A data feedback-based flexible circuit board production quality monitoring method and system

Abstract

The application discloses a flexible circuit board production quality monitoring method and system based on data feedback, collects physical parameters such as line width, line spacing, hole diameter, roundness and copper thickness of developing etching, drilling and copper plating processes, and workshop temperature and humidity and dust concentration data, generates a single variable control chart for single process physical parameters, generates a multivariate T 2 Control chart for multiple coupling parameter processes, calculates process stability index and comprehensive fluctuation index, inputs abnormal data into a BP neural network, identifies abnormal mode types and outputs characteristic parameters, generates parameter correction instructions according to the abnormal mode and the characteristic parameters, adjusts equipment parameters in real time, recalculates the process capability index based on the adjusted data, triggers secondary feedback if the process capability index does not meet the standard, and dynamically adjusts the control chart threshold value according to the standard deviation and mean value of the process capability index of continuous batches. The application comprehensively covers key production parameters, discovers abnormalities in time, accurately and comprehensively detects, adjusts in real time to ensure stability and controllability, and improves product quality.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of circuit board quality detection, in particular to a flexible circuit board production quality monitoring method and system based on data feedback. BACKGROUND

[0002] As a key electronic component, circuit boards have realized the miniaturization and visualization of circuits by integrating various electronic components. In the mass production of fixed circuits and the optimization of electrical appliance layout, circuit boards play a crucial role. With the rapid development of technology, the application range of circuit boards is becoming more and more extensive, from simple household appliances to complex industrial control systems, almost everywhere. This extensive application background makes the market increasingly demanding for the quality of circuit boards.

[0003] Flexible circuit board is a kind of circuit board using flexible insulating substrate and conductive circuit, which is widely used in electronic equipment. Due to its lightness, thinness and bendability, it is favored. With the development of electronic equipment towards miniaturization, lightness and high integration, the quality requirements for flexible circuit board are becoming higher and higher. Traditional quality detection methods mainly rely on manual visual inspection and simple mechanical measurement, which not only have low efficiency, but also are easily affected by human factors, leading to inconsistency of detection results, which is difficult to meet the demand of modern production for high precision and high efficiency.

[0004] In the production process of circuit board, quality detection is an important link to ensure the stability and reliability of product performance. Traditional circuit board detection methods mostly rely on manual inspection, which not only has low efficiency, but also is easily affected by human factors, leading to low accuracy of detection results. The patent with publication number CN118762001B discloses a flexible circuit board quality detection system and method based on data analysis, which first captures the reference image of the flexible circuit board, extracts the bending axis of the flexible circuit board and the position relationship of the electronic components from the reference image, and constructs a model about image change and flexible circuit board quality. Using image recognition technology, the failure rate evaluation value of the flexible circuit board is generated to reduce the workload of manual detection. It can be seen that the detection of circuit board has been concerned. With the continuous development of circuit board industry, the requirement for circuit board detection technology is also getting higher and higher. Modern circuit board detection technology needs to have high efficient and accurate detection capability. SUMMARY

[0005] The purpose of the present application is to overcome the shortcomings of the prior art, provide a flexible circuit board production quality monitoring method and system based on data feedback, adopt multi-source data acquisition, abnormal fluctuation joint detection and feedback process adjustment method to ensure data reliability; timely find single process or multi-process abnormalities in the production process, avoid potential quality problems, and thus improve product quality.

[0006] The object of the present application is achieved by the following technical solutions:

[0007] A flexible circuit board production quality monitoring method based on data feedback, comprising:

[0008] S1. Multi-source data acquisition: Collecting physical parameter data of flexible circuit board developing etching, drilling and copper plating processes, the physical parameters including line width, line spacing, hole diameter, roundness and copper thickness, collecting workshop temperature and humidity and dust concentration data;

[0009] S2. Abnormal fluctuation joint detection: Generating single variable control chart for single process physical parameters to calculate process stability index, generating multivariate T2 control chart for multiple coupled parameter processes to calculate comprehensive fluctuation index, triggering abnormal signal when any index exceeds the preset threshold;

[0010] S3. Neural network mode analysis: inputting abnormal data into pre-trained BP neural network, identifying step, trend or periodic abnormal mode type and outputting characteristic parameters;

[0011] S4. Feedback process adjustment: generating parameter correction instructions according to abnormal mode type and characteristic parameters, and adjusting device temperature, pressure or motion accuracy parameters of corresponding process in real time;

[0012] S5. Quality closed loop evaluation: recalculating process capability index Cp and Cpk based on adjusted process data, triggering secondary feedback if the index is not up to standard, until the process is stable;

[0013] S6. Dynamic threshold updating: dynamically adjusting the threshold of single variable and multivariate control chart according to the standard deviation and mean of process capability index of continuous N batches , and adaptively optimizing the monitoring sensitivity.

[0014] The multi-source data acquisition module comprehensively collects key physical parameters in the flexible circuit board production process, such as line width, line spacing, hole diameter, roundness, and copper thickness in the development etching, drilling, and copper plating processes, as well as workshop environment temperature, humidity, and dust concentration data. These data provide a basis for subsequent anomaly detection and quality evaluation. Next, the anomaly fluctuation joint detection module uses univariate control charts and multivariate T² control charts to calculate the stability index of single-process physical parameters and the comprehensive fluctuation index of multi-coupling parameter processes, respectively. When any index exceeds the pre-set threshold, the system triggers an abnormal signal, indicating that there may be quality problems in the production process. Then, the neural network pattern analysis module inputs the abnormal data into a pre-trained BP neural network, which can identify different abnormal pattern types such as step, trend, or periodicity, and output corresponding feature parameters. These feature parameters help to further understand the specific situation and cause of the anomaly. According to the identified abnormal pattern type and feature parameters, the feedback process adjustment module generates parameter correction instructions to adjust the device temperature, pressure, or motion accuracy of the corresponding process in real time to eliminate the anomaly and restore the stability of the production process. The quality closed-loop evaluation module recalculates the process capability indexes Cp and Cpk based on the adjusted process data to evaluate the adjustment effect. If the indexes do not meet the standards, secondary feedback is triggered to continue adjustment until the process is stable. This process ensures continuous improvement and optimization of production quality. Finally, the dynamic threshold updating module dynamically adjusts the thresholds of univariate and multivariate control charts based on the standard deviation and mean of process capability indexes of the last N batches . This adaptive optimization mechanism can adjust the monitoring sensitivity as the production process changes, ensuring the accuracy and effectiveness of the monitoring system.

[0015] As a preferred mode, the univariate control chart in S2 includes a mean-range control chart, and the control limit calculation formula is:

[0016]

[0017] wherein, is the total sample mean, is the range mean, is the control chart coefficient, and are the upper and lower control limits, respectively.

[0018] As a preferred mode, the upper control limit calculation method of the multivariate T² control chart in S2 is:

[0019]

[0020] wherein, is the variable dimension, is the sample size, is the confidence level F-distribution critical value.

