Circuit board production quality prediction and evaluation method based on big data

Abstract

The invention provides a circuit board production quality prediction and evaluation method based on big data, and the method comprises the steps: building an initial data set through collecting the historical data and real-time operation state of a drill bit in a circuit board processing process, and extracting the wear degree characteristics of the drill bit; the method comprises the following steps: acquiring current production data through a suggestion signal for replacing a drill bit, acquiring and combining time prediction of accumulated use times of the drill bit to reach a life node, and determining an optimal replacement opportunity for immediately replacing the drill bit after the current production batch is completed; acquiring machining data of a new replaced drill bit according to the optimal replacement opportunity of the replaced drill bit, and generating an updated through hole quality influence prediction result; and according to the updated through hole quality influence prediction result, adjusting the vibration frequency change of the drill bit, the collection frequency of residual chip accumulation in the hole and the analysis depth of the hole wall smoothness score, and obtaining the latest circuit board quality evaluation data.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to a method for predicting and evaluating the production quality of printed circuit boards based on big data. Background Technology

[0002] In modern manufacturing, the drilling accuracy in circuit board processing directly affects the performance and reliability of electronic products, its importance being self-evident. Drilling quality not only influences the connection effect of through holes but also plays a decisive role in the success rate of subsequent processes and the overall quality of the board, making it a core element in ensuring product stability and production efficiency. However, current methods for predicting and evaluating drilling accuracy still have significant shortcomings. These methods often overlook the dynamic changes in tool condition and the coupling effect of multiple factors during processing, making it difficult to fully capture the deep-seated patterns affecting drilling quality, especially regarding the continuous impact of tool wear on processing results in long-term production, lacking effective real-time monitoring and prediction methods. Specifically, the key challenge in this field focuses on the core factor of drill bit wear. With increased use, drill bit wear gradually intensifies, leading to a continuous deterioration in the surface finish of the hole walls. This deterioration further affects the consistency of hole position accuracy across different board layers, resulting in decreased processing quality stability. This chain reaction from surface quality to positional accuracy makes relying solely on traditional experience or fixed-cycle drill bit replacement unsuitable for complex production environments and unable to accurately determine the optimal time for drill bit replacement, thus increasing the risk of production defects. Summary of the Invention

[0003] This invention provides a method for predicting and evaluating the production quality of printed circuit boards based on big data, mainly including:

[0004] By collecting historical data and real-time operating status of drill bits during circuit board processing, an initial dataset is constructed, and drill bit wear characteristics are extracted.

[0005] The correlation mapping relationship between the characteristics of drill bit wear is analyzed, and a multi-dimensional quality influence matrix including hole wall smoothness score, hole position accuracy deviation and inter-layer hole position alignment is constructed. The transmission path of drill bit wear depth indirectly affecting hole position accuracy deviation through hole wall smoothness score is analyzed, and a comprehensive quality influence assessment result is generated.

[0006] By dynamically adjusting the parameter weights of hole wall smoothness score and hole position accuracy deviation in the multi-dimensional quality influence matrix based on real-time monitoring data, the risk level of circuit board processing quality stability can be determined.

[0007] Based on the risk level of stable circuit board processing quality, the potential deterioration probability of the rate of change of through hole resistance due to the accumulation of residual material in the hole under the current drill bit wear depth and cutting edge dulling degree is analyzed, and the predicted value of the impact on through hole quality is obtained.

[0008] Based on the predicted value of the impact of through hole quality, a replacement timing optimization decision rule is set, which includes a threshold for hole wall coating integrity. When the predicted value of the impact of through hole quality is lower than the preset hole wall coating integrity standard, a suggested signal for replacing the drill bit is generated.

[0009] By collecting current production data and combining it with the recommended drill bit replacement signal, the optimal replacement time is determined to replace the drill bit immediately after the current production batch is completed.

[0010] Based on the optimal replacement time of the drill bit, the processing data of the newly replaced drill bit is collected to generate updated prediction results of the impact on the quality of the through hole;

[0011] The latest PCB quality assessment data is generated based on the updated via quality impact prediction results.

[0012] Furthermore, by collecting historical data and real-time operating status of the drill bit during circuit board processing, an initial dataset is constructed, and drill bit wear characteristics are extracted, including:

[0013] Data on the number of uses, wear depth, vibration frequency, cutting force, and temperature during drill bit processing are collected. By setting abnormal thresholds for vibration frequency and fluctuation ranges for cutting force, abnormal data points exceeding the normal range are eliminated. An initial dataset containing the number of uses, wear depth, vibration frequency, cutting force, and temperature is constructed. The correlation coefficient between the rate of change of vibration frequency and the degree of edge dulling is calculated. The relationship between the temperature rise and the coating wear thickness is analyzed. The rate of change of vibration frequency and the rate of temperature rise are extracted as core features of drill bit wear. Combined with material crack length data, a comprehensive wear index is calculated using a weighted summation method. The correspondence between the comprehensive wear index and the number of uses is established, and the slope coefficient is extracted as a wear rate feature to generate drill bit wear characteristics.

[0014] Furthermore, the analysis examines the correlation mapping between drill bit wear characteristics, constructs a multi-dimensional quality influence matrix including hole wall smoothness score, hole position accuracy deviation, and inter-layer hole position alignment, analyzes the transmission path of drill bit wear depth indirectly affecting hole position accuracy deviation through hole wall smoothness score, and generates a comprehensive quality influence assessment result, including:

[0015] The core feature data of the drill bit wear degree is obtained and input into a pre-established random forest model. The information gain ratio of each feature in the decision tree splitting process is calculated, and the influence weight value of the core feature of the drill bit wear degree on the hole wall smoothness score is output. The hole wall smoothness score, hole position accuracy deviation value and inter-layer hole position alignment data after actual processing are collected. A three-dimensional quality influence matrix is ​​constructed by normalization, with rows representing the core feature of the drill bit wear degree and columns representing the quality indicators. The Pearson correlation coefficient between the core feature of the drill bit wear degree and the hole wall smoothness score, and the Pearson correlation coefficient between the hole wall smoothness score and the hole position accuracy deviation are calculated to generate a comprehensive quality influence assessment result.

