A medical film quality detection system based on multi-parameter fusion

Abstract

The present application relates to the technical field of medical film, and particularly relates to a medical film quality detection system based on multi-parameter fusion, which comprises an acquisition module, a discrete determination module, an identification module, a risk determination module, an adjustment module and an early warning module. The present application takes a dynamic discrete coefficient as a trigger condition, analyzes the contribution of each process parameter to the current thickness fluctuation to identify the primary root cause, forms a quantitative process risk index by comprehensively considering the out-of-control degree of the root cause parameter and the thickness deviation, and according to the process risk index, an adjustment instruction with clear direction and strength suggestion can be sent, and the adjustment efficiency is evaluated after adjustment, early warning is given when the control loop fails or the problem escalates, and the problems of lagging quality defect discovery and blind and lagging process adjustment caused by the failure to quickly and accurately locate the process abnormal root cause in the complex multi-process coupling production process are effectively solved.

Description

Technical Field

[0001] This invention relates to the field of medical film technology, and in particular to a medical film quality inspection system based on multi-parameter fusion. Background Technology

[0002] The uniformity of thickness in medical films is a core quality attribute determining their clinical functions, such as drug release precision, mechanical compatibility, and biosafety. In continuous manufacturing processes, the nonlinear coupling and temporal transmission characteristics of multiple process parameters make thickness fluctuations complex and difficult to identify in real time. Existing quality monitoring technologies mostly rely on offline sampling or single-parameter threshold judgments, which suffer from common bottlenecks such as response lag, ambiguity in root cause tracing under multiple factors, and insufficient regulatory basis, easily leading to batch quality fluctuations, resource waste, and potential clinical risks. The industry urgently needs an intelligent quality inspection method with multi-dimensional process information fusion capabilities and a dynamic root cause analysis mechanism to achieve accurate early warning of quality anomalies and closed-loop optimization of the process, effectively improving the reliability and industrialization level of high-end medical film products.

[0003] Chinese Patent Publication No. CN118459826A discloses a high oxygen barrier and high transparency medical packaging film and its preparation method. The method includes: a film base layer and a coating layer disposed on at least one surface of the film base layer; the coating layer is formed by a coating liquid; the coating liquid includes: polyvinyl alcohol, a polycarboxylic acid pre-crosslinker, and water; the polycarboxylic acid pre-crosslinker is a crosslinking product of a copolymer and a small molecule compound containing active hydrogen groups; the copolymer includes dicarboxylic acid monomer units and acrylate monomer units; the dicarboxylic acid monomer units are selected from monoene unsaturated dicarboxylic acid monomer units and / or monoene unsaturated dicarboxylic acid anhydride monomer units.

[0004] Therefore, the existing technology has the following problems: the method relies on precise chemical ratios and reaction conditions, but lacks real-time process monitoring, and is prone to unstable coating performance due to fluctuations; the method relies on fixed high-temperature heat treatment for curing, which can easily cause thermal damage to the heat-sensitive film substrate; the method lacks multi-parameter closed-loop control from raw materials to finished products, making it difficult to diagnose and adjust the coating quality in real time, which can easily lead to poor batch consistency of products. Summary of the Invention

[0005] To address this, the present invention provides a medical film quality inspection system based on multi-parameter fusion. This system overcomes the problems in existing technologies, such as delayed quality defect detection, low problem-solving efficiency, and blind and delayed process adjustments, caused by the inability to quickly and accurately locate the root causes of process abnormalities in complex multi-process coupled production processes, through multi-parameter time-delay correlation analysis and root cause localization of thickness time-series fluctuations.

[0006] To achieve the above objectives, the present invention provides a medical film quality inspection system based on multi-parameter fusion, comprising:

[0007] The acquisition module is used to acquire the liquid concentration of polyvinyl alcohol casting liquid, the relative humidity difference between the inlet and outlet of the drying box in the drying and curing process, the transmission speed ratio in the molding process, the supply flow rate of polyvinyl alcohol casting liquid, and the average thickness of the film-formed area during the casting process of water-soluble polyvinyl alcohol medical film.

[0008] A discrete determination module is used to determine dynamic discrete coefficients based on the sliding time-series characteristics of the average thickness;

[0009] The identification module is used to identify primary root cause parameters based on the contribution of the liquid concentration, the relative humidity difference, the transmission speed ratio, and the supply flow rate to the dynamic discrete coefficient, respectively.

[0010] The risk determination module is used to determine the process risk index based on the primary root cause parameters and the dynamic dispersion coefficients;

[0011] The adjustment module is used to issue adjustment instructions based on the degree of deviation of the primary root cause parameters, the process risk index, and the preset risk threshold.

[0012] The early warning module is used to issue early warnings based on the threshold comparison results of the adjustment efficiency index and the primary root cause parameters, wherein the adjustment efficiency index is determined based on the frequency of issuing the adjustment command, the dynamic dispersion coefficient, and the changing trend of the process risk index.

[0013] Furthermore, the discrete determination module includes:

[0014] The mean calculation unit is used to determine the thickness sliding mean and thickness sliding fluctuation value based on the sliding mean and sliding standard deviation of the average thickness, respectively.

[0015] A discrete determination unit is used to determine the dynamic discrete coefficients based on the ratio of the thickness sliding fluctuation value to the thickness sliding mean value.

[0016] Furthermore, the identification module includes:

[0017] The discrete determination unit is used to determine the thickness anomaly based on the threshold comparison result of the dynamic discrete coefficient, and to determine the uniformity anomaly time based on the dynamic discrete coefficient corresponding to the thickness anomaly.

[0018] The relevant determination unit determines the concentration correlation, humidity correlation, speed correlation, and flow correlation based on the determination result of thickness anomaly and the correlation between the liquid concentration, relative humidity difference, transmission speed ratio, and supply flow rate before the time of uniform anomaly and the dynamic dispersion coefficient.

[0019] The identification unit is used to identify the primary root cause parameters based on the concentration correlation, the humidity correlation, the velocity correlation, and the flow rate correlation.

[0020] Furthermore, the relevant determination unit includes:

[0021] A concentration determination subunit is used to determine the maximum value of all concentration time-delay correlations as the concentration correlation, wherein the concentration time-delay correlation is determined based on the liquid concentration and the dynamic dispersion coefficient;

[0022] A humidity determination subunit is used to determine the humidity correlation by the maximum value of all humidity time-delay correlations, wherein the humidity time-delay correlation is determined based on the relative humidity difference and the dynamic dispersion coefficient;

[0023] A speed determination subunit is used to determine the maximum value of all speed time delay correlations as the speed correlation, wherein the speed time delay correlation is determined based on the transmission speed ratio and the dynamic discrete coefficient;

[0024] A flow determination subunit is used to determine the maximum value of the time-delay correlation of all flow rates as the flow correlation, wherein the flow correlation is determined based on the supply flow and the dynamic dispersion coefficient.

[0025] Furthermore, the identification unit includes:

[0026] The contribution determination subunit is used to determine the concentration contribution, humidity contribution, velocity contribution, and flow contribution based on the ratios of the concentration correlation, humidity correlation, velocity correlation, and flow correlation to the overall correlation, respectively, wherein the overall correlation is determined based on the sum of the concentration correlation, humidity correlation, velocity correlation, and flow correlation.

[0027] The identification subunit is used to determine the primary root cause parameter as the liquid concentration, the relative humidity difference, the transmission speed ratio, or the supply flow rate based on the threshold comparison results of the concentration contribution, the humidity contribution, the velocity contribution, and the flow rate contribution.

[0028] Furthermore, the risk determination module includes:

[0029] The deviation calculation unit is used to calculate the relative deviation between the dynamic discrete coefficient and the preset discrete threshold to obtain the discrete deviation degree.

[0030] An index calculation unit is used to determine the degree of parameter runaway based on the time-series changes of the primary root cause parameters;

[0031] A risk determination unit is used to determine the process risk index based on the dispersion deviation and the parameter runaway degree.

[0032] Furthermore, the index calculation unit includes:

[0033] A concentration determination subunit is used to determine the degree of loss of control of the primary root cause parameter based on the statistical characteristics of the degree to which the liquid concentration deviates from its threshold within a preset identification time when the primary root cause parameter is the liquid concentration.

[0034] A humidity determination subunit is used to determine the degree of parameter runaway based on the fluctuation characteristics and instantaneous change characteristics of the relative humidity difference within the preset identification time when the primary root cause parameter is the relative humidity difference.

[0035] A speed determination subunit is used to determine the degree of runaway of the parameter based on the instantaneous change characteristics of the transmission speed ratio within the preset identification time when the primary root cause parameter is the transmission speed ratio.

[0036] The flow determination subunit is used to determine the degree of parameter runaway based on the correlation characteristics and temporal fluctuation characteristics of the dynamic discrete coefficient of the supply flow within the preset identification time when the primary root cause parameter is the supply flow.

