TWI719786B - Data processing system and method - Google Patents

Abstract

A data processing system including a cyclic correlation establishing module, a data pattern establishing module, and a data pattern alignment module is provided. The cyclic correlation establishing module receives a number of first sensing data obtained by performing a first sensing operation on a number of processing devices, and receives a table of processing step and cyclic procedure information. The cyclic correlation establishing module obtains the data correlation of the first sensing data based on the number of sample points in the period of the first sensing data and the table of the processing step and cyclic procedure information. The data correlation of the first sensing data is used to correct the first sensing data. The data pattern establishing module obtains a number of first data pattern features from the first sensing data. A number of second sensing data is obtained by performing a second sensing operation on the processing devices. The data pattern alignment module aligns the first sensing data and the second sensing data according to the first data pattern features.

Description

Data processing system and method

This disclosure relates to a data processing system and method.

In today’s pre-diagnosis technology, for equipment with few failure events and short life cycles, such as Regenerative Thermal Oxidizer (RTO), one of the methods is to combine historical base data with recent new (Fresh ) Data, establish or train an equipment failure prediction model. However, there are two considerations for the above-mentioned pre-diagnosis technology: (1) The data obtained by the actual detection of the machine by the sensor often contains incorrect information, such as: the machine is in an unstable state, or there is noise, omission, or Error values, etc., need to be effectively eliminated; and (2) Historical and recent data patterns have considerations that need to be aligned with each other, and there are also considerations that need to be aligned with each other between recent data in different time periods, leading to retrieval The recent characteristic parameter values at different stages, combined with the misaligned historical base data, affect the accuracy of the equipment failure prediction model. Both of these two considerations will seriously affect the accuracy of subsequent pre-diagnosis. Therefore, how to effectively deal with the above considerations is one of the directions the industry is committed to. In addition, this type of equipment has the characteristics of a cyclic program, but the current pre-diagnosis technology does not make good use of the cyclic program characteristics to effectively deal with the above considerations.

According to the first aspect of the present invention, a data processing system is provided, which includes a cyclic correlation building module, a data pattern building module, and a data pattern alignment module. The loop correlation building module is used to receive a plurality of first sensing data obtained from a first sensing of a plurality of processing devices, and to receive a processing step and a loop procedure information table, and a processing step and a loop procedure information table The information about the multiple processing steps and multiple cycle procedures of these processing equipment is recorded. The cycle correlation building module is further used to obtain the data correlation of the first sensing data according to the number of sample points in a data cycle of the first sensing data and the processing steps and the cycle procedure information table to calibrate the data. Some first sensing data. The data pattern building module is used to obtain a plurality of first data pattern characteristics of the first sensing data. The data pattern alignment module is used for aligning a plurality of second sensing data obtained by performing a second sensing on the processing equipment with the first sensing data according to the characteristics of the first data pattern.

According to the second aspect of the present invention, a data processing method is provided, which includes the following steps. Receive a plurality of first sensing data obtained by performing a first sensing on a plurality of processing devices, and receive a processing step and cycle program information table. The processing step and cycle program information table records the multiple processing steps of the processing equipment and the related cycle operation mode information of the multiple cycle programs. According to the number of sample points in a data cycle of the first sensing data and the processing steps and cycle procedure information table, the data relevance of the first sensing data is obtained to calibrate the first sensing data. According to the first sensing data, a plurality of first data aspect characteristics are obtained. According to the characteristics of the first data, a plurality of second sensing data obtained by performing a second sensing on these processing devices are aligned with the first sensing data.

In order to have a better understanding of the above and other aspects of the present disclosure, the following examples are specially cited, and the accompanying drawings are described in detail as follows:

Tbl: Processing steps and cycle program information table

4: Multi-processing equipment machine

4A: First processing equipment

4B: Second processing equipment

4C: Third processing equipment

4D: Fourth processing equipment

4E: Fifth processing equipment

10.160: data processing system

20: Data Acquisition Module

40: Loop Relevance Build Module

60: Data Style Building Module

80: Data pattern alignment module

100: Feature extraction module

110: Failure prediction module

103: The first stage

105: second stage

107: The third stage

202~206, 41~51, 61~71, 81~91: process steps

302(A), 302(B), 302(C), 302(D), 302(E): heat storage tank

304(A1), 304(A2), 304(B2), 304(C1), 304(D1), 304(E2): valve

306~334: exhaust gas flow direction

402~412: sample points

602, 604, 606, 614, 616: Windows

608, 610, 612: data state

1300: box

S_Data1_A~S_Data1_E: the first sensor data

S_Data2_A~S_Data2_E: second sensing data

F_A~F_E, F_A’~F_E’: Features of data

FIG. 1 is a block diagram of a data processing system according to an embodiment of the disclosure.

FIG. 2 shows a flowchart of a data processing method according to an embodiment of the disclosure.

Figure 3A shows the operating status of the multi-processing equipment in the first stage.

Figure 3B shows the operating status of the multi-processing equipment in the second stage.

Figure 3C shows the operating status of the multi-processing equipment in the third stage.

Figure 4 shows an example of the processing steps and cycle program information table of the multiple processing equipment in Figures 3A to 3C.

Figure 5 shows the processing type S(n) executed by the five loop programs in Figure 4.

FIG. 6 is a block diagram of a data processing system according to another embodiment of the disclosure.

FIG. 7 is a schematic diagram of an example of a plurality of first sensing data obtained by the data capturing module performing the first sensing and capturing.

Figure 8 shows a flow chart of an example of the steps executed by the cyclic association building module.

Figure 9A shows a schematic diagram of calculating the number of sample points in the data period from the first sensing data through the autocorrelation analysis method.

Figure 9B shows that the data period sample points can be calculated from the first sensing data by comparing the displacement data period.

Figure 10 is a schematic diagram of the number of steps in the cycle program in Figure 4.

Fig. 11 is a schematic diagram of the loop correlation building module in Fig. 1 judging whether the sensed data has abnormal points.

Figure 12 is a flowchart showing an example of the detailed steps performed by the data-style building module in Figure 1 or Figure 6.

FIG. 13 is a schematic diagram of the data aspect characteristics of the first sensing data obtained.

Figure 14 shows a flowchart of an example of the detailed steps performed by the data pattern alignment module.

