JPH06273286A - Process state diagnosing system - Google Patents

Abstract

PURPOSE:To discover the process anomaly at early times and lighten the load of an operator for a process by converting the pattern change of the process measurement data to the symbol and utilizing the obtained symbol to the diagno sis for the process state. CONSTITUTION:A process state diagnosing system consists of a symbolic notation part 120 for converting the change pattern of the process measurement data to the symbol, diagnosing part 50 for diagnosing the process state by using the time relation between the change of the symbol through the lapse of time and the symbol of a plurality of time sequence data, and a construction assisting part 180 for assisting the construction of the parameter used in each part. Further, the diagnosing part 50 consists of the single data management part 130 for managing the change of the symbol through the lapse of time, event formation part 140 for generating the event by using the time relation of the symbol between a plurality of data, and a fault mode management part 150 for diagnosing the process state by using the change of the event.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention converts a change state of measurement data obtained from processes of various plants into a linguistic symbol understood by an operator or an engineer and diagnoses the state of the plant using the symbol. About the system to do. In particular, the present invention relates to a system for converting time-series data that changes from moment to moment into a linguistic symbolic representation in real time and diagnosing a process state by inference processing using the symbol.

[0002]

2. Description of the Related Art In recent years, a system for diagnosing a process state has been adopted in many manufacturing industrial processes and processes for managing public resources such as water and sewage. As a typical example, using the state quantity of the process, "if ~
There are some which make inference based on the rule of "then" type. This system handles the values and changes of the data measured from the process, or those that have undergone simple arithmetic processing such as regression equations, as the process state, so it is necessary to fully understand the process state that undergoes complicated changes. Was difficult.

In Japanese Patent Laid-Open No. 3-57917, the current detection amount of the sensor is at a predetermined position of the position, and the change speed of the current detection amount is at a predetermined position of change. There is described a method of checking whether the detector is correct or not by grasping the above and combining them for a plurality of detection amounts.

Further, Japanese Patent Application No. 4-240 by the same applicant.
Japanese Patent No. 647 describes the basic technique of a system for diagnosing a process using a plurality of time-series pattern data. Hereinafter, this technique will be briefly described.
The configuration of the entire system will be described with reference to FIG. This system includes a symbolization unit 4720 that converts changes in time-series data obtained from a process into symbols, and a symbol conversion unit 4730 that converts the appearance order and combination of symbols into event names.
And an inference unit 4740 that diagnoses a process state as a combination of event names, and a symbolization unit, a symbol conversion unit, and a construction support unit 4770 that supports construction of parameters and knowledge used in the inference unit. The symbolization unit 4720 converts the time series data sent from the process measurement data input device 4710 into symbols. The symbol conversion unit 4 will be described with reference to FIG.
The configuration of 730 will be described. The symbol conversion unit 4730 uses the symbol sent from the symbolization unit 4720 to manage the establishment condition of the event to be diagnosed, the event management unit 4
810 and a symbol network storage unit 4820 that stores a symbol network that is constructed by the construction support unit 4770 and that represents a condition for an event to be satisfied. A control system A
In the above, it is assumed that, when a certain abnormality X occurs, three time series data SV2, PV2, MV2 shown in FIG. 37 are obtained. Consider the following as the knowledge used in the inference unit 4740.

[Knowledge 1] "If control system A becomes event 4 after event 2," abnormal X "" Each event used in the knowledge is represented by a network of symbols as shown in FIG. That is, the change of the symbol for each measurement data and the relation of the symbol between the data are expressed by the network of the list structure. Specifically, the phenomenon 2 and the phenomenon 4 included in the abnormality determination knowledge are expressed in a symbol network as shown in FIG.

Actually, the symbol change of the measurement data obtained from the process in the case of the abnormality X is as shown in FIG.
The event management table shown in FIG. 41 is used to manage the condition fulfillment condition in the symbol network for this symbol change.

The operation of the symbol converting unit 4730 will be specifically described in detail. The inference unit 4740 lists the events to be monitored in order to establish the antecedent part of knowledge 1. According to the knowledge 1, the event 2 of the control system A is the event to be monitored.
This list of events to be monitored is sent to the event management unit 4810 of the symbol conversion unit 4730. The symbol conversion unit 4730
Based on the event name sent from the inference unit 4740, the data name and symbol name to be monitored are extracted from the symbol network. In the case of this example, the event to be monitored is the event 2 of the control system A, and the event name and the data name are written in a predetermined position of the event management table 4812. In addition, the event management table 481
The symbol name of the monitoring target is written in the portion of the monitoring symbol of 2. In this example, the SV2 of event 1 is “down from constant”,
PV2 is "step down", MV2 is "up from a certain level"
Is a watch symbol. These monitoring symbols are converted into a symbolization unit 4720.
Sent to.

If the symbol sent from the symbolization unit 4720 is equal to the monitoring symbol of the event management table 4812 and exceeds the allowable similarity within the allowable time, the event management table 48
At the same time as writing a new symbol name in 12, the symbol network 4822 is referred to and the symbol name to be the next symbolization target is written in the field of the monitoring symbol. When there are no new symbol names to be written in the monitoring symbol column, that is, when each branch of the symbol network reaches the end node, the event registered in the event name column is established. It will be. When a new event is established, the established event name is sent to the inference unit 4740.

The inference unit 4740 extracts the event name to be monitored next from the new event name and knowledge antecedent sent from the event management unit 4810 and sends it to the event management unit 4810.
In this example, the event 4 of the control system A is the event to be monitored next. Also, if you have the knowledge that the antecedent department is all satisfied,
The knowledge is established, and the conclusion written in the consequent part is sent to the process status display device 4750.

[0010]

[Patent Document 1] Japanese Patent Application Laid-Open No. 3-57917
The technique of No. can handle the change tendency of the detection amount as a phase change, but cannot handle the temporal change of the phase change, that is, the situation that the upward trend changes to the downward trend.
Further, the technique of Japanese Patent Application No. 4-240647 can handle the occurrence order of symbols assigned to one time-series data as a condition for establishing an event, but cannot handle the occurrence order of symbols between data. For example, it is not possible to incorporate which of the SV2 decrease and the MV2 increase started to change into the condition for establishing the event. However, depending on the control system, the order of changes among a plurality of time series data may be important.

A first object of the present invention is to represent a change pattern for each process data at the time of a failure by a series of symbols representing a change state of time-series data, and further, a mutual relation of symbols among a plurality of process data. It is to provide a process state diagnosis system that enables the process state diagnosis using the. A second object of the present invention is to provide a construction support system that facilitates the setting of internal parameters and knowledge used in the above system.

[0012]

The first object of the present invention is to:
A symbolization unit that converts a change state of pattern data such as time-series data obtained from a process into a symbol, and a symbol obtained from the symbolization unit is used, and a change pattern of time-series data at the time of a failure by a symbol transition sequence. Is represented by a state transition diagram of a single data, and a single data management unit that updates the state (status) in this state transition diagram, and in the state transition diagram of the single data, the statuses of a plurality of data are A fault that expresses the reaching stage of each failure mode by using the event generation unit that generates an "event" unique to the failure mode when the specified time conditions are satisfied, and the event obtained from this event generation unit. It is achieved by providing a failure mode management unit that updates the state transition diagram of the mode.

A second object of the present invention is to provide a feature extraction filter used for feature extraction automatically from a pattern to be encoded, a parameter for feature extraction, and a symbolization parameter construction unit for determining a symbol pattern of a dictionary. A state transition diagram construction unit for determining a state transition diagram of a single data representing a change pattern of time series data at the time of a failure by a transition sequence of symbols obtained from a symbolization unit, and the fault in the state transition diagram. The event generation condition construction that determines the context relationship of the transition start time between the statuses of different measurement data that must be satisfied in the mode and the event generation condition diagram that expresses the event name that is generated when the relationship is satisfied And a state transition diagram of the failure mode, which represents the arrival stage of the failure mode in each failure mode as a transition sequence of the event. Achieved by providing a state transition diagram construction unit for a failure mode to be defined and a consistency check means for checking the consistency between the state transition diagram of the single data and the event generation condition diagram between different failure modes. To be done.

