JP7447758B2 - Plant abnormality prediction system - Google Patents

Description

The present invention relates to a plant abnormality prediction system.

In factories, process data is collected from sensors in real time. Process data is numerical data arranged in chronological order and changes from moment to moment. Generally, a threshold value is set for each sensor, and when process data exceeds the threshold value, it is detected as an abnormal situation. In recent years, technology has been established to identify the cause of an abnormality by analyzing process data. Patent Document 1 discloses that process information is used to detect an abnormality, and maintenance information is used to perform an abnormality diagnosis for an abnormal state found by the abnormality detection.

Japanese Patent Application Publication No. 2011-227706

However, in Patent Document 1 mentioned above, since daily process information is not described in the maintenance information, it is difficult to perform an abnormality diagnosis that takes process information into consideration. Furthermore, since abnormality detection is performed using only process information, it has not been possible to incorporate daily operation information of operators into the detection algorithm.

This invention was made to solve the above-mentioned problems. An object of the present invention is to provide a plant abnormality prediction system that can predict abnormalities in plant equipment by combining process data of the plant equipment and operating information recorded by operators.

In order to achieve the above object, a plant abnormality prediction system according to the present invention is configured as follows.

The plant abnormality prediction system includes a process data storage section, a text data storage section, an abnormality prediction model learning section, an abnormality prediction section, and a display section.

The process data storage unit stores process data including state values of plant equipment measured by sensors at each time.

The text data storage unit stores operator intervention data that is text data recorded by an operator and associates an event based on a change in the process data with a treatment performed by the operator in response to the event. .

The anomaly prediction model learning unit creates a decision tree, which is a prediction model, using the accumulated process data and the operator intervention data as explanatory variables and the presence or absence of an abnormality as an objective variable.

The anomaly prediction unit uses the decision tree to predict whether or not an abnormality will occur depending on whether future measures will be taken for an event based on a newly measured change in the process data.

The display unit displays the newly measured process data and the details of the future treatment when it is predicted that the abnormality has occurred.

Preferably, the operator intervention data further includes a cause for the occurrence of the event. The display unit displays an alarm when the abnormality is predicted to occur, and displays a past event related to the newly measured process data and a past event related to the event based on the operator intervention data. Displays the attached action and cause.

According to the plant abnormality prediction system according to the present invention, the occurrence of abnormalities in plant equipment is predicted by combining process data of plant equipment measured by sensors and operation information (operator intervention data) recorded by operators. can. According to this, even if no abnormality has occurred at the present time, it is possible to predict the abnormality that will occur if future measures are not taken for the current event. Further, according to the present invention, it is possible to display measures that can be taken in advance to prevent abnormalities from occurring in the future.

1 is a block diagram showing the overall configuration of a plant abnormality prediction system according to Embodiment 1 of the present invention. It is a conceptual diagram of data conversion in a text analysis part. FIG. 3 is a conceptual diagram of keyword search in an abnormality factor extraction unit. FIG. 3 is a diagram illustrating an example of predicting the occurrence of an abnormality using a decision tree. FIG. 3 is a diagram for explaining screen information displayed by an abnormality notification section. FIG. 3 is a diagram for explaining screen information displayed by an abnormality notification section. It is a conceptual diagram of model learning. 1 is a conceptual diagram showing an example of the hardware configuration of a processing circuit included in the plant abnormality prediction system according to Embodiment 1 of the present invention.

Embodiments of the present invention will be described in detail below with reference to the drawings. However, in the embodiments shown below, when referring to the number, amount, amount, range, etc. of each element, the reference shall not be made unless it is specifically specified or it is clearly specified to that number in principle. The invention is not limited to this number. Furthermore, the structures and the like described in the embodiments shown below are not necessarily essential to the present invention, unless specifically specified or clearly specified in principle. It should be noted that common elements in each figure are given the same reference numerals and redundant explanations will be omitted.

