Cylinder equipment material data feature extraction method

Abstract

The invention relates to a cylindrical equipment material data feature extraction method, and belongs to the field of tobacco processing. According to the method, a sensor is arranged at a barrel material outlet, and after material information data are collected and cleaned through the sensor, characteristic values are obtained through calculation. The method can solve the problems of uneven materials, data acquisition omission and incapability of reflecting data in an arithmetic mean value calculation mode in the production process, and can conveniently, stably and reliably detect the water content of the materials on line in real time at low cost by combining with a sensor.

Description

Technical Field

[0001] The invention relates to the field of tobacco processing, and in particular to a method for extracting data features of cylinder equipment in the tobacco processing industry. Background Art

[0002] In the field of tobacco processing, different equipment such as barrel equipment, such as feeders, tobacco drying machines, etc., have different process parameters and quality index requirements. Since most of the material flows are in a real-time unstable circulation state, it is necessary to measure the quality of the materials exported by the barrel equipment and whether they meet the process standards. Using the data in the production process is the simplest and most intuitive standard. How to abstract and extract the process data so that only one set of data is used to accurately characterize the data change trend in the process and reflect the overall state and quality of the materials in the production process is of great help in optimizing the production process parameters, inspecting the production quality, and improving the process standards.

[0003] Currently, the operating values ​​of equipment parameters are usually calculated using dot sampling and arithmetic mean to represent the production process data within a batch. However, dot sampling is prone to skip or omit changes in data during the production process and cannot reflect the data characteristics under sudden conditions in the production process. The arithmetic mean calculation method can reflect the overall changes in the data, but it is not accurate enough, especially when the data is relatively discrete and distributed on both sides of the center value, the arithmetic mean calculation method cannot reflect the data characteristics. Summary of the invention

[0004] The present invention provides a method for extracting material data features of barrel-type equipment. A sensor is arranged at the outlet of the barrel-type material. The material information data is collected by the sensor. After the data is cleaned, a characteristic value is calculated. A set of data is generated in combination with the time, batch and brand information of the current production. The set of data is the characteristic value data of this production. The extracted characteristic value is stored. The characteristic value is a reference parameter for quality control of the production of the same brand of material. After the characteristic values ​​of all the features are extracted, the characteristic value vector of the equipment in the production process is obtained. Where w (n represents the minimum frequency cumulative value of the i-th feature, k (n0 represents the number of non-repeated eigenvalues ​​in the nth feature, represents the data of the i-th feature at time j, represents the frequency value of the i-th feature at time j, i represents the i-th feature in the feature vector, and v represents the value of the collected data point;

[0005] A. Data collection: collect data from sensors of drum equipment and upload it to the data collection management platform through TCP / IP protocol;

[0006] B. Data cleaning: sort the data collected in step A by time to form a material set ν; remove relevant data according to the process requirements of different barrel equipment. If there are fewer data or the data standard deviation is low after cleaning, the data is invalid and will not be retained;

[0007] C. Feature extraction:

[0008] (1) Collect the data in step A, and group the data of one or more parameters after cleaning in step B, including ambient temperature, ambient humidity, inlet material moisture, outlet material moisture, hot air temperature, cylinder wall temperature, steam pressure, fan frequency, dehumidification negative pressure, damper opening, storage time, inlet material moisture meter intercept, and outlet material moisture meter intercept. For ease of explanation, the following data set is used for representation:

[0009]

[0010] in, Indicates that the device has a total of n data point parameters, and any data point has m data values. j represents the different moments of the data point collection process, and i represents the number of data point classifications.

[0011] (2) The data point set is recorded in vector form as: in, It represents the first data value of the i-th data point, i represents the number of data point categories, j represents the data value of the data point at time j, and m represents the number of data values ​​of the data point. Column vectors take up more space to write, and are usually written as follows Where T represents the transpose of a vector;

[0012] (3) The data points of a device are called features, and the feature vector of the device is recorded as:

[0013] x=(x (1) ,x (2) ,x (3) ,…,x (i) ,…,x (n) ) T , where x (i) represents the i-th feature of x, and n represents the number of features;

[0014] (4) The characteristic vector of the device at time t is recorded as: in, represents the eigenvalue of the first feature at time t, and n represents the number of features;

[0015] (5) For each feature, take out all its eigenvalues, keep one decimal place, count each eigenvalue and the number of times it appears, and get the following:

[0016]

[0017] sequence, Indicates the eigenvalue v1,C that appears in the i-th feature vj (i) represents the frequency of occurrence of eigenvalue vj in the i-th feature, and k represents the number of non-repeated eigenvalues ​​in the feature;

[0018] The C in the Y sequence vj (i) The values ​​are accumulated. When the accumulated value exceeds a certain threshold Δ, the condition for extracting the feature value is met. (i) The minimum frequency value that satisfies the threshold Δ in the i-th feature; otherwise, continue to accumulate until the threshold is exceeded, that is:

[0019] Formula①

[0020] When formula ① satisfies The minimum w (i) Value and k (i) After the value is obtained, the feature value extraction of this feature is calculated by formula ②. Formula ②

[0021] E. Store feature values: Store the features calculated in step C into a storage device.