[0037] Furthermore, the adjustment module includes:

[0038] The adjustment determination unit is used to determine whether the process needs to be adjusted based on the threshold comparison result of the process risk index, and when the primary root cause parameter is the liquid concentration, the liquid concentration is adjusted according to the theoretical adjustment amount, wherein the theoretical adjustment amount is determined based on the preset concentration threshold and the parameter runaway degree.

[0039] A humidity adjustment unit is used to adjust the relative humidity difference based on the determination result of process requirements when the primary root cause parameter is the relative humidity difference, according to the humidity adjustment amount, wherein the humidity adjustment amount is determined based on the process risk index and the preset risk threshold.

[0040] A speed adjustment unit is used to adjust the transmission speed ratio according to the speed adjustment amount when the primary root cause parameter is the transmission speed ratio, based on the determination result of the process requirement adjustment. The speed adjustment amount is determined based on the process risk index and the preset risk threshold.

[0041] A flow adjustment unit is used to adjust the supply flow based on the determination result of process needs, when the primary root cause parameter is the supply flow, according to the flow adjustment amount, wherein the flow adjustment amount is determined based on the process risk index and the preset risk threshold.

[0042] Furthermore, the early warning module includes:

[0043] An efficiency determination unit is used to determine the adjustment efficiency index based on the frequency of issuing the adjustment command, the dynamic dispersion coefficient, and the changing trend of the process risk index.

[0044] An early warning unit is used to issue an early warning based on the primary root cause parameters when the adjustment efficiency index is greater than a preset index threshold.

[0045] Furthermore, the performance determination unit includes:

[0046] The frequency determination subunit is used to determine the frequency factor based on the ratio of the frequency of issuing the adjustment command to a preset frequency threshold.

[0047] The improvement determination subunit determines the degree of dispersion improvement and the degree of risk improvement based on the difference between the average value of the dynamic dispersion coefficients before and after the adjustment instruction is issued and the degree of difference between the process risk index.

[0048] The efficiency determination subunit is used to determine the adjusted efficiency index based on the weighted fusion result of the frequency factor, the discrete improvement degree, and the risk improvement degree.

[0049] Compared with existing technologies, the advantages of this invention lie in its ability to achieve real-time and sensitive capture of the deterioration trend of film uniformity by calculating dynamic dispersion coefficients based on the sliding time-series characteristics of thickness data. By analyzing the contribution of each process parameter to the current thickness fluctuation to identify primary root causes, quality anomalies can be accurately traced back to specific process steps. By comprehensively considering the degree of loss of control of root cause parameters and thickness deviation, a quantitative process risk index is formed, enabling an objective assessment of the severity of the problem. Based on the process risk index, the system can issue adjustment instructions with clear direction and strength suggestions, achieving adaptive process optimization from diagnosis to execution. By evaluating the frequency and effect of historical adjustment instructions, an adjustment effectiveness index is formed, giving the system the ability to reflect on its own control effectiveness. This allows for early warning when control loops fail or problems escalate, effectively solving the problems of delayed quality defect detection, low problem-solving efficiency, and blind and delayed process adjustments caused by the inability to quickly and accurately locate the root causes of process anomalies in complex multi-process coupled production processes.

[0050] Furthermore, by introducing a sliding window mechanism to calculate the mean and standard deviation of thickness, random measurement noise and instantaneous interference are effectively filtered out, resulting in a robust estimate characterizing the thickness trend level and fluctuation intensity. By dividing the sliding standard deviation by the sliding mean to obtain the dynamic dispersion coefficient, the dimensional differences caused by different product thickness specifications are eliminated. This allows for standardized evaluation and comparison of thickness uniformity across batches and specifications of different calibrated products, making it more sensitive to early and minor uniformity degradation trends while avoiding false alarms caused by normal product specification switching.

[0051] Furthermore, by employing a threshold triggering mechanism, the root cause analysis process is initiated only when statistically significant degradation in thickness uniformity occurs, avoiding ineffective calculations by the system. By quantitatively analyzing the correlation between time-series data of various process parameters and thickness fluctuations within a specific period prior to the occurrence of thickness anomalies, the contribution weight of each parameter to the current quality anomaly is assessed. If a parameter is the true cause of the thickness anomaly, its change pattern should exhibit a significant statistical correlation with thickness fluctuations under appropriate time lag conditions. By calculating and comparing the maximum correlation coefficient between each parameter and thickness fluctuations, the system can identify the core parameters most closely related to the current defect from numerous synchronously changing process variables, improving the accuracy of diagnosis and processing efficiency.

[0052] Furthermore, by setting a time delay window set that conforms to the physical transmission law for each type of parameter, and calculating the time delay correlation coefficient under each window, the maximum value is finally selected as the correlation strength characterization of the parameter. This can adaptively find the strongest correlation moment between each parameter and thickness fluctuation, improve the statistical confidence and accuracy of root cause identification, and overcome the problem that fixed time delay analysis can lead to serious underestimation or even misjudgment of causal correlation due to the inherent and differentiated time delay in the transmission of different process disturbances to the final thickness index. This enables accurate diagnosis of complex time-varying coupled industrial processes.

[0053] Furthermore, by summing the time-delay correlations between each parameter and thickness fluctuations as a comprehensive correlation, a total variation benchmark characterizing the current anomaly that can be explained by the selected process parameters is constructed. The contribution of each parameter is calculated as the ratio of its correlation to this sum, achieving normalization while converting absolute correlation strength into relative responsibility share. This eliminates the interference caused by inherent differences in magnitude and fluctuation range between different parameters, making each parameter comparable on a uniform relative scale. The preset contribution threshold provides an objective and consistent criterion for judging the significance of contributions. Only when the contribution of a parameter exceeds this threshold is its influence considered sufficiently dominant and identified as a primary root cause. This ensures that the focus is on the factors that contribute most significantly to the current anomaly, achieving the engineering goal of robustly and interpretably extracting the core causes from a multivariate coupled system.

[0054] Furthermore, by calculating the relative deviation between the dynamic dispersion coefficient of thickness and a preset threshold, the dispersion deviation is obtained, which precisely quantifies the severity of quality anomalies. For primary root cause parameters, historical time-series data is analyzed to calculate the parameter runaway degree, which characterizes their own operational state and reflects the instability of the underlying process causes leading to quality anomalies. The process risk index is calculated using the dispersion deviation and parameter runaway degree. The basic risk is determined by the dispersion deviation, while the parameter runaway degree acts as a risk amplifier, multiplying the basic risk by (1+λ×C). When the parameter runaway degree is abnormally high, even if the currently observed quality deviation is not significant, the comprehensive risk index will be greatly increased, thus achieving early warning of potential, yet-to-be-fully-manifested, serious risks. Conversely, if the root cause is under control, the risk mainly reflects the current quality fluctuations, achieving a deep correlation between quality inspection and process monitoring.

[0055] Furthermore, four sets of mathematical models quantify the unique physical failure modes of different process parameters into parameter runaway degrees. For liquid concentration, stability is key; therefore, the formula simultaneously penalizes the frequency and persistence of deviations. The logarithmic function reflects the engineering understanding that the harm increases dramatically upon the first sustained deviation, but the rate of increase slows down with subsequent sustained deviations. For relative humidity difference, stability is key; therefore, the formula adopts a multiplicative structure, coupling the amplitude anomaly representing the intensity of fluctuations with the frequency anomaly representing the degree of instability. Frequency exceeding the limit will multiply the total risk through a preset frequency penalty coefficient, capturing the phenomenon that oscillations can exponentially amplify drying defects. For transmission speed ratio, its risk stems from abrupt changes; therefore, the formula uses a linear event counting model to directly count the number of harmful jumps and their continuity, giving higher weight to continuous jumps, simulating the cumulative damage characteristics of a mechanical system that can withstand a single impact but fails due to continuous impacts. The risk associated with supply flow rate is conditional, depending on the sensitivity of the flow rate to thickness under the current process conditions. Therefore, the formula multiplies the absolute value of the real-time correlation between flow rate and thickness by the relative volatility of the flow rate itself, allowing the risk assessment to dynamically adapt to different production conditions. Only when the flow rate is both unstable and has a significant impact on current quality will it be classified as high-risk. Through this differentiated modeling method, the parameter runaway degree can reflect the specific way in which the parameter is going out of control.

[0056] Furthermore, based on preliminary diagnostic conclusions and the identified root cause types, differentiated adjustment strategies are invoked. Concentration adjustments are not only based on the current concentration deviation but also incorporate the runaway rate, a parameter characterizing the historical severity of the problem, as a multiplier for the adjustment intensity. For concentration problems that have been out of control for some time, the system automatically applies a larger correction than a momentary overshoot, thus more effectively pulling the problem back to steady state. For other parameters such as humidity, velocity, and flow rate, linear adjustments are made directly based on the degree to which the comprehensive process risk index deviates from the threshold. This ensures a deep correction capability for slow variables like concentration while also considering the control efficiency for other fast variables, enabling the entire system to implement the most suitable closed-loop correction based on the characteristics of different failure modes.