[0014]

In the symbolization system according to the present invention, the single data management unit uses the state transition diagram of the single data expressing the change pattern of the time-series data at the time of failure by the symbol transition sequence to perform the symbolization. The status of the change stage of the pattern data in each failure mode is updated by the symbol sent from the unit. As a result, it is possible to grasp at which step the change stage of the process data that the current change of the process data changes in a complicated manner in each failure mode is reached.

Further, the event generation unit defines the status of the change stage of the pattern data in each failure mode obtained from the single data management unit and the context that the status between different measurement data should satisfy in each failure mode. By using the event generation condition diagram and
Get the event name that represents the symbolic relationship between data. Thereby, for example, even when the change patterns of the process data themselves are the same in different failure modes, the failure mode can be identified from the context of the change between the plurality of process data. Also, in a system with a large time delay, there is a time difference in the change between process data, but by setting the condition time of the front-back relationship appropriately,
It is possible to handle such a time lag of change between data.

Further, the failure mode management unit uses the state transition diagram of the event and the failure mode obtained from the event generation unit,
Get the status name for each failure mode. As a result, it is possible to grasp at which stage of each failure mode the process state is reached, and a message or the like can be displayed on a display device such as a CRT according to the status.

Further, in the construction support unit, the single data state transition diagram construction unit can give the time change of the symbol in each failure mode as a state transition diagram. The event generation condition construction unit can give the context, which the status of the changing stage of the process data peculiar to each failure mode should satisfy, as a list expression for each failure mode. The failure mode state transition diagram construction unit can provide the reaching stage in the failure mode as a state transition diagram with the event obtained from the event generation unit as a transition condition. As a result, a complex process state is represented by a symbol that represents the change state of individual process data, an event that represents the relationship of changes between different data that is unique to each failure mode, and a failure mode that uses the event as a transition condition. It becomes easy to express by the stepwise description of the arrival stage of each failure mode by the status of the state transition diagram.

[0018]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The configuration of a process state diagnosis system according to the present invention will be described below with reference to FIG.

The process state diagnosis system includes a symbolization unit 120 for converting changes in time-series data obtained from a process into a symbolic representation, and a diagnosis unit 50 for diagnosing the state of a process by using a temporal change and combination of symbols. And a process state display management unit 160 that manages to display the measurement data and the status of the failure mode, and the diagnosis unit 50.
And a construction support unit 180 for supporting construction of parameters, state transition diagrams and knowledge used in.

The diagnosis unit 50 updates the state in the state transition diagram of the single data expressing the change pattern of the time series data at the time of failure by the transition sequence of symbols, and the single data management unit 130, In the state transition diagram of single data, when the status of a plurality of data satisfies a predetermined time condition or the like, an event generation unit 140 that generates an “event” unique to the failure mode.
And a failure mode management unit 150 that updates the state transition diagram of the failure mode that expresses the arrival stage of each failure mode using the event obtained from the event generation unit.

The symbolization section 120 is a process measurement data input device 11 for inputting data representing a process state.
0, the process status display management unit 160 is
It is connected to a process status display device 162 for informing the plant operator of the process status. Also,
The construction support unit 180 is connected to a process database 170 that stores measurement data obtained from the process measurement data input device 110.

The symbolization unit 120 is a symbolization processing unit 122 for converting a variation pattern of measurement data into a symbol, and various parameters used for symbolization constructed by the symbolization parameter construction unit 182 of the construction support unit 180. The parameter storage unit 124 stores a dictionary. The single data management unit 130 uses a symbol sent from the symbolization processing unit 122 of the symbolization unit 120 to manage a temporal change of the symbol in each failure mode, and a single data status management unit 132. It is composed of a single data state transition diagram storage unit 134, which is constructed by the state transition diagram etc. construction unit 184 of the construction support unit 180 and stores the temporal change of symbols in each failure mode. The event generation unit 140 uses a signal such as a switch ON / OFF sent from the control device 100, and pattern data in each failure mode obtained from the single data state transition diagram storage unit 134 of the single data management unit 130. Using the combination of the statuses of the change stages of
An event management unit 142 that manages an event that expresses a relationship of changes between data in each failure mode, and a construction support unit 180.
And an event generation condition storage unit 144 that stores a condition for an event to be established, which is constructed by the state transition diagram construction unit 184 of FIG. The failure mode management unit 150 uses the events sent from the event management unit 142 of the event generation unit 140 to manage the arrival stage of each failure mode, the failure mode status management unit 152, and the construction support unit 180. A state transition diagram storage unit 154 of a failure mode constructed by the state transition diagram etc. construction unit 184 for storing the temporal change of the event. The construction support unit 180 includes a symbolization parameter construction unit 182 that supports construction of various parameters and dictionaries used for symbolization, a state transition diagram expressing temporal changes of symbols in each failure mode, and an event in each failure mode. The state transition diagram construction unit 184 that supports the construction of the state transition diagram that expresses the conditions for generation and the temporal change of the established event in each failure mode, and the management mechanism 196 that integrally manages these operations. . In addition, the management mechanism 196 has C, which is an interface for exchanging information with the user.
A display device 190 such as RT, a keyboard 192, a mouse 194, etc. are connected.

The symbolization unit 12 of the symbolization system
0, the single data management unit 130, the event generation unit 140, and the failure mode management unit 150 operate online to constantly perform process diagnosis. The construction support unit 180 is activated as needed, and supports addition and modification of parameters for process diagnosis, a dictionary, and a state transition diagram.

Next, the operation of the symbolization unit 120 will be described in detail with reference to FIG. The symbolization unit 120 includes a symbolization processing unit 122 and a parameter storage unit 124. further,
The symbolization processing unit 122 includes a feature extraction function 210, a collation function 220, and a symbolization progress file 230. The symbolization progress file 230 is a data file 232 that temporarily stores the measurement data of the process, and a filter result file 2 that stores the sum of products operation result with the feature extraction filter.
34, a vector series file 236 for storing the result converted into the polygonal line vector, and a collation progress file 238 for storing the collation progress with the dictionary. The parameter storage unit 124 also stores a feature extraction filter file 240 for storing a feature extraction filter and a feature extraction threshold file 24 for storing a feature extraction threshold.
2. A conversion coefficient file 244 that stores conversion coefficients that are coefficients that standardize and handle measurement patterns, and a symbolization dictionary file 246 that stores polygonal line vectors that form symbol patterns.

The operation of the feature extraction function 210 will be described in detail with reference to FIG. The operation of the feature extraction function 210 includes the following processes (1) to (6).

(1) Process 310: The latest measurement data is read from the process measurement data input device 110. Also,
With reference to the conversion coefficient file 244, an engineering value is converted into a standardized value and stored in the data file 232. At this time, the oldest data among the data stored in the data file 232 is deleted, and a fixed number of measurement data are always stored cyclically.

(2) Process 320: The feature extraction filter for the measurement data to be processed is read from the feature extraction filter file 240. The feature extraction filter is a sequence of numbers having an example shown in FIG. Also, data file 2
The measurement data required for feature extraction is read from 32. Next, as feature extraction calculation, time series data f (t) (t = -T, -T
+ 1,…, 0) and feature extraction filter W ₂ (x)

[0028]

[Equation 1]

According to the above, the feature amount g ₂ (t) indicating the degree of unevenness of the measurement data is calculated. Where a is the filter W ₂ (x)
Is a constant representing the spread of. The sum of products calculation according to the equation 1 is stored in the filter result file 234.

(3) Process 330: The past product-sum operation result is read from the filter result file 234. Next, using the filter feature amount g ₂ (t) (t is the current time), it is determined by Equation 2 whether a new feature appears.

[0031]

[Equation 2]

When the above condition is satisfied, the time (t-1-b) becomes a new feature point. Here, θ is a feature extraction threshold stored in the feature extraction threshold file 242 for the measurement data to be processed. Also, b is a constant value determined by the filter. By this determination processing, if a new feature point is generated, the processing proceeds to step 340, and if no new feature point is generated, the processing proceeds to processing 350.