[0021] As a preferred mode, the training of the BP neural network in S3 includes: generating a control pattern sample set containing normal and abnormal patterns, extracting the statistical feature vector of the sample as input, the pattern type and the feature parameter as output, optimizing the network weight through the back propagation algorithm, and the statistical features include mean, range and standard deviation.

[0022] As a preferred mode, the parameter correction instruction in S4 includes:

[0023] Adjusting the temperature or pressure parameters of the step anomaly, and the correction amplitude is proportional to the step amplitude;

[0024] Calibrating the mechanical motion accuracy of the trend anomaly, and the correction slope is inversely proportional to the slope of the trend anomaly;

[0025] Replacing the worn parts for the periodic anomaly, and the replacement period matches the fluctuation wavelength.

[0026] As a preferred mode, the secondary feedback in S5 includes: marking the process that does not meet the standard as a high-risk batch, starting the manual review process, and comparing the review results with the neural network diagnosis results to optimize the model weight.

[0027] As a preferred mode, it further includes: environmental coupling compensation, calculating the environmental interference coefficient HJ according to the temperature, humidity and dust data, and dynamically compensating the physical parameters when HJ exceeds the threshold, the formula is:

[0028]

[0029] Among them, is the real-time temperature, is the standard temperature, is the real-time humidity, is the upper limit of humidity, is the dust concentration, , , is the weight coefficient.

[0030] As a preferred mode, the specific formula of the dynamic threshold update in S6 is:

[0031] New threshold = original threshold ×

[0032] Among them, is the process capability index standard deviation, is the mean, is the adjustment factor.

[0033] A flexible circuit board production quality monitoring system based on data feedback, comprising:

[0034] A data acquisition module, integrated with a high-precision microscope, a copper thickness measuring instrument and an environmental sensor, is used to acquire line width, line spacing, aperture, roundness, copper thickness and workshop environment data in real time;

[0035] A statistical analysis module performs single-variable control chart analysis and multi-variable T2 control chart analysis, and outputs process stability and comprehensive fluctuation indicators;

[0036] A neural network module identifies abnormal patterns of control charts and outputs characteristic parameters by using a built-in BP neural network;

[0037] A feedback execution module generates parameter adjustment instructions according to abnormal patterns, and dynamically controls production equipment;

[0038] A closed-loop evaluation module recalculates process capability indexes, determines whether secondary feedback is needed, and stores the results in a database;

[0039] An environmental compensation unit calculates an environmental interference coefficient HJ according to temperature, humidity and dust data, and dynamically corrects physical parameters; the modules communicate through an industrial bus and an Ethernet to realize real-time data interaction and closed-loop control.

[0040] Preferably, the data acquisition module includes a vibration compensation unit and a data verification unit, the vibration compensation unit eliminates the influence of mechanical vibration on detection accuracy through an acceleration sensor, and the data verification unit marks the time stamp of abnormal data and associates it with historical batches.

[0041] The present application has at least the following beneficial effects: The multi-source data acquisition strategy comprehensively covers key physical parameters and environmental variables in the production process, providing detailed data basis for subsequent abnormal detection and pattern analysis. The abnormal fluctuation joint detection mechanism can timely detect single-process or multi-process abnormalities in the production process, effectively avoiding potential quality problems, and the combination of single-variable and multi-variable control charts improves the accuracy and comprehensiveness of abnormal detection. The feedback process adjustment strategy can adjust equipment parameters in real time according to abnormal patterns and characteristic parameters, ensuring the stability and controllability of the production process, thereby improving product quality. BRIEF DESCRIPTION OF DRAWINGS

[0042] In order to reveal the technical details of the embodiments of the present application, the drawings involved in the embodiments will be briefly introduced. It should be emphasized that these drawings only present several embodiments of the present application and should not be regarded as a limitation on the scope of the invention. For those skilled in the art, other related drawings can be derived from these drawings without creative labor.

[0043] Figure 1 A flowchart of the method embodiment of the present application;

[0044] Figure 2 Structure diagram of an embodiment of a flexible circuit board production quality monitoring system based on data feedback;

[0045] Figure 3 Real-time correction process diagram based on data feedback. DETAILED DESCRIPTION

[0046] The technical solutions of the present application will be described in further detail below with reference to the accompanying drawings, but the scope of protection of the present application is not limited to the following description.

[0047] In the following, the embodiments of the present disclosure will be described in detail with the help of the accompanying drawings. However, it should be made clear that the present disclosure is not limited to the specific forms shown here. On the contrary, it should be understood to cover various changes, equivalent forms and / or alternatives of the embodiments of the present disclosure. In the process of describing the accompanying drawings, the same reference signs will be used to indicate similar components.

[0048] It should be noted that although detailed specific details are provided in the following description, the purpose is to help a comprehensive understanding of the example embodiments. However, those skilled in the art should know that the example embodiments can also be implemented without these specific details. For example, the system can be shown in the form of a block diagram, the purpose of which is to avoid excessive details that interfere with the clarity of the examples. In some other cases, in order to maintain the clarity of the examples, unnecessary details of well-known processes, structures and technologies can be omitted.

[0049] As shown in Figure 1 A flexible circuit board production quality monitoring method based on data feedback, comprising:

[0050] S1. Real-time acquisition of multi-source data: through a high-precision microscope detection platform and a copper thickness measuring instrument, physical parameter data of flexible circuit board development etching, drilling and copper plating processes are synchronously acquired, the physical parameters including line width, line spacing, hole diameter, roundness and copper thickness, and workshop temperature and humidity and dust concentration data are also acquired at the same time;

[0051] S2. Joint detection of abnormal fluctuations: a single variable control chart is generated for single-process physical parameters to calculate process stability indicators, and a multivariate T2 control chart is generated for multiple coupled parameter processes to calculate comprehensive fluctuation indicators, and an abnormal signal is triggered when any indicator exceeds a preset threshold;

[0052] S3. Neural network mode analysis: abnormal data is input into a pre-trained BP neural network to identify step, trend or periodic abnormal mode types and output characteristic parameters;

[0053] S4. Feedback process adjustment: generate parameter correction instructions according to the abnormal mode type and characteristic parameters, and real-time adjust the device temperature, pressure or motion accuracy parameters of the corresponding process;

[0054] S5. Quality closed-loop evaluation: recalculate the process capability indexes Cp and Cpk based on the adjusted process data, Cp measures the potential capability of the process in the ideal state, that is, whether the variation range of the process can meet the specification requirements assuming that the process mean and the specification center are completely consistent, Cpk introduces the influence of mean shift, reflects the actual process capability, these two indexes are commonly used and will not be described here. If the indexes do not meet the standard, secondary feedback is triggered until the process is stable;

[0055] S6. Dynamic threshold update: dynamically adjust the threshold of the single variable and multivariate control chart according to the standard deviation and mean of the process capability indexes of the continuous N batches , and adaptively optimize the monitoring sensitivity.