FIG. 15 is a schematic diagram of a plurality of second data aspect characteristics obtained when any one of the second sensing data is captured in different parts by the sliding window method.

Figure 16 shows a schematic diagram of different parts of the second sensing data captured by the sliding window method.

Figure 17 is a schematic diagram of the feature extraction module extracting feature parameters of each processing device according to the data state.

Please refer to FIG. 1, which shows a block diagram of a data processing system 10 according to an embodiment of the present disclosure. The data processing system 10 includes a cyclic correlation building module 40, a data pattern building module 60 and a data pattern alignment module 80. An example of the data processing system 10 is described by taking a computer system with software stored in a computer hard disk as an example, but it is not limited to this. For example, the computer hard disk stores software program codes with corresponding functions of the cyclic association building module 40, the data pattern building module 60, and the data pattern alignment module 80. For example, the software code is the cyclic association building code, the data pattern building code, and the data pattern alignment program. Style code. When the computer processor executes this code, the corresponding functions of the cyclic correlation building module 40, the data pattern building module 60, and the data pattern alignment module 80 can be implemented correspondingly. In addition, the code corresponding to the three modules can also be stored in the computer hardware in a firmware mode, or implemented in an Application Specific Integrated Circuit (ASIC). The loop correlation building module 40 is used for receiving a plurality of first sensing data obtained by performing a first sensing on a plurality of processing devices, and receiving a processing step and a loop program information table Tbl. The cycle correlation building module 40 is further used to obtain the data correlation of the first sensing data according to the number of sample points in a data cycle of the first sensing data and the processing steps and the cycle procedure information table Tbl, so as to Calibrate these first sensing data. The data pattern building module 60 is used to obtain a plurality of first data pattern characteristics of the first sensing data. The data pattern alignment module 80 is used for aligning a plurality of second sensing data obtained by performing a second sensing on a plurality of processing devices with the first sensing data according to the characteristics of the first data pattern.

Please refer to FIG. 2, which shows a flowchart of a data processing method according to an embodiment of the present disclosure. The data processing method includes the following steps. In step 202, a plurality of first sensing data obtained by performing a first sensing on a plurality of processing devices is received, and a processing step and cycle program information table is received. According to the number of sample points in a data cycle of the first sensing data and the processing steps and cycle procedure information table, the data relevance of the first sensing data is obtained to calibrate the first sensing data. Then, step 204 is executed to obtain a plurality of first data aspect characteristics according to the first sensing data. Then, proceed to step 206 to align a plurality of second sensing data obtained by performing a second sensing on these processing devices with the first sensing data according to the characteristics of the first data morphology.

The data processing system 10 can optionally include a screen, and the screen can be connected to a computer system that executes cyclic association building code, data pattern building code, and data pattern alignment code, making this computer the present disclosure One implementation of data processing system 10 example. The computer system outputs the execution results of these three pieces of code to the screen, so that the screen displays the Graphic User Interface (GUI). The graphical user interface is used to display operation options related to the data processing system 10, such as at least one option related to the cycle operation mode, processing steps and cycle procedure information tables, model prediction results, or equipment health indicators. In this way, it is convenient for the user to operate, or for the user to easily understand the relevant information such as the equipment pre-diagnosis results or equipment health indicators. This option includes, but is not limited to, for example, the option to display multiple first sensing data or multiple second sensing data, the option to display the processing steps and cycle program information table Tbl, and the option to display the data relevance of the first sensing data. , Display the options of the first sensing data after calibration, the options of displaying the multiple first data aspect characteristics of the first sensing data, and the multiple second sensing aligned with the first sensing data Data options.

The above-mentioned processing step and cycle program information table Tbl, for example, records information about multiple processing steps of multiple processing devices and related cycle operation modes of multiple cycle programs. This disclosure improves the consistency of (1) the correctness of the data pattern and (2) the consistency of the data pattern alignment by simultaneously considering the data relevance between the "processing step" and the "circular program" of the machine . To improve the correctness of the data pattern, the steps are included: (i) Consider the data relevance of the processing steps and the cycle program at the same time, compare the sampling magnification between the number of processing steps and the number of sample points in the data cycle, and consider it together Calculate the number of data offset points of the correlation data between each processing device (such as: number of steps * sampling magnification), and (ii) build a variety of heterogeneous data patterns , The data pattern is captured according to the number of sample points in the data period, and the point pattern characteristics of each sample point are captured through the positive/negative trend for comparison. To improve the consistency of the alignment between the historical base and the recent new pattern data, it includes the following steps: Use the sliding window method to capture the data patterns of the multiple second sensing data obtained from the second sensing, according to the first The first data sample of the first sensing data obtained by sensing The state features are compared with each other to align the plurality of second sensing data with the plurality of first sensing data, which further improves the accuracy of the pre-diagnosis.

Please refer to FIGS. 3A to 3C, which illustrate a schematic diagram of a multi-processing equipment machine 4 according to an embodiment of the present disclosure. In this embodiment, the multi-processing equipment machine 4 is illustrated with a five-bed regenerative thermal incinerator (Regenerative Thermal Oxidizer, RTO) machine as an example. The five-bed regenerative thermal incinerator machine has, for example, multiple Heat storage tanks (e.g. heat storage tanks 302(A), 302(B), 302(C), 302(D), 302(E)) and their corresponding valves (e.g. valves 304(A1), 304(A2), 304(B2), 304(C1), 304(D1), 304(E2)), through the valve to control the gas flow in the heat storage tank and other procedures. A processing device such as a five-bed regenerative incinerator and a heat storage tank is taken as an example for description, but this embodiment is not limited to this. The multi-processing device machine 4 has, for example, a first processing device 4A, a second processing device 4B, a third processing device 4C, a fourth processing device 4D, and a fifth processing device 4E. Please refer to FIG. 4, which shows an example of the processing steps and cycle program information table Tbl of the first processing device 4A to the fifth processing device 4E in FIGS. 3A to 3C. Ten processing steps and five loop procedures are taken as examples for description, but the embodiments of the disclosure are not limited thereto. The processing step and cycle program information table Tbl, for example, records information about the 10 processing steps of the first processing device 4A to the fifth processing device 4E and the related cycle operation mode information of the 5 cycle programs. The processing step and cycle program information table Tbl may, for example, Obtained by reading files. The processing steps and cycle program information table Tbl can reflect the cycle operation mode of the physical machine. The first processing device 4A to the fifth processing device 4E respectively execute the cycle program A to the cycle program E, and each of the cycle programs A to the cycle program E executes the same sequence of processing steps cyclically. In other words, the first processing device 4A to the fifth processing device 4E respectively execute the cyclic operation mode, the first processing device 4A repeatedly executes the cyclic program A, and the second processing device 4B repeatedly executes the cyclic program B, thereby By analogy, the fifth processing device 4E executes the loop program E repeatedly.