(4) Process 340: The restoration value f _e (τ _p ) at the feature point τ _p newly generated in the process 330 is calculated by the formulas 3 and 4.

[0034]

[Equation 3]

[0035]

[Equation 4]

Here, W ₁ (x) and W ₂ (x) are the feature extraction filters shown in FIG. 4, the feature amount g ₀ (τ _p ) is the feature amount representing the average value, and g ₁ ( τ _p ) is a feature quantity that represents an average slope.

The restored value f _e (τ _p ) and the restored value f _e (τ _p-1 ) at the previous feature point are connected, and a new vector V _k =
(v1, v2) is generated by Equation 5.

[0038]

[Equation 5]

Next, FLAG = 1 is set and the process advances to the process 360.

(5) Process 350: Newest feature point (τ
regression line through _p , f _e (τ _p )) to the latest data f (0)
h (t) is calculated by Equation 6.

[0041]

[Equation 6]

By substituting t = 0 into the equation 6, the restoration value f _e (0) (= h (0)) at the end (t = 0) of the input pattern is obtained, and the equation 5 of the process 340 is obtained. The latest vector can be obtained in the same manner as. Next, process 3 with FLAG = 0
Proceed to 60. FIG. 5 shows an example in which time-series data is approximated as a polygonal line vector. Figure 5 (a) shows time series data f
FIG. 5B shows a series of polygonal line vectors.

(6) Process 360: The latest vector and FLAG are sent to the matching function 220.

The above is the details of the processing in the feature extraction function 210. Next, details of the processing in the matching function 220 will be described. First, the basic concept of the matching process will be described. The collation processing is based on the concept of DP (dynamic programming). DP refers to the principle of optimality, that is, "optimal policy is such that, no matter what the initial state or decision is, the subsequent decision is the optimal policy for the state resulting from this initial decision. Must be constructed. ”The multi-step decision process is formally decomposed into a one-step process to calculate the solution. Here, it is guaranteed that the obtained solution is the optimal policy by Markov property, that is, "the decision at each stage depends only on the state of the system at that time, and the history of past decisions. It has to be done regardless of. "

Utilizing the above properties, matching between vector sequences is calculated as a multi-step decision process shown in FIG. On the horizontal axis, the vector series U _m (m = 1,
2, ..., M) are arranged, and the vertical axis is the vector series V of measurement data.
Arrange _n (n = 1,2, ..., N). Each grid point (m, n) represents the correspondence between the symbol pattern vector U _m on the horizontal axis and the measurement data vector V _n on the vertical axis. Finding the correspondence between vector sequences is performed from the grid point (1, n) (n = 1,2, ..., N) on the network configured as shown in FIG.
The problem is to find the route that reaches (M, N) that maximizes the predetermined evaluation function. In the vector series of the measurement data and the vector series of the vector of the symbol pattern, there is a correspondence relationship before and after, when there is no corresponding vector, and a plurality of dictionary vectors correspond to one measurement data vector. If it does not occur, it suffices to consider only the connection relationship between the grid points and the grid points shown in FIG.

The search for the optimum correspondence relationship involves calculating the optimum route from the lower left grid point to the upper right grid point in FIG. However, in actual calculation, it is sufficient to calculate the path between the latest vector of the measurement data and the vector one older than each time the vector is generated. As shown in FIG. 8, the processing in the matching function 220 is as follows.
It consists of the following processes (1) to (9).

(1) Process 810: If there are unmatched symbols, the process proceeds to process 820, and if not, the process proceeds to process 890.

(2) Process 820: Collation progress file 23
8, the previous collation process for the symbol is read out, and the vector series of symbol patterns to be collated is read out from the symbolization dictionary file 246.

(3) Process 830: FLAG is judged. If FLAG is 1, it means that the new feature point has just appeared and the latest vector of measurement data has been confirmed.
Go to processing 840. If FLAG is 0, the latest vector of the measurement data has been obtained by the regression line, and since it will be a different vector in the next calculation, the process proceeds to step 850.

(4) Process 840: Lattice point (m, N) (m = 0,1,
,, M) performs the matching process. That is, 4 shown in FIG.
The route with the highest similarity is selected from the two routes. Similarity S
Is calculated by the following equation 7. (The meaning of Expression 7 will be described in detail later.)

[0051]

[Equation 7]

Further, the scale K indicating how many times the size of the vector series of the measurement data at this time is larger than the vector series of the symbol pattern can be calculated by the equation 8.

[0053]

[Equation 8]

(5) Process 850: Collation process is performed at the lattice point (M, N). Calculation of similarity S and scale K is processed 8
Similar to 40, the equations 7 and 8 are given.

(6) Process 860: similarity S and scale K
Is temporarily stored.

(7) Process 870: The FLAG is judged. If FLAG is 1, the process proceeds to processing 880 for updating the collation progress file, and if FLAG is 0, the process returns to processing 810.

(8) Process 880: Collation progress file 23
8 update processing is performed. That is, the information on the newly determined vector V _n is rewritten as the information on V _n-1 .

(9) Process 890: The status management unit 1 of the single data sets the similarity and scale of the symbols stored in the process 860.
To 32.

Next, referring to FIG. 9, the single data management unit 1
The operation of 30 will be described in detail. Single data management unit 130
Is a single data status management unit 132 that manages the status of the change stage of the measurement data in each failure mode based on the change of the symbol sent from the symbolization processing unit 122.
And a time transition of the symbol in each failure mode are represented as a state transition diagram in which the pattern changes of the symbol-represented time series data constructed by the state transition diagram construction unit 184 of the construction support unit 180 are connected. Single data state transition diagram 9
Single data state transition diagram storage unit 13 storing 20
4 and. Further, the single data status management unit 132 includes a single data state transition function 910.

State transition diagram of single data with reference to FIG.
The configuration of 20 will be described. The state transition diagram 1020 for each measurement data is obtained by knowing how each measurement data changes in a certain failure mode, that is, how the symbol sent from the symbolization processing unit 122 changes.
Express by. A failure mode is identified by the temporal changes in multiple measurement data and the context between the changes.
Therefore, the single data state transition diagram 1010 for one failure mode is composed of a plurality of state transition diagrams 1020 for each measurement data. Here, the single data state transition diagram 1010 for one failure mode is an independent representation of how each measurement data changes in a certain failure mode. Is not expressed. When there are a plurality of failure modes, a single data state transition diagram 1010 for one failure mode is prepared for each failure mode.

Next, the state transition diagram 1020 for each measurement data will be described in detail with reference to FIG. The state transition diagram 1020 for each measurement data is represented by a linear list of structures that represent the status of the change stage of the measurement data. Each status has a transition flag 1140 that represents a transition state.
A symbol name 1150 that represents a symbol required to make a transition to the current status, a transition minimum time 1160, a transition maximum time 1170 that shows the time condition of a symbol that should be satisfied to make a transition to the current status, and a current status The transition start time 1180 that is the transition time. Of these, the transition flag 1140 and the transition start time 1180 are updated in real time. The status of the state transition diagram 1020 for each measurement data is the initial state 1110 and the final state 1130.
And other intermediate states 1120. At the start of diagnosis, the status is in the initial state 1110. here,
Since there is no transition to the initial state, the symbol name 1150 of the status of the initial state, the minimum transition time 1160, the maximum transition time 1170, and the transition start time 1180 are 0, and the transition flag 1140 is 1 (transition completed). And will not be updated thereafter.

The operation of the single data state transition function 910 will be described in detail with reference to FIG. The operation of the single data state transition function 910 includes the following processes (1) to (11).

(1) Process 1210: Access the state transition diagram 920 of a single data, start from the initial state, and advance the noticed status until the transition flag of the next status becomes 0 or the final state. Go. Now you should pay attention to the current status.

(2) Process 1220: Determine whether the current status is the final state. If the pointer to the next status is NULL, the current status is the final state and the entire processing is terminated. If it is not NULL, the current status is the initial state or the intermediate state, and the processing proceeds to processing 1240.

