A method of monitoring the quality of a tobacco primary processing

Abstract

This invention belongs to the field of tobacco production control technology, specifically relating to a method for quality monitoring in the tobacco processing process. The method includes the following steps: collecting historical process data from multiple batches to obtain process data in a stable state; fitting the process data using an autoregressive moving average model; calculating the residual sequence of the model and testing the autocorrelation and partial correlation of the residuals; if neither is significant, the model fits the data as required; constructing a residual control chart by monitoring the residuals; extending the control limits of the residual control chart to convert it into a control chart for general control, and calculating the residuals at that time; if the residuals are within a set range, the process is considered stable; if they exceed the set range, the process is considered abnormal. This technical solution, by constructing a time series control chart for monitoring the tobacco processing process, can effectively monitor both significant step shifts and slower-changing trend shifts simultaneously.

Description

Technical Field

[0001] This invention belongs to the field of tobacco production control technology, and specifically relates to a method for quality monitoring in the tobacco processing process. Background Technology

[0002] Due to the effective application of automatic control and sensor technologies, the cigarette manufacturing process exhibits characteristics such as large-scale production, rapid pace, and continuity. This results in significant autocorrelation in process monitoring data, making it difficult to effectively apply traditional Statistical Process Control (SPC) methods. Summary of the Invention

[0003] The purpose of this invention is to provide a method for quality monitoring in the tobacco processing process, so as to solve the problems existing in the background art.

[0004] To achieve the above objectives, this application employs the following technical solution:

[0005] A method for quality monitoring in the tobacco processing process includes the following steps:

[0006] S1. Collect multiple batches of historical process data to obtain process data in a stable state;

[0007] S2. The process data are fitted using an autoregressive moving average model;

[0008] S3. Calculate the residual sequence of the model and test the autocorrelation and partial correlation of the residuals. If neither is significant, then the model fits the data as required.

[0009] S4. Construct residual control charts by monitoring residuals;

[0010] S5. Extend the control limits of the residual control chart to convert it into a control chart for control purposes, and calculate the residual at that time. If the residual is within the set range, it indicates that the process is in a stable state. If it exceeds the set range, it is considered that the process is abnormal.

[0011] Furthermore, in step S1, for collecting multiple batches of process historical data, it is necessary to delete the unstable period data of each batch of process historical data.

[0012] Furthermore, in step S2, to avoid the problem of inaccurate determination of the order by observing the autocorrelation plot and partial autocorrelation plot, 16 models from ARMA(1,1) to ARMA(4,4) are tried. Minitab is used to obtain the mean square error (MS) value of each fitting attempt. The model with the smallest MS value is the optimal model, denoted by ARMA(p,q), where p and q represent the autoregression order and the moving average order, respectively.

[0013] Furthermore, to address insufficient sensitivity to small or trending process deviations, the following steps are employed:

[0014] S5. Monitor using the time-series-based ARMA control chart method;

[0015] S6. Obtain the control limits of the time series control chart;

[0016] S7. Extend the control limits of the control chart and convert it into a control chart for control purposes. Obtain the latest data for the process at the latest moment and calculate the monitoring statistics of the time series control chart at that moment. If the monitoring statistics are within the set range, the process is considered to be in a stable state; otherwise, the control chart alarms.

[0017] Furthermore, monitoring statistics Represented as: ,

[0018] in, and These are the parameters for the time series control chart. , .

[0019] Furthermore, and This will directly affect the calculation results of the statistics and variance of the statistics. A search algorithm is adopted to select reasonable parameter values ​​and determine the signal-to-noise ratio of short-term and long-term abnormal offsets.

[0020] Furthermore, this also includes using genetic algorithms in and The calculation is performed within the range of values ​​to obtain the maximum signal-to-noise ratio (SNR) for short-term abnormal offsets. The maximum SNR value is then assigned to the value corresponding to... and Used as parameters for time series control charts.

[0021] Furthermore, if the signal-to-noise ratio (SNR) of short-term anomalous offsets is less than a set value, a genetic algorithm is used to calculate the sum that maximizes the SNR of long-term anomalous offsets. The optimal parameter combination is the combination of parameters that is found in the first parameter combination.

[0022] The beneficial effects of this invention are:

[0023] This technical solution constructs a time series control chart for monitoring the yarn-making process, which can effectively monitor both significant step shifts and slower trend shifts in the process. Attached Figure Description

[0024] Figure 1This is a schematic diagram of the autocorrelation and partial correlation functions of the residual sequence.

[0025] Figure 2 The residual control chart is constructed.

[0026] Figure 3 A single-value control chart for predicting residuals.

[0027] Figure 4 This is a time series control chart.

[0028] Figure 5 This is a control chart used for this control. Detailed Implementation

[0029] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings. The following embodiments are merely exemplary and can only be used to explain and illustrate the technical solution of the present invention, and should not be construed as limiting the technical solution of the present invention.

