation path based on the influence depth level; performing reverse backtracking verification on the forward correlation path to form a reverse verification path; performing consistency comparison between the forward correlation path and the reverse verification path to filter valid transmission segments; and integrating channels based on the valid transmission segments to form cross-process transmission channels.

[0046] Based on the depth of influence hierarchy, a forward transmission tracing is performed to form a forward correlation path. The source process is selected as the starting point for tracing, and the parameter transmission trajectory is traced downstream along the process flow. In the depth of influence hierarchy structure, starting from the source process at the deepest level, it passes through each intermediate level process sequentially until the target process is reached. Each process node and the connection path between nodes is recorded during the tracing process, forming an ordered process sequence from source to target. When insufficient exposure energy in the inner layer pattern transfer process leads to rough circuit edges, the tracing is performed downstream according to the depth of influence hierarchy: the rough circuit edges cause uneven interlayer insulation distance during lamination, which in turn causes uneven drill bit force and hole position displacement during drilling due to material density differences, ultimately resulting in uneven plating thickness within the holes during copper plating. This transmission trajectory from the exposure problem to the plating defect is recorded as a forward correlation path. The transmission coupling degree values ​​between adjacent processes on the forward correlation path are extracted, and the overall transmission strength of the path is calculated. When multiple paths exist, each path is tracked and recorded to form multiple candidate forward correlation paths. The process sequence, transmission strength, and path length of each forward correlation path are summarized. The forward correlation path describes the possible trajectory of process parameters from the source process to the downstream process.

[0047] A reverse backtracking verification is performed on the forward correlation path to form a reverse verification path. Starting from the end process of the forward correlation path, the process is traced upstream in the reverse direction to verify the bidirectional consistency of the transmission relationship. During the backtracking process, it is checked whether the parameter anomalies in the downstream process can indeed be traced back to the parameter changes in the upstream process. Starting from the parameter deviation data at the end of the forward correlation path, the causal relationship between the deviation and the parameters of the previous process is analyzed. If the causal relationship is valid, the backtracking continues to the next upstream. When the final inspection finds a local open circuit fault in the PCB finished product, the reverse backtracking is performed along the forward correlation path: first, the copper plating process is checked, and it is found that the plating layer of the through hole corresponding to the fault location is indeed too thin; continuing to trace upward to the drilling process, it is found that there are burrs on the hole wall of the through hole affecting the copper plating adhesion; then tracing upward to the lamination process, it is found that there are delamination bubbles in the board in this area, which cause burrs to be generated during drilling. The causal chain is verified step by step upward. The process nodes and connections that are verified during the reverse backtracking process are recorded to form a reverse verification path. By comparing the process sequences of the forward correlation path and the reverse verification path, the portions of the two paths that pass through the same process nodes are marked as having passed bidirectional verification. The reverse verification path has a similar structure to the forward correlation path but runs in the opposite direction, and is used to confirm that the transitive relationship obtained through forward tracing also holds true in the reverse causal analysis.

[0048] The forward correlation path and the reverse verification path are compared for consistency to screen valid transfer segments. The process connection relationships in the forward correlation path and the reverse verification path are compared segment by segment to identify overlapping process pairs in the two paths. For each process connection in the forward correlation path, it is checked whether the connection also exists in the reverse verification path; connections existing in both directions are marked as consistent. When analyzing the transfer relationship from drilling to copper plating, if the forward correlation path shows that hole wall roughness affects plating quality, and the backtracking of the reverse verification path confirms that the location of plating defects perfectly matches the location of hole wall roughness, then this connection segment passes the consistency comparison. However, if the inner layer pattern shown in the forward correlation path is transferred to a certain transfer relationship in the lamination stack, and the backtracking in the reverse verification path reveals that the lamination problem actually stems from batch differences in the prepreg rather than the inner layer pattern, then this connection segment fails the consistency comparison. The consistency comparison results of each connection segment are statistically analyzed, and the process connection segments that pass bidirectional verification are extracted as valid transfer segments. Connection segments that exist only in a one-way path but cannot be verified in the other direction are judged as weak or spurious associations and are not included in the valid transmission segments. All process connection segments that pass the consistency comparison are summarized, and valid transmission segments are stored in the form of a process pair list. Each process pair includes the preceding process, the following process, and the transmission strength verified by both directions. Valid transmission segments represent the true process parameter transmission relationship confirmed by two-way verification, excluding spurious associations that are statistically correlated in one direction but have no actual causal relationship.

[0049] Based on the effective transfer segments, channels are integrated to form cross-process transfer channels. Continuous process pairs within the effective transfer segments are connected in series to form complete transfer channels spanning multiple processes. The connection relationship between the beginning and end of each process pair in the effective transfer segment list is checked. If the later process of one process pair is the same as the earlier process of another process pair, the two process pairs are merged into a longer transfer chain. Within the effective transfer segments, the transfer relationship from lamination to drilling and from drilling to copper plating has been verified. Since the drilling process connects two process pairs, it is integrated into a three-process transfer channel from lamination through drilling to copper plating. This channel describes the complete path of how board quality issues progressively affect the final plating quality. The integrated transfer channels are then optimized, merging channels with the same starting and ending points but different intermediate paths, retaining the main channel with the highest transfer strength. The overall transfer efficiency of each cross-process transfer channel is calculated; the overall transfer efficiency is the product of the transfer strengths of each effective transfer segment on the channel. The process sequence, overall transfer efficiency, and channel length of each transfer channel are integrated to form a cross-process transfer channel dataset. A cross-process transfer channel describes the complete path of process parameters from the source process through multiple intermediate processes to the target process. The longer the channel, the wider the scope of influence of parameter transfer.