[0016] Furthermore, the step of dynamically adjusting the parameter weights of the hole wall smoothness score and hole position accuracy deviation in the multi-dimensional quality influence matrix by combining real-time monitoring data includes:

[0017] The multi-dimensional quality influence matrix and the real-time monitored drill bit vibration frequency change value and temperature rise amplitude are obtained. The ratio of the drill bit vibration frequency change value to the preset vibration threshold is calculated as the vibration influence factor, and the ratio of the temperature rise amplitude to the preset temperature threshold is calculated as the temperature influence factor. The vibration influence factor and the temperature influence factor are used to adjust the hole wall smoothness score weight and hole position accuracy deviation weight in the multi-dimensional quality influence matrix.

[0018] Furthermore, based on the risk level of stable circuit board processing quality, the potential deterioration probability of the rate of change of through-hole resistance due to chip accumulation in the hole under the current drill bit wear depth and cutting edge dulling degree is analyzed to obtain the predicted value of the impact on through-hole quality, including:

[0019] The core feature data of the risk level of the circuit board processing quality stability and the wear degree of the drill bit are obtained, input into a pre-established gradient boosting tree model, and output the probability value of the through-hole resistance change caused by the accumulation of debris in the hole; based on the probability value, combined with the statistical relationship between the amount of debris accumulation and the resistance change rate in historical data, the expected change rate of the through-hole resistance is calculated, and the comparison result of the expected change rate with the preset quality deterioration threshold is judged to determine the predicted value of the through-hole quality impact.

[0020] Furthermore, based on the predicted value of the impact of through-hole quality, a replacement timing optimization decision rule is set, which includes a threshold for hole wall coating integrity. When the predicted value of the impact of through-hole quality is lower than the preset hole wall coating integrity standard, a suggested signal for replacing the drill bit is generated, including:

[0021] Obtain the predicted value of the impact on the quality of the through hole and the preset threshold for the integrity of the hole wall coating. Compare the predicted value of the impact on the quality of the through hole with the threshold for the integrity of the hole wall coating to determine whether the replacement conditions are met. Based on the determination result, generate a drill bit replacement suggestion signal.

[0022] Furthermore, the step of collecting current production data through the drill bit replacement suggestion signal, obtaining and combining the predicted time when the drill bit reaches its lifespan milestone based on the cumulative number of uses, and determining the optimal replacement time for the drill bit immediately after the completion of the current production batch includes:

[0023] The data acquisition is triggered by the suggested drill bit replacement signal to obtain the remaining processing volume of the current production batch, the drilling accuracy requirement value of the next batch, and the drill bit inventory turnover efficiency data. The time required to complete the current production batch is calculated, and combined with the drill bit inventory turnover efficiency, the best time to replace the drill bit immediately after the current production batch is completed is determined.

[0024] Furthermore, the step of collecting machining data of the newly replaced drill bit based on the optimal replacement time to generate updated through-hole quality impact prediction results includes:

[0025] After replacing the drill bit at the optimal replacement time, the processing data of the new drill bit is collected. The processing data is matched with the through-hole resistance measurement value of the corresponding time period to construct a new training dataset. The new training dataset is input into the gradient boosting tree model after updating the model parameters to calculate the updated through-hole quality influence prediction result.

[0026] Furthermore, the step of generating the latest circuit board quality assessment data based on the updated via quality impact prediction results includes:

[0027] Based on the updated through-hole quality impact prediction results, the sampling frequency of drill bit vibration frequency change and hole debris accumulation is determined, drill bit temperature rise and hole wall burn trace data are acquired in real time, temperature adjustment factor and burn ratio are calculated, comprehensive adjustment coefficient is obtained, and the latest circuit board quality assessment data is generated.

[0028] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects:

[0029] This invention discloses a big data-based method for predicting and evaluating the production quality of printed circuit boards (PCBs). Addressing the business problem of changes in hole wall smoothness, hole position accuracy, and through-hole resistance caused by drill bit wear during PCB drilling, this invention achieves stable quality and precise control over drill bit replacement timing through a data-driven and model-optimized fusion approach. By collecting multi-dimensional drill bit operational data and combining it with a random forest model to analyze the influence weight of wear characteristics on hole wall smoothness, a multi-dimensional quality influence matrix is ​​constructed, revealing the transmission path of wear depth indirectly affecting hole position accuracy through smoothness. A gradient boosting tree model is used to predict the probability of through-hole resistance degradation, dynamically adjusting monitoring parameters and optimizing replacement decision rules. Real-time data and model updates ensure adaptive matching between quality assessment and the production environment. This invention significantly improves drilling quality stability, reduces the processing defect rate caused by drill bit wear, optimizes drill bit replacement timing and inventory efficiency, and achieves efficient and intelligent PCB processing quality management. Attached Figure Description

[0030] Figure 1 This is a flowchart of a circuit board production quality prediction and evaluation method based on big data according to the present invention. Detailed Implementation

[0031] To further understand the content of this invention, a detailed description of the invention is provided in conjunction with the accompanying drawings and embodiments. The specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention. It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0032] like Figure 1 This embodiment of a big data-based method for predicting and evaluating the production quality of printed circuit boards may specifically include:

[0033] Step S101: By collecting historical data and real-time operating status of the drill bit during the circuit board processing, an initial dataset is constructed, and the wear degree characteristics of the drill bit are extracted.

[0034] Data on the number of uses, wear depth, vibration frequency, cutting force, temperature, coating thickness, crack distribution, and cutting edge condition during drill bit processing were collected. By setting abnormal vibration frequency thresholds and cutting force fluctuation ranges, abnormal data points exceeding the normal range were removed. The cleaned data were arranged in a time series to construct an initial dataset containing eight types of wear parameters. The correlation between the vibration frequency change rate and the cutting edge passivation degree in the initial dataset was calculated to obtain the vibration-passivation correlation coefficient. Simultaneously, the relationship between temperature rise and coating wear thickness was analyzed. If the temperature rise exceeded a preset threshold, the coating was judged to have entered a rapid wear stage, resulting in a temperature-coating wear mapping table. The vibration frequency change rate and temperature rise rate were extracted as the main characteristics of wear degree. Based on these two main characteristics, combined with material crack length data obtained from ultrasonic testing, a comprehensive wear index was calculated using a weighted summation method. The vibration feature had a weight of 0.4, the temperature feature a weight of 0.3, and the crack feature a weight of 0.3, resulting in a quantitative assessment value reflecting the overall wear degree of the drill bit. The relationship between the comprehensive wear index and the number of uses can be used to:

[0035] WI p =a·N+b

[0036] WIp represents the predicted wear index value, N represents the number of times the drill bit has been used, a represents the slope coefficient obtained from linear regression, and b represents the intercept term of linear regression. This formula establishes a linear predictive relationship between the wear index and the number of times the drill bit has been used, and can be used to predict the wear degree of the drill bit under a specific number of uses. The slope coefficient is extracted as the wear rate feature, and combined with the comprehensive wear index, wear rate and the standardized values ​​of each individual wear parameter, the wear degree feature of the drill bit is formed.