[0057] Furthermore, by clarifying that a healthy and effective control system should achieve continuous improvement in process status and quality indicators through reasonable intervention frequency, the frequency factor is penalized for over-adjustment. When the adjustment frequency exceeds a reasonable threshold U0, the factor saturates to 1, preventing the inflated efficiency index from endless adjustments. The calculation design of discrete improvement degree and risk improvement degree ensures that the improvement degree is normalized to a value between 0 and 1, intuitively representing the relative improvement ratio of thickness uniformity and overall process risk after adjustment. A value of 0 indicates no improvement or deterioration, while a value of 1 indicates an ideal complete solution. Finally, the adjustment efficiency index is obtained by weighted fusion of the frequency factor and the two improvement degrees. A low index indicates that the system may be trapped in an ineffective cycle of frequent adjustments without improvement, or even worsen the problem. At this time, triggering an early warning can remind manual intervention to investigate deeper problems such as equipment failure, model mismatch, or process limits, thereby preventing the automated system from continuing to operate in the wrong direction. Attached Figure Description

[0058] Figure 1 This is a schematic diagram of the medical film quality inspection system based on multi-parameter fusion in this embodiment;

[0059] Figure 2 This is the logic diagram for determining thickness anomalies by the discrete determination unit in this embodiment;

[0060] Figure 3 This is a logic diagram for the identification subunit in this embodiment to determine that the liquid concentration is a primary root cause parameter.

[0061] Figure 4 The determination logic diagram that needs to be adjusted for the determination process of the determination unit in this embodiment. Detailed Implementation

[0062] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0063] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0064] Please see Figure 1 As shown, this is a schematic diagram of a medical film quality inspection system based on multi-parameter fusion according to this embodiment. This embodiment provides a medical film quality inspection system based on multi-parameter fusion, including:

[0065] The acquisition module is used to acquire the liquid concentration of polyvinyl alcohol casting liquid, the relative humidity difference between the inlet and outlet of the drying box in the drying and curing process, the transmission speed ratio in the molding process, the supply flow rate of polyvinyl alcohol casting liquid, and the average thickness of the film-formed area during the casting process of water-soluble polyvinyl alcohol medical film.

[0066] A discrete determination module, connected to the acquisition module, is used to determine dynamic discrete coefficients based on the sliding time-series characteristics of the average thickness.

[0067] An identification module, which is connected to the acquisition module and the discrete determination module respectively, is used to identify primary root cause parameters based on the contribution of the liquid concentration, the relative humidity difference, the transmission speed ratio and the supply flow rate to the dynamic discrete coefficient.

[0068] A risk determination module, which is connected to the acquisition module, the discrete determination module and the identification module respectively, is used to determine the process risk index based on the primary root cause parameters and the dynamic discrete coefficients;

[0069] An adjustment module, which is connected to the identification module and the risk determination module respectively, is used to issue adjustment instructions based on the degree of deviation of the primary root cause parameters, the process risk index and the preset risk threshold;

[0070] An early warning module, which is connected to the discrete determination module, the adjustment module, and the risk determination module respectively, is used to issue early warnings based on the threshold comparison results of the adjustment efficiency index and the primary root cause parameters. The adjustment efficiency index is determined based on the frequency of issuing the adjustment command, the dynamic discrete coefficient, and the changing trend of the process risk index.

[0071] In this embodiment, a medical film quality inspection system based on multi-parameter fusion is applied to an industrial production line for continuous casting of water-soluble polyvinyl alcohol (PVA) medical films. The casting liquid preparation process involves heating, stirring, and dissolving PVA resin, plasticizer, and deionized water according to a preset formula, followed by vacuum degassing to form a uniform, stable casting liquid that meets viscosity requirements. In the casting forming process, the casting liquid is coated at a constant flow rate onto a uniformly moving mirror-finished stainless steel strip or roller using a precision metering pump and coating die to form a wet liquid film. In the drying and curing process, several temperature zones and independently adjustable airflow drying chambers are set along the steel strip's running direction. The steel strip or roller carrying the wet film passes through the multi-temperature zone programmable temperature drying chamber, where controlled hot air convection causes the solvent water to gradually and uniformly evaporate, promoting the gelation and crystallization transformation of the PVA molecular chains. Finally, in the forming process, the completely dried film is peeled off the steel strip, cooled, and electrostatically eliminated before being wound up into a finished product. Film thickness directly determines product performance, and any deviation in key processes such as raw materials, drying, molding, and feeding can lead to defects such as uneven thickness and stress concentration. Therefore, the system achieves real-time diagnosis and intelligent control through online fusion analysis of multiple parameters.

[0072] In this embodiment, the liquid concentration of the polyvinyl alcohol casting solution is the mass percentage concentration of PVA in the PVA casting solution, which can be obtained by an online refractometer installed on the casting solution delivery pipeline or storage tank; the relative humidity difference is the difference between the relative humidity measurements of the air at the inlet and outlet of the drying chamber. This is obtained by real-time measurement of the relative humidity by high-precision humidity sensors (such as capacitive or resistive sensors) installed in the air ducts at the inlet and outlet of the drying chamber, and calculation of the difference; the transmission speed ratio is the ratio of the surface linear velocity of the main traction roller driving the steel belt or film in the production line to the surface linear velocity of the coating roller carrying the casting solution. Rotary encoders can be installed on the drive shafts of the main traction roller and the coating roller respectively to collect the rotational speed pulse signals of the two rollers in real time, and convert the rotational speed values ​​into surface linear velocities according to the preset standard diameter of each roller; finally, the ratio of the main traction roller linear velocity to the coating roller linear velocity is calculated in real time. The supply flow rate is the volume of PVA casting liquid delivered from the feeding system to the coating head per unit time, which can be obtained in real time by a precision mass flow meter or electromagnetic flow meter installed on the feeding pipeline; the thickness variation coefficient is the percentage ratio of the standard deviation to the average value of all thickness data scanned along the width direction of the film at any measurement moment, which characterizes the uniformity of the transverse thickness distribution of the film. It can be obtained by scanning the full width of the film transversely with an online scanning thickness gauge, simultaneously collecting the original thickness data, and calculating the standard deviation and average value of these data in real time, and finally obtaining it by (standard deviation / average value) × 100%.

[0073] The preset risk threshold is a critical value used to determine whether the process risk index is high enough to require immediate proactive adjustment. It is determined by the uniformity requirements of the film product and the historical steady-state performance of the production line. First, based on the upper limit of the acceptable thickness variation coefficient, the corresponding risk index constraint value is obtained through actual measurement of the acceptable boundary sample. At the same time, the risk index distribution during historical steady-state operation is statistically analyzed, and its upper limit of fluctuation, such as the mean + 3 times the standard deviation, is used as the normal operating condition boundary. The higher of the two is taken as the threshold benchmark. Then, it is fine-tuned based on the trade-off between quality risk and intervention frequency. Based on the above principles, the preset risk threshold is usually set between 0.8 and 1.2. In this embodiment, it is set to 1.0, which can trigger timely control before the thickness uniformity shows a significant deterioration risk but before it causes substantial defects.

[0074] By calculating dynamic dispersion coefficients based on the sliding time-series characteristics of thickness data, real-time and sensitive capture of thin film uniformity degradation trends is achieved. Primary root causes are identified by analyzing the contribution of various process parameters to current thickness fluctuations, enabling precise tracing of quality anomalies to specific process stages. A quantitative process risk index is formed by comprehensively considering the degree of loss of control of root cause parameters and thickness deviation, achieving an objective assessment of problem severity. Based on the process risk index, the system can issue adjustment instructions with clear direction and strength suggestions, realizing adaptive process optimization from diagnosis to execution. By evaluating the frequency and effectiveness of historical adjustment instructions, an adjustment efficiency index is formed, giving the system the ability to reflect on its own control effectiveness. Early warnings can be issued when control loops fail or problems escalate, effectively solving the problems of delayed quality defect detection, low problem-solving efficiency, and blind and delayed process adjustments caused by the inability to quickly and accurately locate the root causes of process anomalies in complex multi-process coupled production processes.

[0075] Specifically, the discrete determination module includes:

[0076] The mean calculation unit is used to calculate the sliding mean and sliding standard deviation of the average thickness within a preset time period based on a preset sliding window, so as to obtain several thickness sliding mean values ​​and several thickness sliding fluctuation values ​​respectively.

[0077] A discrete determination unit, connected to the mean calculation unit, is used to calculate the ratio of the thickness sliding fluctuation value to the thickness sliding mean value within each preset sliding window, so as to obtain a number of dynamic discrete coefficients.

