JP6749219B2 - Plant operation data analysis system - Google Patents

Description

The present invention relates to a plant operation data analysis system that monitors a plant or the like.

Currently, various plants are operating in various industries to produce various products. If an abnormality occurs in these plants and they stop, production of the products is stopped and repair costs are incurred. In addition, the loss of product production can result in significant losses. It is practically difficult to completely eliminate the occurrence of plant abnormalities and failures. Therefore, it is necessary to understand the sign of abnormality based on the measured value as early as possible.

ART (Adaptive Resonance Theory) is one of the versatile abnormality prediction techniques, for example. This is because the range of normal categories in which normal data exist is automatically determined in advance using the learning data in the normal state, and if the diagnostic data obtained during operation is within the normal categories, it is normal or outside. For example, it is a technology that outputs an alarm by regarding it as an abnormality. At the time of learning, the category range is calculated with the data of n input items, and at the time of diagnosis, whether or not it is within the category is determined with the data of n input items. At the time of diagnosis, it was necessary to have n data input items, and there was a problem that it was not possible to calculate if there was input data that could not be used even one.

The technique described in Patent Document 1 is known for such a problem. The technique described in Patent Document 1 is a system for creating plant diagnostic data. This document proposes a plant diagnostic system for diagnosing a plant abnormality by using adaptive resonance theory, vector quantization, or a neural network based on the correlation of created multidimensional diagnostic data. Then, when the measured value shows an abnormal value or is missing, it is described that it is excluded from the learning data and processed.

On the other hand, the technique described in Patent Document 2 is also known. The technique described in Patent Document 2 is a facility state monitoring method that detects an abnormality based on a time-series sensor signal output from a facility or a device. A learning process of creating a normal model based on the selected learning data and a feature vector extracted based on the sensor signal are compared with the normal model to perform abnormality identification. Then, when the acquired multi-dimensional time series signal is missing, it is described that such data is deleted.

JP, 2015-230576, A JP, 2011-070635, A

By using the techniques described in Patent Documents 1 and 2, it is possible to deal with data loss and abnormal values. However, in order to prevent production stoppage due to equipment failure, an actual plant may be equipped with a plurality of main equipment and used for rotation. In such a case, since the data of the stopped equipment cannot be obtained, any operation data is always lost. When the process of deleting missing data is executed as in the technique described in Patent Document 1 or Patent Document 2, all data of other input items containing valid numerical values will be deleted in some cases. That is, there is a possibility that a large amount of operation data that cannot be used for abnormality prediction may appear, which is inefficient. As a result, abnormality prediction cannot be performed, and as a result, there is a possibility that the plant will be broken down and must be stopped for a long period of time.

Although the above description is limited to the abnormality prediction for ease of understanding, there may be an analysis means for analyzing operation data other than the abnormality prediction when the plant is operating. Even in such a case, if the input data includes a defect or an abnormal value, the valid data of other input items may be deleted, which is inefficient.

It was made in view of such a problem, the object of the present invention, even if the invalid data such as missing or abnormal value when predicting the abnormality of the plant is included, the validity of other input items To provide a plant operation data analysis system that utilizes various operation data.

The plant operation data analysis system of the present invention is a combination of operation data of a plurality of items obtained from the target equipment, extraction means for extracting an item having a valid numerical value from the operation data, and operation data of the extracted plurality of items. The present invention is characterized in that it has an analysis means for performing analysis according to the above, and a display means for displaying information of the extracted item or the item not extracted together with the output of the analyzing means.

According to the present invention, even if all the operation data is not complete and some missing or abnormal value data is included, such as a plant having equipment operating in rotation, it is possible to analyze the operation data and predict the abnormality more. Can continue appropriately and efficiently.

The figure showing the example of composition of the plant operation data analysis system concerning a first embodiment. An example of a display screen that allows the operator to evaluate the analysis result without misunderstanding. An example of the screen which displays an analysis result in a different graph frame according to the combination of the extracted items. The example of the screen which displays the analysis result in the same graph frame with the legend of different shape and color according to the combination of the extracted items. An example of a screen that displays analysis results in the same graph frame with lines of different shapes and colors depending on the combination of the extracted items. The figure which shows the structural example of the plant operation data analysis system which concerns on 2nd embodiment. The figure which shows the structural example of the plant operation data analysis system which concerns on 3rd embodiment. An example of causal relationship data held in a causal relationship storage device of multiple items.

Next, modes for carrying out the present invention (referred to as “embodiments”) will be described in detail with reference to the drawings as appropriate.