[0037] Specifically, the acquisition of eight types of wear parameters requires the coordinated operation of different types of sensors. Vibration frequency data is acquired using an accelerometer mounted on the drill spindle, with a sampling frequency typically set above 1000Hz to capture high-frequency vibration signals during drill cutting. Temperature data is monitored in real-time using an infrared thermometer, with the measurement point chosen near the drill cutting edge, as temperature changes at this location best reflect the wear condition. Coating thickness is measured using optical inspection equipment, and the remaining coating thickness is calculated based on the principle of light reflection.

[0038] In one possible implementation, the normal range of vibration frequency is determined based on the drill bit type and the material being machined. For example, when machining FR-4 circuit boards, the normal vibration frequency range is between 200-800Hz, and data points outside this range are marked as abnormal. The determination of the cutting force fluctuation range is based on statistical analysis of historical stable machining data, typically using the average value plus or minus three times the standard deviation as the boundary of the normal range. This data cleaning method effectively removes erroneous data caused by sensor failure or machining anomalies, improving the accuracy of subsequent analysis.

[0039] Specifically, as the drill bit's cutting edge gradually dulls, the frictional force during the cutting process increases, causing a change in vibration frequency. By calculating the rate of change of vibration frequency, i.e., the increase in frequency per unit time, the degree of cutting edge dulling can be quantified. Meanwhile, the temperature-coating wear mapping table is established based on the principle of heat conduction. After coating wear, the drill bit substrate material directly contacts the workpiece, increasing frictional heat generation and accelerating the temperature rise rate. The weighted calculation of the comprehensive wear index reflects the different contributions of various wear characteristics to the overall wear degree.

[0040] In one embodiment, the vibration feature weight is set to 0.4 because vibration changes most directly reflect the cutting state; the temperature feature weight is 0.3 to reflect the impact of thermal damage on drill bit life; and the crack feature weight is 0.3 to consider the contribution of material fatigue to drill bit failure. This weighting method allows the comprehensive index to fully reflect the actual wear state of the drill bit. The process of establishing a wear prediction function using linear regression is essentially about finding the mathematical relationship between the degree of wear and the number of uses. By fitting a large amount of historical data, the key feature of wear rate can be obtained, which directly reflects the durability of the drill bit under specific machining conditions.

[0041] Step S102: Analyze the correlation mapping relationship between the characteristics of drill bit wear degree, construct a multi-dimensional quality influence matrix including hole wall smoothness score, hole position accuracy deviation and inter-layer hole position alignment, analyze the transmission path of drill bit wear depth indirectly affecting hole position accuracy deviation through hole wall smoothness score, and generate comprehensive quality influence assessment results.

[0042] Three core feature data points—drill bit wear depth, cutting edge dulling degree, and surface coating loss—are acquired and input into a pre-established random forest model. The model calculates the information gain ratio of each feature during the decision tree splitting process and outputs the influence weight values ​​of wear depth, dulling degree, and coating loss on the hole wall finish score, thus obtaining the feature-surface finish influence weight distribution. Based on this distribution, actual hole wall finish scores, hole position accuracy deviations, and inter-layer hole alignment data are collected. Normalization is applied to unify the numerical ranges of each indicator. A three-dimensional quality influence matrix is ​​constructed according to a pre-defined quality indicator ratio, with rows representing wear features, columns representing quality indicators, and numerical values ​​representing the degree of influence. For the data in the three-dimensional quality influence matrix, the Pearson correlation coefficients between wear depth and hole wall finish score, and between finish score and hole position accuracy deviation, are calculated. The product of these two correlation coefficients determines the transmission path strength value of wear depth affecting accuracy deviation through finish score, establishing a transmission path mapping relationship. Based on the transmission path mapping relationship, a weighted summation method is used to integrate the direct impact value and the indirect impact value through the transmission path to calculate the comprehensive quality impact index. If the index exceeds the preset quality threshold, it is determined that the drill bit wear has seriously affected the processing quality, and the comprehensive quality impact assessment result including the degree of impact is output.

[0043] Specifically, the application principle of the random forest model in analyzing the influence of drill bit wear characteristics on weights deserves further exploration.

[0044] It should be noted that this model consists of multiple decision trees, and the information gain is calculated when each tree splits a node. Information gain reflects the ability of a feature to distinguish the target variable. When the wear depth feature can effectively distinguish between samples with high and low surface finish, its information gain will be relatively large. The model determines the influence weight of each wear feature by statistically analyzing the average information gain ratio of each feature across all decision trees.

[0045] In one possible implementation, the construction process of the three-dimensional quality influence matrix embodies the idea of ​​multi-dimensional quality assessment. The rows of the matrix represent different wear characteristic states, such as light wear, moderate wear, and heavy wear; the columns represent three quality indicators: hole wall smoothness score, hole position accuracy deviation, and inter-layer alignment; the numerical values ​​in the matrix represent the degree of influence of the corresponding quality indicator under a specific wear state. Normalization allows quality indicators with different dimensions to be compared on the same scale. For example, the smoothness score, originally in the range of 1-10, and the accuracy deviation, in the range of 0.01-0.1 mm, are uniformly mapped to the 0-1 interval through normalization. The calculation process of the Pearson correlation coefficient reveals the strength of the linear relationship between the variables.

[0046] Specifically, the correlation coefficient is obtained by calculating the covariance of the wear depth sequence and the surface finish score sequence, and then dividing by the product of the standard deviations of the two sequences. As wear depth increases, the surface finish score decreases, showing a negative correlation. The transmission path strength value is obtained by multiplying the two correlation coefficients; this calculation method is based on path analysis theory in statistics and reflects the cumulative effect of indirect influences. The weighted summation process of the comprehensive quality impact index integrates multiple influencing factors.