[0078] The preset sliding window is the number of continuous data points used to calculate the statistical characteristics of thickness in real time. It depends on the thickness gauge's sampling cycle and the production line's operating speed. First, to ensure the statistical validity of the standard deviation calculation, the window must contain at least 20 sampling points. Based on this, the lower limit of the window width is calculated according to the sampling cycle. Second, to avoid misjudging lateral unevenness as longitudinal fluctuation, the physical length corresponding to the window must not exceed half of the scanning cycle in the film width direction. Based on this, the upper limit of the window width is determined. Finally, an integer value that balances noise filtering and real-time anomaly detection is selected within this range. Based on these principles, the preset sliding window is typically set between 30 and 120 seconds. In this embodiment, it is set to 60 seconds to balance noise filtering. In addition to the real-time nature of anomaly detection, the preset duration is the total duration of historical data covered by the sliding window calculation. It depends on the maximum duration of the process disturbance source and the batch switching interval. First, to ensure complete capture of the entire process from occurrence to recovery of the anomaly, the duration must cover at least 1.5 times the longest disturbance period. Second, to avoid mixing in thickness data of different product specifications, the duration must not exceed the minimum interval of formula adjustment between adjacent batches. Within this constraint, an integer value that can cover the duration of most historical anomalies is selected. Based on the above principles, it is usually set between 10 and 30 minutes. In this embodiment, it is set to 20 minutes, which can ensure the capture of the complete anomaly sequence and avoid interference from outdated data.

[0079] By introducing a sliding window mechanism to calculate the mean and standard deviation of thickness, random measurement noise and instantaneous interference are effectively filtered out, resulting in a robust estimate characterizing the trend level and fluctuation intensity of thickness. The dynamic dispersion coefficient is obtained by dividing the sliding standard deviation by the sliding mean, eliminating dimensional differences caused by different product thickness specifications. This allows for standardized evaluation and comparison of thickness uniformity across batches and specifications of different calibrated products, making it more sensitive to early and minor uniformity degradation trends while avoiding false alarms caused by normal product specification switching.

[0080] Please see Figure 2 As shown, this is the logic diagram for determining thickness anomalies by the discrete determination unit in this embodiment. In this embodiment, the identification module includes:

[0081] The discrete determination unit is used to determine the thickness abnormality when the dynamic discrete coefficient is greater than the preset discrete threshold, and to record the end time of the preset sliding window corresponding to the dynamic discrete coefficient as the uniformity abnormality time.

[0082] A related determination unit, connected to the discrete determination unit, is used to calculate the correlation between the liquid concentration, the relative humidity difference, the transmission speed ratio, and the supply flow rate and the dynamic discrete coefficient within a preset identification time before the uniform anomaly time, based on the determination result of the thickness anomaly, so as to obtain the concentration correlation, humidity correlation, speed correlation, and flow rate correlation respectively.

[0083] An identification unit, connected to the correlation determination unit, is used to identify the primary root cause parameters based on the concentration correlation, the humidity correlation, the velocity correlation, and the flow correlation.

[0084] The preset dispersion threshold is a statistical critical value for determining whether abnormal degradation of film thickness uniformity has occurred. It depends on the uniformity grade standards of the film product and the historical steady-state fluctuation level of the production line. First, the acceptable upper limit of the thickness variation coefficient is determined based on the clinical functional requirements of the medical film. This upper limit is then converted into the allowable boundary of the dynamic dispersion coefficient. Second, dynamic dispersion coefficient data is collected during the continuous stable operation of the production line. The mean plus three times the standard deviation is used as the statistical upper limit of normal process fluctuations. The higher of these two values ​​is taken as the threshold benchmark to ensure that uniformity degradation with quality risks can be identified. Based on these principles, the preset dispersion threshold is usually set between 0.5% and 2.0%. In this embodiment, it is set to 1.0%, which can filter out normal process fluctuations while capturing... There is a decreasing trend in uniformity with quality risks; the preset identification time is the duration of historical data backtracked during root cause tracing, which depends on the maximum time delay of the transmission of disturbances in each process parameter to the thickness detection point. Through on-site step response tests or historical abnormal event records, the lag time of liquid concentration, relative humidity difference, transmission speed ratio, and supply flow rate from the moment of parameter change to the moment when the thickness produces a detectable response is calibrated, and the maximum value among them is taken as the benchmark time. On this basis, a coverage margin of not less than 20% is added to ensure complete capture of the entire process of disturbance events from occurrence, transmission to decay. Based on the above principles, it is usually set between 5 minutes and 15 minutes. In this embodiment, it is set to 10 minutes, which can ensure that the complete process disturbance event sequence that leads to the current thickness abnormality is covered.

[0085] By employing a threshold triggering mechanism, the root cause analysis process is initiated only when statistically significant degradation in thickness uniformity occurs, avoiding ineffective calculations. Through quantitative analysis of the correlation between time-series data of various process parameters and thickness fluctuations within a specific period prior to the occurrence of thickness anomalies, the contribution weight of each parameter to the current quality anomaly is assessed. If a parameter is the true cause of the thickness anomaly, its change pattern should exhibit a significant statistical correlation with thickness fluctuations under appropriate time lag conditions. By calculating and comparing the maximum correlation coefficients between each parameter and thickness fluctuations, the system can identify the core parameters most closely related to the current defect from numerous synchronously changing process variables, improving diagnostic accuracy and processing efficiency.

[0086] Specifically, the relevant determination unit includes:

[0087] The concentration determination subunit is used to calculate the time-delay correlation between the liquid concentration and the dynamic discrete coefficient based on different preset concentration time-delay windows within the preset identification time period, so as to obtain a number of concentration time-delay correlations, and determine the maximum value of all concentration time-delay correlations as the concentration correlation.

[0088] The humidity determination subunit is used to calculate the time-delay correlation between the relative humidity difference and the dynamic dispersion coefficient based on different preset humidity time-delay windows within the preset identification time period, so as to obtain several humidity time-delay correlations, and determine the maximum value of all humidity time-delay correlations as the humidity correlation.

[0089] The speed determination subunit is used to calculate the time delay correlation between the transmission speed ratio and the dynamic discrete coefficient based on different preset speed time delay windows within the preset identification time period, so as to obtain a number of speed time delay correlations, and determine the maximum value of all speed time delay correlations as the speed correlation.

[0090] The flow determination subunit is used to calculate the time-delay correlation between the supply flow and the dynamic discrete coefficient based on different preset flow time-delay windows within the preset identification time period, so as to obtain several flow time-delay correlations, and determine the maximum value of all flow time-delay correlations as the flow correlation.

[0091] The preset concentration time delay window is a set of time offsets used to calculate the time delay correlation between liquid concentration and thickness fluctuation. The minimum offset is determined based on the theoretical transmission time of the casting liquid from the preparation tank through the coating die to the thickness gauge, and then incremented at equal intervals using this transmission time as the step size until the complete response cycle from the occurrence of the concentration disturbance to drying and curing is covered. This generates a discrete time delay sequence, typically set between 120 and 600 seconds. In this embodiment, it is set to [120 seconds, 240 seconds, 360 seconds, 480 seconds, 600 seconds], enabling a systematic evaluation of this slow process throughout the entire process. The potential causal relationship within the different periods is identified. The preset humidity time lag window consists of multiple time offsets pre-set to calculate the time lag correlation between relative humidity difference and thickness fluctuation. The minimum offset is determined based on the theoretical residence time of humid air in the drying chamber, and increments at equal intervals using this residence time as a step size, until the complete delay interval of humidity fluctuation from entering the drying chamber to film curing is covered. This is typically set between 60 and 600 seconds; in this embodiment, it is set to [60 seconds, 180 seconds, 300 seconds, 420 seconds, 600 seconds], ensuring that humidity fluctuations are captured throughout the entire film curing process. The preset speed delay window is a set of time offsets used to calculate the time delay correlation between the transmission speed ratio and thickness fluctuation. When selecting values, it first includes zero delay to capture the instantaneous stretching response of speed changes, and then selects multiple candidate delays at equal intervals based on the transmission lag time from the coating roller to the thickness gauge. These delays are typically set between 0 and 600 seconds; in this embodiment, they are set to [0 seconds, 30 seconds, 150 seconds, 300 seconds, 600 seconds]. This allows for a systematic investigation of possible indirect or compound delay effects while focusing on instantaneous effects. The preset flow rate delay window is... Multiple time offsets are pre-set to calculate the time lag correlation between the supply flow rate and thickness fluctuation. The initial offset is determined based on the minimum transmission lag determined by the length of the supply pipeline and the response time of the metering pump. The offset is then increased at equal intervals with this transmission lag as the step size until it covers the complete response cycle from the pump outlet to coating, film formation, drying and thickness measurement of the flow rate disturbance. The offset is usually set between 30 seconds and 600 seconds. In this embodiment, it is set to [30 seconds, 120 seconds, 240 seconds, 360 seconds, 600 seconds]. This allows for a comprehensive evaluation of all possible impacts of the supply system dynamics on coating uniformity within the identification time.