Please refer to Figure 5, which shows the processing type S(n) executed by the 5 loop programs in Figure 4, n=1~10. In this example, the number of processing types is 10 as an example. Of course it is not limited to this. Taking the sequence of the cycle program C as an example, the 10 sorted processing types are in order: processing type S(1) is no action, processing type S(2) is the first stage air intake, and processing type S(3) Is the second stage intake, the treatment type S(4) is the third stage intake, the treatment type S(5) is the fourth stage intake, the treatment type S(6) is suction, and the treatment type S(7) is the first One-stage exhaust, the treatment type S(8) is the second-stage exhaust, the treatment type S(9) is the third-stage exhaust, and the treatment type S(10) is the fourth-stage exhaust. As shown in Figure 4, in processing step 1, when cycle program A executes processing type S(5) (fourth stage intake), cycle program C executes processing type S(1) (no action), and cycle Program E executes processing type S(7) (first stage exhaust). The difference between the cycle program A and the cycle program C is=(5-1)=(4), the cycle program C and the cycle program E are different in the number of steps=(1-7)=(-6), because 10 cycles are executed The processing type of sorting, so the number of phase difference steps=(4) and the number of phase difference steps=(-6) are actually the same, and the number of phase difference steps is 4 steps apart.

Please refer to Figures 3A~3C and Figure 4. Figure 4 also shows three stages 103, 105 and 107 (corresponding to processing steps 1, 2, and 3 respectively) to illustrate that multiple processing devices 4 are different The status of the stage. The dotted arrows in Figures 3A to 3C indicate the flow direction of exhaust gas (for example, exhaust gas flow directions 306 to 334). Please refer to FIG. 3A, which illustrates the operating state of the multi-processing equipment machine 4 in the first stage 103. At this time, the first processing device 4A executes step S(5) (fourth stage intake), the second processing device 4B executes step S(9) (third stage exhaust), and the third processing device 4C executes steps S(1) (no action), the fourth processing device 4D executes step S(3) (second stage intake), and the fifth processing device 4E executes step S(7) (first stage exhaust).

Please refer to FIG. 3B, which shows the operating state of the multi-processing equipment machine 4 in the second stage 105. The first processing device 4A executes step S(6) (exhaust), the second processing device 4B executes step S(10) (fourth stage exhaust), and the third processing device 4C executes step S(2) (the first One-stage intake), the fourth processing device 4D executes step S(4) (third-stage intake), and the fifth processing device 4E executes step S(8) (second-stage exhaust).

Please refer to FIG. 3C, which shows the operating state of the multi-processing equipment machine 4 in the third stage 107. The first processing device 4A executes step S(7) (first stage exhaust), the second processing device 4B executes step S(1) (no action), and the third processing device 4C executes step S(3) ( The second stage intake), the fourth processing device 4D executes step S(5) (fourth stage intake), and the fifth processing device 4E executes step S(9) (third stage exhaust).

Among them, during the operation of air intake, the exhaust gas will flow into the processing equipment, allowing the exhaust gas to burn, oxidize, and release heat. In the treatment equipment for air extraction, air will be blown into the treatment equipment, and the residual, unburned exhaust gas will be guided, so that the exhaust gas will be burned together. The combusted exhaust gas flows into another processing device for exhaust gas to perform the action of accumulating heat generated by the combustion and exhaust the exhaust gas. These operations such as air intake, air extraction (air blowing or air extraction), and exhaust are related to each other. Through the treatment of intake, exhaust, exhaust, etc., combined with the cycle program of multiple treatment equipment, the effect of high exhaust gas removal rate and low fuel consumption can be achieved.

Please refer to FIG. 6, which shows a block diagram of a data processing system 160 according to another embodiment of the present disclosure. The data processing system 160 can be used for pre-diagnosing a plurality of processing devices 4A-4E. Compared with the data processing system 10 shown in FIG. 1, the data processing system 160 further includes a data acquisition module 20, a feature extraction module 100 and a fault prediction module 110. The cyclic correlation building module 40, the data pattern building module 60, the data pattern alignment module 80 of the data processing system 160 and the cyclic correlation building of the data processing system 10 shown in Figure 1 The module 40, the data pattern building module 60, and the data pattern alignment module 80 are the same modules and will not be repeated here.

After the data acquisition module 20 performs the first sensing on the multiple processing devices 4A~4E, it will obtain multiple first sensing data S_Data1, and then the cyclic correlation building module 40 receives the multiple first sensing data S_Data1. Sensing data S_Data1. The plurality of first sensing data S_Data1 includes, for example, the first sensing data S_Data1_A of the first processing device 4A, the first sensing data S_Data1_B of the second processing device 4B, the first sensing data S_Data1_C of the third processing device 4C, and the first sensing data S_Data1_C of the third processing device 4C. The first sensing data S_Data1_D of the fourth processing device 4D, and the first sensing data S_Data1_E of the fifth processing device 4E. After that, after the data acquisition module 20 performs the second sensing on the multiple processing devices 4A~4E, a plurality of second sensing data S_Data2 will be obtained, which are then received by the cyclic correlation building module 40 The second sensing data S_Data2. The plurality of second sensing data S_Data2 includes, for example, the second sensing data S_Data2_A of the first processing device 4A, the second sensing data S_Data2_B of the second processing device 4B, the second sensing data S_Data2_C of the third processing device 4C, and the second sensing data S_Data2_C of the third processing device 4B. The second sensing data S_Data2_D of the fourth processing device 4D, and the second sensing data S_Data2_E of the fifth processing device 4E.