[0047] For example, the wear depth of a drill bit directly leads to an increase in hole wall roughness—this is a direct impact. Simultaneously, the rough hole wall causes the drill bit to shift during subsequent machining, resulting in decreased hole position accuracy—this is an indirect impact. In the weighted summation, direct impacts are assigned a higher weight, while indirect impacts are assigned a corresponding weight after adjustment through the transmission path coefficient. When the comprehensive index exceeds a preset threshold, it indicates that the drill bit wear has reached a point where replacement is necessary. This judgment mechanism helps prevent batch quality problems caused by excessive drill bit wear and improves the yield rate of circuit board production.

[0048] Step S103: Combine real-time monitoring data to dynamically adjust the parameter weights of hole wall smoothness score and hole position accuracy deviation in the multi-dimensional quality influence matrix. If the change in drill bit vibration frequency or the increase in temperature exceeds the preset threshold range, an abnormal warning signal is triggered to determine the risk level of circuit board processing quality stability. If it does not exceed the preset threshold range, the current quality assessment status is maintained and normal circuit board drilling operations continue.

[0049] The system acquires comprehensive quality impact assessment results and real-time monitoring data on drill bit vibration frequency changes and temperature rise. It calculates the ratio of the vibration frequency change to a preset vibration threshold as a vibration impact factor, and the ratio of the temperature rise to a preset temperature threshold as a temperature impact factor. By multiplying the vibration impact factor by a vibration adjustment coefficient and then adding the temperature impact factor multiplied by a temperature adjustment coefficient, the system dynamically adjusts the weights of hole wall smoothness and hole position accuracy deviation in the multi-dimensional quality impact matrix. Based on the adjusted weights and real-time monitoring data, it determines whether the drill bit vibration frequency change or temperature rise exceeds its respective preset threshold. If both the vibration frequency and temperature exceed the threshold, the risk level is determined to be high; if only one exceeds the threshold, it is determined to be medium; and if both are close to but do not exceed the threshold, it is determined to be low. An abnormal warning signal containing the risk level, the exceeding parameter, and the occurrence time is generated. Based on the determined risk level, corresponding actions are taken. If the risk level is high, an abnormal warning signal is triggered and drilling operations are suspended. If it is medium, a warning signal is triggered but operations continue while monitoring frequency is increased. If it is low or all parameters are within the threshold range, the current quality assessment status is maintained and normal drilling operations continue. The risk level assessment result of the stability of the circuit board processing quality is output.

[0050] Specifically, the core of the dynamic weight adjustment mechanism lies in real-time response to changes in drill bit status.

[0051] Specifically, the vibration influence factor and temperature influence factor are calculated based on the ratio principle. When the measured vibration frequency change is 800Hz and the preset threshold is 1000Hz, the vibration influence factor is 0.8. This value reflects the proportion of the current vibration state to the limit state. The vibration adjustment coefficient and temperature adjustment coefficient are empirical values ​​derived from historical data. Generally, vibration has a more direct impact on the surface finish of the hole wall; therefore, the vibration adjustment coefficient is set to 0.6, and the temperature adjustment coefficient is 0.4.

[0052] In one possible implementation, the weight adjustment calculation process embodies the idea of ​​weighted superposition. Assuming the original hole wall smoothness score weight is 0.5, when the vibration influence factor is 0.8 and the temperature influence factor is 0.7, the adjustment amount is calculated as 0.8 × 0.6 + 0.7 × 0.4 = 0.76. This adjustment amount, multiplied by the adjustment amplitude coefficient, is used to correct the original weight. The adjusted weight directly affects the sensitivity of the quality assessment; an increased weight means that the indicator's importance in the comprehensive assessment is enhanced. The three-level risk classification reflects the concept of hierarchical management.

[0053] It should be noted that high-risk situations correspond to simultaneous exceedances of vibration and temperature limits. This dual anomaly indicates that the drill bit is in a state of severe wear, and continued use will lead to batch quality accidents. Medium-risk situations indicate an exceedance of a single parameter; for example, excessive vibration alone may indicate localized chipping of the cutting edge, while excessive temperature alone may indicate localized coating peeling. In such cases, the drill bit can still be used for a short period but requires close monitoring. Low-risk situations involve parameters approaching but not exceeding the threshold, indicating that the drill bit is beginning to show signs of wear. The generation of abnormal warning signals follows a structured principle. Warning signals contain three key elements: the risk level to guide subsequent handling strategies, the type of parameter exceeding the limit to help maintenance personnel quickly locate the root cause of the problem, and the time of occurrence for tracing quality issues and statistical analysis. This structured information enables the production management system to automatically trigger corresponding handling processes. The design of differentiated handling strategies considers the balance between production continuity and quality assurance. High-risk situations trigger immediate shutdown to avoid irreparable quality losses; medium-risk situations allow continued operation but with enhanced monitoring because the quality risk is controllable in the short term; low-risk situations and normal conditions maintain routine operations to avoid excessive intervention affecting production efficiency.

[0054] Step S104: Based on the risk level of stable circuit board processing quality, analyze the potential deterioration probability of the through hole resistance change rate caused by the accumulation of debris in the hole under the current drill bit wear depth and cutting edge dulling degree, and obtain the predicted value of the through hole quality impact.

[0055] Based on the risk level of stable PCB processing quality, current drill bit wear depth and cutting edge dulling data are obtained. The risk level, wear depth, and dulling degree are used as input features and fed into a pre-established gradient boosting tree model. The model outputs the probability value of through-hole resistance change caused by debris accumulation under the current wear state. Based on this probability value, and combined with the statistical relationship between debris accumulation and resistance change rate in historical data, the expected change rate of through-hole resistance is calculated by multiplying the probability value by the mean resistance change rate. It is then determined whether this change rate exceeds a preset quality degradation threshold. Comparing the expected change rate of through-hole resistance with the quality degradation threshold, if the threshold is exceeded, the expected change rate is used as the predicted value of the through-hole quality impact; otherwise, half of the threshold is used as the predicted value. This yields a quantitative prediction result reflecting the potential impact of drill bit wear on through-hole quality.

[0056] Specifically, the application of gradient boosting tree models in predicting via quality degradation demonstrates the advantages of ensemble learning.

[0057] It should be noted that this model constructs a decision tree sequence through multiple iterations, with each new tree fitting the residuals of the previous prediction. When dealing with the relationship between drill bit wear and through-hole quality, the model can capture complex nonlinear mappings. Risk level, as a discrete feature, reflects the overall processing status; wear depth and passivation degree, as continuous features, provide a quantitative description of the drill bit's physical state. The combination of these three features allows the model to assess through-hole quality risk from different dimensions. The mechanism by which chip accumulation affects through-hole resistance warrants further analysis.