[0092] By setting a time-delay window set that conforms to the physical transmission law for each type of parameter, calculating the time-delay correlation coefficient under each window, and finally selecting the maximum value as the correlation strength characterization of the parameter, it is possible to adaptively find the strongest correlation moment between each parameter and thickness fluctuation, improve the statistical confidence and accuracy of root cause identification, and overcome the problem that fixed time-delay analysis will lead to serious underestimation or even misjudgment of causal correlation due to the inherent and differentiated time delay in the transmission of different process disturbances to the final thickness index, thus achieving accurate diagnosis of complex time-varying coupled industrial processes.

[0093] Please see Figure 3 As shown, this is the logic diagram for determining liquid concentration as a primary root cause parameter by the identification subunit in this embodiment. In this embodiment, the identification unit includes:

[0094] The contribution determination subunit is used to calculate the ratios of concentration correlation, humidity correlation, velocity correlation, and flow rate correlation to the overall correlation, so as to obtain the concentration contribution, humidity contribution, velocity contribution, and flow rate contribution. The overall correlation is determined based on the sum of the concentration correlation, humidity correlation, velocity correlation, and flow rate correlation.

[0095] An identification subunit, connected to the contribution determination subunit, is configured to determine the liquid concentration as the primary root cause parameter when the concentration contribution is greater than a preset contribution threshold, determine the relative humidity difference as the primary root cause parameter when the humidity contribution is greater than a preset contribution threshold, determine the transmission speed ratio as the primary root cause parameter when the speed contribution is greater than a preset contribution threshold, and determine the supply flow rate as the primary root cause parameter when the flow rate contribution is greater than a preset contribution threshold.

[0096] The preset contribution threshold is the critical proportion at which the contribution of process parameters is large enough to be identified as a major root cause. The value is determined comprehensively based on the contribution significance level required for root cause identification, the statistical distribution of the contribution of a single root cause in historical failure cases, and the tolerance for multiple root causes. First, to ensure that the identified root cause is dominant, its contribution must be at least 1.5 times the average contribution of each parameter. Second, by analyzing the contribution distribution of root cause parameters in historically confirmed single root cause events, a low critical value that can cover more than 80% of cases, such as the 25th percentile, is taken as the threshold benchmark. It is usually set between 20% and 40%. In this embodiment, it is set to 25%, which can ensure that the identified root cause has a significant relative impact.

[0097] By summing the time-delay correlations of each parameter with thickness fluctuations as a comprehensive correlation, a total variation benchmark characterizing the current anomaly that can be explained by the selected process parameters is constructed. The contribution of each parameter is calculated as the ratio of its correlation to this sum, achieving normalization while converting absolute correlation strength into relative responsibility share. This eliminates the interference caused by inherent differences in magnitude and fluctuation range between different parameters, making each parameter comparable on a uniform relative scale. A preset contribution threshold provides an objective and consistent criterion for judging the significance of contributions. Only when a parameter's contribution exceeds this threshold is its influence considered sufficiently dominant and identified as a primary root cause. This ensures that the focus is on the factors that contribute most significantly to the current anomaly, achieving the engineering goal of robustly and interpretably extracting core causes from a multivariate coupled system.

[0098] Specifically, the risk determination module includes:

[0099] The deviation calculation unit is used to calculate the relative deviation between the dynamic discrete coefficient and the preset discrete threshold to obtain the discrete deviation degree.

[0100] An index calculation unit is used to determine the degree of parameter runaway based on the time-series changes of the primary root cause parameters;

[0101] A risk determination unit, which is connected to the deviation calculation unit and the index calculation unit respectively, is used to determine the process risk index based on the discrete deviation degree and the parameter runaway degree, wherein G=L×(1+λ×C), where G is the process risk index, L is the discrete deviation degree, C is the parameter runaway degree, and λ is a preset adjustment coefficient.

[0102] The preset adjustment coefficient is an adjustment factor used to balance the relative weights of the direct manifestation of thickness anomalies and the degree of loss of control of the primary root cause parameters that cause the anomalies in the comprehensive risk assessment model. It depends on the priority setting of quality deviation phenomena and process parameter instability in the specific production process, and is usually set between 0.5 and 2.0. In this embodiment, it is set to 1.0, which can give the apparent deviation of quality the same importance as the intrinsic instability of the primary root cause parameters in risk quantification.

[0103] By calculating the relative deviation between the dynamic dispersion coefficient of thickness and a preset threshold, the dispersion deviation is obtained, which accurately quantifies the severity of quality anomalies. Furthermore, for primary root cause parameters, historical time-series data is analyzed to calculate the parameter runaway degree, which characterizes the parameter's operational state and reflects the instability of the underlying process causes leading to quality anomalies. A process risk index is calculated using the dispersion deviation and parameter runaway degree. The basic risk is determined by the dispersion deviation, while the parameter runaway degree acts as a risk amplifier, multiplying the basic risk by (1+λ×C). When the parameter runaway degree is abnormally high, even if the currently observed quality deviation is not significant, the comprehensive risk index will be greatly increased, thus achieving early warning of potential, yet-to-be-fully-manifested, serious risks. Conversely, if the root cause is under control, the risk mainly reflects current quality fluctuations, achieving a deep correlation between quality inspection and process monitoring.

[0104] Specifically, the index calculation unit includes:

[0105] A concentration determination subunit is used to calculate the runaway degree of the primary root cause parameter based on the excess ratio and the maximum number of consecutive excesses when the primary root cause parameter is the liquid concentration, wherein...

[0106] ,

[0107] Wherein, B is the excess ratio and O is the maximum number of consecutive excesses. The excess ratio is determined based on the proportion of the number of times the liquid concentration is greater than the preset concentration threshold within the preset recognition time to the total number of times within the preset recognition time. The maximum number of consecutive excesses is determined based on the maximum number of times the liquid concentration is greater than the preset concentration threshold within the preset recognition time.

[0108] A humidity determination subunit is used to calculate the degree of parameter runaway based on amplitude anomaly and frequency anomaly when the primary root cause parameter is the relative humidity difference.

[0109] ,

[0110] Among them, Y F It is the amplitude anomaly, Y F0 It is a preset amplitude threshold, α is a preset frequency penalty coefficient, and Y is a preset amplitude threshold. P It is the frequency anomaly, Y P0 It is a preset frequency threshold, wherein the amplitude anomaly is determined based on the standard deviation of the relative humidity difference within the preset recognition time, and the frequency anomaly is based on the proportion of the number of times the signs of two adjacent times are inconsistent in the first-order difference sequence of the relative humidity difference within the preset recognition time to the total number of times in the preset recognition time.

[0111] A speed determination subunit is used to calculate the runaway degree of the parameter based on the number of risk jumps and the number of additional consecutive jumps when the primary root cause parameter is the transmission speed ratio. Here, C = β × T1 + γ × T2, where β is a preset first weight, γ is a preset second weight, T1 is the number of risk jumps, and T2 is the number of additional consecutive jumps. The number of risk jumps is determined based on the number of times the absolute value of the difference between the transmission speed ratios at two adjacent moments within the preset identification time is greater than a preset speed threshold, and the number of additional consecutive jumps is determined based on the number of moments when the number of consecutive risk jumps occurs.

[0112] The flow determination subunit is used to calculate the degree of parameter runaway based on the flow correlation and the flow variation coefficient when the primary root cause parameter is the supply flow. C=|R|×(X / X0), where R is the flow correlation, X is the flow variation coefficient, and X0 is a preset variation threshold. The flow correlation is determined based on the Pearson correlation coefficient between the supply flow and the dynamic dispersion coefficient within the preset identification time, and the flow variation coefficient is determined based on the ratio of the standard deviation to the mean of the supply flow within the preset identification time.