[0030] like Figures 1 to 5 As shown, this application provides a method for quality monitoring in the tobacco processing process, comprising the following steps:

[0031] S1. Collect multiple batches of historical process data, delete data from unstable periods such as material start-up, material finish-up, and material interruption, and obtain process data in a stable state. Taking the Hongqiqu brand cigarette sheet drying process in Anyang Factory as an example, collect 3 batches of production data, remove non-steady-state process data, and obtain 5000 steady-state data. Descriptive statistics show that the mean and standard deviation are as follows: , .

[0032] S2. An Autoregressive Moving Average (ARMA) model was used to fit the process data. Since the order of the time series was unknown, to avoid the inaccuracy of determining the order by observing autocorrelation and partial autocorrelation plots, 16 models from ARMA(1,1) to ARMA(4,4) were tried. Minitab was used to obtain the mean squared error (MS) value for each fitting attempt. The model with the smallest MS value was the optimal model, denoted as ARMA(p,q), where p and q represent the autoregressive order and the moving average order, respectively. Comparing the MS values ​​of different models fitted to the thin plate drying data, the minimum value was found when using ARMA(2,1), as shown in the table below.

[0033] Comprehensive Autoregressive Moving Average (ARIMA) Model: Export Moisture Content (Final Estimation of Parameters)

[0034]

[0035] Therefore, the ARMA(2,1) model can be represented as:

[0036] (1).

[0037] S3. Calculate the residual sequence of the model and test the autocorrelation and partial correlation of the residuals. If neither is significant, the model is considered to fit the data well, and its residuals are a white noise sequence, denoted as S3. The autocorrelation and partial autocorrelation functions of the residual series are shown in [reference needed]. Figure 1 As shown, both autocorrelation and partial correlation are insignificant within the 95% confidence interval, indicating that the residuals are a white noise sequence. The mean and standard deviation of the residuals are -0.000038 and 0.013836, respectively, and therefore can be approximated as... .

[0038] S4. Since the random variables at any two moments in the white noise sequence are independent and both follow a normal distribution. Therefore, a control chart can be constructed to determine whether the process is a stationary time series by monitoring the residuals. The variance of the residuals is then calculated. This allows for the construction of residual control charts for analysis:

[0039] (2)

[0040] The variance of the residuals is 0.0138, therefore the residual control limits are: The constructed residual control chart is as follows: Figure 2 As shown.

[0041] S5. Extend the control limits of the control chart, convert it into a control chart for control purposes, and obtain the latest time data for this process. The predicted value at that moment is calculated using the fitted ARMA(p,q) model. Calculate the residual at that moment. ,like exist If the control limits are within the specified range, it indicates that the process is in a stable state. If the control limits are exceeded, the control chart will alarm, indicating that the process is abnormal. Figure 3 As shown. Figure 3 The result indicates that the initial performance of the new batch in this process was good, with no abnormal fluctuations. However, the control chart exceeded the control limits at point 167, indicating that an abnormality occurred in the process at this point.

[0042] S6. While residual control charts are effective at monitoring large abnormal fluctuations in time series processes, they are often insufficiently sensitive to smaller or trend-based process deviations. Therefore, further monitoring using time series-based ARMA control charts is necessary. Its monitoring statistics... It can be represented as:

[0043] (3)

[0044] in, and These are the parameters for the time series control chart. , Substituting the expression for the ARMA(2,1) model in equation (1) into equation (3) yields the monitoring statistic for each data point. The variance of this statistic under stationary conditions is:

[0045] (4)

[0046] in, Representing stationary time series of The first-order autocorrelation coefficient can be calculated using the following formula:

[0047] (5)

[0048] in, This represents the total number of samples obtained. For example, based on the 5000 steady-state data obtained in the first step, i.e. ,Pick It can be calculated that: .

[0049] S7. In equation (3), and This directly affects the calculation results of the statistic and its variance; a search algorithm can be used to select appropriate parameter values. Similar to conventional control charts, large abnormal fluctuations, such as... The residual control chart of equation (2) is often easily monitored and identified. Therefore, the ARMA control chart monitors smaller abnormal fluctuations. Here, its monitoring range is defined as follows: When an anomalous fluctuation occurs, it causes short-term and long-term process shifts in the time series process. That is, the process experiences large short-term anomalous fluctuations, but these are quickly masked by the autocorrelation of the time series, leaving only a small long-term fluctuation. Therefore, our primary goal is to detect this anomaly promptly when a large short-term fluctuation occurs. Let the long-term process shift be... , by monitoring statistics As can be seen from the formula, the short-term process offset can be approximated as... Therefore, the signal-to-noise ratios (SNRs) of short-term and long-term anomalous offsets can be defined as follows:

[0050] , (6)

[0051] The larger abnormal fluctuations here are Therefore, it is advisable to first select a suitable one. and Combining improves the signal-to-noise ratio of short-term anomalous shifts. This is large enough that the following optimization design model can be established:

[0052] (7)

[0053] Substituting equation (4) into the model and using a genetic algorithm... and The calculation is performed within the range of values ​​to obtain... The maximum value is 2.973. If If the maximum value is greater than 4, it is generally believed that time series control charts can quickly detect short-term abnormal fluctuations and issue alarms. The maximum value corresponding to and These serve as suitable parameter values ​​for time series control charts.