[0058] Specifically, as drill bit wear intensifies, cutting efficiency decreases, causing metal chips to accumulate inside the hole due to the inability to be discharged in time. These chips result in uneven plating during subsequent electroplating, directly affecting the conductivity of the through-hole. Statistical analysis of historical data shows that for every unit increase in chip accumulation, the through-hole resistance increases by an average of about 15%. This statistical relationship provides an empirical basis for probabilistic prediction.

[0059] In one possible implementation, the calculation of the probability value and the rate of change of resistance follows the principle of conditional probability. Assuming the model outputs a degradation probability of 0.7, and the average rate of change of resistance corresponding to this probability in historical data is 25%, then the expected rate of change is calculated as 0.7 × 25% = 17.5%. This value reflects the most likely range of change in the through-hole resistance under the current drill bit condition. The quality degradation threshold is typically set at 20%, a critical value determined based on industry standards and product reliability requirements. The dual-processing mechanism for threshold comparison reflects the concept of risk-level management. When the expected rate of change exceeds the threshold, it is directly used as the predicted value, indicating that the through-hole quality is in a high-risk state and immediate action is required. When it does not exceed the threshold, half of the threshold is used as the predicted value; this conservative estimation ensures that even in a relatively safe state, a moderate level of vigilance is maintained.

[0060] For example, if the threshold is 20% and the expected rate of change is 15%, the final predicted value is 10%, which reflects the existing risk without causing an overreaction. The value of this predictive mechanism lies in enabling preventative maintenance. By quantifying the potential impact of through-hole quality, production managers can intervene before quality problems actually occur, such as adjusting drill changeover cycles or optimizing processing parameters. This not only reduces the defect rate but also avoids the scrapping of entire batches of circuit boards due to through-hole quality issues, significantly improving production efficiency.

[0061] Step S105: Based on the predicted value of the impact of the through hole quality, set a replacement timing optimization decision rule that includes a threshold for hole wall coating integrity. When the predicted value of the impact of the through hole quality is lower than the preset hole wall coating integrity standard, generate a recommendation signal to replace the drill bit.

[0062] For the predicted value affecting the quality of the through hole, a preset threshold for the integrity of the hole wall coating is obtained. The predicted value is compared with the threshold. If the predicted value is lower than the coating integrity threshold, it is determined that the current drill bit wear has affected the quality of the hole wall coating, and the replacement condition is met. Based on the determination result of meeting the replacement condition, if the replacement condition is met, a drill bit replacement suggestion signal is generated, which includes the current predicted value, the threshold difference, and a replacement identifier. If the replacement condition is not met, no suggestion signal is generated, and the monitoring of changes in the through hole quality continues.

[0063] Specifically, the principle for setting the hole wall coating integrity threshold is based on the quality standards of circuit board manufacturing.

[0064] It should be noted that this threshold reflects the minimum requirement for copper layer coverage on the hole wall, typically expressed as a rate of change in resistance. When drill bit wear increases the roughness of the hole wall, uneven copper layer adhesion occurs during subsequent electroplating processes, creating localized weak points. These weak points lead to increased via resistance, affecting signal transmission quality. Industry standards typically use a resistance change rate of 20% as a critical value; values ​​below this indicate compromised coating integrity.

[0065] In one possible implementation, the comparison between predicted values ​​and thresholds embodies the concept of preventative maintenance. Assuming the predicted impact of through-hole quality is 15%, and the coating integrity threshold is set at 20%, this means the predicted quality level of the current drill bit condition is only 75% of the standard requirement. This quantitative comparison provides clear data support for replacement decisions, avoiding the uncertainty caused by subjective judgment. The calculation of the threshold difference provides a quantitative indicator of the degree of wear.

[0066] Specifically, the difference between the threshold and the predicted value directly reflects the degree of deviation of the current quality status from the standard. The larger the difference, the more severe the drill bit wear, and the higher the urgency of replacement. This difference information is included in the recommendation signal, providing a basis for prioritization in production scheduling. The structured design of the replacement recommendation signal ensures the integrity and operability of the information. The signal contains three key elements: the current predicted value provides a real-time snapshot of the quality status; the threshold difference quantifies the degree of deviation; and the replacement identifier is a clear execution instruction. This structure enables downstream systems to automatically parse the signal and trigger corresponding processing flows, such as automatically generating work orders or adjusting production plans. The dual-path processing mechanism for conditional judgment reflects the flexibility of the system. When the replacement conditions are met, a recommendation signal is generated immediately, ensuring that quality risks are responded to in a timely manner. When the conditions are not met, the system continues to monitor without generating a signal, avoiding frequent false alarms that affect normal production. This design balances the needs of quality assurance and production efficiency. The importance of the continuous monitoring mechanism lies in capturing dynamic changes. Drill bit wear is a gradual process, and the quality of the through hole will gradually deteriorate accordingly. Through continuous monitoring, we can remain vigilant when quality indicators are close to but have not yet reached the threshold, and respond immediately once the critical point is crossed.

[0067] Step S106: By collecting current production data through the drill bit replacement suggestion signal, and combining the time prediction of the cumulative number of drill bit uses reaching the life node, determine the best replacement time to replace the drill bit immediately after the current production batch is completed.

[0068] Data acquisition is triggered by a drill bit replacement suggestion signal to obtain the remaining processing volume of the current production batch, the drilling accuracy requirement of the next batch, and the drill bit inventory turnover efficiency. Based on the current number of times the drill bit has been used and its historical wear rate, the estimated number of uses before reaching the wear limit is calculated, and this is divided by the processing volume per unit time to obtain the remaining available time. Based on the remaining available time and the remaining processing volume of the current batch, the time required to complete the current batch is calculated. If the required time is less than the remaining available time, it is determined that the drill bit can support the batch until completion. By comparing the accuracy requirement of the next batch with the current drill bit wear prediction accuracy, the batch completion time is determined as the initial replacement timing. Based on the initial replacement timing and the drill bit inventory turnover efficiency, the preparation time from issuing the replacement instruction to the new drill bit being in place is calculated. The preparation time can be calculated using the following formula:

[0069]

[0070] T prep Indicates the preparation time for drill bit replacement, I current R represents the current quantity of drill bits in stock. turnover T represents the inventory turnover rate. signal T represents the transmission time when the replacement command was issued. position This indicates the time it takes for the new drill bit to be in place; if the preparation time is less than the downtime window between two batches, then replacing the drill bit immediately after the current production batch is completed is the optimal replacement time.