(3) Process 1240: Determine whether the current status is the initial state. Current status symbol is 0
, The current status is the initial state, and the processing 125
If it is not 0, it means the intermediate state, and the process 12
Proceed to 60.

(4) Process 1250: The current status is the initial state. The symbol sent from the symbolization processing unit 122 is
If it is the same as the symbol name of the next status (= the symbol name required to make a transition to the next status), the process proceeds to the status transition process 1296, and if it is not the same, the entire process ends.

(5) Process 1260: The current status is the intermediate status. The symbol sent from the symbolization processing unit 122 is
If it is the same as the symbol name of the next status, there is a possibility of transition, and the process proceeds to processing 1280. If it is not the same, there is no possibility of transition and processing proceeds to processing 1270.

(6) Process 1270: Symbolization processing unit 122
If the symbol sent from is the same as the symbol name of the current status, it means that it has not changed from the symbol at the time of transition to the current status, and the whole process is terminated. Since different symbols are detected for both the status and the next status, the process proceeds to the status initialization processing 1294.

(7) Process 1280: The current status is an intermediate state, and the same symbol as the symbol name of the next status is detected. After that, the transition can be made if the time condition is satisfied.
Here, there are two types of status transitions. One is the case where the symbol of the current status and the symbol of the next status are different, which corresponds to the transition of the status by changing the symbol. The other is the case where the symbol of the current status and the symbol of the next status are the same, which corresponds to the transition of the status by continuing to detect a certain symbol for a certain period of time. Since the formats of the temporal conditions are different, which of these two patterns is determined. If the symbol of the current status and the symbol of the next status are the same, the latter pattern is entered, and the process proceeds to step 1290. If the symbols are different, the former pattern is entered, and the process proceeds to step 1292.

(8) Process 1290: The time condition regarding the transition when the symbol of the current status and the symbol of the next status are the same is given as follows.

End time of detection symbol-start time of detection symbol> maximum transition time of next status Here, the start time and the end time of the detection symbol are information attached to the symbol sent from the symbolization processing unit 122. , Corresponding to the times at both ends of the vector series of the measurement data having the optimum correspondence with the vector series of the symbol pattern of the dictionary.
The transition using this condition is mainly for grasping that the linearly changing state of the time-series data (for example, constant, rising, or falling) continues for a long time. When the above temporal conditions are satisfied, the status transition processing 1296
If not satisfied, the whole process is terminated.

(9) Process 1292: The temporal condition regarding the transition when the symbol of the current status and the symbol of the next status are different is given by the following.

Minimum transition time of next status <Start time of detection symbol-Transition start time of current status <Maximum transition time of next status This time condition is that a certain symbol is attached and another symbol is attached next. The upper and lower limits are given to the time between. If the above temporal condition is satisfied, the processing proceeds to status transition processing 1296, and if not satisfied, the processing proceeds to status initialization processing 1294.

(10) Process 1294: Initialize the status. This is done when the current status symbol and the next status symbol are different and the sensed symbol is different from the next status symbol or the same but the time condition is not met. In these cases, it is considered that pattern data of a failure mode different from the target failure mode is input, and the state transition of each measurement data included in the single-state state transition diagram 1010 for one failure mode. Regarding FIG. 1020, transition flags of statuses other than the initial state are set to 0 (untransitioned).

(11) Processing 1296: Status transition processing is performed. The transition flag of the next status is set to 1 (transition completed). Also, the start time of the detection symbol is substituted for the transition start time of the next status.

Next, referring to FIG. 13, the event generator 140
The operation of will be described in detail. The event generation unit 140 includes an event management unit 142 and an event generation condition storage unit 144.
Furthermore, the event management unit 142 uses the event conversion function 1310.
And an event generation function 1320 and an event list 1330. The event generation condition storage unit 144 includes an event generation condition diagram 1340.

The operation of the event conversion function 1310 will be described.
The event conversion function 1310 converts a signal such as ON / OFF of a switch, which is sent from the control device 100, into an event name that identifies them, and stores the event name in the event list 1330.

Using FIG. 14, an event generation condition diagram 1340
The configuration of will be described. In a certain failure mode, the temporal transition between the statuses in the single data state transition diagram 1010, which represents the characteristic change relation between the measurement data, is shown as an event generation condition diagram for one failure mode. This is represented by 1410. When a plurality of failure modes exist, an event generation condition diagram 1410 for one failure mode is prepared for each failure mode. Event generation conditions for one failure mode The event generation conditions in FIG. 1410 are event 1 (1420) and event 2 in FIG.
(1430) and event 3 (1440). Event 1 (1420) is generated when the following temporal context condition is satisfied between status 1472 and status 1474.

Minimum condition time <transition start time of status 1474-status 147
2 transition start time <maximum condition time This condition means that the time from the transition to status 1472 to the transition to status 1474 must be at least the minimum condition time and at most the maximum condition time. . The minimum condition time and the maximum condition time are set for each context. Phenomenon 2 (1430) is an example of what is generated when the condition of the temporal context is satisfied between a plurality of statuses. Event 2 (1430) is
It is generated when the temporal context is established between the status 1478 and the status 1476, and the status 1478 and the status 1480. Although shown here between two statuses, this may be between three or more statuses. Event 3
(1440) is the event generation condition that the status 1482 is transited. That is, status 1482
It is used when the characteristic in the failure mode is shown in the transition to.

The internal representation of the event generation condition diagram 1410 for one failure mode will be described in detail with reference to FIG. Structures 1510 to 1512 representing each event are connected in a list, and a structure representing an event generation condition is connected to each structure. The structure 1520 representing the event generation condition for the event 1 has two statuses 1 in the single data state transition diagram 1010 for one failure mode.
472, a pointer that points to a structure representing 1474, a minimum condition time, and a maximum condition time. Since there are two event generation conditions for the above event 2, two structures (structures 1521 and 1522) expressing them are connected in series. The above event 3 is status 1482.
Since the event generation condition is that the status is changed to, the structure 1523 has only a pointer to the structure representing the status 1482 as information.

The operation of the event generation function 1320 will be described in detail with reference to FIG. The operation of the event generation function 1320 includes the following processes (1) to (11).

(1) Process 1610: Access the state transition diagram 920 of the single data and the event generation condition diagram 1340.

(2) Process 1620: In the failure mode currently being checked, if there is an unchecked event, the process proceeds to process 1630, and if not, the process proceeds to process 1692.

(3) Process 1630: In the event currently being checked, if there is an unchecked event generation condition, the process proceeds to process 1640. If not, the process proceeds to process 1680.

(4) Process 1640: When the pointers to the two statuses are stored in the structure representing the event generation condition currently being checked, the condition is in the temporal context and the process 1650 is executed. If only one pointer to one status is stored, the transition to that status itself is the event generation condition, so process 1
Proceed to 660.

(5) Process 1650: When the two statuses in the temporal context currently being checked have both transitioned, the process proceeds to process 1670 to determine the temporal context, and at least one of them is executed. If the transition is not yet made, the determination of the event currently being checked is terminated, and the process 1620 is returned to for checking the next event.

(6) Process 1660: When the status, which is an event generation condition, has been transitioned, the event generation condition currently being checked is satisfied, so the process returns to the process 1630 to check the next event generation condition. If not yet transitioned, the determination of the event currently being checked is ended, and the process returns to the process 1620 to check the next event.

(7) Process 1670: If the temporal context is satisfied during the transition start time of the two statuses currently being checked, the process returns to process 1630 to check the next event generation condition. , If not,
The measurement data is considered to have changed different from the failure mode currently being checked, and the process proceeds to processing 1690.

(8) Process 1680: The event is generated because all the event generation conditions for the event currently being checked are satisfied. The generated event name is temporarily stored, and the process returns to the process 1620 to check the next event.

