Closed-loop quality control method and system for cable production process

Abstract

The invention relates to the technical field of quality control, in particular to a closed-loop quality control method and system for a cable production process. The method comprises the following steps: acquiring a quality parameter sequence, establishing a time sequence model and decomposing the time sequence model into a variation component and a residual component; calculating a dynamic control limit based on weighted fusion of the residual historical variance and the auxiliary environment variable variance; judging whether the process is out of control or not by calculating divergence between a real-time residual probability density function and a reference function; and when the residual error sequence is out of control, identifying abnormal modes such as a single-point pulse type or a step response type presented by the residual error sequence, matching a reference adjustment strategy, calculating adjustment intensity according to the divergence exceeding degree, and generating a final process parameter adjustment quantity. According to the scheme of the invention, the problem of high false alarm rate caused by data autocorrelation is solved, early weak offset can be sensitively detected, and rapid and accurate automatic quality correction is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of quality control. More particularly, the present application relates to a closed-loop quality control method and system for cable production process. BACKGROUND

[0002] In the cable production process, product quality is easily affected by many factors such as raw material properties, equipment status, environmental temperature and humidity, and process parameter settings. In order to ensure the stability of product quality, enterprises usually use statistical process control (SPC) technology to monitor the production process in real time. Among them, the traditional SPC method represented by Shewhart control chart has been widely used in industry due to its simple principle and easy implementation. However, one of the core premises of this method is that the process data should meet the classical assumption of independent and identical distribution. However, in continuous cable production, due to the inertia of material flow and heat transfer processes, the correlation between the sequences of quality characteristic parameters collected continuously is significant, which directly violates the theoretical basis of traditional control charts. Therefore, the fixed control limit of traditional control charts cannot adapt to the normal process fluctuations caused by raw material batch replacement or environmental changes, resulting in a significant decrease in applicability and monitoring sensitivity in complex dynamic environments.

[0003] In order to solve the problem of the failure of the traditional control chart due to data autocorrelation, the existing technology proposes a method of combining time series model with traditional control chart. The basic idea is: first, establish a time series model to fit and describe the dynamic characteristics of the process data, thereby effectively filtering out the autocorrelation components in the data, and then monitor the residual sequence after model fitting. Since the residual sequence approximately meets the assumption of independent and identical distribution, traditional control charts can be applied for analysis. This method to some extent solves the problem of data autocorrelation, making the control chart applicable to more industrial scenarios.

[0004] Although the above improvement scheme has achieved certain effect, there are still obvious deficiencies in actual application. First of all, the order of the time series model is usually identified offline and fixed. When the dynamic characteristics of the production process change due to changes in working conditions, the fixed model will be mismatched, thereby affecting the accuracy of monitoring. Secondly, in the identification of process out-of-control state, these methods still mostly rely on single-point over-limit or simple operating rules, which are not sensitive to small shifts in the overall distribution pattern of the process, and may miss the best opportunity to detect early abnormalities. Moreover, most existing quality control methods stop at the monitoring and alarm level, which is an open-loop control method. When the system detects process abnormalities, manual intervention is usually required to adjust the process parameters based on experience. This approach not only has a lagging response, but also has uneven adjustment accuracy, which is difficult to meet the strict demands of modern industry for high-quality and high-efficiency production. SUMMARY

[0005] The present application aims to provide a closed-loop quality control method and system for cable production process, to solve the problem of response lag and poor adjustment accuracy in the prior art; therefore, the present application provides solutions in the following two aspects.

[0006] In the first aspect, the present application provides a closed-loop quality control method for cable production process, comprising the following steps: obtaining a real-time quality characteristic parameter sequence of the cable production process; establishing a time series model for the parameter sequence, and decomposing the time series model into a variation component and a residual component; when a variance change rate of the residual component within a preset time window exceeds a first preset threshold, re-identifying and updating an order of the time series model; calculating a control upper limit and a control lower limit by weighted fusion based on a historical statistical variance of the residual component and a variance of an auxiliary environmental variable associated with the production process within a current time window; establishing a residual component reference probability density function under a steady-state process, obtaining a real-time residual component sequence by a sliding window method and calculating a current probability density function, and judging whether the production process is out of control by calculating a divergence between the current probability density function and the reference probability density function; when the divergence is greater than a second preset threshold, triggering a closed-loop adjustment, identifying an abnormal pattern presented by a residual component sequence before triggering, matching and determining a corresponding reference adjustment strategy from a preset strategy library containing a single-point pulse type and a step response type adjustment strategy, calculating an adjustment intensity coefficient according to a degree that the divergence exceeds the second preset threshold, and generating a final process parameter adjustment amount to adjust the quality characteristic parameter by combining the reference adjustment strategy and the adjustment intensity coefficient.