[0056] The embodiment is based on data feedback to monitor the production quality of flexible circuit boards. Through a high-precision microscope detection platform and a copper thickness measuring instrument, physical parameter data of each process of the flexible circuit board is collected in real time, such as line width, line spacing, hole diameter, roundness and copper thickness, and workshop environment data is monitored in real time, such as temperature, humidity and dust concentration. The single variable control chart and the multivariate T² control chart are used to calculate the process stability of single process and multi-coupling parameter process. Once abnormal fluctuation is found, an abnormal signal is triggered immediately. Then, the abnormal data is input into the pre-trained BP neural network to identify the abnormal mode type, such as step, trend or periodic abnormality, and output the characteristic parameters. According to the identified abnormal mode and characteristic parameters, parameter correction instructions are generated to real-time adjust the device parameters of the corresponding process, such as temperature, pressure or motion accuracy, to correct the abnormality. After adjusting the parameters, the process capability indexes Cp and Cpk are recalculated. If the indexes do not meet the standard, secondary feedback is triggered for continuous adjustment until the process is stable. Finally, according to the standard deviation and mean of the process capability indexes of the continuous batches, the threshold of the control chart is dynamically adjusted to adaptively optimize the sensitivity of the monitoring system, ensuring the accuracy and effectiveness of the monitoring.

[0057] In a preferred embodiment, the single variable control chart in S2 includes a mean-range control chart, and the control limit calculation formula is:

[0058]

[0059] wherein, is the total sample mean, is the range mean, is the control chart coefficient, and are the upper and lower control limits, respectively.

[0060] The mean-range control chart assesses the stability of a production process by monitoring the sample mean and range. The sample mean reflects the trend of the data set, while the range reflects the dispersion of the data. Control limits are calculated based on the overall mean and average range of historical data. The upper and lower limits are centered on the overall mean, with the addition or subtraction of the product of the coefficient A2 and the average range, forming a reasonable fluctuation interval. When data points exceed this interval, it indicates that there may be abnormal factors that need to be investigated and adjusted.

[0061] The mean-range control chart assesses the stability of a production process by monitoring the sample mean and range. The sample mean reflects the trend of the data set, while the range reflects the dispersion of the data. Control limits are calculated based on the overall mean and average range of historical data. The upper and lower limits are centered on the overall mean, with the addition or subtraction of the product of the coefficient A2 and the average range, forming a reasonable fluctuation interval. When data points exceed this interval, it indicates that there may be abnormal factors that need to be investigated and adjusted.

[0062] The control limits are set based on the overall mean and range of historical data. The upper and lower control limits of the mean chart are centered on the overall mean, with the addition or subtraction of a certain multiple (determined by the control chart coefficient) of the range mean, forming a reasonable fluctuation interval. The control limits of the range chart are also calculated based on the range mean and coefficient. When equipment abnormalities or material fluctuations occur in the production process, the sample mean or range may exceed the control limits, triggering a warning from the system, prompting the need to investigate process parameters or equipment status. This control chart can effectively identify sudden abnormalities or gradual trends in the process, providing a guarantee for the quality stability of high-precision flexible circuit boards.

[0063] In a preferred embodiment, the upper control limit of the multivariate T2 control chart in S2 is calculated as follows:

[0064]

[0065] wherein, is the variable dimension, is the sample size, is the confidence level is the F-distribution critical value at the confidence level.

[0066] Multivariate T2 control chart can monitor the comprehensive fluctuation of multiple related quality parameters simultaneously. In the production of flexible circuit board, such as line width and line spacing in developing etching process, hole diameter and roundness in drilling process, these parameters often have mutual correlation, and single parameter control chart cannot fully reflect the overall process state. Control chart integrates the fluctuation of multiple variables into a statistical quantity by calculating the "multidimensional distance" of each sample from the center of historical data, forming a unified quality fluctuation index. T2 statistic is essentially "multivariate distance", which is used to measure the "comprehensive deviation" between sample data and process mean. This distance is always non-negative, and only needs to focus on whether it exceeds the safety threshold (upper control limit). When the statistical quantity exceeds the pre-set threshold, it indicates that the process may be disturbed by abnormal factors.

[0067] The threshold of T2 control chart is set by considering the variable dimension, sample size and confidence level. The variable dimension refers to the number of parameters monitored simultaneously (such as 2 dimensions when monitoring line width and line spacing), the sample size refers to the amount of data collected in each batch, and the confidence level determines the sensitivity to abnormal fluctuations. When the process requires strictness, a higher confidence level (such as 99%) can be selected, and the threshold will be correspondingly increased to reduce the risk of false positives. This embodiment can capture both cooperative abnormalities between variables and avoid the problem of error accumulation when monitoring multiple parameters separately, and is particularly suitable for high-precision control scenarios of flexible circuit board parameters.

[0068] In a preferred embodiment, the training of the BP neural network in S3 includes: generating a control chart sample set containing normal and abnormal patterns, extracting the statistical feature vector of the sample as input, the pattern type and feature parameters as output, optimizing the network weight through the back propagation algorithm, and the statistical features include mean, range and standard deviation.

[0069] The BP neural network-based control chart abnormal pattern recognition realizes intelligent diagnosis through network training based on simulation data. The system first generates a control chart sample set containing normal and abnormal patterns, simulates step, trend, periodic and other abnormal patterns, such as step abnormality showing sudden jump of data, trend abnormality showing continuous rise or fall, and periodic abnormality showing regular fluctuation. The mean, range and standard deviation of each sample are extracted as three types of statistical features, the mean reflects the trend of the data set, the range captures the fluctuation range, and the standard deviation quantifies the dispersion degree, which together constitute the input feature vector. The network uses a three-layer structure, the input layer corresponds to three feature nodes, the hidden layer learns the feature association through nonlinear transformation, and the output layer is divided into pattern classification and parameter estimation two parts: the classification module outputs the probabilities of normal, step, trend and periodic, and the parameter module predicts the numerical values such as step amplitude, trend slope or periodic wavelength. During training, the back propagation algorithm is used to compare the predicted results with the true labels and adjust the weights layer by layer, and after tens of thousands of iterations, the network establishes the mapping relationship between features and patterns. In practical application, the system collects process data in real time and calculates statistical features, and inputs the trained network to obtain the abnormal type and quantitative parameters.

[0070] In a preferred embodiment, the parameter correction instruction in S4 comprises:

[0071] For step abnormality, adjust the equipment temperature or pressure parameters, and the correction amplitude is proportional to the step amplitude Δ;

[0072] For trend abnormality, calibrate the mechanical motion precision, and the correction slope is inversely proportional to the trend abnormality slope k;

[0073] For periodic abnormality, replace the worn parts, and the replacement period matches the fluctuation wavelength λ.

[0074] For different abnormal types in flexible circuit board production, the system adopts differentiated process adjustment strategies. For step abnormality, which is characterized by sudden jump of physical parameters (such as line width increasing by 1.5 μm), the system adjusts the equipment temperature or pressure parameters for correction. The correction amplitude is proportional to the step amplitude Δ, for example, the larger Δ is, the temperature compensation amount increases in proportion, so as to quickly offset the mutation effect. Such abnormality is usually caused by incorrect equipment parameters or material batch fluctuation, and the process stability can be restored through linear compensation.