(9) Process 1690: It is considered that the measurement data is not the failure mode currently being checked, and the state transition diagram 1020 for all the measurement data in the current failure mode is initialized. This processing is the status initialization processing 1 of FIG.
Same as 294. Further, the generated event name temporarily stored in the process 1680 is cleared, the check of the current failure mode is finished, and the process 1694 is performed.

(10) Process 1692: Since all events in the current failure mode have been judged, process 168 is performed.
In 0, the generated event name temporarily stored is the event list 1
It is stored in 330.

(11) Process 1694: If there is a failure mode that has not been checked yet, the process returns to process 1620 to check the next failure mode, and if all failure modes have been checked, the entire The process ends.

Next, the operation of the failure mode management unit 150 will be described in detail with reference to FIG. Failure mode management unit 15
0 includes a failure mode status management unit 152 and a failure mode state transition diagram storage unit 154. Furthermore, the failure mode status management unit 152
A failure mode status management function 1710 for managing the status of the failure mode by the failure mode status transition diagram 1730 constructed by the construction section 184 such as the state transition diagram of the construction support unit 180, and the current failure mode status are stored. The failure mode list 1720 is stored. Further, the failure mode state transition diagram storage unit 154 includes the failure mode state transition diagram 1730.

The configuration of the failure mode state transition diagram 1730 will be described with reference to FIG. Failure mode state transition diagram 1
730 indicates a failure mode status 18 represented by a circle.
01 to 1806, status transitions 1811 to 1815 represented by arrows, and event names necessary for status transition. FIG. 18 shows a state transition diagram of failure modes for a certain three failure modes.
A failure mode α is generated with events A, B, ..., A failure mode β is generated with events C, D, ..., A failure mode γ is generated with events C, EandF (in no particular order). is there. At the start of diagnosis, the failure mode status is in the initial state. When the event A is generated in the initial state, the failure mode status transits to the state a. When the event B is generated in the state a, the status of the failure mode transits to the state b. When both the event A and the event C are generated in the initial state, the state transitions to both the state a and the state c. Event E is required for transition from state c to state e
And event F both need to be generated. Also,
The state c is a state that is common in the temporary periods in which the failure modes β and γ are present. As described above, the failure modes can be grasped at various levels by subdividing and managing the statuses of the failure modes by using the sequence of events generated in the respective failure modes.

FIG. 19 is a state transition diagram 173 of the failure mode.
The internal representation of 0 is shown. The internal representation of the failure mode state transition diagram 1730 connects the node structures that represent the status of the failure mode and the node structures,
It is composed of a transition structure that represents a status transition and an event structure that is connected to the transition structure and that represents an event that is a condition of the status transition. When it is possible to transit from one status to multiple statuses, the transition structure is connected by the number of transitionable statuses.

The operation of the failure mode status management function 1710 will be described in detail with reference to FIG. The operation of the status management function 1710 in the failure mode is as follows (1)-
It consists of the processing of (10).

(1) Process 2010: The generated event is read from the event list.

(2) Process 2030: For the initial state node, the process proceeds to the end node determination routine.

(3) Process 2040: If the current node is the end node (the pointer to the transition structure is NULL), proceed to process 2060. If it is not the end node, proceed to process 2050.

(4) Process 2050: Since the current node is not the end node, the process proceeds to a search routine to search for a node further inside.

(5) Process 2060: The status name of the current node (end node) is stored in the failure mode list 1720.

(6) Process 2070: If an event that is a condition of the current transition exists, the process can be transited to the process 2080, and if no event exists, the process cannot proceed to a node further in the process 2090. Proceed to.

(7) Process 2080: With respect to the node at the back, proceed to the end node determination routine.

(8) Process 2090: The status name of the current node is stored in the failure mode list 1720, and the process proceeds to process 2092 to check whether the transition can be made from the current node via another transition.

(9) Process 2092: If there is an unsearched transition among the transitions connected to the current node, proceed to process 2094. If not, terminate the search routine.

(10) Process 2094: The noticed transition is moved to an unsearched transition, and the process proceeds to the search routine.

Next, the operation of the process status display management section 160 will be described. The process status display management unit 160 reads out the status of the failure mode obtained by the failure mode management unit 150 from the failure mode list 1720. If there is an important status in the process diagnosis, the operation of the plant corresponding to the status is performed. A message to a member, a single data state transition diagram 1010 for one failure mode, and a portion of the failure mode state transition diagram 1730 for the failure mode are sent to the process status display device 162. Process status display device 16
2 displays the process status sent from the process status display management unit 160 on a display device such as the CRT 164. An example of the display screen is shown in FIG. A failure mode name 2104, a message 2106 to an operator of the plant, and a failure mode state transition diagram 1730 regarding a plurality of statuses of high importance.
108 and a single data state transition diagram 2110 are displayed. Further, by displaying a trend graph 2112 of measurement data at the bottom of the screen, the operator of the plant can easily confirm the work. In addition, the process status display device 16
2 can also notify the plant operator by voice using the voice output device 166 or the like.

Hereinafter, how the state transition diagram changes will be described using a specific example. FIG. 22 shows an example of the target process. The target process includes piping and a valve 2120 that adjusts the flow rate of the piping. The flow rate is controlled by the flow rate controller 2130.
Controlled by. The state of the flow rate controller 2130 is represented by the following three quantities.

SV: Flow rate 2132 (set value) given from another control device or other computer PV: Flow rate 2134 (control amount) obtained from the flow rate sensor 2110 provided in the pipe MV: Adjusting the opening of the valve 2120 Amount 2136 (manipulation amount) In the process of FIG. 22, two abnormal states are detected: when the flow rate sensor 2110 is stuck at the upper limit (sensor upper limit sticking) and when the valve 2120 is stuck fully open (valve fully open sticking). think of. FIG. 23 and FIG. 24 show changes in each state quantity of the flow rate controller 2130 when the sensor upper limit is fixed and the valve is fully opened, and FIG. 23 and FIG. 24 show a single data state transition diagram for each failure mode and event generation. 25 and 26 show the related transition series in the condition diagram and the state transition diagram of the failure mode. In FIG. 25 and FIG. 26, when the transition minimum time, the transition maximum time, and the minimum condition time are written as “−”, the condition regarding the time is ignored.
A message to the plant operator is also shown in the important statuses in the failure mode state transition diagram.

FIG. 27 and FIG. 27 show a part of the time change between the symbol sent from the symbolization processing unit 122 in the case of the sensor upper limit fixation and the valve fully open fixation, the symbol start time, and the symbol end time. 28. The detection symbol "-" written in FIGS. 27 and 28 indicates that there is no detection symbol. Regarding the sensor upper limit fixation of FIG. 27, FIG.
We will follow how the state transition diagram of changes.

Time 45: S in the state transition diagram of single data
V, PV, and MV change to "constant".

Time 83: PV of state transition diagram of single data
Changes to "increase".

Time 84: MV of state transition diagram of single data
Changes to "fall". Event 19 holds. The failure mode status transitions to state 14.

Time 100: P in the state transition diagram of single data
V transits to "step up". Event 20 holds.

Time 101: M in the state transition diagram of single data
V transits to “step down”. Event 21 is established. The failure mode status transitions to state 15. The message "Sensor upper limit stuck" is output.

Regarding the valve upper limit fixation in FIG. 28, FIG.
We will follow how the state transition diagram of changes.

Time 88: S in the state transition diagram of single data
V, PV, and MV change to "constant".

Time 91: MV of state transition diagram of single data
Changes to "fall".

Time 92: PV of state transition diagram of single data
Transitions to "increase (3710)".

Time 101: P in the state transition diagram of single data
V transits to “increase (3720)”. Event 25 holds.

Time 114: M in the state transition diagram of single data
V transits to “step down”. Event 26 holds. The failure mode status transitions to state 18. The message "Valve fully open stuck" is output.

Hereinafter, the construction support unit 18 for determining various parameters of the process state diagnosis system according to the present invention.
0 will be described in detail.

First, the symbolization parameter construction unit 182 will be described in detail. There are four (a) feature extraction filter, (b) transform coefficient, (c) feature extraction threshold, and (d) symbol pattern of the symbolization dictionary, which are determined by the symbolization parameter construction unit 182. Hereinafter, they will be described in order.