[0007] Preferably, the auxiliary environmental variable includes a raw material melt flow rate or an environmental temperature and humidity.

[0008] Preferably, the re-identification and update of the order of the time series model comprises: using an autoregressive moving average model as the time series model; setting a length of the preset time window as 100 sampling points, calculating a residual component variance within a current time window and a residual component variance within a previous time window, and defining a change rate as a ratio of an absolute value of a difference between the two variances to the variance of the previous time window; when the change rate exceeds the first preset threshold, re-determining an autoregressive order and a moving average order of the time series model by Akaike information criterion.

[0009] Preferably, the calculation of the control upper limit and the control lower limit by weighted fusion comprises: using a formula = ×​​​ + × Calculate dynamic variance ,in The statistical variance of the residual components is collected from no less than 500 historical samples. The variance of the ambient temperature within the sliding window used to calculate the current probability density function. , The weighting coefficient; the upper limit of control and control lower limit Based on dynamic standard deviation Confirm, set as and .

[0010] Preferably, the determination of whether the production process is out of control includes: using a Gaussian kernel function to perform kernel density estimation, establishing the baseline probability density function based on 1000 steady-state residual component samples from the production process startup phase; setting a sliding window length of 50 sample points, and calculating the current probability density function using samples within the window; when the calculated... When the divergence value is greater than the second preset threshold, the production process is judged to be out of control.

[0011] Preferably, the identification of abnormal patterns presented by the residual component sequence before the trigger includes: detecting the residual component sequence composed of the 20 data points before the trigger alarm time; when there is an isolated data point in the sequence whose absolute value exceeds 3 times the dynamic standard deviation, and the two adjacent points are both within 1 times the dynamic standard deviation, it is identified as a single-point pulse mode; when there are 5 consecutive data points in the sequence whose mean value deviates from the mean value of the first 10 data points in the sequence by more than 1.5 times the dynamic standard deviation, it is identified as a step mode.

[0012] Preferably, the matching and determining of the corresponding reference adjustment strategy includes: when the identified abnormal mode is a single-point pulse mode, the matched reference adjustment strategy is to apply a pulse adjustment to the traction speed with the opposite sign to the residual component and a duration of 1 sampling period; when the identified abnormal mode is a step mode, the matched reference adjustment strategy is to apply a step-type continuous adjustment to the extruder screw speed with the opposite offset direction to the residual component.

[0013] Preferably, the calculation of the adjustment intensity coefficient includes: calculating the adjustment intensity coefficient using the linear gain formula, wherein the calculation formula is: ,in, To adjust the strength coefficient, For the currently calculated divergence value, The second preset threshold is set; and the following settings are made: The maximum value is 3.0. When the calculated value exceeds 3.0, it is taken as 3.0.

[0014] Preferably, the step of combining the baseline adjustment strategy and the adjustment intensity coefficient to generate the final process parameter adjustment amount includes: for a single-point pulse adjustment strategy, setting the baseline adjustment amount as a change in traction speed of 0.1 m / min, and the final adjustment amount as... m / min; For the step-response adjustment strategy, the baseline adjustment is set to a change of 0.5 rpm in the extruder screw speed, and the final adjustment is rpm; apply the final adjustment amount to the corresponding process equipment actuator.

[0015] In the second aspect, a closed-loop quality control system for a cable manufacturing process includes: The processor; the memory storing computer instructions for closed-loop quality control of the cable production process, which, when executed by the processor, cause the system to perform the aforementioned closed-loop quality control method for the cable production process.