Among them, the sensing data is, for example, the sensing data of the machine parts of the processing equipment. The sensing data can be, for example, using the method of reading files, using the data capture card interface to read files, or by accessing a database system to The way to obtain the file, to obtain the sensing data of the machine parts. The sensing data is, for example, the temperature of the processing equipment, the windmill motor speed (Round Per Minute, RPM), frequency (Hertz, Hz), current, bearing vibration, valve opening and closing action, exhaust gas concentration value, exhaust gas flow, burner pressure or other Parameters for detecting the operating status of processing equipment. A sensor for a regenerative tank of a regenerative incinerator machine As an example, the original sensing data for a period of time is retrieved by reading the file as the above-mentioned first sensing data.

Please refer to FIG. 7, which shows a schematic diagram of an example of a plurality of first sensing data S_Data1 obtained by the data capturing module 20 performing the first sensing and capturing. In a first period of time, the data acquisition module 20 generates the first sensing data S_Data1_A of the first processing device 4A after the first processing device 4A cyclically executes 10 processing steps multiple times, and the second processing device 4B cyclically executes the first sensing data S_Data1_A After 10 processing steps are executed multiple times, the first sensing data S_Data1_B of the second processing device 4B is generated, and so on. The first sensing data S_Data1_E of the fifth processing device 4E is generated in response to the fifth processing device 4E cyclically executing 10 processing steps multiple times.

Please refer to FIG. 8, which shows a flowchart of an example of the steps performed by the cyclic association building module 40. Figure 8 illustrates how the cycle correlation building module 40 obtains the data association of the first sensing data based on the number of sample points in a data cycle and the processing steps and the cycle procedure information table Tbl of the first sensing data. It is an example of the detailed steps of calibrating the first sensing data based on the data relevance. In the process of calibrating the first sensing data by the loop correlation building module 40, the loop correlation building module 40 confirms the number of steps between the loop procedures to obtain the first sensing data Data relevance. The data relevance of the first sensing data includes a data offset point between one of the first sensing data and the other of the first sensing data. The cyclic correlation building module 40 is further used for the point pattern characteristics of a first sample point of one of the first sensing data corresponding to the number of data offset points and the first sensing data The point pattern feature of another second sample point is compared with the first sample point of one of the first sensing data and the second sample point of the other of the first sensing data. The correctness of the data is used to calibrate the first sensing data. Among them, in the process of calculating the data relevance of these first sensing data, the cycle is closed The linkage building module 40 is used, for example, to obtain the number of data cycle sample points from the first sensing data through an autocorrelation analysis method, and to obtain the number of these processing steps and the data cycle sample points According to the sampling magnification, the data offset points between the first sensing data are obtained according to the sampling magnification, so as to obtain the data relevance of the first sensing data.

In step 41 in FIG. 8, the loop correlation building module 40 obtains the number of processing steps from the processing steps and loop program information table Tbl, as shown in FIG. 4. In this embodiment, the number of processing steps in the cycle programs A to E is equal to 10. In step 42, it is determined whether there is a sensor sampling magnification. If not, go to step 43; if yes, go to step 45.

In step 43, please refer to FIG. 9A at the same time, which illustrates a schematic diagram of calculating the number of data period sample points from the first sensing data S_Data1 through the autocorrelation analysis method. Please refer to Figure 9B, which shows that the data period can be calculated by comparing the displacement data, and the data period is obtained from the first sensing data S_Data1 and the data period sample points are calculated. Choosing one of these two methods can calculate the number of data period sample points, but it is not limited to these two methods.

In step 44, the sampling rate is obtained by dividing the number of sample points in the data period by the number of processing steps based on the formula of "Sampling Magnification = Number of Sample Points in the Data Period/Number of Processing Steps". As shown in Figures 9A to 9B, the first sensing data S_Data1 can be obtained by autocorrelation analysis method or comparing the displacement data period, and it is known that the sample points of the data period are 5 sample points. According to the number of processing steps obtained in step 41 is 10, the sampling magnification between the two can be calculated to be equal to 5/10=1/2. In one embodiment, it can also be obtained through the sampling magnification of the actual sensor.

In step 45, the number of different steps between these cycle programs is confirmed. Please also refer to Fig. 10, which shows a schematic diagram of the number of steps in the cycle program in Fig. 4. FIG. 10 is an example of the loop program A executed by the first processing device 4A, the loop program C executed by the third processing device 4C, and the loop program E executed by the fifth processing device 4E. The cycle program A executed by the first processing device 4A is 4 steps different from the cycle program C executed by the third processing device 4C, and the cycle program C executed by the third processing device 4C is the same as the cycle executed by the fifth processing device 4E. Procedure E differs by 4 steps.

In step 46, the number of data offset points between the first sensing data S_Data1 is calculated according to the formula of "data offset point number = phase difference step number * sampling magnification" to obtain data relevance. As shown in Figure 10, since the number of phase difference steps is 4, the sampling magnification obtained by step 44 is equal to 1/2. In this way, the number of data offset points between the first sensing data S_Data1_A of the first processing device 4A and the first sensing data S_Data1_C of the third processing device 4C can be calculated=4*(1/2)=2, and the number of data offset points between the first sensing data S_Data1_A of the first processing device 4A and the first sensing data S_Data1_C of the third processing device 4C is calculated. The number of data offset points between the first sensing data S_Data1_C of the third processing device 4C and the first sensing data S_Data1_E of the fifth processing device 4E=4*(1/2)=2.

In step 47, calculate the trend of each sample point in the first sensing data S_Data1 and the other two sample points before and after the corresponding processing equipment, so as to calculate the other of the first sensing data S_Data1 The first sample point of one corresponds to the positive trend or the negative trend of the other two sample points before and after the first sample point to obtain the point pattern characteristics of the first sample point. Please refer to FIG. 11, which illustrates a schematic diagram of the loop correlation building module 40 in FIG. 1 determining whether the sensing data has abnormal points. The sample point 408 of the first sensing data S_Data1_A of the first processing device 4A is the same as the one on the left and the value obtained by subtracting the value trend is positive, and the sample point 408 is the same as the one on the right and the value of the value obtained by subtracting the trend is negative. The trend between the sample point 408 and the other two sample points before and after is [+,-], which is used as the point pattern characteristic of the sample point 408 Levy. The sample point 406 of the first sensing data S_Data1_E of the fifth processing device 4E is the same as the one on the left and the value obtained by subtracting the trend is negative, and the sample point 406 is the same as the one on the right and the value of the subtracting trend is positive. Therefore, the trend between the sample point 406 and the other two sample points before and after is [-,+], which is used as the point pattern feature of the sample point 406. Calculate the trend of each sample point and the corresponding two sample points before and after, and complete the sampling of each sample point of the first sensing data S_Data1_A of the first processing device 4A to the first sensing data S_Data1_E of the fifth processing device 4E Calculation of state characteristics.