[0113] The preset concentration threshold is the process critical value for determining whether the liquid concentration exceeds the standard. It depends on the target concentration of the PVA solution and the allowable deviation of the process, and is usually set between ±0.5% and ±1.5% of the standard value. In this embodiment, it is set to the nominal value +1.0%, which can identify concentration deviations that may cause significant changes in viscosity. The preset amplitude threshold is the engineering limit value for determining whether the fluctuation amplitude of the relative humidity difference is abnormal. It depends on the steady-state fluctuation range of the humidity control system of the drying oven and the upper limit of the tolerance of the film drying process to humidity fluctuations. This upper limit of tolerance is determined by the critical value of the impact of humidity fluctuations on key indicators such as film crystallinity and haze. It is usually set between 1% and 5%, and in this embodiment, it is set to 3%, which can distinguish between normal random fluctuations and harmful severe fluctuations. The preset frequency penalty coefficient is a penalty factor used to amplify the impact of high-frequency oscillations on humidity runaway. It depends on the stringency of the drying process stability requirements. First, the sensitivity of the film product to drying oscillations is divided into three levels: low, medium, and high, corresponding to initial penalty coefficients of 1.0, 2.0, and 3.0, respectively. Second, adjustments are made based on the drying system's ability to suppress oscillations. If the system is equipped with a fast-response actuator or has oscillation attenuation functionality, the coefficient is lowered by 0.5 from the initial value in the previous step; conversely, if the system has high inertia and slow recovery, the coefficient is increased by 0.5. Finally, 3 to 5 historical humidity oscillation events are selected, and the corrected coefficient is substituted into the runaway calculation formula to verify whether it can significantly increase the runaway degree of oscillation events compared to stable periods. If the discrimination is insufficient... If the requirement is met, the frequency is continuously adjusted in increments of 0.5 until it is satisfied. Based on the above principle, it is usually set between 1 and 3. In this embodiment, it is set to 2, which can impose a significant additional risk penalty on excessive oscillation frequencies. The preset frequency threshold is the critical value for determining whether the reversal frequency of the relative humidity difference change direction is too high. It depends on the response speed of the drying system and the allowable adjustment frequency. First, the closed-loop time constant of the humidity in the drying chamber from the execution of the action to the sensor feedback change is measured. The reciprocal of the time constant is normalized to the range of 20% to 40% as the initial benchmark. Second, the natural reversal frequency distribution of the humidity difference during historical stable operation is statistically analyzed, and the 85th percentile is taken as the lower limit threshold. Finally, combined with the thin film stress test results, if alternation has occurred below this threshold, the threshold is considered. Shrinkage marks are adjusted upwards in 5% increments until no stress defects are generated. Based on the above principle, the preset frequency threshold is usually set between 20% and 40%. In this embodiment, it is set to 30%, which can effectively identify abnormal oscillation modes that may cause alternating changes in film stress. The preset first weight is a coefficient used to quantify the degree of harm of a single transmission speed jump. It depends on the sensitivity of the mechanical system to instantaneous impact. First, a base weight of 1.0 is used, corresponding to the acceptable amplitude of thickness fluctuation caused by a single jump. Second, through a step response test of the transmission system, the peak value of thickness fluctuation caused by a 1% speed jump and the recovery time are measured. For every 0.1 μm of fluctuation amplitude exceeding the base, the weight is increased by 0.2, and for every 2 seconds of recovery time exceeding the base, the weight is increased by 0.1. Finally, round the corrected weights to the nearest multiple of 0.5 within the range of 0.5 to 2.0. Based on the above principles, the preset first weight is usually set between 0.5 and 2. In this embodiment, it is set to 1.0, which can reasonably assess the basic harm of a single jump to transmission stability. The preset second weight is a coefficient used to quantify the additional harm caused by continuous transmission speed ratio risk jumps. It depends on the cumulative damage characteristics of continuous jumps. First, it is set to twice the first weight as the initial value. Second, through comparative tests of three consecutive speed jumps and a single speed jump, the peak multiple of thickness fluctuation and the multiple of recovery time extension are measured. The average of the two is used as the weight multiple benchmark. Finally, this multiple is rounded to an integer multiple of 0.5 in the range of 2 to 5. Based on the above principles, the preset second weight is usually set between 2 and 5. In this embodiment, it is set to 3.0, which can significantly amplify the penalty for continuous jumps, a serious precursor to failure. The preset speed threshold is the critical difference for determining whether the change in transmission speed ratio between adjacent moments constitutes a risk jump. It depends on the control resolution and mechanical backlash of the transmission system. First, the pulse equivalent of the rotary encoder and the minimum speed command step of the servo driver are read, and the larger value of the two is taken as the lower limit of the resolution. Speed ​​fluctuation data are collected during steady-state operation of the production line under no-load and load conditions. The upper limit of the 99% confidence interval is used as the normal noise boundary. Finally, the larger of the resolution lower limit and the noise boundary is rounded up to an integer multiple of 0.1%. Based on the above principles, the preset speed threshold is usually set between 0.5% and 2%. In this embodiment, it is set to 1.0%, which can filter out small fluctuations and capture risky significant speed mutations. The preset variation threshold is the limit of the relative coefficient of variation for determining whether the fluctuation rate of the supply flow itself is too high. It depends on the stability performance of the feeding system and the metering pump. First, the volumetric efficiency pulsation rate of the metering pump and the residual pulsation after damper attenuation are measured across the entire flow range. The combined value of these two values ​​is taken as the lower limit of the system's own fluctuation. Second, based on the flow rate-wet film thickness transfer gain of the coating die, the maximum allowable flow rate variation coefficient within the upper limit of the thickness tolerance is calculated. Finally, the smaller of the lower limit of the system's own fluctuation and the calculated value of the thickness tolerance is taken and rounded up to an integer multiple of 0.1%. Based on the above principles, the preset variation threshold is usually set between 1% and 3%. In this embodiment, it is set to 2.0%, which can define the acceptable boundary of flow rate stability.

[0114] Four sets of mathematical models quantify the unique physical failure modes of different process parameters into parameter runaway degrees. For liquid concentration, stability is key; therefore, the formula penalizes both the frequency and persistence of deviations. The logarithmic function reflects the engineering understanding that the harm increases dramatically upon the first sustained deviation, but the rate of increase slows down with subsequent sustained deviations. For relative humidity difference, stability is key; therefore, the formula uses a multiplicative structure to couple the amplitude anomaly, which characterizes the intensity of fluctuations, with the frequency anomaly, which characterizes the degree of instability. Frequency exceeding the limit is penalized exponentially by a preset frequency penalty coefficient, capturing the phenomenon that oscillations exponentially amplify drying defects. For transmission speed ratio, the risk stems from abrupt changes; therefore, the formula uses a linear event counting model to directly count the number of harmful jumps and their continuity, assigning higher weight to continuous jumps, simulating the cumulative damage characteristics of a mechanical system that can withstand a single impact but fails due to continuous impacts. The risk associated with supply flow rate is conditional, depending on the sensitivity of the flow rate to thickness under the current process conditions. Therefore, the formula multiplies the absolute value of the real-time correlation between flow rate and thickness by the relative volatility of the flow rate itself, allowing the risk assessment to dynamically adapt to different production conditions. Only when the flow rate is both unstable and has a significant impact on current quality will it be classified as high-risk. Through this differentiated modeling method, the parameter runaway degree can reflect the specific way in which the parameter is going out of control.

[0115] Please see Figure 4 As shown, this is a logic diagram for determining the process adjustment required by the adjustment determination unit in this embodiment. In this embodiment, the adjustment module includes:

[0116] An adjustment determination unit is used to determine that the process needs adjustment when the process risk index is greater than the preset risk threshold, and to adjust the liquid concentration according to a theoretical adjustment amount when the primary root cause parameter is the liquid concentration. The theoretical adjustment amount is determined based on a preset concentration threshold and the parameter runaway degree, wherein E... C =k C ×(N-N0)×(1+η×C), where, E C It is the theoretical adjustment amount, k C η is the preset concentration adjustment coefficient, N is the liquid concentration, N0 is the preset concentration threshold, η is the preset runaway coefficient, and C is the parameter runaway degree when the primary root cause parameter is the liquid concentration.

[0117] A humidity adjustment unit, connected to the adjustment determination unit, is used to adjust the relative humidity difference based on the determination result of process requirements, when the primary root cause parameter is the relative humidity difference, according to the humidity adjustment amount. The humidity adjustment amount is determined based on the process risk index and the preset risk threshold, wherein E... S =k S ×(G-G0), where ES It is the humidity adjustment amount, k S G0 is the preset humidity adjustment coefficient, and G0 is the preset risk threshold.

[0118] A speed adjustment unit, connected to the adjustment determination unit, is used to adjust the transmission speed ratio based on the determination result of process requirements, when the primary root cause parameter is the transmission speed ratio, according to the speed adjustment amount. The speed adjustment amount is determined based on the process risk index and the preset risk threshold, wherein E... D =k D ×(G-G0), where E D It is the speed adjustment amount, k D It is the preset speed adjustment coefficient;

[0119] A flow adjustment unit, connected to the adjustment determination unit, is used to adjust the supply flow rate according to the determination result of process requirements, when the primary root cause parameter is the supply flow rate. The flow adjustment amount is determined based on the process risk index and the preset risk threshold, wherein E... L =k L ×(G-G0), where E L It is the flow adjustment amount, k L It is the preset flow rate adjustment coefficient.