[0054] S8. If the signal-to-noise ratio of short-term abnormal offsets If the value is less than 4, there is a possibility that the anomalous offset may not be detected quickly. For example, in this case, the maximum value is 2.973, requiring further consideration of the ability to identify long-term anomalous offsets. This is based on the signal-to-noise ratio of long-term anomalous offsets. Consider whether smaller anomalous offsets can be detected more quickly, i.e. Choose the appropriate and , making To maximize its size, the following optimization design model can be established:

[0055] ;

[0056] The same genetic algorithm is used to calculate and find the result that... The largest and The optimal parameter combination is the combination of parameters found when... (The calculation is incomplete and requires further context.) , hour, The value is the largest. Therefore, , As the optimal parameter combination for time series control charts, it can be obtained from equation (3). 0.644, And monitoring statistics From equation (4), we can obtain that... 0.0077, that is .

[0057] S9. Thus, the control limits of the time series control chart are obtained. .in, For monitoring statistics The mean of , as can be seen from equation (3), . To control the limit parameters, following the design principles of conventional control charts, a value can typically be taken as... ,Right now . Figure 4 The analytical control chart, constructed using steady-state data, shows the observed values. If the data is within the upper and lower control limits, it indicates process stability, and this chart can be used to monitor newly acquired production data. To increase the sensitivity of the control chart, the number of... The value can be adjusted, but this will increase the probability of false alarms.

[0058] S10. Extend the control limits of the control chart, convert it into a control chart for control purposes, and obtain the latest time data for this process. Calculate the time series control chart monitoring statistics at that moment. ,like In If the flow rate is within the acceptable range, the process is considered to be in a stable state without abnormal fluctuations; otherwise, the control chart will issue an alarm, indicating that some abnormal deviation has occurred in the process. Figure 5 The monitoring image for the new batch shows that the initial performance of the new batch in this process is good, with no abnormal fluctuations. However, starting at point 68, a gradual upward shift occurs, and the control chart exceeds the control limit at point 83, indicating that this abnormal fluctuation is detected by the control chart.

[0059] Using the calculation obtained from step S7 or step S8 and By combining the parameter combinations with the control limits obtained in step S9, a time series control chart can be constructed. Using this control chart and the residual control chart constructed in step S5 simultaneously for monitoring the same yarn-making process allows for the effective monitoring of both significant step shifts and slower-changing trend shifts.

[0060] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method of monitoring the quality of a tobacco primary processing process, characterised in that, Includes the following steps: S1. Collect multiple batches of historical process data to obtain process data in a stable state; S2. The process data are fitted using an autoregressive moving average model; S3. Calculate the residual sequence of the model and test the autocorrelation and partial correlation of the residuals. If neither is significant, then the model fits the data as required. S4. Construct residual control charts by monitoring residuals; S5. Extend the control limits of the residual control chart to convert it into a control chart for control purposes, and calculate the residual at that time. If the residual is within the set range, it indicates that the process is in a stable state. If it exceeds the set range, it is considered that the process is abnormal.

2. A tobacco primary processing quality monitoring method according to claim 1, characterised in that, In step S1, for collecting multiple batches of historical process data, it is necessary to delete the unstable period data of each batch of historical process data.

3. The tobacco primary processing quality monitoring method according to claim 1, characterized in that, In step S2, to avoid the problem of inaccurate determination of the order by observing the autocorrelation plot and partial autocorrelation plot, 16 models from ARMA(1,1) to ARMA(4,4) are tried. Minitab is used to obtain the mean square error (MS) value of each fitting attempt. The model with the smallest MS value is the optimal model, denoted by ARMA(p,q), where p and q represent the autoregression order and the moving average order, respectively.

4. The tobacco primary processing quality monitoring method according to claim 1, characterized by, It also includes the following steps to address insufficient sensitivity to small or trending process deviations: S5. Monitor using the time-series-based ARMA control chart method; S6. Obtain the control limits of the time series control chart; S7. Extend the control limits of the control chart and convert it into a control chart for control purposes. Obtain the latest data for the process at the latest moment and calculate the monitoring statistics of the time series control chart at that moment. If the monitoring statistics are within the set range, the process is considered to be in a stable state; otherwise, the control chart alarms.

5. A tobacco primary processing quality monitoring method according to claim 4, characterised in that, Monitoring statistics is represented as: ​ wherein and are parameters of a time series control chart, , .

6. A tobacco primary processing quality monitoring method according to claim 5, characterised in that, and The search algorithm is adopted to select reasonable parameter values and determine the signal-to-noise ratio of short-term abnormal offsets and the signal-to-noise ratio of long-term abnormal offsets, which will directly affect the calculation results of the statistical quantity and the statistical quantity variance.

7. A tobacco primary processing quality monitoring method according to claim 6, characterised in that, Also included is the use of a genetic algorithm in and is iteratively calculated within the range of values of and as time series control chart parameter values.

8. A tobacco primary processing quality monitoring method according to claim 7, characterised in that, If the signal-to-noise ratio of the short-term abnormal deviation is less than a set value, a genetic algorithm is used to calculate the parameter combination that maximizes the signal-to-noise ratio of the long-term abnormal deviation is the optimal parameter combination.

Images