In step 48, it is determined whether the point pattern characteristics of the first sample point and the second sample point corresponding to the number of data offset points are the same, for example, the sample point of the first sensing data S_Data1_A of the processing device 4A is selected, The first sensing data S_Data1_C of the third processing device 4C and the first sensing data S_Data1_E of the fifth processing device 4E are shifted by the number of sample points of the data to compare whether the trend between the sample points is the same. When the trend of each sample point is the same, it is determined that there is no abnormal point. When the trend of a few sample points is inconsistent with the trend of most sample points, it is determined that an abnormal point has occurred. As shown in Fig. 11, corresponding to the cycle program A of the first processing device 4A, the trends of the first sample point 402 and the sixth sample point 408 of the first sensing data S_Data1_A of the first processing device 4A (ie, point The modal feature) is [+,-]. Corresponding to the loop program C of the third processing device 4C, the trends of the third sample point 404 and the eighth sample point 410 of the first sensing data S_Data1_C of the third processing device 4C are [+,-]. Corresponding to the loop program E of the fifth processing device 4E, the trend of the fifth sample point 406 of the first sensing data S_Data1_E of the fifth processing device 4E is [-,+], and the trend of the tenth sample point 412 is [ +,-]. Under the assumption that most sample points are normal and only a few sample points are abnormal, step 48 determines the first sample point 402 and the sixth sample of the first sensing data S_Data1_A of the first processing device 4A based on the trend of the majority of sample points Point 408, the third sample point 404 and the eighth point of the first sensing data S_Data1_C of the third processing device 4C The sample point 410 and the tenth sample point 412 of the first sensing data S_Data1_E of the fifth processing device 4E are normal sample points, but the fifth sample point of the first sensing data S_Data1_E of the fifth processing device 4E 406 is an abnormal point. That is, the first sensing data S_Data1_A of the processing device 4A and the first sensing data S_Data1_C of the third processing device 4C have no abnormalities, but the first sensing data S_Data1_E of the fifth processing device 4E has an abnormality 406. In this way, one of the first sensing data (for example, the first sensing data S_Data1_A of the first processing device 4A or the first sensing data S_Data1_C of the third processing device 4C) can be achieved. The point pattern feature of this point (for example, sample point 402 or 404) and a second sample point (for example, the first sensor data S_Data1_E of the fifth processing device 4E) of the other of the first sensing data ( For example, the point pattern feature of the sample point 406) is compared with the first sample point of one of the first sensing data (for example, the sample point 402 or 404) and the other of the first sensing data. The correctness of the data of the second sample point (for example, sample point 406).

In step 49 and step 50, when the sample point of the processing equipment is confirmed to be abnormal, the wrong sample point is corrected or eliminated. For example, in Figure 11, the fifth sample point 406 of the first sensing data S_Data1_E of the fifth processing device 4E has been regarded as an abnormal point, it can be excluded and not included in the subsequent calculation, or the fifth sample point 406 can be referenced to the sample The points 402 and 404 are calibrated to obtain a corrected first sensing data S_Data1_E' of the fifth processing device 4E. In step 51, the cyclic correlation building module 40 collects correct sample points or erroneous sample points that have been corrected, such as the corrected first sensing data S_Data1_E' of the fifth processing device 4E.

Please refer to FIG. 12, which shows a flowchart of an example of the detailed steps performed by the data-style building module 60 in FIG. 1 or FIG. Figure 12 illustrates the execution of the data pattern building module 60 to obtain the first data based on the first sensing data S_Data1 The step of pattern characterization includes retrieving a plurality of data patterns of the first sensing data S_Data1 according to the number of sample points in the data period to obtain the first data patterns of the first sensing data S_Data1 State characteristics. These data patterns are obtained, for example, by calculating the positive trend or the negative trend of a plurality of sample points of the first sensing data S_Data1. For example, according to the first sensing data S_Data1_A of the processing device 4A to the first sensing data S_Data1_D of the fourth processing device 4D and the corrected first sensing data after the abnormal point is excluded, the fifth processing device 4E The first sensing data S_Data1_E is used to obtain the first data aspect characteristics of the first sensing data S_Data1_A~S_Data1_E.

In step 61 of Figure 12, initialize the data modal characteristics and sample points, set i=1 and i is a positive integer. In step 62, the data pattern of the processing equipment is acquired by the number of data period sample points={S ₁ ,S ₂ ,...S _{data period sample points} }, where each S ₁ ,S ₂ ,...sample The point value system is a real number. In this embodiment, the number of sample points in the data period is 5, so sampling {S ₁ , S ₂ ,...S ₅ }.

In steps 63 to 64, if (S _i -S _i+1 )>0, it is determined that the numerical trend = -. In steps 65-66, if (S _i -S _i+1 )<0, it is determined that the numerical trend=+. In steps 67 to 68, if (S _i -S _i+1 )=0, it is determined that the numerical trend=0. In steps 69 to 71, the numerical trends are added to the data pattern characteristics of the corresponding processing equipment one by one. In step 71, the value of the sample point parameter i is increased by 1, and the sliding window continues to move one sample point to the right, and the loop is repeated until the numerical trend analysis of the 5 sampling points in each first sensing data is completed. Here, the numerical trend is illustrated by taking the slope of the line segment as an example, but it is not limited to this. For example, the numerical trend can also be a rising and falling feature. Please refer to Figure 13, which shows that each of these first sensing data S_Data1 takes 5 sample points and subtracts them from the values of the corresponding sample points before and after, and then 4 numerical trends can be calculated as Corresponding data style characteristics. In this embodiment, with respect to the first sensing data S_Data1 shown in FIG. 7, the pattern characteristics corresponding to the data patterns of each first sensing data selected by block 1300 in FIG. 13 can be obtained . For example, the data pattern feature of the first sensing data S_Data1_A corresponding to the cycle program A of the first processing device 4A is F_A1=[-,-,+,+], corresponding to the first of the cycle program B of the second processing device 4B The data pattern characteristic of the sensing data S_Data1_B is F_B1=[+,+,+,-], and the data pattern characteristic of the first sensing data S_Data1_C corresponding to the cycle program C of the third processing device 4C is F_C1=[+, +,-,-], the data pattern characteristic of the first sensing data S_Data1_D corresponding to the cycle program D of the fourth processing device 4D is F_D1=[+,-,-,+], corresponding to the fifth processing device 4E cycle program The data aspect characteristic of E's first sensing data S_Data1_E is F_E1=[-,+,+,+]. In this way, the data pattern building module 60 obtains each first data pattern feature of each first sensing data according to the five first sensing data S_Data1_A~S_Data1_E.