[0016] The beneficial effects of this invention are as follows: First, by establishing and dynamically updating a time series model to monitor residuals, this invention fundamentally solves the problem of high false alarm rate caused by data autocorrelation in traditional SPC control charts; second, it employs... Divergence is used to measure the difference between real-time data distribution and baseline distribution, enabling sensitive detection of early and subtle process deviations, thus improving the timeliness and accuracy of early warnings. Moreover, this invention constructs a complete intelligent closed loop from anomaly detection and pattern recognition to adaptive adjustment. It can not only automatically diagnose specific fault modes such as single-point pulses or continuous step jumps, but also dynamically calculate the adjustment intensity according to the severity of the loss of control, thereby achieving rapid, accurate, and automated quality correction without human intervention, significantly improving the stability of the cable production process and the consistency of product quality. Attached Figure Description

[0017] Figure 1 This schematically illustrates the steps of the closed-loop quality control method for the cable production process in this embodiment; Figure 2 The schematic diagram illustrates the structure of the closed-loop quality control system for the cable production process in this embodiment. Detailed Implementation

[0018] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0019] like Figure 1 As shown in this embodiment, a closed-loop quality control method for a cable production process includes the following steps: Step S1, obtaining a real-time quality characteristic parameter sequence of a cable production process; establishing a time series model for the parameter sequence, and decomposing the time series model into a variation component and a residual component; when a variance change rate of the residual component within a preset time window exceeds a first preset threshold, re-identifying and updating an order of the time series model.

[0020] Specifically, first, a real-time insulating outer diameter of the cable is continuously obtained through a laser diameter gauge installed after a cooling section of a cable extrusion production line, and the measurement values are transmitted to an industrial control computer in time sequence to form a raw quality characteristic parameter time series. Then, a time series model is established for the parameter time series. Through the above model, the raw sequence is decomposed into a variation component composed of model predicted values, and a residual component composed of differences between actual measurement values and predicted values.

[0021] A preset time window is set with a length of N=100 sampling points, and the variance of the residual component sequence within the current window and the variance of the last time window are continuously calculated. When the variance change rate exceeds a first preset threshold, the system determines that the potential characteristics of the production process have changed significantly, and the existing model is no longer applicable. At this time, the model updating mechanism is triggered: the system automatically intercepts the latest M=500 data points, and uses Akaike information criterion (AIC) or Bayesian information criterion (BIC) method to traverse different model order combinations to re-identify and determine the model order that can best fit the current data, and finally uses the new order to update the time series model to ensure its description and prediction accuracy for the current production state. In the embodiment, the first preset threshold is set to 0.3.

[0022] In an optional embodiment, re-identifying and updating the order of the time series model includes: using an autoregressive moving average model as the time series model; setting the length of the preset time window to 100 sampling points, calculating the residual component variance within the current time window and the residual component variance within the previous time window, and the change rate is defined as the ratio of the absolute value of the difference to the variance of the previous time window; when the change rate exceeds the first preset threshold, the autoregressive order and the moving average order of the time series model are re-determined using Akaike information criterion.

[0023] Specifically, the collected continuous residual data is divided into a plurality of non-overlapping time windows, each window containing 100 data points. The residual variance of the previous time window and the residual variance of the current latest time window are calculated, and the change rate of the two variances is obtained accordingly. Exemplarily, when the calculated change rate is greater than a preset first threshold value, it indicates that the characteristics of the production process may have changed, and the existing autoregressive moving average model is no longer applicable, and the model updating program will be automatically started. By traversing a series of possible combinations of autoregressive order and moving average order, the Akaike information criterion is used to find the optimal model order for the current production data. Select the pair of autoregressive order and moving average order value that makes the Akaike information criterion function value minimum, and construct a new autoregressive moving average model to represent and predict the behavior of the current process.

[0024] Step S2, based on the historical statistical variance of the residual component and the variance of the auxiliary environmental variable associated with the production process within the current time window, the upper and lower control limits are calculated by weighted fusion. The auxiliary environmental variables include raw material melt flow rate or environmental temperature and humidity.

[0025] Specifically, in the stable operation state of the production line, the historical residual data is collected, and the variance is calculated as the reference residual variance. The environmental temperature of the production workshop is monitored in real time by a temperature and humidity sensor, and the variance of the temperature readings within the current time window is calculated.

[0026] The upper and lower control limits are calculated by weighted fusion, including: The dynamic variance is calculated by the formula = × + × , wherein is the statistical variance of the residual component collected from not less than 500 historical samples, is the variance of the environmental temperature within the sliding window time window used to calculate the current probability density function, , is the weight coefficient; The upper control limit and the lower control limit are determined according to the dynamic standard deviation , and are set to and .