<Example 1>
FIG. 1 is a diagram showing a configuration example of a plant operation data analysis system according to the first embodiment. The target facility 02 may be a device, a facility, a facility, a plant, or a large-scale system as long as it can obtain the operation data 04 of a plurality of items. The operation data 04 of a plurality of items obtained from the target equipment 02 is given to the extraction means 06. The extraction means 06 has a function of extracting only the items for which valid numerical values exist from the operation data 04 of the plurality of items. Here, the "effective number" is not missing the number in the operation data, the character string such as "N/A""DIV/0""-""notmeasured" is not entered, It also means the numerical value when it does not exceed the set upper and lower limits. The operation data of the items for which effective numerical values are present (that is, the extracted operation data 08) is given to the analysis means 10. The analysis means 10 processes the extracted operation data 08 and gives the output 12 of the analysis means to the display means 16. Further, the information 14 of the extracted item or the item not extracted is given from the analysis means 10 to the display means 16. In the display means 16, at the same time as the output 12 of the analysis means, it is possible to know which item of data was used to obtain the result, or which item of data was not used, so that the extracted item or the item not extracted Information 14 is displayed.

The analysis means 10 for analyzing the operation data 04 may be any as long as it uses the operation data 04 of a plurality of items as an input, performs some information processing, and outputs the output 12 of the analysis means. For example, Mahalanobis Taguchi method, adaptive resonance theory, vector quantization, neural network, subspace method, deep learning, etc. are listed as typical analysis means 10. Among them, Mahalanobis Taguchi method, adaptive resonance theory, vector quantization, and subspace method have anomaly prediction function.

In general, the analyzing means 10 is premised on that the operation data 04 of a plurality of items have all valid numerical values of all items. For example, it can be analyzed that data is missing even in only one item of the plurality of items. Since it cannot be used, the operation data 04 at that time may not be used for analysis.

In reality, the target facility 02 is composed of a plurality of devices, and the devices may be used in rotation. In such a case, one of the devices will be put to sleep without fail. If the analysis means 10 tries to analyze the operation data 04 for such a target facility 02, the analysis cannot be executed because the data is missing at any time. With the configuration shown in FIG. 1, it is possible to solve such a problem. In FIG. 1, extraction means 06 is provided. If even a part of the operation data 04 obtained from the target equipment 02 has an effective numerical value, it is possible to reduce wasteful data by analyzing using only that numerical value.

However, if the input items of the extracted operation data 08 given to the analysis means 10 are different, it is general that the outputs 12 of the analysis means cannot be compared side by side. Therefore, the screen as shown in FIG. 2 is displayed by the display means 16 so that the operator can evaluate the analysis result properly without misunderstanding. In the first line of FIG. 2, the unused input item “input item 2” is displayed. It is also good to display why the input item was not used and the reason for it, as shown by "missing" on the right side. Further, instead of the first line in FIG. 2, the input item “used in the analysis” may be shown. The second line of FIG. 2 is the information of the output 12 of the analysis means, and here, as an example, the result of the abnormality prediction calculation is shown. The figure below it is an example of a trend graph, and the dotted line on the right shows the latest time. For example, when the analysis means 10 has a function of predicting an abnormality, the alarm level is gradually increased or decreased with respect to time by displaying information to the operator in a trend graph as shown in this figure. It is possible to convey intuitively easy-to-understand information such as the presence or the situation has not changed much.

The input items of the operation data 04 may change in a pattern with time. That is, if there are 3 pumps from Unit 1 to Unit 3 in a certain plant and they are operating by rotation one by one, the analysis result when Unit 1 is operating, the analysis when Unit 2 is operating As a result, three patterns of analysis results when Unit 3 is moving are assumed. When the patterns are displayed together in a trend graph as shown in FIG. 2, numerical values that cannot be compared are originally arranged side by side, which may cause misunderstanding.

In order to avoid this, for example, as shown in FIG. 3, there is a method of displaying the analysis result in a different graph frame depending on the combination of the extracted items. FIG. 3 shows an example in which three types of trend graphs are displayed on one screen. The text (text) in the upper left shows the latest status, and in this example, the data for pumps 2 and 3 are currently missing. It also shows that the calculation result of the abnormality prediction is the warning level 2. The trend graph on the right side shows the transition of the alarm level in the pattern "the data of pumps 2 and 3 are missing". The dotted line on the right shows the latest time. The blank space in the plot is the period of the pattern that "the data of pump 2 and pump 3 were not missing". Similarly, the lower left corresponds to the “data of pumps 2 and 1 were missing” pattern, and the lower right is the pattern of “data of pumps 1 and 3 were missing”. In this way, by displaying the trend graph in different graph frames when there are three types of patterns, it is possible to accurately grasp the past results and transitions corresponding to the combination of devices currently operating. ..