[0050] A defect propagation link is formed by superimposing the propagation intensity of cross-process transmission channels. The propagation intensity of each channel in the cross-process transmission channel is analyzed to assess the cumulative effect of defect propagation along the channel. The comprehensive propagation efficiency of each channel is retrieved, reflecting the proportion of defect signal intensity retention after multiple processes. At each process node in the cross-process transmission channel, the defect signal may be amplified, retained, or attenuated; the node propagation coefficient is determined based on the process characteristics of each node. When a local undervoltage occurs in the lamination process, resulting in micro-bubbles inside the board, the defect propagates along the cross-process transmission channel to the drilling process. Due to the heat and vibration generated by the high-speed rotation of the drill bit, the material in the bubble area fractures, and the defect is amplified into hole wall damage. Continuing to the copper plating process, the damaged hole wall cannot form a complete conductive seed layer, and the defect is further amplified into plating voids, potentially leading to open-circuit vias. The propagation coefficients of each node in the cross-process transmission channel are superimposed to calculate the comprehensive defect propagation intensity of the channel. Cross-process transmission channels with propagation intensity exceeding a set threshold are marked as significant defect propagation paths, forming a defect propagation link. The defect propagation chain is stored in a linked list structure, with each node serving as a process identifier. The connections between nodes carry information about the propagation strength. The defect propagation chain clearly identifies the path and intensity of a defect's propagation from the source process to downstream processes, providing a basis for root cause localization and propagation prevention.

[0051] A process correlation matrix is ​​formed by cascading and accumulating process deviation parameters along the defect propagation chain. At the source process node of the defect propagation chain, four fields of information are extracted from the process deviation parameters of that process: the deviating process determines the location of the deviation, the deviation parameter name determines the parameter type, the offset direction indicates whether the parameter is too high or too low, and the offset amount serves as the initial deviation value. The defect propagation continues downstream along the defect propagation chain. At each intermediate process node, the upstream cumulative deviation value is superimposed with the offset amount of the process deviation parameter of the current process, taking into account the propagation strength coefficient. The cumulative deviation value is calculated using the formula A_n = A_{n-1} × T_n + D_n, where A_n is the cumulative deviation value of the nth process, A_{n-1} is the upstream cumulative value, T_n is the propagation strength coefficient, and D_n is the offset amount of the current process. All terms are signed normalized values, and the positive or negative sign indicates the offset direction. In actual production, when a process deviation parameter such as low exposure energy occurs in the inner layer pattern transfer process, this deviation is propagated downwards along the defect propagation chain. In the lamination process, the deviation is compounded by the temperature deviation inherent in this process due to interlayer alignment issues, and the two deviations in opposite directions partially cancel each other out. In the drilling process, the hole position offset is further compounded due to the impact on the positioning reference. The cumulative deviation value reflects the linkage effect and directional superposition result of deviations from multiple processes. The cumulative deviation values ​​of each process along the defect propagation chain are recorded, along with the corresponding deviation parameter names, forming a cumulative sequence of deviation values ​​propagated along the process. A matrix structure is constructed with processes as rows and columns, and the matrix element values ​​are the cumulative deviation propagation amounts between corresponding process pairs. The cascaded cumulative results of each defect propagation chain are filled into the corresponding positions in the matrix, forming a complete process correlation matrix. In the process correlation matrix, rows represent source processes, columns represent target processes, and element values ​​represent the impact of the deviation from the source process on the target process after accumulation through the defect propagation chain. The positive or negative sign of the element reflects the direction of the cumulative offset.

[0052] Step S140: Defect propagation backtracking and influence domain aggregation are performed on the process correlation matrix to determine the defect impact range. Process status parameters are generated based on the defect impact range. Trend turning features are extracted from the process status parameters to construct a process feature map.

[0053] In some embodiments, determining the defect impact range by performing defect propagation backtracking and influence domain aggregation on the process association matrix includes: generating node contribution degrees based on the process association matrix; tracking defect development based on the node contribution degrees to obtain a propagation trajectory; performing source backtracking from the propagation trajectory and combining it with the node contribution degrees to generate a traceability chain; and performing influence domain classification and aggregation along the traceability chain to determine the defect impact range.

[0054] Node contribution is generated based on the process association matrix. Row and column vectors of each process node are extracted from the process association matrix. The output contribution of each process node is calculated, which equals the sum of the element values ​​in the row containing that process in the process association matrix, reflecting the process's ability to transmit influence downstream as a defect source. The input contribution of each process node is calculated, which equals the sum of the element values ​​in the column containing that process in the process association matrix, reflecting the degree to which the process receives influence from upstream defects. The output and input contributions are then weighted to obtain the comprehensive node contribution of each process, N = α × O + β × I, where O is the output contribution, I is the input contribution, α and β are weighting coefficients with α + β = 1, and all terms are normalized dimensionless values. In PCB production lines, the drilling process holds a unique position in terms of its contribution to the overall process. This process not only handles the internal stress and interlayer alignment issues inherited from lamination, but also transmits problems related to hole wall roughness and hole position accuracy to the copper plating and solder mask printing processes. Its input and output contributions are both high, placing it among the top processes in terms of overall contribution. Like a crossroads in the production flow, any defects passing through this process can be amplified and spread in multiple directions. By ranking the contribution values ​​of each process by value, the critical process nodes with the highest contribution can be identified.