Please refer to FIG. 14, which shows a flowchart of an example of the detailed steps performed by the data pattern alignment module 80. Figure 14 will explain in detail how the data pattern alignment module 80 can perform a second sensing on a plurality of processing devices 4A-4E based on the characteristics of the first data pattern. An example of the detailed steps of aligning S_Data2 with these first sensing data S_Data1. The data pattern alignment module 80 can, for example, use a sliding window method to capture a plurality of second data patterns of the second sensing data S_Data2, and obtain a plurality of second data patterns of the second data patterns Features, and compares the second data-like features with the corresponding first data-like features. When the similarity between the second data pattern features and the corresponding first data pattern features is greater than a threshold value, it is determined that the second data patterns are aligned with the corresponding first data patterns .

In steps 81 to 83 in Figure 14, the sample points are initialized, set i=1, and the first data modal characteristics and threshold values of the built processing equipment are obtained. Among them, i is a positive integer. In step 85, the data patterns of a plurality of second sensing data are captured by the sliding window method = {S _i ,S _i+1 ,...S _{i+data period sample points} }. In step 87, the modal characteristics of each of the plurality of second sensing data are calculated to obtain the plurality of second data modal characteristics. In step 89, the first data aspect feature is compared with the second data aspect feature. In step 90, compare whether the ratio of the same modal feature exceeds the threshold value, if yes, it is determined that the aligned part has been found; if not, then step 91 is executed. In step 91, the value of the sample point parameter i is increased by 1, and the sliding window continues to move one sample point to the right.

Please refer to FIG. 15, which shows a schematic diagram illustrating a plurality of second data aspect characteristics obtained when any one of the second sensing data S_Data2 is captured in different parts by the sliding window method in step 85. For windows 602, 604, and 606 in different positions, the second data pattern features of the captured data patterns 608, 610, and 612 are, for example, F_1=[+,+,+,-], F_2=[+ ,+,-,-] and F_M=[-,-,+,+].

Please refer to FIG. 16, which is a schematic diagram illustrating that in step 90, when the sliding window method is used to capture different parts of the second sensing data S_Data2. In the process of sliding the window, it is necessary that the 5 second data modal features captured by the sliding window from the second sensing data S_Data2 are similar to the corresponding 5 first data modal features greater than a threshold. When the value is, for example, 80% to 90%, the data pattern alignment module 80 will determine that the second data patterns are aligned with the corresponding first data patterns. As shown in Figure 16, for the 5 data patterns of the 5 second sensing data obtained by the window 614, the second data pattern feature of the second sensing data S_Data2_A of the first processing device 4A is [+, +,-,-], the second data aspect characteristic of S_Data2_B in the second sensing data of the second processing device 4B is [-,-,+,+], and so on, the second of the fifth processing device 4E The second data aspect characteristic of the sensing data S_Data2_E is [+,-,-,+]. It is not similar to the corresponding first data pattern feature shown in Figure 13, so it is known that 5 of the 5 second sensing data obtained by the window 614 are The data patterns are not aligned with the 5 data patterns of the 5 first sensing data obtained by the block 1300 in FIG. 13.

When the sliding window is the window 616, the data pattern characteristic is calculated from the slope of the first line segment completely covered by the window 616. The second data pattern characteristic of the second sensing data S_Data2_A of the first processing device 4A is F_A2=[- ,-,+,+], the second data pattern feature of the second sensing data S_Data2_B of the second processing device 4B is F_B2=[+,+,+,-], and so on, the fifth processing device 4E The second data aspect characteristic of the second sensing data S_Data2_E is F_E2=[-,+,+,+]. In this way, the similarities between the five second data pattern features corresponding to the window 614 and the corresponding five first data pattern features shown in Figure 13 are greater than a threshold value (for example, F_A2~ as shown in Figure 16). The values of F_E2 are respectively the same as F_A1~F_E1 shown in Figure 13), it is determined that the part of the second sensing data S_Data2 that can be correctly aligned with the first sensing data S_Data1 has been found, that is, for example, the part of the first sensing data S_Data1. The portion of the second sensing data S_Data2 indicated by the window 616 may be aligned with the portion of the first sensing data S_Data1 indicated by the dashed box 1300 in FIG. 13. The portion of the second sensing data S_Data2 indicated by the window 614 in FIG. 16 is regarded as not aligned with the portion of the first sensing data S_Data1 indicated by the dashed box 1300 in FIG. 13.

As shown in FIG. 6, the data processing system 160 further includes a feature extraction module 100 for extracting the aligned first feature parameters of the first sensing data S_Data1 and the second sensing data S_Data2. A plurality of second characteristic parameters are combined, and these first characteristic parameters and these second characteristic parameters are combined.