Alternatively, as shown in FIG. 4, there is a method of displaying in the same graph with legends of different shapes and colors depending on the combination of the extracted items. FIG. 4 shows an example in which one trend graph is displayed on one screen. Legends with different shapes and colors are assigned to combinations of operating equipment. With such a display, the past result corresponding to the combination of devices currently operating can be grasped without error. Further, the number of trend graphs to be displayed can be reduced as compared with the case of FIG. 3, and it is possible to effectively use the screen of the display to be displayed and at the same time eliminate erroneous selection of graphs to be viewed.

Alternatively, as shown in FIG. 5, there is a method of displaying in the same graph as lines of different shapes and colors depending on the combination of the extracted items. FIG. 5 shows an example in which one trend graph is displayed on one screen. Lines with different shapes are assigned to combinations of operating equipment. With such a display, the past result corresponding to the combination of devices currently operating can be grasped without error. Further, the number of trend graphs to be displayed can be reduced as compared with the case of FIG. 3, and it is possible to effectively use the screen of the display to be displayed and at the same time eliminate erroneous selection of graphs to be viewed. Unlike the case where only the legend is shown as in FIG. 4, the trend of increase and decrease can be more easily grasped by the line corresponding to the combination of operating devices. However, if there are too many combinations of operating devices, the number of displayed lines will increase, which may make it difficult to see.

As an output, it is also possible to use the combination of FIG. 4 and FIG. 5, that is, display the legend and the line at the same time. Thereby, there is a possibility that the information can be presented to the operator in a more understandable manner.

As described above, by adopting the first embodiment, it is possible to reduce the waste of the operation data 04 which has been removed because data loss or effective data does not exist in the past. Further, even when the analysis results cannot be evaluated side by side with respect to the data set used in the analysis, it is possible to output information that the operator can judge without misunderstanding.

<Example 2>
FIG. 6 is a diagram showing a configuration example of a plant operation data analysis system according to the second embodiment. The target facility 02 may be a device, a facility, a facility, a plant, or a large-scale system as long as it can obtain the operation data 04 of a plurality of items. The operation data 04 of a plurality of items obtained from the target equipment 02 is given to the extraction means 06. Further, the past driving data 22 of a plurality of items obtained from the storage device of the past driving data is also given to the extracting means 06.

The extraction means 06 has a function of extracting only the items for which valid numerical values exist from the operation data 04 of the plurality of items. Here, the valid numerical value means that the numerical value is missing in the operation data, or there is no character string such as "N/A", "DIV/0", "-", "not measured", and it is set. It means the numerical value when the upper and lower limits are not exceeded. Different from the first embodiment, in the second embodiment, an item in which the above-mentioned effective numerical value exists in the past operation data 22 obtained by the extraction means 06 from the storage device 18 of the past operation data. It also has the function of extracting the driving data of

The extracted operation data 08 (data extracted from the present operation data and the past operation data) is given to the analysis means 10. The analysis means 10 processes the extracted operation data 08 and gives the output 12 of the analysis means to the display means 16. Further, the information 14 of the extracted item or the item not extracted is given from the analysis means 10 to the display means 16. On the display means 16, at the same time as the output 12 of the analysis means, it is possible to know which item of data was used to obtain the result, or which item of data was not used, so that the extracted item or the item not extracted Information 14 is displayed.

By adopting this form, it becomes possible to use the analysis means 10 that also uses the past operation data 22 as compared with the first embodiment. For example, it is possible to apply the analysis means 10 that machine-learns the range of normal operation using the past operation data 22 and applies the current operation data 20 to it to judge whether it is abnormal or normal. Becomes

<Example 3>
FIG. 7 is a diagram showing a configuration example of a plant operation data analysis system according to the third embodiment. The target facility 02 may be a device, a facility, a facility, a plant, or a large-scale system as long as it can obtain the operation data 04 of a plurality of items. The operation data 04 of a plurality of items obtained from the target equipment 02 is given to the extraction means 06. Further, the past driving data 22 of a plurality of items obtained from the storage device of the past driving data is also given to the extracting means 06.