[0071] Specifically, the data acquisition mechanism triggered by the replacement suggestion signal embodies the event-driven design philosophy.

[0072] It's important to note that when the predicted impact of drill bit quality falls below the coating integrity threshold, a suggested signal containing key parameters is generated. This signal is not merely a simple replacement reminder, but a trigger to initiate a comprehensive evaluation process. The predicted value and threshold difference information contained in the signal provide fundamental data support for subsequent timing optimization. The historical wear rate is calculated based on statistical analysis of a large amount of production data.

[0073] Specifically, by recording the relationship between the number of uses and the wear depth of similar drill bits under similar processing conditions, a wear curve can be fitted. Assuming that a certain model of drill bit increases its wear depth by an average of 0.1 mm per 1000 drilling cycles, and the ultimate wear depth is 2 mm, its total lifespan can be calculated to be approximately 20,000 cycles. A drill bit that has already been used 15,000 times has 5,000 usable cycles remaining. Assuming it processes 500 holes per hour, the remaining usable time is 10 hours.

[0074] In one possible implementation, the calculation of batch completion time needs to consider multiple factors. The base time is obtained by dividing the remaining processing volume by the actual production rate, but in actual production, auxiliary times such as material changeover and inspection also need to be considered.

[0075] For example, machining the remaining 2000 holes theoretically requires 4 hours, but with auxiliary time, it actually takes 4.5 hours. This time, compared to the remaining 10 hours of available time, leaves ample margin, indicating that the drill bit can last until the batch is completed. The core of the accuracy requirement comparison lies in the application of predictive models. Based on the current wear state and wear rate, the drill bit accuracy level at the time of batch completion can be predicted. If the next batch requires a hole position accuracy deviation of no more than 0.05 mm, but the prediction shows that the drill bit can only guarantee an accuracy of 0.08 mm when the batch is completed, then it must be replaced between batches. This predictive judgment avoids batch rework due to insufficient drill bit accuracy. Inventory turnover efficiency reflects the level of spare parts management. This indicator includes the entire process time from issuing a requisition request to the drill bit arriving at the production line. An efficient inventory management system can control this time within 30 minutes, while the downtime window between batches is typically 1-2 hours. Sufficient time margin ensures that the replacement process does not affect production continuity. The downtime window between two batches is a natural interval in circuit board production. This period is used for necessary work such as cleaning equipment, changing materials, and adjusting parameters. By scheduling drill bit replacements within this window, maintenance activities and production rhythms were perfectly integrated. This arrangement minimized additional downtime losses caused by drill bit replacements, reflecting the principles of lean manufacturing. Through precise calculations and rational scheduling, drill bit replacements were transformed from reactive emergency measures to proactive optimization, significantly improving production efficiency and product quality stability.

[0076] Step S107: Collect the processing data of the newly replaced drill bit according to the optimal replacement time, and generate the updated prediction results of the impact on the quality of the through hole.

[0077] After replacing the drill bit at the optimal time, the cumulative usage data of the new drill bit during the initial processing stage, the surface coating thickness measurement, and the material status information from ultrasonic flaw detection are collected. These three types of data are matched with the through-hole resistance measurement values ​​for the corresponding time periods to construct a new training dataset containing input features and output labels. Based on the new training dataset, the data is input into a gradient boosting tree model for parameter updates. By comparing the model's predicted values ​​with the actual through-hole resistance values, the weight coefficients of each decision tree in the model are adjusted. If the difference exceeds a preset threshold, a new decision tree is added, and the splitting gain values ​​of the features are reallocated, resulting in an updated model parameter set. Based on the updated model parameter set, real-time status data of each drill bit on the production line is obtained, including cumulative usage, remaining coating thickness, and crack detection results. This data is input into the updated gradient boosting tree model to calculate the updated through-hole quality impact prediction results.

[0078] Specifically, the importance of initial data acquisition for new drill bits lies in establishing performance benchmarks.

[0079] It should be noted that although the new drill bit has not yet shown obvious wear, its initial condition data is crucial for model optimization. The initial coating thickness is typically in the range of 0.05-0.08 mm, serving as a baseline for subsequent wear calculations. Material condition information is obtained through ultrasonic testing; the test results for the new drill bit should show a uniform reflection waveform with no abnormal signal points. These initial data, along with the resistance values ​​of the machined through-holes, form a control group, providing a standard reference for the model under a "zero wear" state.

[0080] In one possible implementation, the construction of the training dataset requires precise time matching. Each drilling operation is timestamped, and the corresponding through-hole resistance measurement is performed within 30 seconds of drilling completion. This tight temporal correlation ensures a causal correspondence between input features and output labels.

[0081] For example, drill bit status data used at 10:00 AM must be paired with through-hole quality data completed at 10:00 AM to avoid model training bias caused by data mismatch. The parameter update mechanism of the gradient boosting tree model is based on the principle of error backpropagation.

[0082] Specifically, when the model predicts a through-hole resistance of 50 ohms under a certain drill bit condition, while the actual measured value is 55 ohms, a prediction error of 5 ohms will trigger parameter adjustment. The decision tree weight coefficients in the model will be corrected according to the magnitude of the error; the weight of decision trees with accurate predictions will increase, while the weight of decision trees with large prediction deviations will decrease. This dynamic adjustment allows the model to gradually adapt to new production conditions. The redistribution of feature split gain values ​​reflects changes in feature importance. As production conditions evolve, the degree of influence of different wear features on through-hole quality will change. Assuming that the original split gain value for coating loss was 0.3 and the number of uses was 0.5, after training with new data, it may be adjusted to 0.4 for coating loss and 0.45 for the number of uses. This adjustment reflects that under the current production environment, the impact of coating condition on quality is relatively enhanced. The acquisition of real-time status data reflects the monitoring capability of the entire production line. Multiple drill bits may be running simultaneously on the production line, each drill bit being at a different wear stage. Through a sensor network, the system can collect the status parameters of all drill bits in use. These parameters are input into the updated model to obtain the corresponding quality prediction value for each drill bit. This parallel prediction mechanism enables production managers to have a comprehensive understanding of the distribution of quality risks and to identify potentially problematic drill bits in advance. Continuous model optimization forms a closed-loop improvement mechanism. Each drill bit change provides the model with new learning samples, and the model's prediction accuracy continuously improves with the accumulation of data.