[0027] ​Specifically, based on a historical stable production data set containing more than 500 data points, the long-term statistical variance of the residual component is calculated. The ambient temperature readings in the current sliding window time window are monitored, and the variance reflecting the volatility of the current environment is calculated. Then, the above weighting formula is used to fuse the two variances. In this embodiment, the residual variance weight representing the intrinsic process fluctuation is set to 0.7, the temperature variance weight representing the external environmental influence is set to 0.3, and the dynamic variance can be calculated. Then, the value of the dynamic standard deviation is obtained according to the calculation formula of the dynamic standard deviation, and the control upper limit value and the control lower limit value are obtained accordingly.

[0028] Step S3, establishing a residual component reference probability density function under a stable state process, obtaining a real-time residual component sequence by using a sliding window method and calculating a current probability density function, and judging whether the production process is out of control by calculating the divergence between the current probability density function and the reference probability density function.

[0029] Specifically, a large number of residual samples generated under normal production state are collected, and a reference probability density function is established for these residual samples by using kernel density estimation method, which can accurately depict the distribution form of residual data under ideal state. In continuous generation monitoring, a fixed width sliding window is maintained, which is used to store the latest real-time residual sequence. When a new residual is generated, the sliding window slides forward, and the kernel density estimation method is used again for the residual data in the window. A current probability density function is calculated for the real-time data in the window.

[0030] The divergence value of the reference probability density function and the current probability density function is calculated to accurately quantify the difference between the two probability distributions. The calculated divergence value is compared with a second preset threshold value, and if the divergence value is greater than the second preset threshold value, it indicates that the overall distribution of the current production process has deviated significantly from the ideal state, and the system determines that the production process is out of control and triggers the subsequent adjustment action.

[0031] In an optional embodiment, judging whether the production process is out of control comprises: using Gaussian kernel function for kernel density estimation, establishing the reference probability density function based on 1000 stable residual component samples in the production process starting stage, setting the sliding window length to 50 sample points, and calculating the current probability density function by using the samples in the window; and when the calculated divergence value is greater than the second preset threshold value, judging that the production process is out of control.

[0032] ​​​​​​Specifically, after production is initialized and operation is stable, 1,000 residual component data points are collected. Using the Gaussian kernel density estimation method, a baseline probability density function curve representing the ideal production state is generated. This baseline probability density function will be stored as the gold standard for subsequent comparisons.

[0033] Next, during continuous production monitoring, a sliding window with a length of 50 sample points is maintained to capture the latest residual data in real time. Whenever a new data point is generated, the window is updated, and Gaussian kernel density estimation is performed on the 50 sample points within the window in real time to generate a current probability density function that reflects the current process state.

[0034] Then, the relationship between the current probability density function and the baseline probability density function is calculated. Divergence. For example, the divergence calculated at a certain moment. The divergence value is 0.15. If this value is greater than the preset second threshold, it is determined that the current production process has gone out of control and an alarm is triggered immediately.

[0035] Step S4, when the When the divergence exceeds a second preset threshold, closed-loop regulation is triggered, and an abnormal pattern is identified in the residual component sequence before triggering. Based on the abnormal pattern, a corresponding benchmark regulation strategy is matched and determined from a preset strategy library containing single-point pulse and step response regulation strategies. The degree to which the divergence exceeds the second preset threshold is used to calculate the adjustment intensity coefficient; combined with the benchmark adjustment strategy and the adjustment intensity coefficient, the final process parameter adjustment amount is generated to adjust the quality characteristic parameter.

[0036] Specifically, once calculated If the divergence value exceeds the second preset threshold, the system triggers the closed-loop adjustment program. First, the residual sequence within the sliding window before the alarm is triggered is analyzed to identify abnormal patterns: if the value of a single residual point in the sequence is far beyond the control limit, it is identified as a single-point pulse anomaly; if multiple consecutive data points are found to continuously deviate from one side of the center line, it is identified as a step anomaly.

[0037] Next, based on the identified anomaly pattern, a corresponding baseline adjustment strategy is matched and determined from a pre-defined strategy library. This strategy library is predefined as follows: for single-point pulse anomalies, the matched baseline adjustment strategy is to temporarily and slightly adjust the traction machine speed; for step anomalies, the matched baseline adjustment strategy is to continuously adjust the extruder screw speed.

[0038] Subsequently, to achieve precise adjustment, an adjustment intensity coefficient is calculated, which is based on the current... The degree of divergence beyond the second preset threshold is determined. For example, the adjustment strength is set to be proportional to the degree of process deviation.