The extraction means 06 has a function of extracting only the items for which valid numerical values exist from the operation data 04 of the plurality of items. The valid numerical value here means that the numerical value is missing in the operation data, or the character string such as "N/A", "DIV/0", "-", "not measured" is not included, and it is set. It means the numerical value when the upper and lower limits are not exceeded. In addition to this, the extraction means 06 extracts, from the past operation data 22 obtained from the past operation data storage device 18, the operation data of the item in which the above-mentioned effective numerical value exists.

Unlike the first and second embodiments, the causal relationship data 26 is given from the causal relationship storage device 24 to the extraction means 06. An example of the causal relationship data 26 is shown in FIG. Although FIG. 8(a) is shown as a table for easy understanding, it may actually be given as data in csv format or text format. In this table, "Cause" of "causal relationship" is shown on the left and "Cause" of "causal relationship" is shown above. This is a table showing whether the items on the left side affect the items on the upper side. Has become. In the table, "0", "1" and "2" are shown as numerical values. Of these, "0" indicates that there is no causal relationship, "1" indicates that there is a causal relationship, and "2" indicates that they are the same item. The operation of the extraction means 06 for the causal relationship data 26 is as follows.

(1) If the value is "2", ignore it because it is the same item.

(2) If the numerical value is “0”, it means that there is no causal relationship, so ignore it.

(3) If the numerical value is “1”, it means that there is a causal relationship, so if the “cause” data is not a valid numerical value, even if the “result” data is a valid numerical value, The data is considered indefinite and is not treated as a valid number.

For example, when the causal relationship data 26 of FIG. 8A is set, the obtained data is (flow rate 1, flow rate 2, flow rate 3, total flow rate)=(100, N/A, N/A, 450). It is also assumed that the upper limit of flow rate is 1000 and the lower limit is 0. First, the extraction means 06 extracts effective data from this. According to the rules described above, the values of "Flow rate 1" and "Total flow rate" that have no character string and are within the upper and lower limits are valid, and the values of "Flow rate 2" and "Flow rate 3" that contain the character string are valid. Judge that the value is not valid. After that, the causal relationship data 26 is referred to. First, referring to the first line, no processing is performed because "flow rate 1" is valid. Then refer to the second line. Since "flow rate 2" is not valid, the causal relationship is investigated using this table. Since the value "1" is "total flow rate", the value of "total flow rate", which is a numerical value of 450 in this case, is treated as ineffective. Next, referring to the third line, similarly, the process of considering that the value of "total flow rate" is not valid is performed. Finally, refer to the 4th line. Since "total flow" has been processed as not valid, search this line. However, since the item with the numerical value "1" is not found, the processing ends. After all, the extraction means 06 selects only "flow rate 1" as an item in which an effective numerical value exists among the operation data 04 of a plurality of items, and supplies it to the analysis means 10 as the extracted operation data 08.

It is assumed that the total flow rate=flow rate 1+flow rate 2+flow rate 3. Among these, when the flow rate 2 and the flow rate 3 are unknown (N/A), there are innumerable combinations. Even if data processing is performed with the total flow rate = 450, the actual operation modes (flow rate 2, flow rate 3) = (0, 450) and (225, 225) and (450, 0) may have completely different states. However, if they are collectively recognized as the same state and analyzed, a problem may occur in the analysis result. When using the abnormality prediction function that uses the learning data and the diagnostic data, the learning data (the flow rate 2, the flow rate 3 )=(0,450), if the diagnostic data is (flow rate 2, flow rate 3)=(450,0), there is a possibility of failure. At that time, "I was able to analyze the data and determined that it was normal. Despite that, the reality has broken down."

On the other hand, when the causal relationship data 26 of FIG. 8(a) is set, the obtained data is (flow rate 1, flow rate 2, flow rate 3, total flow rate)=(100, 120, 80, N/A) It was. It is also assumed that the upper limit of flow rate is 1000 and the lower limit is 0. First, the extraction means 06 extracts effective data from this. According to the rules described above, it is determined that the values of “flow rate 1”, “flow rate 2”, and “flow rate 3” are valid, and the value of “total flow rate” containing a character string is not valid. After that, the causal relationship data 26 is referred to. First, referring to the first line, no processing is performed because "flow rate 1" is valid. Then refer to the second line. Since "flow rate 2" is also effective, no processing is performed. Similarly, no processing is performed on "flow rate 3". Finally, refer to the 4th line. Since "total flow rate" is not valid, search this line. However, since the item with the numerical value "1" is not found, the processing ends. Eventually, the extraction means 06 selects "flow rate 1", "flow rate 2", and "flow rate 3" as the items having valid numerical values in the operation data 04 of the plurality of items, and the analysis means 10 outputs the extracted operation data 08 to the analysis means 10. give. In this case, the problem of innumerable combinations as described above does not occur, and it can be said that the evaluation that "there is a failure in reality despite the fact that the data can be analyzed and is judged to be normal" does not occur. ..