Embodiment 1.
<Overall configuration>
FIG. 1 is a block diagram showing the overall configuration of a plant abnormality prediction system 1 according to Embodiment 1 of the present invention.

The plant abnormality prediction system 1 is a system for predicting future abnormalities in plant equipment in a factory and preventing the abnormalities from occurring. The plant abnormality prediction system 1 includes a process data input section 10, a process data storage section 11, an operation record input section 12, a text data storage section 13, a text analysis section 14, an abnormality factor extraction section 15, an abnormality prediction model learning section 17, and an abnormality prediction section 17. It includes a prediction section 18 and a display section 21. The display section 21 includes a monitor that displays a screen, an abnormality cause display section 16, an abnormality notification section 19, and a cooperation section 20.

Two types of data are input into the plant abnormality prediction system 1: process data and text data. Process data is state information of plant equipment measured by sensors. The text data is an operation record in which an operator intervenes in the operation of plant equipment.

The process data input unit 10 inputs process data output from each sensor placed at various locations in the factory. The process data includes state values of plant equipment measured by sensors at each time. Specifically, process data includes measurement time, equipment identifiers that identify plant equipment (e.g., equipment number, equipment name), and state values of plant equipment (measured values of temperature, pressure, concentration, flow rate, and valves). (including equipment status such as opening/closing and pump operation).

The process data storage unit 11 stores process data input to the process data input unit 10 in real time. The process data storage unit 11 is, for example, a database.

By the way, in addition to process data, important operational information remains in the operation diary in which operators record daily operation information. In this system, information on driving logs, which were conventionally kept on a paper basis, is accumulated as text data (operation record input section 12, text data accumulation section 13). Furthermore, the accumulated text data is used to analyze the wording of the text (text analysis section 14, abnormal factor extraction section 15).

The operation record input unit 12 inputs text data recorded by an operator who operates plant equipment. The text data is, for example, a driving diary, and includes operation intervention data to be described later. Note that for the same plant equipment, process data and operator intervention data are associated by being assigned a common equipment identifier (equipment number, equipment name).

The text data storage unit 13 stores the text data input to the operation record input unit 12. The text data storage unit 13 is, for example, a database.

The text analysis unit 14 analyzes the accumulated text data and converts it into operator intervention data. Operator intervention data includes "events" based on changes in the state values of plant equipment detected by sensors, the "cause" of the occurrence of the event, and the manual intervention performed by the operator on the plant equipment in response to the event. "Treatment" is included. Further, the "event" is associated with the time of occurrence of the event, the "treatment" is associated with the time when the treatment is implemented, and the plant equipment described in the event/cause/treatment is associated with the above-mentioned equipment identifier. That is, the process data and text data (operator intervention data) described above are associated by time and equipment identifier.

FIG. 2 is a conceptual diagram of data conversion in the text analysis section 14. The text analysis unit 14 can divide the text data into an event, cause, and treatment using, for example, a predetermined delimiter string. In FIG. 2, text data 201 associated with time is divided into "event", "cause", and "treatment" and converted into converted text data 202 (operator intervention data). The converted text data 202 is stored in the text data storage section 13.

The abnormality factor extraction unit 15 organizes the converted text data 202 (operator intervention data) for each "event". FIG. 3 is a conceptual diagram of keyword search in the abnormality factor extraction unit 15. By keyword search, "events" that match the keywords are identified, and "causes" and "treatments" associated with the identified "events" are extracted. Note that an equipment identifier (for example, equipment number, equipment name) is input in the TAG column of FIG. 3 .

The abnormality factor display section 16 can display "event", "cause", and "treatment" extracted by keyword search using the abnormality factor extraction section 15.

The abnormality prediction model learning unit 17 creates a decision tree, which is a prediction model, using the accumulated past process data and operator intervention data as explanatory variables, and using the presence or absence of an abnormality as an objective variable. Specifically, the explanatory variables are the process data at each time, the "event" at each time, the presence or absence of a "treatment" performed for the event, the content of the treatment, and the subsequent process depending on whether the treatment was implemented. Including changes in data. The objective variables include the presence or absence of subsequent abnormality occurrence depending on whether or not the treatment is performed.