[0055] Defect propagation tracing is performed based on node contribution to obtain the diffusion trajectory. The process with the highest node contribution is selected as the starting point for tracing, and the defect propagation path is tracked along the process flow. During tracing, downstream processes with higher node contributions are prioritized as the next hop node to construct the main path of defect propagation. When air bubbles appear in the dry film bonding of the inner layer pattern transfer process, resulting in pinhole defects in the circuit pattern, the defect is traced downstream based on node contribution: the pinhole defect first propagates to the lamination process. Due to poor bonding between the copper foil and resin at the pinhole location, a tiny interlayer void is formed at this location after lamination. The void defect continues to propagate to the drilling process. When the drill bit passes through the void area, the sudden change in material density causes the drill hole to deviate. The deviated through-hole propagates to the copper plating process. The tilted hole wall causes uneven current distribution during plating, resulting in differences in plating thickness, which ultimately manifests as excessive through-hole resistance during finished product testing. The process of recording and tracking each node in the process forms a diffusion trajectory. The diffusion trajectory contains the node contribution value of each node. For cases where there are branches, each branch path is tracked separately. All tracking paths are integrated to form a complete diffusion trajectory that describes the overall development of the defect.

[0056] For example, the step of generating a traceability chain by tracing back the source from the diffusion trajectory and combining the node contribution includes: establishing a time-reverse correlation based on the diffusion trajectory; setting multiple backtracking verification points along the time-reverse correlation; constructing a defect transmission chain in reverse through the backtracking verification points; and forming a traceability chain by weighting and connecting the defect transmission chain with the node contribution.

[0057] A time-reverse correlation is established based on the diffusion trajectory. Following the chronological order of the process flow, a reverse correlation is established starting from the final node of the diffusion trajectory. The timestamps of each process node in the diffusion trajectory are arranged in ascending order, and a time-reverse correlation is established between adjacent process nodes, with the connection direction pointing from the later process to the earlier process. When the final inspection process discovers large-area blistering and peeling of the solder mask layer in a batch of PCBs, a time-reverse correlation is established based on the diffusion trajectory: tracing back from the final inspection time to the curing time of the solder mask printing process, which is responsible for coating and curing the protective ink on the board surface; then tracing back to the exit time of the copper plating process, which is responsible for forming a conductive copper layer on the inner wall of the through-holes and the board surface; continuing to tracing back to the completion time of the drilling process, which is responsible for drilling various through-holes and blind holes on the board; finally, tracing back to the exit time of the lamination process, which is responsible for laminating the multilayer circuit board into a whole. The time-reverse correlation links each process together, forming a complete time-reverse correlation structure arranged in reverse chronological order, recording the timestamp information of each process node in the correlation.

[0058] Multiple backtracking verification points are set up along a time-reverse correlation. Key locations are selected within the time-reverse correlation process sequence to set backtracking verification points, which are used to verify whether defects actually exist at those locations. Processes with high node contribution in the time-reverse correlation are selected as backtracking verification points, as these processes play a pivotal role in defect propagation. Appropriate intervals are maintained between adjacent backtracking verification points to avoid overly dense verification points causing verification redundancy and increasing traceability costs, and also to avoid overly sparse verification points missing key defect propagation links. For the backtracking analysis of solder resist blistering defects, the first backtracking verification point is set in the copper plating process. This verification point needs to check whether there is oxide film or oil residue on the plating surface causing poor adhesion to the solder resist ink; the second backtracking verification point is set in the drilling process. This verification point needs to check whether there are burrs, edge curls, or copper shavings around the hole opening affecting the uniformity of subsequent coating; the third backtracking verification point is set in the lamination process. This verification point needs to check whether there is resin precipitation or copper foil wrinkling on the board surface affecting coating adhesion. Configure verification rules for each backtracking verification point. The verification rules clearly specify the types of parameters to be checked, the threshold conditions for determining the existence of defects, and the source of verification data.

[0059] A defect propagation chain is constructed by backtracking verification points. Defect existence verification is performed at each backtracking verification point, and the verification results determine whether the location is a valid node in the defect propagation chain. Starting from the backtracking verification point closest to the defect discovery point, historical process data and quality inspection records for that process are retrieved and compared with the verification rules. Data sources include process parameter acquisition devices, online inspection equipment, and manual sampling records. If the verification passes, the process is confirmed as a node in the defect propagation chain, and verification continues upstream to the next backtracking verification point. If the verification fails, it indicates that the defect source is located between this verification point and the previous verification point, requiring the addition of temporary verification points within this interval to further narrow down the scope. In the retrospective analysis of solder resist blistering defects, the copper plating process's verification point retrieved the surface roughness and cleanliness test records. It was found that the batch of plating did indeed have a slight oxide film on the surface, and the cleanliness index was at the lower limit. The verification passed, confirming the copper plating process as a node in the defect propagation chain. The drilling process's verification point retrieved the hole burr inspection images. It was found that some through-holes had flanged burrs at the hole openings, and the burr height exceeded the standard limit. The verification passed, confirming the drilling process as a node in the defect propagation chain. The lamination process's verification point retrieved the board appearance inspection records and surface flatness data. All indicators were within the normal range, and the verification failed, determining that the defect source was located in the drilling process rather than the upstream lamination process. Connecting the verified process nodes in reverse chronological order forms a complete defect propagation chain from the defect discovery point to the defect source.