A case where the causal relationship data 26 is as shown in FIG. 8B is also assumed. The difference from FIG. 8A is that “1” is included in the lower left cell of the numerical value “2”. This means that the search needs to be performed retrospectively or iteratively searched. As an example, it is assumed that the obtained data is (flow rate 1, flow rate 2, flow rate 3, total flow rate)=(100, N/A, 600, 950). According to the rules described above, it is determined that the values of "flow rate 1", "flow rate 3" and "total flow rate" are valid, and the value of "flow rate 2" containing a character string is not valid. After that, the causal relationship data 26 shown in FIG. 8B is referred to. First, referring to the first line, no processing is performed because "flow rate 1" is valid. Then refer to the second line. Since "flow rate 2" is not valid, a causal relationship is searched for. Since the value "1" is "flow rate 3", the value of "flow rate 3" is assumed to be a valid value in terms of the upper and lower limits of 600 here, but this value is considered to be invalid. Apply processing. Next, refer to the third line. Since "flow rate 3" has been processed as not valid, this line is searched. Then, it is understood that the numerical value being "1" is "flow rate 1". Therefore, a process for considering "flow rate 1" as ineffective is performed. Next, referring to the 4th line, no processing is performed because "1" is not included. Then, it returns to the first line again. Since "flow rate 1" was processed as not valid, "1" is described in "total flow rate" when searched, so processing that considers the value of "total flow rate" as not valid. Next, refer to the second line and confirm that new processing does not occur. Similarly, confirm that no new processing occurs on the 3rd and 4th lines. Furthermore, it is confirmed that no new processing will occur from the first line to the fourth line, and the process ends. As a result, no valid data remains. In this way, when "1" is included in the lower left cell of the numerical value "2", the search is performed retroactively or repeated.

As described above, by adopting the third embodiment, it is possible to analyze the operation data 08 extracted by excluding the items affected by the indefinite condition and increase the possibility of obtaining a more accurate result. be able to.

02 Target equipment
04 Operation data
06 Extraction means
08 Extracted operation data
10 Analytical means
12 Output of analysis means
14 Information on extracted items or items not extracted
16 Display means
18 Past operation data storage device
20 Current operation data
22 Past operation data
24 Causal memory device
26 causal data

Claims (7)

Operation data of multiple items obtained from the target equipment,
Extraction means for extracting items in the operation data for which valid numerical values exist,
Analyzing means for analyzing the extracted operation data of a plurality of items in combination,
Display means for displaying the information of the extracted items or the items not extracted, together with the output of the analysis means,
A plant operation data analysis system comprising: Current operation data of multiple items obtained from the target equipment,
A storage device that stores past operation data of multiple items obtained from the target equipment,
Extraction means for extracting from the current operation data and past operation data the items for which valid numerical values exist among the operation data,
Analyzing means for analyzing the present operation data and past operation data of a plurality of extracted items in combination,
Display means for displaying the information of the extracted items or the items not extracted, together with the output of the analysis means,
A plant operation data analysis system comprising: 3. The extraction unit according to claim 1, wherein the extraction unit searches for and extracts an item affected by an item for which a valid numerical value does not exist, based on causal relationship data obtained from a causal relationship storage device that stores a causal relationship of a plurality of items. A plant operation data analysis system having a function of removing from data. 4. The plant operation data analysis system according to claim 1, wherein the analysis unit has an abnormality prediction function of predicting an abnormality in the target equipment. 5. The plant operation data according to claim 1, wherein at least one of Mahalanobis Taguchi method, adaptive resonance theory, vector quantization, neural network, subspace method, and deep learning is used as the analysis means. Analysis system. The plant according to any one of claims 1 to 5, wherein when the output of the analysis unit is displayed as a trend graph as the display unit, the output is displayed in a different graph frame depending on the combination of the extracted items. Operation data analysis system. 7. The display unit according to claim 1, wherein when the output of the analysis unit is displayed as a trend graph, a legend of a different shape or color or a different shape or color according to the combination of the extracted items is displayed. A plant operation data analysis system characterized by being displayed as a line in a graph.

Images