(A) Feature Extraction Filter The feature extraction filter has the form shown in Expression 9.

[0125]

[Equation 9]

Since the calculation is actually executed as a discrete value, it is necessary to obtain it as a value for the discrete value of x. Hereinafter, a method of generating a feature extraction filter, which is stored in the feature extraction filter file 240 in advance as necessary before performing the process state identification, will be described in detail with reference to FIG.

(1) Process 4110: A program for executing the following process is started by using the input means such as the keyboard 192, and the constants a and σ that determine the size and shape of the filter and the maximum order n _max to be obtained are set. input.

(2) Process 4120: Gaussian function E (x) (x
= -a, -a + 1, ..., 0) is calculated.

(3) Process 4130: The order n of the filter to be calculated is set to 0.

(4) Process 4140: A polynomial of degree n is calculated. Hereinafter, the polynomial calculation formulas up to the second order will be shown as an example. By repeating the same calculation, a higher-order polynomial can be easily calculated if necessary. Since the 0th-order polynomial is a constant,

[0131]

[Equation 10]

The process ends as. The first-order polynomial

[0133]

[Equation 11]

If we say that in order to establish orthogonality with the 0th-order polynomial H ₀ (x),

[0135]

[Equation 12]

The following must be satisfied. Therefore,

[0137]

[Equation 13]

It can be calculated as follows. Quadratic expansion polynomial

[0139]

[Equation 14]

[0140] In order to establish orthogonality with a first-order polynomial,

[0141]

[Equation 15]

The following must be satisfied. Considering that 0th-order and 1st-order polynomials are orthogonal,

[0143]

[Equation 16]

Must be satisfied. Therefore,

[0145]

[Equation 17]

Is calculated. Furthermore, a quadratic polynomial and 0
The condition that the orthogonality of the following polynomial holds

[0147]

[Equation 18]

From

[0149]

[Formula 19]

Is calculated.

(5) Process 4150: If the polynomial to be calculated has reached the maximum degree n _max , proceed to the next process. Otherwise, in process 4160, n is incremented to n + 1 and the process 4140 is performed.

(6) Process 4170: Normalization is performed based on the polynomial obtained in process 4140. The polynomial of each degree is
You have to satisfy the number 20. For each degree polynomial,

[0153]

[Equation 20]

By calculating, the normalized feature extraction filter W _m (x) becomes

[0155]

[Equation 21]

Is calculated.

(7) Process 4180: The generated feature extraction filter is stored in the feature extraction filter file 240.

An example of the generated feature extraction filter is as shown in FIG. In this example, the constant representing the range of the filter is a = 16, and the constant determining the shape is σ = 4.0, and W ₀ (x), W ₁ (x), and W ₂ (x) in FIG. 0 order, 1 respectively
It is a generation result of the secondary and secondary feature extraction filters.

(B) Conversion coefficient The conversion scale stored in the conversion coefficient file 244 is a constant for converting, for each measurement data, an engineering value thereof into a standardized value handled inside the computer. This constant is
When the plant operator or technician displays a trend graph of data on a display device such as a CRT, the display ratio in the vertical and horizontal directions on the screen when the change for which a symbol is desired is displayed in the most visible form I can decide. For example, when it is desired to add a symbol “step change” to a change state of certain data, and the screen display shown in FIG. 30 is closest to the “step change” considered by the operator, the conversion coefficient of the data is

[0160]

[Equation 22]

Is given by Here, t is the number of dots on the screen corresponding to one sampling interval in the horizontal axis direction on the screen, and h is the number of display dots for the engineering value 1.0 in the vertical axis direction.

(C) Feature Extraction Threshold The feature extraction threshold stored in the feature extraction threshold file 242 consists of the following six processes shown in FIG.

(1) Process 4310: Read measurement data from the process database 170.

(2) Process 4320: The time series data f (t) is expanded into the frequency spectrum X (ω) by using the fast Fourier transform (FFT) or the like.

(3) Process 4330: An appropriate high frequency component is designated (for example, (1/2) π to (3/2) π) and inverse Fourier transform is performed to restore the noise component fn (t). To do.

(4) Process 4340: Noise component fn (t)
On the other hand, a quadratic feature extraction filter W ₂ (x) is applied to obtain a quadratic expansion coefficient.

(5) Process 4350: The maximum absolute value of the quadratic expansion coefficient is obtained and set as the feature extraction threshold θ.

(6) Process 4360: The feature extraction threshold θ is stored in the feature extraction threshold file 242.

FIG. 32 shows an example of the above processing. Figure 32
(a) is the original time series data f (t). FIG. 32 (b) shows the absolute value of the spectrum | X (ω) | when the time series data f (t) is Fourier expanded by FFT. When the spectrum information and the phase information in the section of (1/2) π to (3/2) π as the high frequency component are inverse Fourier transformed, the noise component fn (t) can be obtained as shown in FIG. 32 (c). . When the quadratic expansion term A ₂ (t) of the noise component fn (t) is obtained,
It becomes as shown in (d). The maximum absolute value of the second-order expansion term is obtained and used as the feature extraction threshold θ.

(D) Symbolic Dictionary The symbol pattern stored in the symbolic dictionary file 246 is determined by the following processes (1) to (5) shown in FIG.

(1) Process 4510: The user specifies the reference pattern of the dictionary by using the mouse on the CRT screen,
Read it into the calculator.

(2) Process 4520: An arbitrary part of the process data stored in the process database 170 is converted into a polygonal line and displayed on the CRT screen, and the user selects the part indicated by the corresponding symbol from the mouse. Specify using etc. Read the specified part and temporarily store it as a pattern file.

(3) Process 4530: The reference pattern is collated with the pattern stored in the pattern file, and an average pattern is generated from the corresponding relationship to make a symbol pattern.

(4) Process 4540: The symbol pattern and the pattern stored in the pattern file are collated again to obtain the similarity and scale for each pattern. The calculated similarity and minimum value of the scale are calculated and used as thresholds of the similarity and the scale, respectively.

(5) Process 4550: The symbol pattern, the similarity threshold, and the scale threshold are stored in the symbolization dictionary file 246.

Next, the configuration of the state transition diagram etc. construction section 184 will be described with reference to FIG. State transition diagram construction unit 184
Symbolizes the measurement data read from the process database 170, and the symbolization result is the symbol string file 462.
The encoding unit 4610 for storing in 0 and the encoding file 46
20. A symbol string is read from 20, a single data state transition diagram is constructed, and stored in the single data state transition diagram storage unit 134. A single data state transition diagram construction function 4630 and a single data state transition diagram. Using the state transition diagram of the single data constructed by the construction function 4630, an event generation condition diagram is constructed from the characteristic change relation between the measurement data and stored in the event generation condition storage unit 144. Construction function 4640
And the event generation condition diagram constructed by the event generation condition diagram construction function 4640 to construct a state transition diagram of the failure mode and store the state transition diagram of the failure mode stored in the state transition diagram storage unit 154 of the failure mode. Function 184 and a consistency check function 4660 that checks the consistency of single data state transition diagrams between failure modes and event generation condition diagrams so that different failure modes are correctly identified and refines those diagrams.
Consists of. The symbolization unit 4610 is equivalent to the symbolization unit 12
This is the same as the processing of 0.

Although one embodiment of the process state diagnosing method and apparatus according to the present invention has been described above, the data that can be handled by the present invention is not limited to data such as flow rate, temperature, water level, etc., and vibration, light, etc. are measured. The data of various sensors can be handled in the same way. Further, the data may be received via a public line, a dedicated LAN, or the like, or various recording media (such as a magnetic disk) may be used. Further, the present invention may be configured as hardware using a dedicated processing device, or may be configured as hardware using an LSI such as a DSP (digital signal processor).
Alternatively, it can be realized as software that performs processing by a microcomputer or the like.