[0039] Finally, the reference adjustment strategy is combined with the adjustment strength coefficient to generate the final process parameter adjustment amount. For example, if the matched reference adjustment strategy is to reduce the screw speed by 5 r / min, and the calculated adjustment strength coefficient is 0.6, the final process parameter adjustment amount is 3 r / min, a control instruction of "reduce the screw speed by 3 r / min" is generated, and the instruction is automatically issued to the control system of the extruder for execution, thereby completing the entire closed-loop adjustment.

[0040] In an optional embodiment, identifying the abnormal pattern exhibited by the residual component sequence before triggering includes: detecting the residual component sequence composed of 20 data points before the alarm triggering moment; when there is an isolated data point in the sequence, whose absolute value exceeds 3 times the dynamic standard deviation, and the adjacent two points before and after it are within 1 times the dynamic standard deviation, it is identified as a single-point pulse mode; when there are 5 consecutive data points in the sequence, whose mean value deviates from the mean value of the previous 10 data points by more than 1.5 times the dynamic standard deviation, it is identified as a step mode.

[0041] For example, when the out-of-control alarm is triggered, the 20 residual data points before the alarm occurs are retrieved for analysis. Assuming that the dynamic standard deviation at this time is 0.1, 3 times the standard deviation is 0.3. Check if there is a single-point pulse mode, scan the sequence point by point, if the value of the 15th data point is greater than 0.3, and the absolute values of the previous and next points are within 0.1, this abnormality is identified as a single-point pulse mode. If no single-point pulse mode is detected, it will continue to check if there is a step mode, which calculates the mean value of the first 10 data points in the 20-point sequence, and then calculates the mean value in groups of 5 data points in the sequence. If the difference between the mean value of the 13th to 17th five consecutive data points in the sequence and the mean value of the first 10 data points in the sequence exceeds 0.15 (1.5 times the dynamic standard deviation), this abnormality is identified as a step mode.

[0042] In an optional embodiment, matching and determining the corresponding reference adjustment strategy include: When the identified abnormal pattern is a single-point pulse mode, the matched reference adjustment strategy is to apply a pulse adjustment to the traction speed with a duration of 1 sampling period, which is opposite in sign to the residual component; when the identified abnormal pattern is a step mode, the matched reference adjustment strategy is to apply a stepwise sustained adjustment to the screw speed of the extruder, which is opposite in direction to the residual component deviation.

[0043] If the abnormal pattern identified in the previous step is a single-point pulse pattern, and the residual value of that pulse point is positive, it usually means that the product size is momentarily too large. In this case, the corresponding adjustment strategy library will be automatically matched, and the traction speed will be adjusted. The adjustment command is a negative pulse with the opposite sign to the residual. That is, in the next sampling period, the traction speed is briefly increased and then immediately restored to the original speed, thereby quickly thinning the product size to counteract the pulse interference.

[0044] If the identified abnormal pattern is a step pattern, and the direction of the step is positive (e.g., the mean residual value continuously increases from 0.02 to 0.18), indicating a persistently large product size, then another adjustment strategy will be applied. This involves adjusting the extruder screw speed. In this case, the adjustment command is a negative step, opposite to the direction of the residual offset, i.e., continuously reducing the extruder screw speed setpoint. The new lower speed will be maintained until process monitoring indicators show that the product size has returned to normal levels, thus correcting the persistent deviation.

[0045] In an optional embodiment, calculating the adjustment intensity coefficient includes: calculating the adjustment intensity coefficient using a linear gain formula, the formula being: , in, To adjust the strength coefficient, For the currently calculated divergence value, The second preset threshold is set; and the following settings are made: The maximum value is 3.0. When the calculated value exceeds 3.0, it is taken as 3.0.

[0046] Specifically, the second preset threshold is preferably 0.1, calculated when the runaway alarm is triggered. The actual divergence value. Substituting this value into the calculation formula for the adjustment intensity coefficient above, we can obtain the adjustment intensity coefficient, which represents the amplitude of the adjustment action. The larger the divergence value, the more serious the deviation of the process, and the greater the adjustment required.

[0047] To prevent over-adjustment leading to oscillation, an upper limit is set for the adjustment intensity coefficient; in this embodiment, it is set to 3.0. For example, in one event, if... With a divergence value of 1.1, the calculated adjustment intensity coefficient is 3.5. Since this value exceeds the set maximum value of 3.0, the adjustment intensity coefficient is adjusted accordingly. The value is forcibly set to 3.0.