[0075] In the production process of flexible circuit board, physical parameter jump (such as line width sudden increase of 1.5 pm) is usually caused by sudden change of process condition or misalignment of equipment parameter, and adjustment of temperature or pressure parameter can correct such abnormality. Taking temperature as an example, etching rate is positively correlated with temperature. In the process of developing and etching, temperature of etching liquid directly affects chemical reaction rate. Temperature rise accelerates corrosion speed of etching liquid to copper layer, resulting in line width narrowing; on the contrary, temperature drop slows down corrosion speed, resulting in line width widening. Flexible substrate material (such as polyimide) is sensitive to temperature. Temperature fluctuation can cause micron-level deformation of substrate, indirectly change alignment accuracy of etching mask and copper layer, and thus affect line width. When line width sudden increase (Delta>0) is detected, the system corrects through the following steps. Referring to Figure 3 ① Abnormality attribution (identify abnormality): step signal indicates that etching rate suddenly decreases, which can be caused by temperature sensor failure, abnormality of cooling system or sudden drop of ambient temperature. ② Temperature compensation formula (apply temperature compensation): adjustment amount is calculated through proportional model: Delta T=k*Delta, wherein, Delta is step amplitude, k is temperature correction coefficient, and Delta T is temperature adjustment amount. ③ Real-time feedback (provide real-time feedback): increase etching tank temperature (such as from 25 DEG C to 26.5 DEG C), accelerate etching liquid reaction, and make line width gradually return to target value in subsequent process. Temperature adjustment directly affects line width accuracy through dual effects of chemical reaction rate and material deformation. Based on real-time correction mechanism of data feedback, combined with quantitative model of process parameter and physical parameter, the system can quickly respond to step abnormality and guarantee stability of micron-level manufacturing of flexible circuit board.

[0076] In the process of developing and etching of flexible circuit board, adjustment of pressure parameter can correct line width sudden increase. Etching uniformity control: pressure directly affects contact tightness between etching roller and substrate. When pressure is insufficient, etching liquid is not evenly distributed, copper layer in local area is not fully removed, resulting in line width widening; when pressure is too high, it can squeeze mask, cause etching liquid penetration abnormality, and thus result in line width fluctuation. Mechanical stability: pressure fluctuation can cause equipment vibration, resulting in misalignment of mask and substrate. For example, imbalance of roller pressure can cause misalignment of etching area, directly magnify line width deviation. Pressure correction: when line width sudden increase Delta is detected, the following steps are executed: ① Abnormality attribution: through comparison of historical data and process model, it is judged whether it is caused by pressure abnormality. If etching rate decrease and mask displacement alarm are accompanied, pressure factor can be locked. ② Correction amount calculation: linear proportional model is adopted: , wherein, is pressure adjustment amount (such as unit: MPa), is pressure correction coefficient (calibrated through experiment, for example, k=0.3 MPa / pm), is the line width sudden increase amplitude (e.g. 1.5 pm). ③ Real-time feedback: through the servo system dynamic adjustment of hydraulic device, the roll pressure is increased, the contact efficiency of etching liquid and copper layer is improved, and the line width continues to be wide is inhibited. The pressure parameter directly acts on the copper layer removal rate by regulating the etching liquid distribution and the equipment mechanical stability, thereby quickly correcting the line width abnormality. Combined with the proportional model and closed-loop feedback, the automatic control from "detecting abnormality" to "dynamic pressure regulation" is realized, which provides a key guarantee for micron-level manufacturing of high-density flexible circuit board.

[0077] When the trend anomaly (e.g. the hole diameter continues to decrease) is detected, the system determines that it is a gradual deviation caused by mechanical motion precision deviation. At this time, the equipment motion mechanism needs to be calibrated, and the correction slope is inversely proportional to the trend anomaly slope k, that is, the steeper the trend (k value is large), the smaller the calibration amplitude to avoid overshoot; the trend is gentle (k value is small) and the calibration amount is increased. For example, the linear change of the hole diameter is caused by the wear of the guide rail of the drilling equipment, and the cumulative error can be eliminated by compensating the motion precision in the opposite direction.

[0078] When the trend anomaly (e.g. the hole diameter continues to decrease) is detected, the system corrects the deviation by calibrating the equipment motion mechanism. The trend slope k reflects the deviation accumulation rate of the process parameter. The k value is large (the trend is steep): it indicates that the deviation accumulates rapidly in a short time, which may be caused by sudden wear of mechanical parts or instantaneous interference (such as vibration). At this time, if the adjustment is large, it may exceed the dynamic response capability of the equipment, causing overshoot or oscillation. The k value is small (the trend is gentle): it indicates that the deviation accumulates slowly, which is usually caused by long-term wear or environmental gradual change (such as temperature drift), and a larger correction amount is needed to offset the long-term cumulative effect.

[0079] This embodiment adopts an inverse relationship (the correction amplitude ∝ 1 / |k|) to realize stable compensation: steep trend (large k): small adjustment to avoid drastic disturbance. For example, the sudden wear of the guide rail of the drilling machine causes the hole diameter to decrease by 2 pm per hour (k=2 pm / h), and the system slightly adjusts the guide rail pressure (e.g. reduces 0.5%), which gradually approaches the target value through multiple iterations. Gentle trend (small k): increase the correction amount to cover the long-term effect. If the hole diameter decreases by 0.5 pm per month (k=0.5 pm / month), the system may calibrate the mechanical transmission ratio (e.g. increase by 3%), which completely compensates for the systematic deviation caused by wear.

[0080] Dynamic lag compensation, the mechanical system has response lag, and direct correction according to the k value proportion may cause overshoot. The inverse ratio strategy balances the response speed and stability through adaptive adjustment. The formula is: correction amount correction amount Δ= , where λ is the reference correction coefficient, to prevent zero constant. The inverse correction strategy quantifies the deviation accumulation rate, dynamically matches the response characteristics of the mechanical system, effectively eliminates the deviation while suppressing overshoot, and ensures the stability of micron-level manufacturing.

[0081] Periodic anomalies (such as copper thickness regular fluctuations) are mostly caused by periodic wear or vibration of parts. The system dynamically sets the replacement cycle of parts according to the fluctuation wavelength λ, and increases the replacement frequency when λ is short, and prolongs the maintenance interval when λ is long. If the copper plated roller shaft has a fluctuation of amplitude A = 0.8 μm every λ = 8 hours, then match the 8-hour replacement cycle to block the abnormal fluctuation from the source. The three adjustment strategies form a closed-loop control to ensure that the production quickly returns to a stable state.

[0082] In a preferred embodiment, the data in S1 is transmitted through a C / S architecture, the client is deployed in the process detection terminal, and the server integrates historical batch data and automatically compares with the Gerber production standard file. The historical data includes process parameter setting values and defect records. Through the client-server (C / S) architecture, distributed management of flexible circuit board production data is realized. The client is deployed at each process detection terminal (such as near the developing etching, drilling, and copper plating equipment), responsible for real-time acquisition of physical parameters such as line width and aperture obtained by high-precision microscopes and copper thickness measuring instruments, and records the workshop temperature and humidity and dust concentration data. The collected data is transmitted to the server through industrial bus or Ethernet, ensuring the immediacy and integrity of the detection results.