Further, according to the present invention, the process state diagnosis is performed by collating a plurality of process data with each other.
The knowledge representation method of the present invention, which regards pattern data as a series of element patterns such as rising and falling, and regards the relation between data as a temporal relation between element patterns, is not limited to process state diagnosis, It can also be used as a method of expressing a plurality of one-dimensional data in the field of storing and collating.

As is apparent from the configuration and processing of the present invention, the operating state is accurately determined by storing the measured data of the process components at the normal time and various abnormal times and using them as knowledge. It has the feature that it can.

Mathematical Background In order to better understand the embodiments of the process condition diagnosis method and apparatus according to the present invention and to evaluate their inherent possibilities and advantages, they are used below for implementing the embodiments. It provides a brief mathematical background and foundation for the method. The formulas introduced also give an insight into the functionality of the embodiment.

The feature extraction filter W ₀ (x) stored in the feature extraction filter file 240 and used in the feature extraction filtering 320 in the feature extraction function 210,
W ₁ (x) and W ₂ (x) are filters that find the average value of the pattern W ₀ (x), filters that find the average slope of W ₁ (x), and W ₂ (x) This is a filter for obtaining the average unevenness of the pattern. Below is a summary of how to determine W ₀ (x), W ₁ (x), and W ₂ (x).

As a pattern data expansion method,

[0183]

[Equation 23]

The polynomial H _m (x) defined by is used. Where E (x) is a suitable differentiable function, and a and b
Is a constant indicating a defined section. At this time,

[0185]

[Equation 24]

Putting it aside,

[0187]

[Equation 25]

Forms an orthonormal function system.

If the orthonormal function system ψ _m (x) is used as it is and the time series data f (t) is expanded in the vicinity of time t, higher order terms are obtained in order to obtain a good approximation of the original data. Will be required. To avoid this, time-series data centered around time t

[0190]

[Equation 26]

It will be converted into and treated.

If the expansion using the orthogonal function system of the equation 25 is performed on the equation 26,

[0193]

[Equation 27]

It becomes Where the expansion coefficient a _m (t) (m = 0,1,
…, ∞) is

[0195]

[Equation 28]

It is calculated by According to Equations 27 and 28, the time series data f (t) is near the time t.

[0197]

[Equation 29]

It means that it has been expanded as follows.

Differentiable function E (x)

[0200]

[Equation 30]

In the case of the exponential function represented by, H _m (x) is a m-th order Hermite polynomial according to the equation (23). Also, from Equation 29, the time series data f (t) is expanded into a polynomial in the vicinity of time t. Expansion filter W at this time
By substituting equations 25 and 26 into equation 28, _m (x) is

[0202]

[Equation 31]

Since,

[0204]

[Equation 32]

It is expressed as follows.

The restored value f _e (t) of the time series data at the time t 1 is obtained using the result of the polynomial expansion by the feature extraction filter represented by the equation 32. Substituting x = 0 into Equation 29, f _e (t) becomes

[0207]

[Expression 33]

Is obtained.

The quadratic term a ₂ (t) of the expansion coefficient according to Equation 33
Indicates the unevenness of the time series data near time t. Therefore, for time series data f (t),
The expansion coefficient a ₂ (t) of the quadratic polynomial is calculated according to Equation 31,
By extracting the time having the extreme value, the time when the gradient of the time series data changes can be obtained.

The method of obtaining the degree of similarity and the scale will be described below. The vector sequence of one symbol pattern stored in the symbolization dictionary file 246 is U _i (i = 1,2, ..., m), and the vector sequence of measurement data is V _i (i = 1,2, ..., m). ). When the vector series of the symbol pattern is multiplied by K, the squared error E between both vector series is

[0211]

[Equation 34]

It is represented as follows. First, find the scale K that minimizes the squared error E. Expanding number 34,

[0213]

[Equation 35]

It becomes Here, (,) indicates the dot product of the vector. Since Equation 35 is a quadratic equation in which the quadratic coefficient of K is positive, the scale K that gives the minimum value of the squared error E is

[0215]

[Equation 36]

From the above,

[0217]

[Equation 37]

Given as The squared error E at that time is
Substituting equation 37 into equation 35,

[0219]

[Equation 38]

It is given by. Squared error E in Eq. 38
Is normalized with respect to the vector series of the time series pattern,

[0221]

[Formula 39]

It becomes Therefore, the evaluation index of the degree of correspondence by the squared error E normalized by the scale K is given by
It is defined by the similarity S using the square root of the second term of 9.

[0223]

[Formula 40]

In Expression 39, the squared error E is the same whether the scale K according to Expression 37 is positive or negative. However, when collating time-series data, the degree of similarity must be small when they are inverted and coincident. The number 40 is
An appropriate degree of similarity can be obtained by considering the square root of the second term on the right side of Expression 39. The similarity S represented by the equation 39 satisfies -1.0≤S≤1.0, has a formula similar to the ordinary correlation coefficient, and can be considered as a correlation coefficient between vector sequences.

As described above, the equations of the similarity S and the scale K are obtained on the assumption that the vector series of the measurement data and the vector series of the symbol pattern have a one-to-one correspondence. In general, there is a case where one vector and a plurality of vectors correspond between the vector series of the symbol pattern and the vector series of the measurement data. In this case, the similarity S and the scale K shown in Equations 40 and 37 are

[0226]

[Formula 41]

[0227]

[Equation 42]

It is used after being modified. Here, the vector series of the measurement pattern corresponding to the vector U _i of the symbol pattern is defined as V _ij (j = 1,2, ..., n _i ).

[0229]

By using the system according to the present invention, it becomes possible to diagnose the state of the process in consideration of the change pattern of the process, which leads to the stable service of the plant and the improvement of the quality or the reliability of the information processing equipment for customer service. Can bring a dramatic improvement to the improvement of.

(1) By converting the measurement data obtained from the process into a symbol, the causal relationship between a plurality of data can be treated as the context of the time when the symbol appears, and the process model construction by the conventional method is possible. It is possible to handle the interference between variables, which has been a bottleneck of the process, and to diagnose the state of the process that overcomes the limitation of numerical calculation by the multivariable model.

(2) As a criterion for diagnosing the behavior of a complicated process, the process of converting individual measurement data into symbols, and the characteristics of each failure mode are managed from the time change of symbols for a plurality of measurement data and the context. Process to understand the event
By hierarchically performing the process of grasping the arrival stage for each failure mode by the transition of events, knowledge for diagnosing the state of the process can be easily constructed.

(3) A process state diagnostic system, which has been possible only by an expert in the past, can be constructed easily and by an unskilled person, and a large amount of manpower is required to maintain a plant or process. Can be reduced.

(4) By using the data of past process states, the knowledge can be increased with time, and the function can be gradually increased so that the maintenance staff gains experience and experience.

(5) Knowledge and know-how possessed by skilled operators,
Alternatively, it becomes possible to construct a control model using causal relationships between variables that are determined by the structure of the process, and extract knowledge and empirical rules that exceed the expressive power of experts in a consistent and consistent manner. Available.

[Brief description of drawings]

FIG. 1 is a diagram illustrating a configuration of a process state diagnosis device according to the present invention.

FIG. 2 is a diagram illustrating a symbolization unit 120 in detail.

FIG. 3 is a diagram illustrating a processing flow in a feature extraction function 210.

FIG. 4 is a diagram showing an example of a feature extraction filter.

FIG. 5 is a diagram showing an example of approximation by a polygonal line vector.

FIG. 6 is a diagram illustrating a matching process as a multi-step determination process.

FIG. 7 is a diagram illustrating a range in which an optimum route to each grid point is searched.

FIG. 8 is a diagram illustrating the flow of processing in the matching function 220.

FIG. 9 is a diagram illustrating the configuration of a single data management unit 130.

FIG. 10 is a diagram illustrating the configuration of a state transition diagram 920 for single data.

FIG. 11 is a diagram for explaining the state transition diagram 1020 for each measurement data in detail.

FIG. 12 is a diagram illustrating a flow of processing in a single data state transition function 910.

FIG. 13 is a diagram illustrating a configuration of an event generation unit 140.