[0048] In an optional embodiment, the final process parameter adjustment amount is generated by combining the reference adjustment strategy and the adjustment intensity coefficient, including: for the single-point pulse adjustment strategy, setting the reference adjustment amount as a change of 0.1 m / min in the pulling speed, and the final adjustment amount as 0.2 m / min; for the step response adjustment strategy, setting the reference adjustment amount as a change of 0.5 rpm in the screw rotation speed of the extruder, and the final adjustment amount as 1.5 rpm; and applying the final adjustment amount to the corresponding process equipment actuator.

[0049] For example, it is assumed that the single-point pulse mode is identified, and the adjustment intensity coefficient is calculated as 2.0. The reference adjustment amount of the single-point pulse mode, i.e. a change of 0.1 m / min in the pulling speed, is called, and the final adjustment amount is obtained by multiplying the reference adjustment amount by the adjustment intensity coefficient, i.e. 0.2 m / min.

[0050] In another scenario, if the step mode is identified, and the adjustment intensity coefficient is calculated as 3.0. Then the reference adjustment amount of the step mode, i.e. a change of 0.5 r / min in the screw rotation speed of the extruder, is called, and the final adjustment amount is also obtained by multiplying the reference adjustment amount by the adjustment intensity coefficient, i.e. 1.5 r / min.

[0051] After the final adjustment amount is calculated, a control instruction is generated according to the final adjustment amount, and is sent to a frequency converter or servo motor driver for controlling the screw rotation speed of the extruder through a communication interface. After the actuator receives the instruction, the rotation speed adjustment can be accurately completed.

[0052] The present application also provides a closed-loop quality control system for a cable production process. As shown in Figure 2 the system includes a processor and a memory, and the memory stores computer program instructions which, when executed by the processor, implement the closed-loop quality control method for the cable production process according to the present application.

[0053] The system also includes a communication bus and a communication interface and other components well known to those skilled in the art, the settings and functions of which are known in the art, and thus will not be described here.

[0054] In this description, the term "application" also means any computer program product storing such a program for use with or in connection with a computer system, apparatus or device. The program can be stored on any apparatus-readable medium, for example, but not limited to, any volatile memory or non-volatile memory. In this description, the term "memory" also means any computer program product storing such a program for use with or in connection with a computer system, apparatus or device. The program can be stored on any apparatus-readable medium, for example, but not limited to, any volatile memory or non-volatile memory. In this description, the term "computer-readable medium" means any tangible medium that stores, communicates, or otherwise provides data that can be used by an instruction execution system, apparatus or device. The computer-readable medium can be any suitable magnetic storage medium or magneto-optical storage medium, such as, for example, resistive random access memory (RRAM), dynamic random access memory (DRAM), static random access memory (SRAM), enhanced dynamic random access memory (EDRAM), high-bandwidth memory (HBM), hybrid memory cube (HMC), and the like, or any other medium that can be used to store the desired information and that can be accessed by an application, module, or both. Any such computer storage media can be part of the device or accessible or connectable thereto. Any application or module described in this description can be implemented by computer-readable / executable instructions stored or otherwise held by such computer-readable media.

[0055] In the description of the present description, the meaning of "a plurality of" is at least two, for example, two, three or more, and the like, unless otherwise explicitly specified.

[0056] Although the present description has shown and described a number of embodiments of the present application, it will be apparent to those skilled in the art that many modifications, changes and substitutions can be made thereto without departing from the spirit and scope of the present application.

Claims

1. A closed loop quality control method for a cable production process, characterized by, The method comprises the following steps: obtaining a real-time quality characteristic parameter sequence of a cable production process; establishing a time series model for the parameter sequence, and decomposing the time series model into a variation component and a residual component; when a variance change rate of the residual component within a preset time window exceeds a first preset threshold, re-identifying and updating an order of the time series model; based on a historical statistical variance of the residual component and a variance of an auxiliary environmental variable associated with the production process within a current time window, calculating a control upper limit and a control lower limit through weighted fusion; The residual component reference probability density function under the steady state process is established, the real-time residual component sequence is obtained by using the sliding window mode and the current probability density function is calculated, the divergence between the current probability density function and the reference probability density function is calculated, and whether the production process is out of control is judged divergence When the When the divergence exceeds a second preset threshold, closed-loop regulation is triggered, and an abnormal pattern is identified in the residual component sequence before triggering. Based on the abnormal pattern, a corresponding benchmark regulation strategy is matched and determined from a preset strategy library containing single-point pulse and step response regulation strategies. The degree to which the divergence exceeds the second preset threshold is used to calculate the adjustment intensity coefficient; combined with the benchmark adjustment strategy and the adjustment intensity coefficient, the final process parameter adjustment amount is generated to adjust the quality characteristic parameter.