FIG. 7 is a conceptual diagram of predictive model learning. Features of the process data corresponding to the operator intervention data extracted from the text data by abnormal factor extraction (72) are extracted (71). The feature amount is extracted using the equipment identifier and numerical value included in the event of the operator intervention data. Process data features and operator intervention data are input into the learning model (73). In the learning model, as shown in FIG. 4, which will be described later, patterns are learned in which an abnormality is avoided by taking appropriate measures against an event, and patterns in which an abnormality occurs without taking appropriate measures (74 ).

Note that the abnormality prediction model learning unit 17 repeats learning every time new process data and operator intervention data are acquired. By repeating learning, it becomes possible to review the prediction model based on new abnormality information when a new abnormality occurs.

The abnormality prediction unit 18 uses the decision tree, which is the prediction model described above, to detect abnormalities according to the implementation or non-implementation of future "measures" for "events" based on changes in newly measured process data. Predict whether or not it will occur.

FIG. 4 is a diagram illustrating an example of predicting the occurrence of an abnormality using a decision tree. In factory processes, there is often not just one factor that causes an abnormal state, but multiple factors that are intricately related. Therefore, for abnormality determination in the present invention, a machine learning decision tree was used, which allows for easy-to-understand visual analysis using multiple factors as explanatory variables. In the example shown in FIG. 4, the event that the temperature of equipment A (tank a of the chemical plant) has increased by X [° C.] is obtained from the process data collected in real time. When treatment A (increasing the flow rate) is implemented in response to this event, it is predicted that the temperature of equipment A will decrease, and product quality can be maintained. On the other hand, if treatment A is not implemented, it is predicted that an event will occur in which the temperature of facility B (tank b provided downstream of tank a) increases by Y [° C.]. Furthermore, if action B (increasing the flow rate) is implemented in response to this event, it is predicted that the temperature of equipment B will decrease, and product quality can be maintained. On the other hand, if treatment B is not implemented, it is predicted that the temperature of equipment B will exceed the upper limit threshold, and it is predicted that an abnormality related to product quality will occur. In this manner, the abnormality prediction unit 18 can predict the occurrence of future abnormalities based on whether or not past actions have been taken with respect to events based on newly measured process data.

The abnormality notification unit 19 displays newly measured process data and details of future treatment. 5 and 6 are diagrams for explaining one type of screen information displayed by the abnormality notification unit 19. As shown in Figure 5, for example, in a graph, the horizontal axis is the time axis, and the vertical axis is the value of process data. The prediction of the process data in the case where the data is not used is displayed, as well as the upper and lower limits of the process data if necessary. In addition, the contents of past "measures" to be performed for the current event classified into a decision tree are displayed in text. Operators can prevent abnormalities from occurring by intervening in the process equipment according to the displayed treatment details. In FIG. 5, only one treatment and its result prediction are shown, but multiple treatments and their resulting predictions may be shown on one screen or divided into multiple windows. However, it may be divided into multiple pages (sheets) and displayed.

Furthermore, as shown in FIG. 6, the abnormality notification unit 19 displays alarm details on the screen when it is predicted that an abnormality will occur. The alarm contents include, for example, the time when the alarm is displayed, the contents related to the "event", and the time when the event is predicted to occur. Furthermore, the content related to the "event" includes, for example, process values, device states, and identifiers of related devices. The abnormality notification section 19 cooperates with the abnormality cause display section 16 via the cooperation section 20. As a result, when the alarm content related to the predicted abnormality occurrence displayed on the screen by the abnormality notification unit 19 is selected, the abnormality cause display unit 16 displays the past “event” related to the alarm content and the “event” linked to the event. "Cause" and "Treatment" are displayed. The alarm display screen may display the time at which the prediction was displayed, the predicted event occurrence time, the importance of the event, the identifier of the related equipment, the order of the process, etc., and may be able to be sorted by these.