[0060] A traceability chain is formed by weighted chaining based on the defect propagation chain and the node contribution. The node contribution value of each process node in the defect propagation chain is added to the chain as a node weight, and the node with a higher contribution value bears greater responsibility in the defect formation process. The weighted contribution value of each node in the defect propagation chain is calculated, and adjacent nodes on the chain are chained together. When chaining, the node contribution value is used as the connection weight. A connection with a higher weight indicates that the defect propagation relationship of that segment is more certain and the transmission efficiency is higher. In the defect propagation chain of solder resist blistering defect, the drilling process is identified as the defect source. When the burr problem at the hole opening is propagated to the copper plating process, the flow of plating solution in the burr area is obstructed, resulting in a thin and rough plating layer that is prone to oxidation. This segment of the propagation relationship is clear and the cause and effect are clear. When the oxidized plating layer is propagated to the solder resist printing process, the adhesion between the ink and the copper oxide layer decreases, resulting in blistering and peeling after curing. This segment of the propagation relationship is also clear and the cause and effect are clear. Based on the node contribution data, drilling, as the source process, has the highest node contribution and significantly impacts multiple subsequent processes; its connection with copper plating is assigned a high weight. Copper plating, as an intermediate transmission process that transforms hole wall defects into plating defects, has the second highest node contribution; its connection with solder mask printing is assigned a medium weight. Integrating the weighted nodes and connections forms a traceability chain. This chain clearly identifies the complete transmission path of defects from their source to the discovery point, making the responsibility weight of each link in the traceability chain readily apparent.

[0061] The scope of defect impact is determined by classifying and aggregating the impact domains along the traceability chain. Taking each node in the traceability chain as the center, the direct impact domain of that node is determined based on the process association matrix. The direct impact domain includes all downstream processes that have a direct transmission relationship with that node. The direct impact domains of each node in the traceability chain are summarized to form the overall impact domain set of the chain. Processes in the impact domain set are classified and analyzed. Processes jointly affected by multiple chain nodes suffer from the cumulative effect of defects from multiple sources, and are marked as high-risk processes, thus falling within the core area of ​​the defect impact range. Processes affected by only a single node have a single source of defect, and the risk is relatively controllable, marked as generally affected processes, thus falling within the peripheral area of ​​the defect impact range. In the traceability chain with drilling as the source, the impact domain of the drilling process covers three downstream processes: copper plating, outer layer pattern transfer, and solder mask printing; the impact domain of the copper plating process covers two downstream processes: outer layer pattern transfer and solder mask printing. The simultaneous appearance of outer layer pattern transfer and solder mask printing in two influence domains indicates that these two processes are simultaneously affected by both hole wall defects and plating defects, thus being marked as high-risk processes requiring close monitoring. Boundary aggregation was performed on each influence domain, merging overlapping areas and integrating scattered areas to eliminate redundant boundaries and form a unified defect impact range. The defect impact range is presented in two layers: a list of high-risk processes and a list of generally affected processes.

[0062] Process status parameters are generated based on the defect impact range. For each affected process within the defect impact range, real-time and historical parameter statistics are retrieved from the production monitoring database. The current parameter values ​​of each process within the defect impact range are compared with their normal baseline values, and the deviation percentage is calculated. The deviation percentage reflects the degree of deviation of the current state of the process from the normal state. Parameters with deviation percentages exceeding a preset threshold are marked as abnormal parameters and included in the process status parameters, while parameters that do not exceed the threshold are marked as normal parameters. When the defect impact range includes two high-risk processes, drilling and copper plating, the drilling process needs to monitor whether the spindle speed is forced to increase due to tool wear, whether the feed rate is adjusted due to changes in the hardness of the board material, and whether the hole position deviation shows a unidirectional drift trend. The copper plating process needs to monitor whether the current density needs to be compensated due to plating solution aging, whether the plating solution temperature fluctuates due to environmental changes, and whether there are batch-to-batch differences in plating thickness. By combining the impact information recorded within the defect's affected area, the deviation percentage of each process parameter is weighted and adjusted. Processes with a higher impact are given higher weights for their parameter deviations, and these are integrated to form process status parameters. The process status parameters fully record the parameter deviation status of each process within the affected area.

[0063] In some embodiments, the step of extracting trend reversal features from the process state parameters to construct a process feature map includes: generating a process change curve based on the process state parameters; performing gradient mutation detection to track the trajectory of process parameter changes from the process change curve; locating the inflection point value where the trajectory of process parameter changes changes from normal to abnormal as a trend reversal feature; and performing boundary clustering processing on the trend reversal feature to generate a process feature map.

[0064] A process variation curve is generated based on the process status parameters. Time series data for each parameter is extracted from the process status parameters, and curves showing the change of each parameter over time are plotted with time on the horizontal axis and parameter value on the vertical axis. Each data point on the curve corresponds to the measured value of the parameter at a sampling time. Multiple related parameter curves for the same process are overlaid on the same process variation curve to facilitate observation of the linkage between parameters. For the process status parameters of the drilling process, the spindle speed curve, feed rate curve, and hole position deviation curve are plotted on the same process variation curve. When the spindle speed is forced to increase due to tool wear to maintain cutting efficiency, it is observed whether the feed rate is simultaneously reduced to reduce tool load, and whether the hole position deviation shows a systematic drift due to the change in speed. The upper and lower boundary lines of the normal range for each parameter are marked on the process variation curve. These boundary lines divide the curve space into normal and abnormal intervals, facilitating intuitive identification of periods when parameters exceed the normal range. Moving average filtering and smoothing are applied to the curve data to eliminate the interference of sampling noise on trend analysis, and the variation curves of each parameter for each process are integrated to form a complete process variation curve.