[0022] This method can solve the problems of uneven materials in the production process, omissions in data collection, and the inability of the arithmetic mean calculation method to reflect the data. Combined with sensors, it is convenient, stable, reliable, and low-cost to detect the moisture content of materials online in real time.

[0023] Due to the adoption of the above scheme, the present invention has the following advantages:

[0024] 1) The data before and after the start of production are usually different from the data during the production process, and there are large differences. Through data cleaning, invalid data from the early and late stages and dirty data under abnormal conditions can be eliminated.

[0025] 2) Taking into account the storage cost, the storage method has few restrictions, and the accuracy of the data is not lost. Using this solution can effectively reduce the storage cost of data, extend the data retention period, and the characteristic value can accurately reflect the data changes and the overall trend of the data.

[0026] 3) The more times the same type of data appears, the larger its proportion in the eigenvalue calculation process, and the more obvious the overall trend of the data reflected. Even if the data is relatively discrete and distributed on both sides of the central value, the eigenvalue calculation method of this scheme can also reflect the data characteristics, which can provide a basis and help for the optimization of production process parameters, production quality review, and improvement of process standards. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Attached Figure 1 Overall flow chart of data feature extraction for cylinder equipment;

[0028] Attached Figure 2 Flowchart of feature value calculation for data feature extraction of cylinder equipment. DETAILED DESCRIPTION

[0029] The following is a clear and complete description of the technical solution of the present invention, taking the stem cutting machine of Sichuan China Tobacco Xichang Cigarette Factory as an example, in conjunction with the flow chart. The stem cutting machine is the last production section on the tobacco equipment silk making line. The moisture content of the exported stem cutting is the main evaluation index for measuring the production quality of the stem cutting machine. This time, the moisture content of the exported stem cutting in the stem cutting machine data is taken as an example for feature extraction.

[0030] A. Data collection: Through the industrial computer IPC equipped with the data collection management platform, the TCP / IP protocol is used to access the address from the PLC controller through the Ethernet switch to obtain the moisture content data of the outlet shredded stems collected by the sensor of the shredded stem drying machine. The sensor is located at the outlet of the material of the shredded stem drying machine.

[0031] B. Data cleaning: The moisture content of the exported cut stems collected in step A is sorted by time to form a set, and the data is screened and extracted. The data extraction rules are extracted according to the extraction rules of the tobacco processing industry, including

[0032] ① Extraction starting point: when the material instantaneous flow rate is >100kg / h, the time delay is 3 minutes; extraction end point: when the material instantaneous flow rate is <100kg / h, the time delay is 3 minutes;

[0033] ② Extraction starting point: delay 3 minutes after the moisture content of exported tobacco is greater than 8%, extraction end point: go back 3 minutes when the moisture content of exported tobacco is less than 8%. Since the data has been sorted by time in step B, any moisture content greater than 8% can be used as the starting point.

[0034] The data collected in this specific implementation method uses the time delay of 3 minutes when the moisture content of the shredded stems is greater than 8% as the extraction starting point, and uses the time traced back 3 minutes when the moisture content of the shredded stems is less than 8% as the extraction ending point. The above-mentioned extraction rules are used for introduction, and m=3559 at this time. A rule can be selected for each data point, and different data points can select different extraction rules. The same rules are used for screening and extraction of the extraction starting point and the end point. Compared with the dot sampling calculation method, sampling is performed at fixed intervals or fixed frequencies, which easily skips or omits changes in data during the production process. This method extracts all data from the start to the end point of the collection, eliminates abnormal data values ​​that may exist in the materials in the batch, and retains the changes in data values ​​in the original production process. The number of measurements does not need to be determined by multiple comparisons.

[0035] C. Feature extraction: The moisture content of the exported shredded stems after cleaning in step B is retained to one decimal place, and the shredded stems are grouped by value, and the number of times the value appears is calculated. At this time, the total number of groups k = 7, and the obtained <value, frequency> sequence is:

[0036] T

[0037] ={<12.7,99>,<12.8,927>,<12.9,1709>,<13.0,632>,<13.1,120>,<12.6,41>,<12.5,31>}

[0038] Assume Δ = 0.8, and set C v The values ​​are accumulated. When the frequency starts from 99 at 12.7 and reaches 1709 at 12.9, according to formula ①, the accumulated value w = 2735. Less than the set Δ value; when adding the frequency 632 of 13.0, the cumulative value of formula ① w=3,367, at this time is greater than the set Δ value, so the minimum w=3,367, the minimum k=4. According to formula ② The calculation result is v=12.88536, and v=12.9 with one decimal place retained.