As explained above, according to the plant abnormality prediction system according to the present embodiment, process data of plant equipment measured by sensors is combined with operation information (operator intervention data) recorded by the operator. It is possible to predict the occurrence of abnormalities in plant equipment. According to this, even if no abnormality has occurred at the present time, it is possible to predict the abnormality that will occur if future measures are not taken for the current event. Furthermore, according to the present invention, it is possible to display possible measures to be taken against an abnormality that may occur in the future, thereby preventing the occurrence of an abnormality.
Furthermore, by analyzing text data, when an abnormality is predicted to occur, the cause and treatment of the abnormality can be displayed in an easy-to-understand manner. This system not only improves the accuracy of abnormality detection within a factory, but also enables analysis of the causes of abnormal conditions.

<Modified example>
Incidentally, in the system of the first embodiment described above, the text analysis unit 14 divides the text data into "event", "cause", and "treatment" using delimiters, but the present invention is not limited to this. It's not something you can do. An "event" column, a "cause" column, and a "treatment" column may be prepared in advance in the operation record template filled in by the operator. According to this, the processing in the text analysis section 14 can be simplified, and the operator intervention data can be stored in the text data storage section 13.

Further, in the system of the first embodiment described above, the text data (operator intervention data) includes the "cause", but the "cause" may not be included in some cases.

<Hardware configuration example>
FIG. 8 is a conceptual diagram showing an example of the hardware configuration of a processing circuit included in the plant abnormality prediction system according to the embodiment. Each part shown in FIG. 1 shows a part of the function, and each function is realized by a processing circuit. In one aspect, the processing circuitry includes at least one processor 81 and at least one memory 82. In other aspects, the processing circuitry includes at least one dedicated hardware 83.

When the processing circuit includes a processor 81 and a memory 82, each function is realized by software, firmware, or a combination of software and firmware. At least one of the software and firmware is written as a program. At least one of software and firmware is stored in memory 82. The processor 81 implements each function by reading and executing programs stored in the memory 82.

If the processing circuitry comprises dedicated hardware 83, the processing circuitry may be, for example, a single circuit, a composite circuit, a programmed processor, or a combination thereof. Each function is realized by a processing circuit.

Although the embodiments of the present invention have been described above, the present invention is not limited to the above-described embodiments, and can be implemented with various modifications without departing from the spirit of the present invention.

1 Plant abnormality prediction system 10 Process data input section 11 Process data accumulation section 12 Operation record input section 13 Text data accumulation section 14 Text analysis section 15 Abnormal factor extraction section 16 Abnormal factor display section 17 Abnormality prediction model learning section 18 Abnormality prediction section 19 Abnormality notification section 20 Cooperation section 21 Display section 81 Processor 82 Memory 83 Hardware 201 Text data 202 Text data after conversion

Claims (2)

a process data storage unit that stores process data including state values of plant equipment measured by sensors at each time;
a text data storage unit that stores operator intervention data that is text data recorded by an operator and associates an event based on a change in the process data with a treatment performed by the operator in response to the event;
an anomaly prediction model learning unit that creates a decision tree that is a prediction model using the accumulated process data and the operator intervention data as explanatory variables and the presence or absence of an abnormality occurrence as an objective variable;
an anomaly prediction unit that uses the decision tree to predict whether or not an abnormality will occur depending on whether future measures will be taken for an event based on a change in the newly measured process data;
a display unit that displays the newly measured process data and the details of the future treatment when the occurrence of the abnormality is predicted;
A plant abnormality prediction system comprising: The operator intervention data further includes the cause of the occurrence of the event,
The display unit displays an alarm when the abnormality is predicted to occur, and displays a past event related to the newly measured process data and a past event related to the event based on the operator intervention data. displaying the treatment and cause of the
The plant abnormality prediction system according to claim 1, characterized in that:

Images