[0065] Gradient abrupt changes are detected and the trajectory of process parameter changes is tracked using process change curves. The gradient value is calculated for each curve in the process change curve, where the gradient value G = (V_{t+1} - V_t) / Δt reflects the rate of parameter change over time. V_t and V_{t+1} are the parameter values ​​at adjacent moments, Δt is the time interval, and the gradient is measured in parameter units divided by time units. The gradient sequence of each curve in the process change curve is scanned to identify locations where sudden changes in gradient values ​​occur. Gradient abrupt changes indicate a significant shift in the rate of parameter change, potentially corresponding to equipment state switching, manual intervention, or abnormal events. When the gradient difference between adjacent moments exceeds a threshold, it is marked as an abrupt change point in the process parameter trajectory. In the current density curve of the copper plating process, during normal production, the current density decreases slowly as copper ions in the plating bath are consumed, with a negative gradient and a small absolute value. When the operator adds replenishing solution to the plating bath to adjust the concentration, the current density rapidly recovers, the gradient suddenly changes from negative to positive, and the absolute value increases. This gradient abrupt change point is detected and included in the process parameter change trajectory. If the amount of replenishing solution added is too large, causing the concentration to exceed the limit, the current density will continue to climb, exceeding the upper limit specified by the process, forming another abrupt change point, indicating that the parameter has changed from normal to abnormal. By connecting the gradient abrupt change points detected on each curve in chronological order, a complete process parameter change trajectory describing the evolution of the parameter from one state to another is formed.

[0066] The inflection point value at which the process parameter change trajectory changes from normal to abnormal is used as the trend reversal feature. Key inflection points with state transition significance are selected from each abrupt change point in the process parameter change trajectory. The parameter state before and after each abrupt change point is analyzed to determine whether the abrupt change point corresponds to a parameter crossing from the normal range to the abnormal range. For cases where the parameter value is within the normal range before the abrupt change point but enters the abnormal range after the abrupt change point, this abrupt change point is the state transition inflection point, representing the critical moment when the process state changes from controllable to uncontrollable. The parameter value at the state transition inflection point is extracted as the characteristic value of the trend reversal feature, representing the critical level at which the parameter changes from normal to abnormal. In the temperature curve of the lamination process, the temperature rises steadily according to the preset heating curve during the heating stage. When the heating element experiences local aging and power decay, the temperature rise rate slows down and deviates from the preset curve. At a certain moment, the temperature stops rising and stabilizes at a level below the minimum temperature required by the process. This moment is the state transition inflection point, and the temperature value at the inflection point is the characteristic value of the trend reversal feature. Temperatures below this value will lead to incomplete resin curing and affect the performance of the board. The state transition inflection points of each parameter are summarized to form a trend reversal feature dataset, which fully records the critical reversal level of each parameter.

[0067] Boundary clustering is performed to generate process feature maps based on trend reversal characteristics. Cluster analysis is then performed on the feature values ​​in the trend reversal feature dataset according to similarity to identify parameter groups with similar trend reversal characteristics or influenced by the same factors. Parameters with similar reversal characteristics are grouped into the same cluster, where parameters often have inherent physical correlations or are driven by common external factors. In multiple processes of the PCB production line, the exposure energy for inner layer pattern transfer is sensitive to ambient humidity, the curing temperature for lamination is sensitive to workshop temperature, and the ink viscosity for solder mask printing is sensitive to both ambient temperature and humidity. The trend reversal characteristics of these three parameters have similar deviation patterns and are all related to environmental conditions. They are clustered into an environmentally sensitive parameter cluster and included in the process feature map. When the workshop air conditioning malfunctions, causing uncontrolled temperature and humidity, these parameters may simultaneously exhibit reversal trends. The trend reversal characteristics of the spindle speed, feed rate, and cathode movement speed in copper plating are all related to the mechanical state of the equipment. They are clustered into an equipment state parameter cluster and included in the process feature map. When the equipment runs continuously for several months without maintenance, these parameters may successively exhibit reversals. Boundary ranges are defined for each parameter cluster. The lower boundary is the minimum value of the trend inflection characteristic value of each parameter within the cluster, and the upper boundary is the maximum value. Using parameter type as one dimension and process position as another dimension, each parameter cluster is mapped to a two-dimensional map space to form a process feature map. The process feature map intuitively presents the distribution and boundary range of the inflection characteristics of various parameters in each process.

[0068] Step S150: Sensitivity decoupling is performed on the process deviation parameters and process correlation matrix to extract the defect amplification factor. Based on the defect amplification factor, the process critical node is identified. A control rule library is constructed around the process critical node. The process feature map is integrated through the control rule library to output the process control instruction.

[0069] In some embodiments, the step of decoupling the process deviation parameter from the process association matrix based on sensitivity to extract the defect amplification factor includes: performing orthogonal decomposition of the process deviation parameter and the process association matrix to form a sensitivity coefficient; determining abrupt change boundary points by locating critical points from the sensitivity coefficients; performing nonlinear gain processing through the abrupt change boundary points to form a gain ratio; and forming a defect amplification factor based on the gain ratio.

[0070] Sensitivity coefficients are generated by orthogonal decomposition of process deviation parameters and process correlation matrices. Four fields of information for each process are extracted from the process deviation parameters: the deviation process and deviation parameter names determine the analysis object, the offset direction determines the vector sign, and the offset amount determines the vector magnitude. Based on this, the process deviation parameters are constructed as deviation vectors. The transitive relationship data in the process correlation matrix is ​​constructed as correlation vectors. Orthogonal decomposition is performed on the deviation vectors and correlation vectors to decouple the process deviation parameters from the transitive relationships in the process correlation matrix. The projection component of the deviation vector along the direction of the correlation vector is calculated; this projection component reflects the part of the process deviation parameters related to the transitive relationships. The orthogonal component of the deviation vector perpendicular to the correlation vector is also calculated; this orthogonal component reflects the part of the deviation parameters independent of the transitive relationships. In the lamination process, temperature deviation, pressure deviation, and holding time deviation have a compensating relationship. When the temperature is too low, the holding time can be extended to compensate. These deviation parameters are constructed as multidimensional vectors and orthogonally decomposed with the influence vectors in the correlation matrix to separate the sensitive components of each parameter. The sensitivity coefficients of each parameter are calculated based on the orthogonal decomposition results. The sensitivity coefficient S = |P_parallel| / |D|, where P_parallel is the projection length of the deviation vector in the associated direction, and D is the magnitude of the deviation vector. All terms are dimensionless values. The higher the sensitivity coefficient, the easier it is for the deviation of the parameter to be amplified and transmitted.