[0083] The server integrates all the historical batch data uploaded by the clients, and builds a comprehensive database containing process parameter setting values, defect records, and environmental parameters. The system automatically compares real-time detection data with Gerber production standard files. Gerber files serve as the digital blueprint of circuit board design, storing nominal values and tolerance ranges of parameters such as line width and aperture. If the detection data exceeds the threshold defined in the Gerber file, the server immediately marks the anomaly and triggers an early warning, helping operators quickly locate process deviations. The linkage analysis of historical data and real-time data is one of the core functions of the system. The server optimizes the current production parameter settings by analyzing the process parameter fluctuation rules in historical batches. For example, when the aperture roundness anomaly frequently occurs in a batch of drilling processes, the system can trace back to historical data, correlate equipment wear cycles or environmental parameter changes, and generate targeted maintenance recommendations. At the same time, the accumulation of defect records provides training data for neural network models, continuously improving the accuracy of anomaly pattern recognition.

[0084] The authority management module ensures data security. Operators of different processes can only access detection data and statistical analysis results of their own processes, and administrators can view global data and adjust system parameters through the server. This hierarchical design not only guarantees data isolation, but also supports cross-process collaborative optimization. Through the closed-loop mechanism of real-time acquisition, intelligent analysis, and dynamic adjustment, the system significantly improves the stability and yield of flexible circuit board manufacturing.

[0085] In a preferred embodiment, the secondary feedback in S5 includes marking the non-compliant process as a high-risk batch, starting a manual review process, and comparing the review results with the neural network diagnosis results to optimize the model weights. The secondary feedback mechanism plays a key role in quality closed-loop evaluation. When the process capability index (Cp / Cpk) recalculated after process adjustment is still not up to standard, the system automatically marks the batch as high-risk, triggering a manual review process. The operator performs a full inspection on the high-risk batch, records the actual defect types and locations, such as specific values of line width deviation or hole position shift. These manual review results are compared with the abnormal pattern and parameter estimates previously diagnosed by the neural network. If it is found that the neural network misjudges (such as identifying a trend anomaly caused by mechanical wear as environmental interference), the system will use the difference data as new training samples to dynamically adjust the network weights. For example, if a batch review shows that the amplitude of periodic fluctuations in hole diameter is 1.2 pm, while the original estimate of the neural network is 0.9 pm, the system corrects the neuron connection strength through the back propagation algorithm, so that the parameter estimation error of similar cases is reduced. This closed-loop mechanism of automatic adjustment, manual verification, and model iteration not only avoids the limitations of relying solely on algorithms, but also continuously optimizes the model with real data, ultimately achieving a dual improvement in the accuracy and adaptability of the monitoring system.

[0086] In a preferred embodiment, it also includes environmental coupling compensation, calculating the environmental interference coefficient HJ according to temperature, humidity, and dust data, and dynamically compensating the physical parameters when HJ exceeds the threshold value, with the formula being:

[0087]

[0088] wherein, is the real-time temperature, is the standard temperature, is the real-time humidity, is the upper limit of humidity, is the dust concentration, , , is the weight coefficient. HJ (environmental interference coefficient) is the trigger condition for determining whether compensation needs to be started.

[0089] In the production process of flexible circuit boards, fluctuations in workshop environmental parameters directly affect the measurement accuracy of micron-level process parameters. For example, an increase in temperature may cause the substrate material to expand thermally, causing the line width detection value to deviate from the true value; too low humidity may exacerbate dust adsorption, interfering with the imaging clarity of high-precision microscopes; and too high dust concentration may contaminate the etching liquid, causing copper thickness measurement deviation. The environmental coupling compensation mechanism quantifies these interference factors, builds a comprehensive evaluation model, and calculates the environmental interference coefficient HJ in real time.

[0090] The calculation of HJ combines three key environmental parameters: temperature, humidity, and dust. Temperature deviation reflects the difference between actual temperature and standard process temperature. The larger the deviation, the more significant the impact on detection accuracy. The humidity deficiency index measures the degree of air dryness. The lower the humidity, the stronger the electrostatic adsorption effect. Dust concentration is directly related to the cleanliness of the workshop. The higher the concentration, the more serious the optical detection interference. The three parameters are weighted and summed through pre-set weight coefficients. The weight values are calibrated based on historical data, reflecting the sensitivity differences of different processes to environmental factors. For example, the development etching process is more sensitive to temperature changes, The drilling process is more affected by dust, The weight is correspondingly increased.

[0091] When the HJ value exceeds the set threshold (the threshold of HJ is set according to historical data or process requirements), the system automatically triggers dynamic compensation. For example, if HJ = 1.2 (threshold HJ = 1.0) is detected, it indicates that environmental interference has affected process stability. At this time, the system locates the main disturbance source according to the contribution value of each item: if the temperature deviation accounts for 60%, the etching tank constant temperature system is adjusted first; if the dust concentration is abnormally prominent, the workshop fresh air system is linked to reduce dust. The compensation amplitude is proportional to the HJ overrun value, ensuring that the micron-level parameter measurement value restores the true process state. This mechanism effectively solves the misjudgment problem caused by environmental noise in traditional detection, allowing high-density flexible circuit boards to maintain a detection accuracy of ±0.5 μm under complex workshop conditions.

[0092] The compensation amount of each physical parameter is related to the environmental parameter it is sensitive to. The compensation amount of each parameter is the weighted sum of the environmental deviation components, and the weight is determined by the sensitivity coefficient: .

[0093] For line width: , , , The sensitivity coefficients of line width to temperature, humidity, and dust are respectively (experimentally calibrated). The compensation direction: when the temperature rises and the line width becomes larger, Δ line width needs to be subtracted during correction; when the temperature decreases and the line width becomes smaller, Δ line width needs to be added during correction. When the humidity is insufficient and the line width becomes smaller, Δ line width needs to be added during correction; when the humidity is too high and the line width becomes larger, Δ line width needs to be subtracted during correction. When the dust concentration increases and the line width becomes smaller, Δ line width needs to be added during correction; when the dust concentration decreases and the line width becomes larger, Δ line width needs to be subtracted during correction.

[0094] For line spacing: ; , , The sensitivity coefficients of line distance to temperature, humidity, and dust, respectively. Compensation direction: when temperature rises and causes line distance to be too small, add Δ line distance during correction; when temperature drops and causes line distance to be too large, subtract Δ line distance during correction. When humidity is insufficient and causes line distance to be too small, increase Δ line distance during correction; when humidity is too high and causes line distance to be too large, decrease Δ line distance during correction. When dust concentration rises and causes line distance to be too small, add Δ line distance during correction; when dust concentration drops and causes line distance to be too large, subtract Δ line distance during correction.