[0083] Step S108: Based on the updated through-hole quality impact prediction results, adjust the sampling frequency of drill bit vibration frequency change and hole debris accumulation, as well as the analysis depth of hole wall smoothness score, to obtain the latest circuit board quality assessment data.

[0084] Based on the updated through-hole quality impact prediction results, the predicted values ​​are divided into three risk ranges: high, medium, and low. High risk corresponds to acquiring vibration data once per second and detecting chip status once per minute; medium risk corresponds to acquiring data every five seconds and every five minutes; and low risk corresponds to acquiring data every ten seconds and every ten minutes. Simultaneously, the surface finish rating is expanded from a single dimension of surface roughness to a multi-dimensional analysis including texture depth and uniformity. Based on the determined acquisition frequency and analysis dimensions, real-time data on drill bit temperature rise and borehole wall burn marks are acquired. The percentage difference between the temperature rise and the average of the previous hour is calculated as a temperature adjustment factor. Image processing identifies the proportion of burned area pixels to the total number of pixels on the borehole wall. The temperature adjustment factor is multiplied by the burn proportion to obtain a comprehensive adjustment coefficient. The temperature and burn threshold are dynamically adjusted by multiplying the comprehensive adjustment coefficient by the original threshold. If the measured temperature rise or the proportion of burn area exceeds the dynamic threshold, the current temperature, burn, vibration and debris data are substituted into the weighted summation formula, and the values ​​of each element of the quality influence matrix are recalculated with a weight of 0.3 for temperature, 0.3 for burn, 0.2 for vibration and 0.2 for debris, to obtain the latest circuit board quality assessment data.

[0085] Specifically, the risk-based data collection frequency adjustment mechanism embodies the concept of optimal resource allocation.

[0086] It's important to note that a high-risk state means the drill bit wear is nearing its critical point. At this stage, acquiring vibration data every second can capture subtle abnormal fluctuations. Sudden changes in vibration frequency often indicate impending cutting edge breakage, and frequent data acquisition ensures no critical signals are missed. In contrast, the drill bit performs stably under low-risk conditions, and a data acquisition interval of every ten seconds meets monitoring requirements while avoiding data redundancy and processing burden.

[0087] In one possible implementation, the multi-dimensional expansion of surface finish scoring significantly enhances the comprehensiveness of quality assessment. Surface roughness reflects the average undulation of the hole wall, typically represented by the Ra value; texture depth focuses on the maximum peak-to-valley difference, reflecting the severity of local defects; and uniformity assesses the consistency of the overall hole wall surface quality. These three dimensions complement each other, jointly constructing a three-dimensional quality profile.

[0088] For example, if the Ra value of a hole wall is acceptable but the texture depth exceeds the standard, it indicates the presence of individual deep scratches. Such defects may be masked by the average value in a single roughness assessment. The temperature regulation factor is calculated based on the principle of the coefficient of variation in statistics.

[0089] Specifically, the average temperature of the previous hour represents a stable production state, while the deviation of the current temperature from the average reflects the degree of abnormality. Assuming the average temperature of the previous hour was 85 degrees Celsius, and the current detected temperature is 95 degrees Celsius, the difference of 10 degrees represents 11.8% of the average; this percentage is the temperature adjustment factor. This relative value calculation method eliminates the influence of different reference temperatures, making the adjustment more precise. The image recognition process for burn marks utilizes computer vision technology. The system first converts the hole wall image into a grayscale image, and then identifies burn areas with abnormal color through threshold segmentation. Burned areas are usually dark brown or black, forming a sharp contrast with the normal copper color. The pixel statistics method is simple and effective; for example, if the total number of pixels is 10,000 and the burned area occupies 500 pixels, then the burn proportion is 5%. This value intuitively reflects the severity of the burn. The multiplication operation of the comprehensive adjustment coefficient reflects the cumulative effect of risk factors. Temperature anomalies and burns often occur together, and their product reflects the composite risk level. When the temperature adjustment factor is 0.118 and the burn ratio is 0.05, the overall adjustment coefficient is 0.0059. This coefficient is used to dynamically adjust the monitoring threshold, achieving adaptive control. The weighted summation quality influence matrix reconstruction process considers the relative importance of each factor. Temperature and burn each account for 0.3 weights because they directly affect the physicochemical properties of the hole wall metal; vibration accounts for 0.2, reflecting the mechanical state; and residue accounts for 0.2, affecting the subsequent electroplating quality. This weight allocation is based on correlation analysis of a large amount of production data, ensuring the scientific validity and practicality of the evaluation results.

[0090] The above description is merely a specific implementation of this specification. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the scope of protection of this specification is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this specification, and these modifications or substitutions should all be covered within the scope of protection of this specification.

Claims

1. A method for predicting and evaluating the production quality of printed circuit boards based on big data, characterized in that, The method includes: By collecting historical data and real-time operating status of drill bits during circuit board processing, an initial dataset is constructed, and drill bit wear characteristics are extracted. The correlation mapping relationship between the characteristics of drill bit wear is analyzed, and a multi-dimensional quality influence matrix including hole wall smoothness score, hole position accuracy deviation and inter-layer hole position alignment is constructed. The transmission path of drill bit wear depth indirectly affecting hole position accuracy deviation through hole wall smoothness score is analyzed, and a comprehensive quality influence assessment result is generated. By dynamically adjusting the parameter weights of hole wall smoothness score and hole position accuracy deviation in the multi-dimensional quality influence matrix based on real-time monitoring data, the risk level of circuit board processing quality stability can be determined. Based on the risk level of stable circuit board processing quality, the potential deterioration probability of the rate of change of through hole resistance due to the accumulation of residual material in the hole under the current drill bit wear depth and cutting edge dulling degree is analyzed, and the predicted value of the impact on through hole quality is obtained. Based on the predicted value of the impact of through hole quality, a replacement timing optimization decision rule is set, which includes a threshold for hole wall coating integrity. When the predicted value of the impact of through hole quality is lower than the preset hole wall coating integrity standard, a suggested signal for replacing the drill bit is generated. By collecting current production data and combining it with the recommended drill bit replacement signal, the optimal replacement time is determined to replace the drill bit immediately after the current production batch is completed. Based on the optimal replacement time of the drill bit, the processing data of the newly replaced drill bit is collected to generate updated prediction results of the impact on the quality of the through hole; The latest PCB quality assessment data is generated based on the updated via quality impact prediction results.