[0071] Critical point location is determined using sensitivity coefficients to identify abrupt change boundaries. Distribution analysis is performed on the coefficient values ​​in the sensitivity coefficient dataset. The parameters are sorted from smallest to largest sensitivity coefficient value, and the changing trends of the coefficient values ​​after sorting are analyzed. In the sensitivity coefficient sequence, most parameters have low and relatively stable coefficient values, indicating that deviations in these parameters are largely maintained and not amplified after being passed through processes. A few parameters experience sudden jumps to higher levels, indicating that deviations in these parameters are significantly amplified; the location of these jumps is the abrupt change boundary. In the sensitivity analysis of all process parameters, the sensitivity coefficients of most parameters, such as the ink thickness of solder resist printing and the plating solution temperature of copper plating, are concentrated in the low range. Even if these parameters deviate, their impact is basically controllable, like ripples caused by pebbles in a river that quickly subside. However, the sensitivity coefficients of a few parameters, such as the feed rate of drilling and the slope of the heating curve of lamination, suddenly rise to several times the low range. Slight deviations of these parameters will snowball and become increasingly larger. Locating the location of these parameters is the abrupt change boundary point. The abrupt change boundary point clearly divides the parameters into low-sensitivity and high-sensitivity groups. Parameters exceeding the abrupt change boundary point need to be given special attention.

[0072] Gain ratios are generated by performing nonlinear gain processing based on abrupt change thresholds. Parameters are divided into different sensitivity ranges according to these thresholds. Parameters below the threshold are treated linearly, while those above the threshold are analyzed using a nonlinear function relationship between deviation and influence. Nonlinear gain curves typically exhibit an exponential or power function shape; when the deviation exceeds the threshold, the influence increases exponentially. In drilling operations, when the feed rate fluctuates slightly within the normal range, the hole wall quality changes linearly with the feed rate, similar to how a slight change in speed has little impact on the travel distance when driving on a flat road. However, when the feed rate exceeds a critical value and continues to increase, the drill bit cutting force rises sharply, causing tearing rather than cutting of the hole wall, similar to how excessive speed leads to brake failure and a rapidly deteriorating consequence; the hole wall quality decays exponentially. The gain ratio of each highly sensitive parameter at the current deviation level is calculated based on the nonlinear gain curve. The gain ratio is equal to the ratio of the nonlinear gain value to the linear gain value. A gain ratio greater than 1 indicates an amplification effect, and a higher gain ratio indicates a more significant amplification effect.

[0073] Defect amplification factors are formed based on gain ratio. Parameters with gain ratios greater than a set threshold are selected; these parameters have significant defect amplification characteristics and are candidates for defect amplification factors. The selected high-gain parameters are sorted according to their gain ratios, and a defect amplification factor level is assigned to each high-gain parameter. In the sensitivity analysis of the PCB production line, the feed rate parameter for drilling has the highest gain ratio. Even a slight deviation in this parameter can cause serious hole wall tearing problems, which are amplified along the transfer path to plating voids in copper plating and ink leakage in solder mask printing. Like the butterfly effect, a small disturbance leads to a major disaster, and this parameter is marked as a level three defect amplification factor. The slope parameter of the heating curve for lamination has the second highest gain ratio. Deviation in this parameter can lead to uneven resin flow, which in turn causes a chain reaction of problems in multiple subsequent processes, and this parameter is marked as a level two defect amplification factor. The development time parameter for inner layer pattern transfer has a relatively low gain ratio but still exceeds the threshold, and is marked as a level one defect amplification factor. The process identifier, parameter name, gain ratio, and amplification factor level of each parameter are integrated to form a complete defect amplification factor dataset.

[0074] Critical process nodes are identified based on defect amplification factors. The number and level distribution of defect amplification factors in each process are statistically analyzed, and processes containing high-level or multiple defect amplification factors are marked as candidate critical process nodes. The location of these candidate critical process nodes within the production flow is analyzed; those located in the middle of the flow and simultaneously affecting multiple downstream processes have higher criticality. The drilling process contains two high-level defect amplification factors: feed rate and spindle speed. Furthermore, this process connects the sheet metal processing section and the surface treatment section, much like a hub in a transportation network; its failure would disrupt multiple lines. Its processing quality directly determines whether subsequent copper plating can achieve reliable interlayer conductivity, whether outer layer pattern transfer can achieve precise pattern alignment, and whether solder mask printing can achieve good ink adhesion. Therefore, the drilling process is identified as a critical process node. Considering the number, level, and location of defect amplification factors, the criticality score of each candidate process is calculated. Processes with scores exceeding the threshold are identified as critical process nodes, which are the focus of subsequent control.