[0039] D. Store the characteristic values ​​extracted in step C into a storage device, such as a text file, database, etc., to provide data support for subsequent third-party system quality analysis, intelligent predictive control, and process optimization. When collecting data to calculate characteristic values, it is irrelevant to the production batch, brand, and other data of the material. After the characteristic value is calculated, a set of data is generated in combination with the time, batch, and brand information of the current production. This set of data is the characteristic value data of this production, reflecting the overall state of the production process. Save this set of data to a storage device. When there are multiple characteristic value record data in the storage device, the characteristic value data record can characterize the data change trend, that is, the production data changes at different times for the same brand, or the production data changes for different brands within a given time period. For each set of characteristic value data recorded, the process data characteristic values ​​that meet the process quality can be screened out according to the requirements of one or more process indicators, providing data reference for the setting of parameters in the next production control of the brand. This can help companies better optimize production process parameters for different tobacco brands, and can also analyze production changes for the same brand in different time periods, provide data support for the adjustment of production parameters, and provide a basis for production quality review and improvement of process standards.

[0040] Although some specific embodiments of the present application have been described in detail by way of example, it should be understood by those skilled in the art that the above examples are only for illustration, not for limiting the scope of the present application. It should be understood by those skilled in the art that the above embodiments may be modified without departing from the scope and spirit of the present application. The scope of the present application is defined by the appended claims.

Claims

1. A method for extracting material data features of a cylindrical device, wherein a sensor is provided at the outlet of the cylindrical material, and the method is characterized in that: The method collects material information data through sensors, cleans the data, calculates the characteristic value, combines the time, batch and brand information of the current production to generate a set of data, which is the characteristic value data of this production, stores and extracts the characteristic value, and the characteristic value is a reference parameter for quality control of the same brand of material. After the characteristic values ​​of all characteristics are extracted, the characteristic value vector of the equipment in the production process is obtained. Where w n represents the minimum frequency cumulative value of the i-th feature, k n represents the number of non-repeated eigenvalues ​​in the nth feature, represents the data of the i-th feature at time j, represents the frequency value of the i-th feature at time j, i represents the i-th feature in the feature vector, and v represents the value of the collected data point; A. Data collection: collect data from sensors of drum equipment and upload it to the data collection management platform through TCP / IP protocol; B. Data cleaning: sort the data collected in step A by time to form a material set ν; remove relevant data according to the process requirements of different barrel equipment. If there are fewer data or the data standard deviation is low after cleaning, the data is invalid and will not be retained; C. Feature extraction: (1) Collect the data in step A, and group the data of one or more parameters after cleaning in step B, including ambient temperature, ambient humidity, inlet material moisture, outlet material moisture, hot air temperature, cylinder wall temperature, steam pressure, fan frequency, dehumidification negative pressure, damper opening, storage time, inlet material moisture meter intercept, and outlet material moisture meter intercept. For ease of explanation, the following data set is used for representation: in, Indicates that the device has a total of n data point parameters, and any data point has m data values. j represents the different moments of the data point collection process, and i represents the number of data point classifications. (2) The data point set is recorded in vector form as: in, It represents the first data value of the i-th data point, i represents the number of data point categories, j represents the data value of the data point at time j, and m represents the number of data values ​​of the data point. Column vectors take up more space to write, and are usually written as follows Where T represents the transpose of a vector; (3) The data points of a device are called features, and the feature vector of the device is recorded as: x=(x (1) ,x (2) ,x (3) ,…,x (i) ,…,x (n) ) T , where x (i) represents the i-th feature of x, and n represents the number of features; (4) The characteristic vector of the device at time t is recorded as: in, represents the eigenvalue of the first feature at time t, and n represents the number of features; (5) For each feature, take out all its eigenvalues, keep one decimal place, count each eigenvalue and the number of times it appears, and get the following: sequence, Indicates the eigenvalue v1,C that appears in the i-th feature vj (i) represents the frequency of occurrence of eigenvalue vj in the i-th feature, and k represents the number of non-repeated eigenvalues ​​in the feature; The C in the Y sequence vj (i) The values ​​are accumulated. When the accumulated value exceeds a certain threshold Δ, the condition for extracting the feature value is met. (i) The minimum frequency value that satisfies the threshold Δ in the i-th feature; otherwise, continue to accumulate until the threshold is exceeded, that is: Official① When formula ① satisfies The minimum w (i) Value and k (i) After the value is obtained, the feature value extraction of this feature is calculated by formula ②. Formula ② D. Store feature values: Store the features calculated in step C into a storage device.

2. The method for extracting material data features of drum equipment according to claim 1 is characterized in that: The data cleaning is performed using standard extraction rules in accordance with the tobacco processing industry.

3. The method for extracting material data features of drum equipment according to claim 1, characterized in that: The vector value is a database or a text file.

4. The method for extracting material data features of drum equipment according to claim 1, characterized in that: The eigenvector and eigenvalue vector V formula can be simplified to 5. The method for extracting material data features of drum equipment according to claim 4 is characterized in that: A rule may be selected for each data point, and different data points may select different extraction rules. The same rule is used for screening and extraction of the extraction start point and the end point.

Images