2. The method for predicting and evaluating the production quality of printed circuit boards based on big data according to claim 1, characterized in that, The process involves collecting historical data and real-time operating status of the drill bit during circuit board processing to construct an initial dataset and extracting drill bit wear characteristics, including: Data on the number of uses, wear depth, vibration frequency, cutting force, and temperature during drill bit processing are collected. By setting abnormal thresholds for vibration frequency and fluctuation ranges for cutting force, abnormal data points exceeding the normal range are eliminated. An initial dataset containing the number of uses, wear depth, vibration frequency, cutting force, and temperature is constructed. The correlation coefficient between the rate of change of vibration frequency and the degree of edge dulling is calculated. The relationship between the temperature rise and the coating wear thickness is analyzed. The rate of change of vibration frequency and the rate of temperature rise are extracted as core features of drill bit wear. Combined with material crack length data, a comprehensive wear index is calculated using a weighted summation method. The correspondence between the comprehensive wear index and the number of uses is established, and the slope coefficient is extracted as a wear rate feature to generate drill bit wear characteristics.

3. The method for predicting and evaluating the production quality of printed circuit boards based on big data according to claim 1, characterized in that, The analysis examines the correlation mapping between drill bit wear characteristics, constructs a multi-dimensional quality influence matrix including hole wall smoothness score, hole position accuracy deviation, and inter-layer hole alignment, analyzes the transmission path of drill bit wear depth indirectly affecting hole position accuracy deviation through hole wall smoothness score, and generates a comprehensive quality influence assessment result, including: The core feature data of the drill bit wear degree is obtained and input into a pre-established random forest model. The information gain ratio of each feature in the decision tree splitting process is calculated, and the influence weight value of the core feature of the drill bit wear degree on the hole wall smoothness score is output. The hole wall smoothness score, hole position accuracy deviation value and inter-layer hole position alignment data after actual processing are collected. A three-dimensional quality influence matrix is ​​constructed by normalization, with rows representing the core feature of the drill bit wear degree and columns representing the quality indicators. The Pearson correlation coefficient between the core feature of the drill bit wear degree and the hole wall smoothness score, and the Pearson correlation coefficient between the hole wall smoothness score and the hole position accuracy deviation are calculated to generate a comprehensive quality influence assessment result.

4. The method for predicting and evaluating the production quality of printed circuit boards based on big data according to claim 1, characterized in that, The method of dynamically adjusting the parameter weights of hole wall smoothness score and hole position accuracy deviation in the multi-dimensional quality influence matrix by combining real-time monitoring data includes: The multi-dimensional quality influence matrix and the real-time monitored drill bit vibration frequency change value and temperature rise amplitude are obtained. The ratio of the drill bit vibration frequency change value to the preset vibration threshold is calculated as the vibration influence factor, and the ratio of the temperature rise amplitude to the preset temperature threshold is calculated as the temperature influence factor. The vibration influence factor and the temperature influence factor are used to adjust the hole wall smoothness score weight and hole position accuracy deviation weight in the multi-dimensional quality influence matrix.

5. The method for predicting and evaluating the production quality of printed circuit boards based on big data according to claim 1, characterized in that, Based on the risk level of stable circuit board processing quality, the analysis examines the potential deterioration probability of the through-hole resistance change rate due to chip accumulation under the current drill bit wear depth and cutting edge dulling degree, yielding predicted values ​​for the impact on through-hole quality, including: The core feature data of the risk level of the circuit board processing quality stability and the wear degree of the drill bit are obtained, input into a pre-established gradient boosting tree model, and output the probability value of the through-hole resistance change caused by the accumulation of debris in the hole; based on the probability value, combined with the statistical relationship between the amount of debris accumulation and the resistance change rate in historical data, the expected change rate of the through-hole resistance is calculated, and the comparison result of the expected change rate with the preset quality deterioration threshold is judged to determine the predicted value of the through-hole quality impact.

6. The method for predicting and evaluating the production quality of printed circuit boards based on big data according to claim 1, characterized in that, The predicted value based on the impact of through-hole quality is used to set a replacement timing optimization decision rule that includes a threshold for hole wall coating integrity. When the predicted value of the through-hole quality impact is lower than the preset hole wall coating integrity standard, a suggested signal for replacing the drill bit is generated, including: Obtain the predicted value of the impact on the quality of the through hole and the preset threshold for the integrity of the hole wall coating. Compare the predicted value of the impact on the quality of the through hole with the threshold for the integrity of the hole wall coating to determine whether the replacement conditions are met. Based on the determination result, generate a drill bit replacement suggestion signal.

7. The method for predicting and evaluating the production quality of printed circuit boards based on big data according to claim 1, characterized in that, The process involves collecting current production data, obtaining and combining it with the predicted time when the drill bit reaches its lifespan milestone based on cumulative usage, to determine the optimal replacement time immediately after the current production batch is completed. This includes: The data acquisition is triggered by the suggested drill bit replacement signal to obtain the remaining processing volume of the current production batch, the drilling accuracy requirement value of the next batch, and the drill bit inventory turnover efficiency data. The time required to complete the current production batch is calculated, and combined with the drill bit inventory turnover efficiency, the best time to replace the drill bit immediately after the current production batch is completed is determined.

8. The method for predicting and evaluating the production quality of printed circuit boards based on big data according to claim 1, characterized in that, The process of collecting machining data of the newly replaced drill bit based on the optimal replacement time to generate updated through-hole quality impact prediction results includes: After replacing the drill bit at the optimal replacement time, the processing data of the new drill bit is collected. The processing data is matched with the through-hole resistance measurement value of the corresponding time period to construct a new training dataset. The new training dataset is input into the gradient boosting tree model after updating the model parameters to calculate the updated through-hole quality influence prediction result.

9. The method for predicting and evaluating the production quality of printed circuit boards based on big data according to claim 1, characterized in that, The process of generating the latest circuit board quality assessment data based on the updated via quality impact prediction results includes: Based on the updated through-hole quality impact prediction results, the sampling frequency of drill bit vibration frequency change and hole debris accumulation is determined, drill bit temperature rise and hole wall burn trace data are acquired in real time, temperature adjustment factor and burn ratio are calculated, comprehensive adjustment coefficient is obtained, and the latest circuit board quality assessment data is generated.

Images