[0075] A control rule library is constructed around the critical process nodes. For each process step within the critical process node, the normal range and critical boundary of its defect amplification factor parameter are analyzed, and trigger conditions for the control rule library are set based on the distance between the current parameter state and the boundary. When the parameter value of the critical process node approaches the critical boundary, an early warning control is triggered; when the parameter value crosses the critical boundary, a forced control is triggered. These control logics are all recorded in the control rule library. Specific rules in the control rule library are formulated for the drilling process critical node: when the feed rate parameter is increased by the operator to the upper limit boundary due to rushing demand, like a driver speeding and approaching a dangerous area to save time, a speed reduction control command is triggered to reduce the feed rate to a safe range and a capacity warning is sent to the scheduling platform; when the spindle speed continues to rise due to tool wear compensation, like an engine continuously increasing its speed to compensate for insufficient power, indicating an impending fault, a tool change reminder command is triggered to suggest replacing the drill bit to prevent the speed from continuing to rise; when the hole position deviation shows a unidirectional drift trend for multiple consecutive batches, like a vehicle steering wheel drifting to one side, indicating that the chassis needs calibration, an equipment calibration command is triggered to arrange for the machine tool to perform positioning accuracy testing. Establish a mapping relationship between control rules and defect types in the control rule library. Configure temperature control adjustment strategies for temperature amplification factors and motion parameter adjustment strategies for speed amplification factors. Index and organize all control rules according to process critical nodes, procedures and parameter types to form a complete control rule library.

[0076] The process control instructions are output by integrating the process feature map through the control rule library. The trend inflection value of each process parameter in the process feature map is matched with the trigger conditions in the control rule library to identify the processes and parameters requiring control and generate process control instructions. When the process feature map shows that the slope of the temperature rise curve of the lamination process is approaching the trend inflection boundary, similar to an oil temperature approaching the red line area on a dashboard, it indicates that the heating element may experience power attenuation. A temperature anomaly control rule is matched in the control rule library, and the heating power compensation action specified by this rule is extracted to generate the corresponding process control instruction. When the process feature map shows that the hole position deviation of the drilling process has crossed the inflection boundary and entered the abnormal range, similar to an alarm light indicating a problem has occurred, a precision loss control rule is matched in the control rule library, and the emergency shutdown and maintenance action specified by this rule is extracted to generate the corresponding process control instruction. The control action is converted into a specific process control instruction, which includes the target process, target parameter, adjustment direction, and adjustment range. The generated process control instructions are sorted according to priority, with instructions for critical process nodes being executed first. The sorted process control instructions are then sent to the execution devices of each process. The execution devices adjust process parameters according to the instructions to achieve closed-loop control of process deviations.

[0077] To implement the above-described method embodiment, a collaborative control method for multi-process PCB manufacturing is provided to achieve the corresponding functions and technical effects. See also... Figure 2 , Figure 2 This diagram illustrates a structural block diagram of a collaborative control system 200 for multi-process PCB manufacturing according to an embodiment of this application. For ease of explanation, only the parts relevant to this embodiment are shown. The collaborative control system 200 for multi-process PCB manufacturing provided in this embodiment includes:

[0078] Data acquisition module 201 is used to collect multi-source process data of each process in the PCB production line, and to perform multi-source cross-verification and filtering on the multi-source process data to form a process status baseline.

[0079] The benchmark detection module 202 is used to perform process capability review and identify process drift characteristics based on the process state baseline, perform dynamic threshold hierarchical construction of process benchmark library in the process drift characteristics, and perform offset detection through the process benchmark library to generate process deviation parameters.

[0080] The correlation analysis module 203 is used to perform inter-process correlation enhancement processing on the process state baseline to determine the transmission coupling degree of adjacent processes, use the transmission coupling degree to perform cross-process penetration tracking to locate the defect transmission link, and cascade and accumulate the process deviation parameters along the defect transmission link to form a process correlation matrix.

[0081] The graph construction module 204 is used to perform defect diffusion backtracking and influence domain aggregation on the process association matrix to determine the defect range, generate process status parameters based on the defect range, and extract trend turning features from the process status parameters to construct a process feature graph.

[0082] The instruction output module 205 is used to perform sensitivity decoupling on the process deviation parameters and the process correlation matrix to extract the defect amplification factor, identify the process critical node based on the defect amplification factor, construct a control rule library around the process critical node, and output the process control instruction by integrating the process feature map through the control rule library.

[0083] The aforementioned collaborative control system 200 for multi-process PCB manufacturing can implement the collaborative control method for multi-process PCB manufacturing described in the above method embodiments. The options described in the above method embodiments are also applicable to this embodiment and will not be detailed here. The remaining content of this application's embodiments can be referred to the content of the above method embodiments, and will not be repeated in this embodiment.

[0084] The purpose of the above embodiments is to reproduce and derive the technical solution of the present invention by way of example, and to fully describe the technical solution, purpose and effect of the present invention. The purpose is to enable the public to have a more thorough and comprehensive understanding of the disclosure of the present invention, and not to limit the scope of protection of the present invention.

[0085] The above embodiments are not an exhaustive list based on the present invention, and there may be many other embodiments not listed. Any substitutions and improvements made without departing from the concept of the present invention are within the protection scope of the present invention.