Please refer to FIG. 17, which shows a schematic diagram of the feature extraction module 100 extracting feature parameters of each processing device according to the data pattern. In this embodiment, five processing devices 4A-4E are taken as an example for illustration, but this embodiment is not limited to this. As shown in Figure 17, for example, multiple sets of characteristic parameters X [A,1] ~X _{[A,5] of the first processing device 4A and multiple sets of characteristic parameters X [B,1] of the} second processing device 4B are taken out _. ~X _[B,5] ..., and multiple sets of characteristic parameters X _[E,1] ~X _{[E,5] of the} fifth processing device 4E. The above-mentioned characteristic parameters are, for example, the sensing values periodically captured by a specific window. For example, the first 100 sets of X _[A,1] ~X _[E,5] data may be captured by the window 1300 shown in Figure 13 100 sets of sensing values of, and the other 50 sets of X _[A,1] ~X _[E,5] data may be 50 sets of sensing values captured by window 616 as shown in Figure 16. For example, the first set of data patterns X _[A,1] ~X _[E,5] shown in the first column of Figure 17, where X _[A,1] ~X _[A,5] is [129.9,112.7] ,84.6,56.5,112], which correspond to the value of S_Data1_A captured in the window 1300 shown in Figure 13. Where X _[B,1] ~X _[B,5] is [77.7,79.7,105.7,132.8,111.7] which corresponds to the value of S_Data1_B captured by window 1300 in Figure 13, and so on, where X _[E,1] ~X _[E,5] is [154.2,111.1,118.5,162.3,187.1] which corresponds to the value of S_Data1_E captured by window 1300 in Figure 13. In this way, the first data set of the 1*25 matrix is obtained after the fields are merged. Then complete the second to 100th sets of sensing values captured by the window 1300 shown in FIG. 13, and then complete the 50 sets of sensing values captured by the window 616 shown in FIG. The 100 sets of sensing values captured by the window 1300 are combined with the 50 sets of sensing values captured by the window 616 in FIG. 16, so that a 150*25 matrix is finally obtained. A plurality of first characteristic parameters and a plurality of second characteristic parameters are extracted from the first sensing data S_Data1 and the second sensing data S_Data2 respectively, and the respective corresponding characteristic parameters are combined. Whenever the data processing system 160 performs one more window capture action on the five processing devices 4A~4E, each processing device will get 5 updated characteristic parameters, so the 5 processing devices will get 25 updated characteristic parameters in total. . The feature extraction module 100 will continue to merge the fields to obtain a new set of data (1 row * 25 columns) for subsequent test data.

As shown in FIG. 6, the data processing system 160 also includes a fault prediction module 110, which is used to perform a fault prediction module 110 based on the combined samples (such as the first sample set and the aligned second sample set). Set) and the characteristic parameter module (1 row * 25 column) obtained by merging the fields to perform prediction model training to construct a fault prediction model. And use a plurality of third characteristic parameters extracted from a test data to perform failure prediction based on the failure prediction model. The fault prediction module 110 participates in training the fault model according to the combined first feature parameter and the second feature parameter output by the feature extraction module 100. After that, the failure prediction module 110 can predict when the multiple processing devices 4A-4E may fail based on the trained failure model. The fault prediction module 110 can use the characteristic parameters output after the above-mentioned Fig. 8 to Fig. 16 related process calculations as training data, and train the model through machine learning. The failure prediction module 110 can input test data into the model to predict failure events, and then output equipment health indicators and provide warnings. The test data is, for example, a plurality of third characteristic parameters corresponding to a third sensing data obtained during normal operation of the processing devices 4A-4E after the failure model training is completed. The aforementioned plurality of third characteristic parameters are, for example, the sensing values of the third sensing data periodically captured by a specific window. The waveform of the third sensing data is similar to the waveform of the first sensing data S_Data1 shown in FIG. 13 or the waveform of the second sensing data S_Data2 shown in FIG. 16. The test data acquisition method here is the same as the feature extraction method of the modeling data, except that the extraction period is different. The test data does not need to be processed by the circular correlation building module, the data pattern building module, and the data pattern alignment module. After the data acquisition module outputs the test data to the feature extraction module, the feature extraction module can acquire a set of test data, and then input this set of test data into the fault prediction module for fault prediction. The above-mentioned machine learning training model is, for example, the SOM (Self Organizing Map) model, or other similar machine learning models, such as: Density-Based Anomaly Detection, Or it is a cluster-based anomaly detection model (Clustering-Based Anomaly Detection) and other related machine learning training models that are well known in the industry.

The simulation data shown in Table 1 below is a simulation performed by the data processing system 160 for the five-bed regenerative incinerator jump event. This simulation is based on a normal model (Normal Model) classifier established by SOM. Compared with the method of training the model only by merging historical data and new data, and the method of training the model by only providing relevant data with multiple loop processing steps, this disclosure corrects the wrong data and The method of training the model after aligning the historical data with the newly sensed data can greatly increase the true positive rate (TPR) of the fault prediction to 90.91%, and obtain a high-precision prediction accuracy result.

The data processing system and method of the embodiment of the present disclosure ensure the correctness of the received sensing data by judging the point pattern characteristics of the sample points. In addition, the sliding window method is used to capture the data patterns and compare the patterns according to the characteristics of the data patterns, which can make The historical base data and the new sensing data are aligned. In this way, traditional methods can be avoided because the actual processing data often has errors, such as in an unstable state, or generating noise, data omissions or errors, etc., or because the historical base data and the new sensing data are not correct Alignment leads to the inability to effectively establish the problem of incorrect abnormal condition information in the equipment failure prediction system. Therefore, in the embodiment of the present disclosure, data alignment is performed through pattern comparison and the use of pattern features, and a more accurate fault model can be trained, which can effectively greatly improve the hit rate of equipment failure prediction.

To sum up, although the present disclosure has been disclosed as above through the embodiments, it is not intended to limit the present disclosure. Those with ordinary knowledge in the technical field to which this disclosure belongs can make various changes and modifications without departing from the spirit and scope of this disclosure. Therefore, the scope of protection of this disclosure shall be subject to the scope of the attached patent application.

202~206: Process steps

Claims (24)