[0120] The preset concentration adjustment coefficient is a proportional factor that determines the intensity of the response of the concentration adjustment amount to the current concentration deviation. It depends on the correction response speed and over-adjustment risk of the casting liquid preparation system. First, the lag time and overshoot amount from issuing the concentration adjustment command to the actual concentration reaching the target value are measured through a step response test. The longer the lag time and the larger the overshoot amount, the coefficient should tend to the lower limit of the range. Second, with a step of 0.1, several candidate values ​​are selected in the range of 0.1 to 0.3 for closed-loop simulation or small-batch production line testing. The number of adjustments and fluctuation amplitude required for the concentration to reach a steady state after each adjustment are recorded. Finally, the candidate value with the fewest adjustment times and without causing continuous oscillation is selected as the coefficient. Based on the above principles, the preset concentration adjustment coefficient is usually set between 0.1 and 0.3. In this embodiment, it is set to 0.2, which can ensure the effectiveness of the correction while avoiding system oscillation caused by excessive adjustment. The preset runaway coefficient is a gain coefficient used to convert the runaway degree of the parameter into an additional adjustment force. It depends on the urgency of correcting the problem of continuous concentration deviation. First, the correlation coefficient between the duration of continuous concentration exceeding the standard and quality indicators such as film haze and crystallinity in historical fault data is analyzed. The stronger the correlation, the higher the urgency of correction. The coefficient should be shifted towards the upper limit of the range. Second, candidate values ​​are selected in the range of 0.5 to 1.5 with a step of 0.25. They are substituted into the concentration adjustment formula to evaluate their ability to accelerate the recovery of continuous deviation events. Finally, the smallest candidate value that shortens the average recovery time of continuous exceedance events by more than 30% is selected. Based on the above principles, the preset runaway coefficient is usually set between 0.5 and 1.5. In this embodiment, it is set to 1.0, which can linearly amplify the basic adjustment amount according to the runaway degree. The preset humidity adjustment coefficient is the proportional coefficient in the proportional control model that maps the part of the process risk index exceeding the threshold to the humidity adjustment amount. It depends on the sensitivity and stability of the humidity adjustment mechanism of the drying box. First, the rate of change of humidity per unit time when the humidity adjustment valve jumps from 0% opening to 100% is measured, and the minimum controllable adjustment step size is calculated. Second, with a step size of 0.05, candidate values ​​are selected in the range of 0.05 to 0.2, and PID closed-loop disturbance test is performed. The number of adjustments after the risk index overshoot and the steady-state recovery time are recorded. Finally, the candidate value with the fewest adjustment times, the shortest recovery time and no continuous oscillation is selected.Based on the above principles, the preset humidity adjustment coefficient is usually set between 0.05 and 0.2. In this embodiment, it is set to 0.1, which can achieve a stable and gradual correction of the drying environment. The preset speed adjustment coefficient is used to map the risk index to the transmission speed ratio. It depends on the adjustment accuracy and response speed of the transmission system. First, the minimum speed command step of the servo driver and the resolution of the rotary encoder are read, and the larger value of the two is taken as the theoretical minimum adjustment amount. Second, with a step of 0.01, candidate values ​​are selected in the range of 0.01 to 0.05, and a speed step test is performed to measure the peak value of thickness fluctuation and the amplitude of tension fluctuation. Finally, the largest candidate value with thickness fluctuation within 1 / 3 of the tolerance and no tension impact oscillation is selected. Based on the above principles, the preset speed adjustment coefficient is usually set between 0.01 and 0.05. In this embodiment, it is set to 0. 02. It can fine-tune the speed without mechanical shock. The preset flow adjustment coefficient is used to map the risk index to the adjustment amount of the supply flow. It depends on the accuracy and repeatability of the metering pump or regulating valve. First, the volumetric efficiency pulsation rate of the metering pump and the repeatability of the regulating valve are measured in the full flow range. The combined error of the two is taken as the uncertainty of the flow control. Second, candidate values ​​are selected in increments of 0.1 from 0.1 to 0.5. According to the transfer gain of the coating die flow-wet film thickness, the single adjustment thickness change corresponding to each candidate value is calculated. Finally, the largest candidate value with the thickness change within the tolerance of 1 / 2 and without triggering the flow alarm is selected. Based on the above principles, the preset flow adjustment coefficient is usually set between 0.1 and 0.5. In this embodiment, it is set to 0.3, which can directly and appropriately correct the flow.

[0121] Based on preliminary diagnostic conclusions and the identified root cause types, differentiated adjustment strategies are invoked. Concentration adjustments are not only based on the current concentration deviation but also incorporate the runaway rate, a parameter representing the historical severity of the problem, as a multiplier for the adjustment intensity. For concentration problems that have been out of control for some time, the system automatically applies a larger correction than a momentary overshoot, thus more effectively pulling the problem back to steady state. For other parameters such as humidity, velocity, and flow rate, linear adjustments are made directly based on the degree to which the comprehensive process risk index deviates from the threshold. This ensures deep correction capability for slow variables like concentration while also considering the control efficiency for direct and rapid responses to other fast variables. This allows the entire system to implement the most suitable closed-loop correction based on the characteristics of different failure modes.

[0122] Specifically, the early warning module includes:

[0123] An efficiency determination unit is used to determine the adjustment efficiency index based on the frequency of issuing the adjustment command, the dynamic dispersion coefficient, and the changing trend of the process risk index.

[0124] An early warning unit, connected to the performance determination unit, is used to issue an early warning based on the primary root cause parameters when the adjusted performance index is greater than a preset index threshold.

[0125] The preset index threshold is a critical value used to determine whether the adjustment efficiency index is too high and requires triggering an early warning. It depends on the system's tolerance for control loop failures or persistent process problems. First, a tolerance number is set for a control loop failure if several consecutive adjustments fail to produce significant improvement. The product of this tolerance number and the single adjustment cycle is mapped to the corresponding quantile of the adjustment efficiency index. Second, the adjustment efficiency index of historical events confirmed as persistent faults or actuator failures is collected, and the lower limit value that covers more than 90% of the events is taken as an empirical benchmark. Finally, based on the trade-off between production interruption due to false alarms and defects caused by missed alarms, the benchmark value is finely adjusted to the range of 0.6 to 0.9. If the cost is biased towards missed alarms, the threshold is lowered; if it is biased towards false alarms, the threshold is raised. Based on the above principles, the preset index threshold is usually set between 0.6 and 0.9. In this embodiment, it is set to 0.8, which can promptly escalate the alarm when the system adjustment action is continuously ineffective or leads to deterioration.

[0126] Specifically, the performance determination unit includes:

[0127] The frequency determination subunit is used to determine the frequency factor based on the ratio of the frequency of issuing the adjustment command to a preset frequency threshold, where Z = min(1, U / U0), where Z is the frequency factor, U is the frequency of the adjustment command, and U0 is the preset frequency threshold.

[0128] An improvement determination subunit is used to determine the degree of dispersion improvement based on the relative deviation between the average value of all dynamic dispersion coefficients within a preset determination time period and the average value of all dynamic dispersion coefficients within a preset adjustment time period, wherein Q L =max{0,min[1,(A1-A2) / A1]}, where Q L The degree of improvement in dispersion is defined as follows: A1 is the average value of all dynamic dispersion coefficients within a preset time period; A2 is the average value of all dynamic dispersion coefficients within a preset adjustment time period; and the degree of improvement in risk is determined based on the relative deviation between the process risk index before the adjustment instruction is issued and the average value of all process risk indices within the preset adjustment time period. Where Q... F =max{0,min[1,(A3-A4) / A3]}, where Q F A3 is the process risk index before the adjustment instruction is issued, and A4 is the average value of all process risk indices within the preset adjustment period.

[0129] An effectiveness determination subunit, which is connected to the frequency determination subunit and the improvement determination subunit respectively, is used to determine the adjusted effectiveness index based on the weighted fusion result of the frequency factor, the discrete improvement degree and the risk improvement degree.

[0130] The preset frequency threshold is a reasonable upper limit reference value for determining whether the number of adjustment commands issued per unit time is too high. It depends on the inherent stability of the production process and the expected intervention frequency of automatic control. First, the time from receiving the command to the controlled parameter entering the steady-state allowable error band is measured by step response test. The reciprocal of this recovery time is taken as the theoretical maximum intervention frequency. Second, the actual adjustment frequency triggered by normal process drift during the continuous stable operation of the production line is collected, and its 85th percentile is taken as the empirical benchmark. Finally, the smaller value between the theoretical upper limit and the empirical benchmark is rounded down to an integer multiple of 5 to 20 times per hour to ensure that the threshold can identify over-adjustment without suppressing necessary intervention. Based on the above principles, the preset frequency threshold is usually set between 5 and 20 times per hour. In this embodiment, it is set to 10 times per hour, which can effectively identify over-adjustment caused by improper control parameters or actuator problems. Adjustment and system oscillation; the preset adjustment duration is the length of the time window used to evaluate the changes in process state after the adjustment measures take effect. It depends on the time required from the execution of the adjustment to the generation of observable and stable responses in process parameters and quality indicators. First, the response lag time from the issuance of the adjustment command to the completion of the actuator action is measured. Second, the physical transmission lag time from the position of the actuator to the detection point of the thickness gauge is calculated. Third, the time constant required for the controlled object to go from parameter change to thickness stability is obtained through a step response test. Finally, the sum of the response lag, transmission lag and 3 times the time constant is rounded up to an integer minute in the range of 10 to 30 minutes to ensure that the observation window can cover the entire process from the adjustment taking effect to steady-state convergence. Based on the above principles, the preset adjustment duration is usually set between 10 and 30 minutes. In this embodiment, it is set to 20 minutes to ensure that enough data reflecting the adjustment effect is collected.