FIG. 14 is a diagram illustrating the configuration of an event generation condition diagram 1340.

FIG. 15: Event generation condition diagram for one failure mode
The figure which illustrates 410 in detail.

FIG. 16 is a diagram illustrating a processing flow of an event generation function 1320.

FIG. 17 is a diagram illustrating the configuration of a failure mode management unit 150.

FIG. 18 is a diagram illustrating a state transition diagram 1730 of a failure mode.

FIG. 19 is a diagram for explaining the internal representation of the failure mode state transition diagram 1730.

FIG. 20 is a diagram illustrating the flow of processing in a failure mode status management function 1710.

FIG. 21 is a diagram showing an example of an output screen.

FIG. 22 is a block diagram of a process illustrating an embodiment according to the present invention.

FIG. 23 shows a change in state quantity due to sensor upper limit fixation.

FIG. 24 shows a change in state quantity due to the valve being fully opened and stuck.

FIG. 25 is a state transition diagram of single data with respect to sensor upper limit fixation, an event condition generation diagram, and a state transition diagram of a failure mode.

FIG. 26 is a state transition diagram of single data with respect to the valve fully opened and stuck, an event condition generation diagram, and a state transition diagram of a failure mode.

FIG. 27 is a view for explaining symbol changes when the sensor upper limit is fixed.

FIG. 28 is a view for explaining a symbol change when the valve upper limit is fixed.

FIG. 29 is a diagram illustrating a feature extraction filter generation process.

FIG. 30 is a diagram illustrating a method of determining a conversion coefficient.

FIG. 31 is a diagram illustrating a method for determining a feature extraction threshold.

FIG. 32 is a diagram illustrating an example of determining a feature extraction threshold.

FIG. 33 is a diagram illustrating a method of determining a symbolic dictionary.

FIG. 34 is a diagram illustrating the configuration of a state transition diagram etc. construction unit 184.

FIG. 35 is a diagram illustrating a system configuration of a conventional method.

FIG. 36 is a block diagram illustrating a configuration of a symbol conversion unit 4730.

FIG. 37 shows a change in state quantity due to a certain abnormality X.

FIG. 38 is a diagram for explaining the configuration of the symbol network in detail.

FIG. 39 is an explanatory diagram of a symbol network that expresses a change event of measurement data.

FIG. 40 is a view for explaining a symbol change when the control system A is abnormal X.

FIG. 41 is a diagram for explaining the configuration of an event management table in detail.

[Explanation of Codes] 50 ... Diagnosis unit, 100 ... Control device, 110 ... Process measurement data input device, 120 ... Symbolization unit, 122 ... Symbolization processing unit, 124 ... Parameter storage unit, 130 ... Single data management unit, 132 ... Single data status management unit, 1
34 ... Single data state transition diagram storage unit, 140 ... Event generation unit, 142 ... Event management unit, 144 ... Event generation condition storage unit, 150 ... Failure mode management unit, 152 ... Failure mode status management unit, 154 ... Failure mode state transition diagram storage unit, 160 ... Process state display management unit, 162 ... Process state display device, 170 ... Process database, 1
80 ... Construction support unit, 182 ... Symbolization parameter construction unit,
184 ... State transition diagram construction unit, 196 ... Management mechanism.

Claims (10)

[Claims] 1. A system for diagnosing a state of a process by processing various data obtained from a process, wherein a symbol for converting a changed state of pattern data such as time series data obtained from the process into a symbol. And a diagnosing means for diagnosing the state of the process by using a temporal change of the symbols obtained from the symbolizing means and a temporal relationship between a plurality of symbols. Process status diagnostic system. 2. A system for diagnosing a state of a process by processing various data obtained from a process, wherein a symbol is used to convert a change state of pattern data such as time series data obtained from the process into a symbol. And a diagnostic means for diagnosing the state of the process by using the temporal change of the symbol obtained from the symbolizing means and the temporal relationship of the symbols of a plurality of time series data, and the diagnostic means obtained from the diagnostic means. And a process state display management means for displaying the diagnostic result. 3. The diagnostic means used in claim 1 uses a symbol obtained from the symbolizing means to express a change pattern of time-series data at the time of a failure by a transition sequence of the symbol. Single data management means for updating a state (status) in a state transition diagram of one data, and statuses of a plurality of data in the state transition diagram of the single data satisfy predetermined time conditions and the like. At this time, the state transition diagram of the failure mode expressing the reaching stage of each failure mode is updated using the event generation means for generating an "event" specific to the failure mode and the event obtained from the event generation means. A process state diagnosis system comprising: a failure mode management unit. 4. A symbol, a transition start time, a transition minimum time, and a transition maximum time required for each status of a single data state transition diagram used in claim 3 to transit to that status. And if the condition for transitioning from the current status to the next status is different from the symbol required to transit to the next status, obtain from the encoding means. The latest symbol that is displayed is equal to the symbol that is required to transition to the next status, and the start time of the latest symbol is from the transition start time of the current status to the transition minimum time or more and the transition maximum time or less, and the next status If the symbol required to transition to is equal to the symbol required to transition to the current status, the latest symbol above is required to transition to the next status. A process state diagnosis system, which is equal to a required symbol and has a start time of the latest symbol that is equal to or longer than a transition maximum time from a transition start time of a current status. 5. The transition between the statuses of different measurement data that the event generation means used in claim 3 must satisfy in its failure mode in the state transition diagram of the single data. An event generation condition diagram storage unit that stores an event generation condition diagram that expresses a context of start time and an event name that is generated when the relation is satisfied, a state transition diagram of the single data, ON / OFF of the switch obtained from the above event generation condition diagram and control device
And a event management means for generating an event in each failure mode from signals such as. 6. The event generation condition diagram used in claim 5 has a minimum condition time and a maximum condition time, and a condition between a pair of single data is used as a condition for generating an event. There are three types of cases: the difference between the transition start times is equal to or more than the minimum condition time and is equal to or less than the maximum condition time, the condition is satisfied between a plurality of pairs of statuses, and a status is in a transition state A process state diagnosis system characterized by being able to express. 7. The failure mode management means used in claim 3 stores a failure mode state transition diagram representing a failure mode arrival stage in each failure mode as the transition sequence of the event. A failure mode state transition diagram storage means, and a failure mode status for managing the arrival stage of the failure mode using the failure mode state transition diagram according to the change of the event obtained from the event generation means. A process state diagnosis system comprising: a management unit. 8. Feature extraction used automatically in feature extraction from a symbolized pattern by means for constructing internal parameters in the symbolization means used in the process state diagnosis system according to claim 1. A state transition diagram of the single data used in the process state diagnosis system according to claim 3, comprising a filter and a parameter for feature extraction, and a symbolized parameter construction means for determining a symbol pattern of the dictionary. And a means for constructing knowledge stored in the event generation condition diagram and the failure mode state transition diagram reads the data accumulated in the process database,
A symbolization means for converting into a symbol and storing in a symbol string file; a single data state transition diagram construction means for reading the symbol string from the symbol file and creating a state transition diagram for the single data; Using the state transition diagram of the single data obtained from the state transition diagram construction means of one data, the event generation condition construction means for producing the event generation condition diagram, and the above-mentioned thing obtained from the event generation condition construction means A failure mode state transition diagram construction means for creating the failure mode state transition diagram by using an image generation condition diagram; a state transition diagram of the single data between different failure modes; and an event generation condition diagram And a consistency check means for checking the consistency of the process status diagnosis system. 9. The single data management means used in claim 3 stores a state transition diagram of single data expressing a change pattern of time-series data at the time of failure by a transition sequence of symbols. A single data state transition diagram storage unit, and a single data status management unit that updates the single data state transition diagram according to a change in a symbol obtained from the symbolization unit. Characteristic process state diagnosis system. 10. When expressing a plurality of one-dimensional data, each one
A method of expressing a plurality of one-dimensional data, characterized in that a pattern of the dimensional data is expressed as a series of element patterns constituting the pattern, and a relation between the one-dimensional data is expressed as a constraint between the element patterns.