[0095] For aperture: ; The sensitivity coefficients of aperture to temperature and dust, respectively (dust has a higher weight). Compensation direction: when temperature rises and causes aperture to be too large, subtract Δ aperture during correction; when temperature drops and causes aperture to be too small, add Δ aperture during correction. When dust concentration rises and causes aperture to be too small, increase Δ aperture during correction; when dust concentration drops and causes aperture to be too large, decrease Δ aperture during correction.

[0096] For roundness: , The sensitivity coefficients of roundness to humidity and dust, respectively. Compensation direction: when humidity is insufficient and causes roundness to decrease, increase Δ roundness during correction; when humidity is too high and causes roundness to abnormally increase, decrease Δ roundness during correction. When dust concentration rises and causes roundness deviation to increase, increase Δ roundness during correction; when dust concentration drops and causes roundness deviation to decrease, maintain Δ roundness during correction.

[0097] For copper thickness: , The sensitivity coefficients of copper thickness to temperature and humidity, respectively. Compensation direction: when temperature rises and causes etching rate to accelerate, copper thickness to be too thin, increase Δ copper thickness during correction; when temperature drops and causes etching rate to slow down, copper thickness to be too thick, decrease Δ copper thickness during correction. When humidity is insufficient and causes copper thickness to be too thick, decrease Δ copper thickness during correction; when humidity is too high and causes copper thickness to be too thin, increase Δ copper thickness during correction.

[0098] In a preferred embodiment, the specific formula of the dynamic threshold value update in S6 is:

[0099] New threshold value = original threshold value ×

[0100] Wherein, is the process capability index standard deviation, is the mean value, is the adjustment factor.

[0101] The dynamic threshold updating mechanism automatically adjusts the alarm threshold of the control chart by quantifying the fluctuation degree of the production process. The traditional fixed threshold is prone to misjudgment when the process stability fluctuates. For example, when small fluctuations frequently occur in the production line (such as material batch differences or slight equipment aging), the fixed threshold may frequently trigger false alarms. Conversely, if the process capability is continuously optimized (such as after process improvement, the Cp value is improved), the original threshold may be too loose to cause missed alarms. This mechanism dynamically amplifies or narrows the threshold range by calculating the standard deviation and mean ratio of the recent process capability index (Cp). The larger the standard deviation, the more intense the process fluctuation, and the threshold tolerance is proportionally increased to avoid false alarms. The smaller the standard deviation, the tighter the threshold, and the sensitivity of abnormal detection is improved.

[0102] The adjustment factor β plays a balancing role in this process. When β takes a larger value (such as β = 1.5), the system is more sensitive to fluctuations, and the threshold adjustment range is large, which is suitable for scenarios with extremely high stability requirements and low tolerance. When β takes a smaller value (such as β = 0.5), the adjustment range is reduced, and the threshold is relatively stable, which is suitable for the production of conventional consumer products. This implementation can adapt to natural fluctuations in the process and match the control requirements of different production stages through parameter configuration, realizing the intelligent upgrade from rigid threshold to flexible adaptation.

[0103] In a preferred embodiment, the characteristic parameters in S3 include: amplitude Δ and occurrence time point t of the step anomaly, slope k and duration τ of the trend anomaly, wavelength λ and amplitude A of the periodic anomaly, and the error rate of anomaly pattern recognition and parameter estimation is less than 5%. The anomaly pattern recognition can accurately determine the type of control chart anomaly and quantify its key parameter characteristics through the trained neural network model. The system first models the characteristics of the three anomaly patterns of step, trend and periodic: the step anomaly shows that the data suddenly changes at a certain time point, and the neural network identifies the device parameter error or batch material anomaly by analyzing the data jump amplitude Δ and mutation position t; the trend anomaly reflects the gradual deviation of the data that continuously increases or decreases, and the system extracts the slope k and duration τ to determine the long-term influencing factors such as mechanical wear or environmental drift; the periodic anomaly presents regular fluctuations, and the algorithm captures the wavelength λ and amplitude A to identify periodic failure sources such as device vibration or component fatigue. In order to ensure that the error rate is less than 5%, the system uses a two-stage learning strategy. In the first stage, the network is trained with tens of thousands of simulated data, covering different combinations of amplitude, slope and period, to enhance the model's ability to distinguish abnormal characteristics; in the second stage, actual production data is injected to fine-tune the weights, eliminating the deviation between simulation and reality. For example, for the step anomaly, the network compares historical mutation cases to control the amplitude error within ±0.1 μm; for the trend slope, the model is calibrated in combination with device maintenance records to ensure that the slope estimation error is less than 2%. This dual mechanism of simulation training and actual verification enables the system to maintain high-precision recognition in complex working conditions. In actual application, the system receives the control chart data stream in real time, and compares it with the preset pattern characteristics point by point. When an abnormal signal is detected, the type judgment and parameter estimation results are output simultaneously. For example, when the periodic wavelength λ=8 hours and the amplitude A=1.2 μm are identified, the device maintenance period is automatically associated, triggering a bearing replacement warning. Through parameterized feedback, the system converts abstract fluctuations into executable process instructions, forming a closed-loop control link from anomaly detection to precise intervention.

[0104] A data feedback-based flexible circuit board production quality monitoring system, referring to Figure 2 , comprising:

[0105] A data acquisition module integrated with a high-precision microscope, a copper thickness measuring instrument and an environmental sensor for real-time acquisition of line width, line spacing, aperture, roundness, copper thickness and workshop environment data;

[0106] A statistical analysis module for performing univariate control chart analysis and multivariate T² control chart analysis and outputting process stability and comprehensive fluctuation indicators;

[0107] A neural network module with a built-in BP neural network for identifying control chart anomaly patterns and outputting characteristic parameters;

[0108] a feedback execution module for generating parameter adjustment instructions according to the abnormal mode to dynamically control the production equipment;

[0109] a closed-loop evaluation module for recalculating the process capability index to determine whether secondary feedback is needed and storing the result in a database;

[0110] an environmental compensation unit for calculating an environmental interference coefficient HJ according to the temperature and humidity and dust data to dynamically correct the physical parameters; the modules realize real-time data interaction and closed-loop control through industrial bus and Ethernet communication.

[0111] The flexible circuit board production quality monitoring system realizes intelligent monitoring of the whole process through multi-module cooperation. The data acquisition module is equipped with a high-precision microscope and a copper thickness measuring instrument, which can capture micron-level parameters such as line width and hole diameter in processes such as development etching, drilling and copper plating in real time, and synchronously integrate temperature and humidity and dust sensors to monitor the workshop environment. The raw data collected are processed by the statistical analysis module, the individual variable control chart tracks the independent fluctuations of each parameter, and the multivariate T2 control chart evaluates the comprehensive abnormalities of related parameters, such as the simultaneous monitoring of the cooperative shift of line width and line spacing.