[0131] By clearly defining that a healthy and effective control system should achieve continuous improvement in process status and quality indicators through reasonable intervention frequency, the frequency factor is penalized for over-adjustment. When the adjustment frequency exceeds a reasonable threshold U0, the factor saturates to 1, preventing the inflated efficiency index from endless adjustments. The calculation design of discrete improvement degree and risk improvement degree ensures that the improvement degree is normalized to a value between 0 and 1, intuitively representing the relative improvement ratio of thickness uniformity and overall process risk after adjustment. A value of 0 indicates no improvement or deterioration, while a value of 1 indicates an ideal complete solution. Finally, the adjustment efficiency index is obtained by weighted fusion of the frequency factor and the two improvement degrees. A low index indicates that the system may be trapped in an ineffective cycle of frequent adjustments without improvement, or even worsen the problem. At this time, triggering an early warning can remind manual intervention to investigate deeper problems such as equipment failure, model mismatch, or process limits, thereby preventing the automated system from continuing to operate in the wrong direction.

[0132] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A medical film quality inspection system based on multi-parameter fusion, characterized in that, include: The acquisition module is used to acquire the liquid concentration of polyvinyl alcohol casting liquid, the relative humidity difference between the inlet and outlet of the drying chamber in the drying and curing process, the transmission speed ratio in the forming process, the supply flow rate of polyvinyl alcohol casting liquid, and the average thickness of the film-formed area during the casting process of water-soluble polyvinyl alcohol medical film. The transmission speed ratio is the ratio of the surface linear velocity of the main traction roller that drives the steel belt or film in the production line to the surface linear velocity of the coating roller that carries the casting liquid for coating. A discrete determination module is used to determine dynamic discrete coefficients based on the sliding time-series characteristics of the average thickness; The identification module is used to identify primary root cause parameters based on the contribution of the liquid concentration, the relative humidity difference, the transmission speed ratio, and the supply flow rate to the dynamic discrete coefficient, respectively. The risk determination module is used to determine the process risk index based on the primary root cause parameters and the dynamic dispersion coefficients; The adjustment module is used to issue adjustment instructions based on the degree of deviation of the primary root cause parameters, the process risk index, and the preset risk threshold. The early warning module is used to issue early warnings based on the threshold comparison results of the adjustment efficiency index and the primary root cause parameters, wherein the adjustment efficiency index is determined based on the frequency of issuing the adjustment command, the dynamic dispersion coefficient, and the changing trend of the process risk index.

2. The medical film quality detection system based on multi-parameter fusion according to claim 1, characterized in that, The discrete determination module includes: The mean calculation unit is used to determine the thickness sliding mean and thickness sliding fluctuation value based on the sliding mean and sliding standard deviation of the average thickness, respectively. A discrete determination unit is used to determine the dynamic discrete coefficients based on the ratio of the thickness sliding fluctuation value to the thickness sliding mean value.

3. The medical film quality detection system based on multi-parameter fusion according to claim 2, characterized in that, The identification module includes: The discrete determination unit is used to determine the thickness anomaly based on the threshold comparison result of the dynamic discrete coefficient, and to determine the uniformity anomaly time based on the dynamic discrete coefficient corresponding to the thickness anomaly. The relevant determination unit determines the concentration correlation, humidity correlation, speed correlation, and flow correlation based on the determination result of thickness anomaly and the correlation between the liquid concentration, relative humidity difference, transmission speed ratio, and supply flow rate before the time of uniform anomaly and the dynamic dispersion coefficient. The identification unit is used to identify the primary root cause parameters based on the concentration correlation, the humidity correlation, the velocity correlation, and the flow rate correlation.

4. The medical film quality detection system based on multi-parameter fusion according to claim 3, characterized in that, The relevant determination unit includes: A concentration determination subunit is used to determine the maximum value of all concentration time-delay correlations as the concentration correlation, wherein the concentration time-delay correlation is determined based on the liquid concentration and the dynamic dispersion coefficient; A humidity determination subunit is used to determine the humidity correlation by the maximum value of all humidity time-delay correlations, wherein the humidity time-delay correlation is determined based on the relative humidity difference and the dynamic dispersion coefficient; A speed determination subunit is used to determine the maximum value of all speed time delay correlations as the speed correlation, wherein the speed time delay correlation is determined based on the transmission speed ratio and the dynamic discrete coefficient; A flow determination subunit is used to determine the maximum value of the time-delay correlation of all flow rates as the flow correlation, wherein the flow correlation is determined based on the supply flow and the dynamic dispersion coefficient.

5. The medical film quality detection system based on multi-parameter fusion according to claim 4, characterized in that, The identification unit includes: The contribution determination subunit is used to determine the concentration contribution, humidity contribution, velocity contribution, and flow contribution based on the ratios of the concentration correlation, humidity correlation, velocity correlation, and flow correlation to the overall correlation, respectively, wherein the overall correlation is determined based on the sum of the concentration correlation, humidity correlation, velocity correlation, and flow correlation. The identification subunit is used to determine the primary root cause parameter as the liquid concentration, the relative humidity difference, the transmission speed ratio, or the supply flow rate based on the threshold comparison results of the concentration contribution, the humidity contribution, the velocity contribution, and the flow rate contribution.

6. The medical film quality detection system based on multi-parameter fusion according to claim 5, characterized in that, The risk determination module includes: The deviation calculation unit is used to calculate the relative deviation between the dynamic discrete coefficient and the preset discrete threshold to obtain the discrete deviation degree. An index calculation unit is used to determine the degree of parameter runaway based on the time-series changes of the primary root cause parameters; A risk determination unit is used to determine the process risk index based on the dispersion deviation and the parameter runaway degree.

7. The medical film quality detection system based on multi-parameter fusion according to claim 6, characterized in that, The index calculation unit includes: A concentration determination subunit is used to determine the degree of loss of control of the primary root cause parameter based on the statistical characteristics of the degree to which the liquid concentration deviates from its threshold within a preset identification time when the primary root cause parameter is the liquid concentration. A humidity determination subunit is used to determine the degree of parameter runaway based on the fluctuation characteristics and instantaneous change characteristics of the relative humidity difference within the preset identification time when the primary root cause parameter is the relative humidity difference. A speed determination subunit is used to determine the degree of runaway of the parameter based on the instantaneous change characteristics of the transmission speed ratio within the preset identification time when the primary root cause parameter is the transmission speed ratio. The flow determination subunit is used to determine the degree of parameter runaway based on the correlation characteristics and temporal fluctuation characteristics of the dynamic discrete coefficient of the supply flow within the preset identification time when the primary root cause parameter is the supply flow.

8. The medical film quality detection system based on multi-parameter fusion according to claim 7, characterized in that, The adjustment module includes: The adjustment determination unit is used to determine whether the process needs to be adjusted based on the threshold comparison result of the process risk index, and when the primary root cause parameter is the liquid concentration, the liquid concentration is adjusted according to the theoretical adjustment amount, wherein the theoretical adjustment amount is determined based on the preset concentration threshold and the parameter runaway degree. A humidity adjustment unit is used to adjust the relative humidity difference based on the determination result of process needs, when the primary root cause parameter is the relative humidity difference, according to the humidity adjustment amount, wherein the humidity adjustment amount is determined based on the process risk index and the preset risk threshold. A speed adjustment unit is used to adjust the transmission speed ratio according to the speed adjustment amount when the primary root cause parameter is the transmission speed ratio, based on the determination result of the process requirement adjustment. The speed adjustment amount is determined based on the process risk index and the preset risk threshold. A flow adjustment unit is used to adjust the supply flow based on the determination result of process needs, when the primary root cause parameter is the supply flow, according to the flow adjustment amount, wherein the flow adjustment amount is determined based on the process risk index and the preset risk threshold.

9. The medical film quality inspection system based on multi-parameter fusion according to claim 8, characterized in that, The early warning module includes: An efficiency determination unit is used to determine the adjustment efficiency index based on the frequency of issuing the adjustment command, the dynamic dispersion coefficient, and the changing trend of the process risk index. An early warning unit is used to issue an early warning based on the primary root cause parameters when the adjustment efficiency index is greater than a preset index threshold.

10. The medical film quality inspection system based on multi-parameter fusion according to claim 9, characterized in that, The performance determination unit includes: The frequency determination subunit is used to determine the frequency factor based on the ratio of the frequency of issuing the adjustment command to a preset frequency threshold. The improvement determination subunit determines the degree of dispersion improvement and the degree of risk improvement based on the difference between the average value of the dynamic dispersion coefficients before and after the adjustment instruction is issued and the degree of difference between the process risk index. The efficiency determination subunit is used to determine the adjusted efficiency index based on the weighted fusion result of the frequency factor, the discrete improvement degree, and the risk improvement degree.

Images