2. The closed loop quality control method of a cable production process according to claim 1, characterized in that, the auxiliary environmental variable comprises a raw material melt flow rate or an environmental temperature and humidity.

3. The closed loop quality control method of a cable production process according to claim 1, characterized in that, The re-identifying and updating of the order of the time series model comprises: using an autoregressive moving average model as the time series model; setting a length of the preset time window as 100 sampling points, calculating a residual component variance within the current time window and a residual component variance within a previous time window, and defining a change rate as a ratio of an absolute value of a difference between the two variances to the variance within the previous time window; when the change rate exceeds the first preset threshold, re-determining an autoregressive order and a moving average order of the time series model using the Akaike information criterion.

4. The closed loop quality control method of a cable production process according to claim 1, characterized in that, The calculating of the control upper limit and the control lower limit through weighted fusion comprises: The dynamic variance is calculated using the formula = × + × where is the statistical variance of the residual component taken from at least 500 historical samples, is the variance of the ambient temperature in the time window used to calculate the current probability density function, , are weighting coefficients; the control upper limit and the control lower limit are determined according to the dynamic standard deviation set to and .​ 5. The closed loop quality control method of a cable production process according to claim 1, characterized in that, The judging of whether the production process is out of control comprises: The Gaussian kernel function is used for kernel density estimation, and the baseline probability density function is established based on 1000 steady-state residual component samples in the production process starting stage; the sliding window length is set to 50 sample points, and the current probability density function is calculated by using the samples in the window; when the calculated The divergence value is greater than the second preset threshold, and it is judged that the production process is out of control.

6. The closed loop quality control method of a cable production process according to claim 1, characterized in that, The identifying of an abnormal pattern presented by a residual component sequence before triggering comprises: detecting a residual component sequence composed of 20 data points before a moment of triggering an alarm; when there is an isolated data point in the sequence, and an absolute value of the data point exceeds 3 times of a dynamic standard deviation, and adjacent two points before and after the data point are within 1 times of the dynamic standard deviation, identifying a single-point pulse mode; when there are 5 continuous data points in the sequence, and a mean value of the 5 data points is offset by more than 1.5 times of the dynamic standard deviation compared with a mean value of 10 data points before the sequence, identifying a step mode.

7. The closed loop quality control method of a cable production process according to claim 6, characterized in that, The matching and determining of a corresponding reference adjustment strategy comprises: when the identified abnormal pattern is the single-point pulse mode, the matched reference adjustment strategy is to apply a pulse adjustment with a duration of 1 sampling period and a sign opposite to that of the residual component to a traction speed; when the identified abnormal pattern is the step mode, the matched reference adjustment strategy is to apply a stepwise continuous adjustment with a direction opposite to that of a residual component offset to a screw rotation speed of an extruder.

8. The closed loop quality control method of a cable production process according to claim 7, characterized in that, The calculating of an adjustment intensity coefficient comprises: calculating the adjustment intensity coefficient using a linear gain formula, and a calculation formula is: ， wherein, is the intensity coefficient, is the currently calculated divergence value, is the second preset threshold value; and setting the maximum value of 3.0, when the calculated value exceeds 3.0, taking 3.

0.

9. The closed loop quality control method of a cable production process according to claim 8, characterized in that, The generating of a final process parameter adjustment amount by combining the reference adjustment strategy and the adjustment intensity coefficient comprises: For the single-point pulse regulation strategy, the reference regulation amount is set as a change of 0.1 m / min in the pulling speed, and the final regulation amount is 0.5 m / min; for the step response regulation strategy, the reference regulation amount is set as a change of 0.5 rpm in the screw rotation speed of the extruder, and the final regulation amount is 2 rpm; the final regulation amount is applied to the corresponding process equipment actuator.

10. A closed loop quality control system for a cable production process, characterized by comprises: a processor; a memory storing computer instructions for closed-loop quality control of a cable production process, when the computer instructions are run by the processor, causing the system to perform the closed-loop quality control method of the cable production process according to any one of claims 1-9.

Images