Claims

1. A collaborative control method for multi-process PCB manufacturing, characterized in that, include: Collect multi-source process data from each process of the PCB production line, and perform multi-source cross-verification and filtering on the multi-source process data to form a process status baseline; Based on the process status baseline, process capability review is performed to identify process drift characteristics. Dynamic threshold hierarchical construction of process benchmark library is performed on the process drift characteristics. Offset detection is performed through the process benchmark library to generate process deviation parameters. The process status baseline is subjected to inter-process correlation enhancement processing to determine the transmission coupling degree of adjacent processes. The transmission coupling degree is then used to perform cross-process penetration tracking to locate defect transmission links. This includes: mapping the transmission coupling degree to the process correlation space to generate a process topology map; performing transmission level analysis from the process topology map to obtain the influence depth level; performing bidirectional correlation verification through the influence depth level to extract cross-process transmission channels; superimposing the transmission intensity of the cross-process transmission channels to form a defect transmission link; and cascading and accumulating the process deviation parameters along the defect transmission link to form a process correlation matrix. The step of performing bidirectional correlation verification through the influence depth level to extract cross-process transmission channels includes: performing forward transmission tracking based on the influence depth level to form a forward correlation path; performing reverse backtracking verification on the forward correlation path to form a reverse verification path; comparing the consistency of the forward correlation path and the reverse verification path to filter effective transmission segments; and integrating the effective transmission segments to form a cross-process transmission channel. Defect propagation backtracking and influence domain aggregation are performed on the process association matrix to determine the defect range. Process status parameters are generated based on the defect range. Trend turning features are extracted from the process status parameters to construct a process feature map. Sensitivity decoupling is performed on the process deviation parameters and the process correlation matrix to extract the defect amplification factor. Based on the defect amplification factor, the process critical node is identified. A control rule base is constructed around the process critical node. The process feature map is integrated through the control rule base to output the process control command.

2. The method according to claim 1, characterized in that, The step of performing multi-source cross-verification filtering on the multi-source process data to form a process status baseline includes: Based on the multi-source process data, a process data sequence is formed by expanding the process flow. The core process anchor point is located by performing multi-source cross-verification filtering from the process data sequence. The core process anchor points are used as trajectory references to establish process feature imprints; A process status baseline is generated based on the process feature imprints.

3. The method according to claim 1, characterized in that, The step of constructing a process benchmark library by performing dynamic threshold hierarchical construction in the process drift characteristics includes: The drift amplitude of the obtained process drift characteristics is compared with the standard process specification to determine the process standard value; An adaptive threshold boundary is set based on the rate of change of the process standard value and the process drift characteristics. The process drift features are hierarchically divided into drift level classifications based on the adaptive threshold boundaries. A hazard assessment process is performed based on the drift level classification to form a process benchmark library.

4. The method according to claim 1, characterized in that, The step of determining the defect impact range by performing defect propagation backtracking and influence domain aggregation on the process correlation matrix includes: Node contribution is generated based on the process association matrix. Based on the node contribution, defect development tracking is performed to obtain the diffusion trajectory; A tracing chain is generated by tracing back to the source from the diffusion trajectory and combining the node contribution. The scope of the defect is determined by classifying and aggregating the affected domains along the traceability chain.

5. The method according to claim 1, characterized in that, The step of extracting trend reversal features from the process state parameters to construct a process feature map includes: Generate a process change curve based on the process status parameters; Gradient mutation detection is performed on the process change curve to track the trajectory of process parameter changes; The inflection point value at which the change trajectory of the process parameters changes from normal to abnormal is used as a trend reversal feature; Boundary clustering is performed on the trend reversal features to generate a process feature map.

6. The method according to claim 1, characterized in that, The step of performing sensitivity decoupling and defect amplification factor extraction on the process deviation parameter and the process correlation matrix includes: Sensitivity coefficients are formed by orthogonally decomposing the process deviation parameters and the process correlation matrix. The critical point is determined by locating the mutation boundary from the sensitivity coefficients; A gain ratio is formed by performing nonlinear gain processing at the aforementioned mutation boundary point; The defect amplification factor is formed based on the gain ratio.

7. The method according to claim 4, characterized in that, The step of tracing back to the source from the diffusion trajectory and combining the node contribution to generate a tracing chain includes: A time-reverse correlation is established based on the aforementioned diffusion trajectory; Multiple backtracking verification points are set up by retroactively tracing back along the aforementioned time. The defect propagation chain is constructed in reverse through the backtracking verification points; A traceability chain is formed by weighting and connecting the defect transmission chain with the node contribution.

8. A collaborative control system for multi-process PCB manufacturing, characterized in that, include: The data acquisition module is used to collect multi-source process data of each process in the PCB production line, and to perform multi-source cross-verification and filtering on the multi-source process data to form a process status baseline. The benchmark detection module is used to review process capability and identify process drift characteristics based on the process status baseline, perform dynamic threshold hierarchical construction of process benchmark library in the process drift characteristics, and perform offset detection through the process benchmark library to generate process deviation parameters. The correlation analysis module is used to strengthen the inter-process correlation of the process state baseline to determine the transmission coupling degree of adjacent processes, and to use the transmission coupling degree to perform cross-process penetration tracking to locate defect transmission links. This includes: mapping the transmission coupling degree to the process correlation space to generate a process topology map; performing transmission level analysis from the process topology map to obtain the influence depth level; performing bidirectional correlation verification through the influence depth level to extract cross-process transmission channels; superimposing the transmission intensity of the cross-process transmission channels to form a defect transmission link; and cascading and accumulating the process deviation parameters along the defect transmission link to form a process correlation matrix. The step of performing bidirectional correlation verification through the influence depth level to extract cross-process transmission channels includes: performing forward transmission tracking based on the influence depth level to form a forward correlation path; performing reverse backtracking verification on the forward correlation path to form a reverse verification path; comparing the consistency of the forward correlation path and the reverse verification path to filter valid transmission segments; and integrating the valid transmission segments to form a cross-process transmission channel. The graph construction module is used to perform defect diffusion backtracking and influence domain aggregation on the process association matrix to determine the defect range, generate process status parameters based on the defect range, and extract trend reversal features from the process status parameters to construct a process feature graph. The instruction output module is used to perform sensitivity decoupling on the process deviation parameters and the process correlation matrix to extract the defect amplification factor, identify the process critical node based on the defect amplification factor, construct a control rule base around the process critical node, and output the process control instruction by integrating the process feature map through the control rule base.

Images