[0112] When an abnormal signal is detected, the neural network module starts analysis. The pre-trained BP neural network identifies abnormal modes such as step, trend and cycle, and outputs specific parameters. For example, if a step abnormality is detected, the jump amplitude 1.5 μm and the time point are given simultaneously. The feedback execution module then links the production equipment, such as adjusting the etching temperature to compensate for line width mutation, or calibrating the drilling machine pressure to correct hole diameter deviation, to form an immediate response of detection, diagnosis and adjustment.

[0113] The environmental compensation unit dynamically corrects measurement errors. For example, when high temperature causes the substrate to expand, the actual line width is recalculated according to the temperature and humidity weight coefficient to ensure data authenticity. The closed-loop evaluation module verifies whether the process capability index meets the standard after process adjustment, and if not, triggers manual re-inspection and optimizes the neural network model. The modules interact with data in real time through industrial networks to build a complete closed loop from parameter collection to process optimization, significantly improving the stability and yield of high-precision flexible circuit board production.

[0114] In a preferred embodiment, the data acquisition module includes a vibration compensation unit and a data verification unit. The vibration compensation unit eliminates the influence of mechanical vibration on detection accuracy through an acceleration sensor, and the data verification unit marks the timestamp of abnormal data and associates it with historical batches.

[0115] The data acquisition module of the flexible circuit board manufacturing system ensures detection accuracy and data reliability through a vibration compensation unit and a data verification unit. The vibration compensation unit uses an acceleration sensor to monitor mechanical vibrations in real time during equipment operation. When a high-precision microscope or copper thickness measuring instrument produces micron-level shifts due to equipment vibration, the sensor captures the vibration amplitude and frequency signals, generates a reverse compensation instruction through an algorithm, and dynamically adjusts the position of the detection platform or the optical focal length. For example, when a drilling machine is running at high speed, the system detects a 10Hz horizontal vibration signal, immediately drives the servo motor to reverse the fine adjustment of the stage position, eliminates a displacement error of ±0.8μm, and ensures the accuracy of aperture measurement.

[0116] The data verification unit performs multiple verifications when collecting data. When a sudden increase or drop in parameters such as line width and aperture is detected, the unit compares the fluctuation range of historical data for the same batch. If the line width measurement value deviates from the historical mean value by more than 3σ, the system automatically marks the abnormal timestamp and associates the production equipment number, environmental parameters, and operator information for that batch. For example, if a batch detects a line width jump of 1.2μm for three consecutive times, the verification unit correlates it with the vibration compensation log of the same model substrate from the previous week, quickly locates the systemic deviation caused by loose guide rail screws, and provides structured data support for subsequent quality traceability. Through the dual mechanisms of vibration suppression and data cross-validation, the system can still maintain micron-level detection accuracy in complex industrial environments.

[0117] In a preferred embodiment, the feedback execution module includes an instruction priority queue and a parameter correction verification unit. The priority queue is sorted by the severity of the abnormality and allocates adjustment resources. The verification unit collects 3 sets of verification data after adjustment, and if it does not meet the standard, it triggers a manual intervention signal. The feedback execution module ensures the accuracy of process adjustment through intelligent instruction scheduling and effect verification mechanism. After the system receives the abnormal type and parameters determined by the neural network, the instruction priority queue sorts the adjustment instructions according to the preset rules. For example, if a step abnormality causes a line width jump of more than ±2μm, it is classified as high priority and triggers immediate adjustment of equipment temperature or pressure; if the slope of a trend abnormality is less than 0.1μm / batch, it is classified as medium priority and the mechanical precision is calibrated during the idle period of the equipment. This hierarchical strategy can avoid resource conflicts caused by simultaneous adjustment of multiple processes, and prioritize high-risk abnormalities to reduce scrap rates.

[0118] The parameter correction verification unit initiates a closed-loop verification process after the execution of the instruction. The system automatically collects three sets of continuous process data after adjustment, recalculates the mean, range and other statistical quantities and compares them with the target value. If all three sets of data are within the control limit, it is determined that the adjustment is effective and the process parameters are updated; if any one set is out of limit, an immediate manual intervention signal is triggered and the abnormal details are pushed to the engineer terminal. For example, the verification data after a certain hole diameter correction shows that the fluctuation amplitude has not been reduced, the system will associate the equipment maintenance log, prompt that the compensation may be invalid due to guide rail wear, and guide manual intervention for deep maintenance. The dynamic adjustment and multiple verification mechanism effectively balances the automation efficiency and process reliability.

[0119] In a preferred embodiment, the database stores data in a time sequence structure, supports multi-dimensional retrieval by batch, time, and process, and automatically compares differences with production standard files Gerber. The database of the system uses a time-sequential structured storage method, which records various data in the production process of flexible circuit boards. The detection values of physical parameters such as line width and hole diameter of each batch are accompanied by accurate time stamps, and are classified and archived according to processes such as development etching, drilling, and copper plating. This time sequence storage mechanism makes historical data traceable, for example, when a hole diameter deviation occurs in a batch, the detection records of the batch at different processes can be quickly traced back to analyze the root cause of the problem. The database supports flexible multi-dimensional retrieval functions. Users can query the production data of a specific product throughout its life cycle by batch number, or filter the quality trend of a certain week or month by time range, or extract all related parameters for a single process (such as drilling). This multi-angle query capability helps engineers quickly locate abnormal periods, such as filtering all records of line width exceeding the standard within the past 24 hours, and analyzing the impact of equipment temperature fluctuations in combination with equipment logs. The system automatically compares the detection data with the Gerber design file. The Gerber file serves as a digital blueprint for circuit board design, storing the nominal values and tolerance ranges of parameters such as line width and hole diameter. Each time new detection data is entered, the system matches the corresponding design standard in real time and marks abnormal points that exceed the tolerance. For example, if a hole is designed to be 20 μm in diameter and the measured value is 19.2 μm, the system automatically triggers an early warning and associates the processing equipment number of the hole, providing accurate positioning for process adjustment. This dynamic comparison mechanism realizes closed-loop verification from design standards to production practice, ensuring that micron-level process parameters strictly meet quality requirements.

[0120] In a preferred embodiment, a mobile monitoring terminal is also included, which receives high-risk alerts from the closed-loop evaluation module and displays real-time video streams, parameter curves, and neural network diagnostic reports, supporting remote confirmation and adjustment instructions. The mobile monitoring terminal provides real-time remote monitoring and decision support capabilities for the flexible circuit board production quality monitoring system. The terminal is connected to the main control system through a wireless network. When the closed-loop evaluation module detects high-risk batches or process abnormalities, the system automatically pushes alerts to the terminal interface. The alert information includes the type of abnormality (such as sudden changes in line width, periodic fluctuations in hole diameter), the location of the abnormality (development etching process or drilling equipment number), real-time parameter curves (such as temperature fluctuation trends), and neural network diagnostic reports (abnormal pattern classification and characteristic parameter estimation). The operator can view the real-time camera footage of the workshop through the terminal to observe the equipment operation status or process adjustment effect, such as confirming whether the drilling machine is causing hole diameter deviation due to guide rail wear.