A data processing system includes: a cyclic correlation building module for receiving a plurality of first sensing data obtained by performing a first sensing on a plurality of processing devices, and receiving cyclic program information of a processing step The cycle correlation building module is further used to obtain the data correlation of the first sensing data according to the number of sample points in a data cycle of the first sensing data and the cycle procedure information table of the processing step , To calibrate the first sensing data; a data pattern building module for obtaining a plurality of first data pattern features of the first sensing data; a data pattern alignment module for According to the first data aspect characteristics, a plurality of second sensing data obtained by performing a second sensing on the processing devices are aligned with the first sensing data. For the data processing system described in item 1 of the scope of patent application, the processing step and cycle program information table records the multiple processing steps of the processing equipment and the related cycle operation mode information of the multiple cycle programs. For example, the data processing system described in item 2 of the scope of patent application, wherein the cyclic correlation building module is used to obtain the data cycle sample points from the first sensing data through an autocorrelation analysis method And obtain a sampling magnification between the number of the processing steps and the number of sample points of the data period, and obtain the data offset points between the first sensing data according to the sampling magnification, to Obtain the data relevance of the first sensing data. The data processing system described in item 2 of the scope of patent application, wherein the cycle correlation building module is further used to confirm the difference steps between the cycle procedures Number to obtain the data relevance of the first sensing data, the data relevance of the first sensing data includes one of the first sensing data and the other of the first sensing data The number of data offset points between. According to the data processing system described in item 4 of the scope of patent application, the cyclic correlation building module is further used to shift the data according to the first one of the first sensing data corresponding to the number of offset points of the data. The point pattern feature of a sample point and the point pattern feature of another second sample point of the first sensing data are compared with the first sample of one of the first sensing data Point and the data correctness of the second sample point of the other of the first sensing data to calibrate the first sensing data. For the data processing system described in item 5 of the scope of patent application, wherein the data pattern construction module further calculates the first sample point of one of the first sensing data to be the same as the first This point corresponds to the positive trend or the negative trend of the other two sample points before and after, and the characteristic of the point of the first sample point is obtained. For example, the data processing system described in item 1 of the scope of patent application, wherein the data pattern construction module is further used to retrieve a plurality of data patterns of the first sensing data according to the number of sample points of the data period, To obtain the first data aspect characteristics of the first sensing data. For example, in the data processing system described in item 7 of the scope of patent application, the data pattern construction module further obtains the data by calculating the positive trend or the negative trend of a plurality of sample points of the first sensing data. Data style. For example, in the data processing system described in item 1 of the scope of patent application, the data pattern alignment module is further used for using a sliding window method to capture a plurality of second data patterns of the second sensing data, and Obtain a plurality of second data aspect characteristics of the second data aspect, and compare the second data aspect characteristics with the corresponding ones The first data pattern feature, when the similarity between the second data pattern features and the corresponding first data pattern features is greater than a threshold value, the data pattern alignment module is used to determine the second data pattern features The data pattern is aligned with the corresponding first data patterns. The data processing system described in item 1 of the scope of patent application further includes a feature extraction module for extracting a plurality of first feature parameters of the aligned first sensing data and the second sensing data A plurality of second characteristic parameters, and merge the first characteristic parameters and the second characteristic parameters. The data processing system described in item 10 of the scope of patent application further includes a fault prediction module for performing prediction model training based on the merged first characteristic parameters and the second characteristic parameters to construct a fault A prediction model, and a plurality of third characteristic parameters extracted from a test data are used to perform failure prediction based on the failure prediction model. The data processing system described in item 1 of the scope of patent application further includes a Graphic User Interface (GUI) for displaying at least one option related to the cyclic operation mode. A data processing method includes: receiving a plurality of first sensing data obtained by performing a first sensing on a plurality of processing devices, and receiving a processing step and cycle program information table, and according to the first sensing data The data relevance of the first sensing data is obtained by the number of sample points in a data cycle and the processing step and cycle program information table, so as to calibrate the first sensing data; and obtaining a complex number based on the first sensing data First data aspect characteristics; according to the first data aspect characteristics, a plurality of second sensing data obtained by a second sensing of the processing equipment is aligned with the first sensing data. The data processing method described in item 13 of the scope of patent application, wherein the processing step and cycle program information table records the multiple processing steps of the processing equipment and the related cycle operation mode information of the multiple cycle programs. For the data processing method described in item 14 of the scope of patent application, wherein the number of sample points of the data period is obtained from the first sensing data through an autocorrelation analysis method, and the number of processing steps and the number of processing steps A sampling magnification between the sample points of the data period, and the data offset points between the first sensing data are obtained according to the sampling magnification to obtain the data relevance of the first sensing data. For example, the data processing method described in item 14 of the scope of patent application, wherein, according to the data cycle sample points of the first sensing data and the processing step and cycle program information table, the data of the first sensing data is obtained Data relevance. The step of calibrating the first sensing data includes: confirming the step number of the phase difference between the cyclic procedures to obtain the data relevance of the first sensing data, the first sensing data The data relevance of the data includes a data offset point between one of the first sensing data and the other of the first sensing data. According to the data processing method described in item 16 of the scope of patent application, the data cycle sample points of the first sensing data and the processing step and cycle program information table are used to obtain the data of the first sensing data Data relevance, the step of calibrating the first sensing data further includes: according to the point pattern feature of a first sample point of one of the first sensing data corresponding to the number of offset points of the data Compared with the point pattern feature of another second sample point of the first sensing data, the first sample of one of the first sensing data Point and the data accuracy of the second sample point of the other of the first sensing data to calibrate the first sensing data. The data processing method according to item 17 of the scope of patent application, wherein the first sample point of one of the first sensing data is calculated by corresponding to the first sample point before the other two The positive or negative trend of the sample point is used to obtain the point pattern characteristics of the first sample point. According to the data processing method described in item 13 of the scope of patent application, wherein the step of obtaining the first data pattern characteristics according to the first sensing data includes extracting the first data sample points according to the data period. A plurality of data patterns of the sensing data are obtained to obtain the characteristics of the first data patterns of the first sensing data. The data processing method described in item 19 of the scope of patent application, wherein the data patterns are obtained by calculating the positive or negative trends of a plurality of sample points of the first sensing data . According to the data processing method described in item 13 of the scope of patent application, the plurality of second sensing data obtained by performing the second sensing on the processing devices and the plurality of second sensing data obtained from the second sensing are performed on the processing devices according to the characteristics of the first data The step of aligning the first sensing data includes: using a sliding window method to capture a plurality of second data patterns of the second sensing data, and obtain a plurality of second data of the second data patterns Modal characteristics; and comparing the second data modal characteristics with the corresponding first data modal characteristics, when the similarity between the second data modal characteristics and the corresponding first data modal characteristics When it is greater than a threshold value, it is determined that the second data patterns are aligned with the corresponding first data patterns. The data processing method described in item 13 of the scope of patent application further includes: extracting a plurality of first characteristic parameters of the aligned first sensing data and a plurality of second characteristic parameters of the second sensing data. , And merge the first characteristic parameters and the second characteristic parameters. The data processing method described in item 22 of the scope of patent application further includes: performing prediction model training based on the merged first feature parameters and the second feature parameters to construct a fault prediction model, and use the The plurality of third characteristic parameters extracted from the test data are used for failure prediction based on the failure prediction model. The data processing method described in item 13 of the scope of patent application further includes: displaying at least one option related to the cyclic operation mode in a graphical user interface.

Images