[0121] The terminal supports interactive operations. After receiving the alert, the engineer can directly retrieve historical data for comparative analysis on the interface. If the system indicates that the abnormal slope of the line width trend of a batch exceeds the standard, the engineer can remotely view the line width control chart of the same process over the past 24 hours and, combined with the equipment maintenance record, determine whether to shut down for maintenance. After confirming the abnormality, the terminal provides a one-key instruction sending function, such as sending an adjustment command of "copper thickness compensation +0.5 μm" to the copper plating process. The device automatically executes the received instruction and feeds back the execution status. To adapt to complex industrial environments, the terminal has a multi-level permission management system. The workshop director can view global data and authorize emergency operations, while the process supervisor can only access information about the process. All operation records and data changes are encrypted and stored to ensure traceability and security. The closed-loop remote management mechanism of early warning, analysis, decision-making, and execution significantly improves the response speed and cross-regional collaboration efficiency of high-density flexible circuit board production.

[0122] The flexible circuit board production quality monitoring system of the application realizes whole-process monitoring from data acquisition to intelligent optimization through multi-module cooperative construction of closed-loop control system. The data acquisition module obtains physical parameters such as line width and aperture and environmental data in real time through high-precision microscopes, copper thickness measuring instruments and environmental sensors, and transmits the data to the statistical analysis module through industrial networks, and synchronously detects single-process stability and multi-parameter coupling abnormalities through single-variable mean-range control chart and multi-variable T2 control chart. The neural network module performs pattern recognition on abnormal data, distinguishes step, trend or periodic type and quantifies characteristic parameters, the feedback execution module dynamically adjusts temperature, pressure or motion accuracy according to abnormal patterns, the closed-loop evaluation module recalculates process capability index to verify the adjustment effect, and triggers manual re-inspection and optimizes the model when the effect is not up to standard. The environmental compensation unit dynamically corrects measurement deviation based on temperature, humidity and dust data, the mobile terminal realizes remote real-time monitoring and instruction issuing, and the database supports time sequence storage and compares differences with Gerber standards. The modules are organically connected through data flow and feedback chain, forming a closed-loop self-optimization mechanism of perception, analysis, decision-making and verification, which significantly improves the precision and yield of micron-level manufacturing of flexible circuit boards.

[0123] Although the preferred embodiments of the present application have been described, those skilled in the art who understand the basic inventive concept can make additional changes and modifications to the embodiments once they are aware of the basic inventive concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application. The above description is only the preferred embodiments of the present application and is not intended to limit the present application. It should be noted that any modification, equivalent replacement and improvement within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A data feedback-based flexible circuit board production quality monitoring method, characterized by, Comprise: S1. Multi-source data acquisition: collect physical parameter data of flexible circuit board development etching, drilling and copper plating process, the physical parameters include line width, line spacing, hole diameter, roundness and copper thickness, collect workshop temperature and humidity and dust concentration data; S2. Abnormal fluctuation joint detection: generate single variable control chart for single process physical parameter to calculate process stability index, generate multivariate T2 control chart for multiple coupled parameter process to calculate comprehensive fluctuation index, trigger abnormal signal when any index exceeds preset threshold; S3. Neural network mode analysis: input abnormal data into pre-trained BP neural network, identify step, trend or periodic abnormal mode type and output characteristic parameters; S4. Feedback process adjustment: generate parameter correction instruction according to abnormal mode type and characteristic parameters, real-time adjust equipment temperature, pressure or motion accuracy parameters of corresponding process; The parameter correction instruction in S4 comprises: Adjust equipment temperature or pressure parameters for step abnormality, the correction amplitude is proportional to the step amplitude; Calibrate mechanical motion accuracy for trend abnormality, the correction slope is inversely proportional to the trend abnormality slope; Replace worn parts for periodic abnormality, the replacement period matches the fluctuation wavelength; S5. Quality closed loop evaluation: recalculate process capability index Cp and Cpk based on adjusted process data, Cp measures the potential capability of process in ideal state, Cpk introduces the influence of mean value offset on the basis of Cp, reflects the actual process capability, if the index is not up to standard, trigger secondary feedback until the process is stable; S6. Dynamic threshold updating: according to the process capability index standard deviation of the last N batches with the mean , dynamically adjust the threshold of univariate and multivariate control charts, and adaptively optimize the monitoring sensitivity; Further comprise: environmental coupling compensation, calculate environmental disturbance coefficient HJ according to temperature and humidity and dust data, dynamically compensate physical parameters when HJ exceeds threshold, the formula is: wherein, is the real-time temperature, is the standard temperature, is the real-time humidity, is the upper limit of humidity, is the dust concentration, , , is the weight coefficient.

2. The method for monitoring production quality of a flexible circuit board based on data feedback according to claim 1, wherein, The single variable control chart in S2 comprises mean-range control chart, the control limit calculation formula is: wherein, is the total mean of the sample, is the range mean, is the control chart coefficient, and are the upper and lower control limits, respectively.

3. The method of claim 1, wherein the method comprises: The upper control limit calculation method of multivariate T2 control chart in S2 is: wherein, is the variable dimension, is the sample size, is the confidence level is the F-distribution critical value at the alpha level.

4. The method of claim 1, wherein the method comprises: The training of BP neural network in S3 comprises: generate control chart sample set containing normal and abnormal modes, extract statistical feature vector of sample as input, mode type and characteristic parameters as output, optimize network weight through back propagation algorithm, the statistical features include mean, range and standard deviation.

5. The method of claim 1, wherein the method further comprises: The secondary feedback in S5 comprises: mark the process not up to standard as high-risk batch, start manual review process, and compare review results with neural network diagnosis results to optimize model weight.

6. The method of claim 1, wherein, The specific formula of dynamic threshold update in S6 is: New threshold = old threshold x wherein, is the process capability index standard deviation, is the mean, is the adjustment factor.

7. A data feedback based flexible circuit board production quality monitoring system, implemented by using the data feedback based flexible circuit board production quality monitoring method of any one of claims 1-6, characterized in that, Comprise: Data acquisition module, used for acquiring line width, line spacing, hole diameter, roundness, copper thickness and workshop environment data; Statistical analysis module, performing single variable control chart analysis and multivariate T2 control chart analysis, outputting process stability and comprehensive fluctuation index; Neural network module, built-in BP neural network, identifying control chart abnormal mode and outputting characteristic parameters; Feedback execution module, generating parameter adjustment instruction according to abnormal mode, dynamically controlling production equipment; Closed loop evaluation module, recalculating process capability index, determining whether secondary feedback is needed, and storing the results to database; Environmental compensation unit, calculating environmental disturbance coefficient HJ according to temperature and humidity and dust data, dynamically correcting physical parameters; The modules communicate through an industrial bus and Ethernet, and realize real-time data interaction and closed-loop control.

8. The data feedback-based flexible circuit board production quality monitoring system according to claim 7, wherein, The data acquisition module comprises a vibration compensation unit and a data verification unit, the vibration compensation unit eliminates the influence of mechanical vibration on detection accuracy through an acceleration sensor, and the data verification unit marks a time stamp for abnormal data and associates the abnormal data with historical batches.

Images