JPH0434165B2 - - Google Patents

Description

[Detailed description of the invention]

The present invention relates to a plant diagnostic device that identifies abnormal events in a plant online in real time and predicts the future in order to contribute to improving operational reliability and availability of large-scale plants. A conventional device of this type is shown in FIG. In the figure, 1 is a data acquisition device (for example, an analog-to-digital converter) for reading process data, and 2 is a data collection device that compares the process data with a reference value. arithmetic processing device 1 for converting into “1”;
3 is a storage device 1 that stores the results of arithmetic processing by the arithmetic processing unit 12; 4 is a cause-and-effect tree (hereinafter referred to as
The storage devices 2 and 5 that store the process information stored in the storage devices 13 and 24 and the CCT are diagnostic processing devices that fix the first cause of an abnormality and predict the future. , 6 is a cathode ray tube display device for displaying the diagnostic results. Next, the operation will be explained. Process data Xi (i = 1, 2, ..., N)
shall be. Xi is quantized by the data acquisition device 1. Using the quantized data Xi as input, the arithmetic processing unit 2 performs the processing shown in the formula, and the result is
Si (i=1, 2, . . . , N) and the event occurrence time are stored in the storage device 3. x ^L _i xix ^U _i =>Si=0 x ^U _i <xi or xi<x ^L _i =>Si=1 i=1, 2,..., N x ^L _i ; lower limit alarm level, x ^U _i ; upper limit Alarm level Here, Si is called the status of xi. Further, the time when the status becomes 1 is called the event occurrence time. An example of a portion of the CCT stored in the storage device 4 is shown in FIG. In FIG. 2, M6 and M7 are diagnostic messages, τ is a time delay, G ₁₁ and G ₁₂ are AND gates, and G ₂₁ and G ₂₂ are OR gates, respectively.
In CCT, the position where Si is defined is called a node. S ₁ , ..., S ₇ in Fig. 2 are storage devices 3
is stored in. CCT describes the abnormality propagation sequence as a logical relationship between statuses, but when the event propagation time between statuses is important, a time delay (τ in Figure 2) is used. CCT
The execution process is performed by the diagnostic processing device 5 based on information from the storage devices 3 and 4. The following process is performed to search for the first cause of disturbance. First, the subs that should be searched among the many CCT counties
In order to extract the CCT portion, an entry node whose status has changed (designated in advance as a node to start analysis) stored in the storage device 3 is detected, and a search starts from that node to lower nodes. And the first gate?
Find out whether it is an AND gate or an OR gate. Next, it is checked whether the relationship between the state of each input of the gate and the event occurrence time, which is the plant observation data written in the storage device 3, matches the CCT model written in the storage device 4. If the logic is established, the next node to be searched is determined according to the CCT model stored in the storage device 4 in order to expand to lower nodes. The order of search nodes is to start with the entry node, search for all nodes that are children of the entry node, and then search for all nodes that are grandchildren of the entry node. In this way, in order, for each gate connected below the entry node, it is determined whether the node on the input side of each gate is false (mismatch) or true (match), and the true branches are spliced together. . When the expansion reaches the primary node, the logic of this entry node is established, and it is determined that the cause has been successfully identified. Time delays between events are set in various parts of the CCT, but if an event actually progresses faster than this, nodes below that will not be expanded. That is, when it is found that there is no cut set for the entry node during the expansion, the diagnosis is stopped at that point. For example, if one side of the AND logic is "false", the following expansion will not be performed. Therefore, once the diagnosis is aborted or finished, analysis from the next entry node is performed taking priority into account. The above processing is performed using the CCT shown in Figure 8.
I will explain about it. In Figure 8, MT, M ₁ , M ₄ ,
M ₁₇ , M ₁₈ are diagnostic messages, G ₁₃ , G ₁₄ , G ₁₅ ,
G ₁₆ is an AND gate, G ₂₃ , G ₂₄ , G ₂₅ , G ₂₆ , G ₂₇ ,
G ₂₈ , G ₂₉ are OR gates, ● are observation nodes, ○
is a non-observation node, VT, V ₀ , V ₁ to V28 are observation device names, and H indicates a requirement condition. FIG. 4 shows an example of the status of observable nodes corresponding to the CCT of FIG. 3. In Figure 4, E indicates that the variable is within the alarm level, H indicates that the variable exceeds the high alarm level.
T indicates that the observed node matches the required conditions, and F indicates that the observed node does not match the required conditions. When V6 exceeds the alarm level, a search begins to find out what the cause of this is.
V6: Entry node The node status in FIG. 4 indicates whether the condition of the node is satisfied (T: true) or not (F: false). If true, this indicates that the process quantity exceeds the alarm level. Starting from V6, expand the CCT downwards,
Search for a set (minimum cut set) for which V6 is true. Expressed as a character expression, it can be written as follows.

[Table] For entry nodes for which the cause has been successfully identified, the logic is traced upwards from that node, and the progress state of the disturbance is analyzed. This is called prediction. That is, check how far the disturbance has spread in the upward direction, and assuming that the disturbance has spread further, check to determine whether the observable node one level higher than the topmost active node will change its state after a while. . Therefore, first check the type of gate immediately above the entry node. Next, check the input status of that gate and search to see if its output status is currently False but will become True over time. (If there is no gate, the status of the observable node above the entry node) When the above prediction process is applied to the CCT in Figure 3, the result is as follows. Starting with V6, search CCT toward the top. Make sure that the gate directly above V6 is an AND gate. Output node U4 becomes true because V7 is true due to V12 and V6 is true. As we go further up, it is an OR gate, so theoretically V1 is true, but from Figure 4 V1
The observed value of is not true due to the time delay. As a general rule, predictions search upwards, but for AND gates, processing is performed downwards. (For example, when checking the status of V7, V12, and V18 to determine the status of U4). The prediction process in the CCT shown in Figure 8 can be described using a literal expression as follows.

[Table] Through the cause identification and prediction processing described above, a tree that sequentially represents the disturbance from its occurrence to its future propagation can be determined. Restore the message contents and variables connected to this relevant tree. Then, the following information is restored on the cathode ray tube display device 6. (1) Messages specified in the tree section that has progressed up to now (ex, M17, M4 messages) (2) Messages that will become active in the near future (example: M1) (3) Follow the steps up to now Variable list related to the obtained tree part In the conventional CCT diagnosis device, the variable is configured as above, so the variable list is one level above the highest observable node among the active nodes. It could only predict up to observable nodes, and if a message was attached to that node, it would simply issue it as a potential active message. Thus, for example, when a signal to a node cannot be detected due to a failure, the node is determined to be a potentially active node, no matter how long the event is propagating. During the prediction process, if even one item is determined to be potentially active, the prediction process ends and the search is not performed further upstream, so even if the plant is in the worst condition, the operator cannot Cannot be displayed as a message. To explain the case in Figure 8, node N1
Even if node N0 becomes active, it does not become active due to a failure.In the actual plant, an abnormality in the final state called VT/H occurs without displaying the message MT indicating future prediction. Sometimes. Therefore, there is a drawback in that the operator cannot check the displayed prediction message and take corrective action according to the abnormality. This invention was made in order to eliminate the drawbacks of the conventional ones as described above, and for each execution cycle,
The CCT AND and OR gates are calculated from the primary node observation results and the CCT logic diagram, and the status of the logical operation values of all nodes is determined.
During prediction processing, this logical operation value is diagnosed as correct. If a node's logical operation value and observed value are different, it does not immediately determine that it is a potentially active node, but instead checks the status of the observed node on the higher side on the corresponding CCT. The present invention aims to provide a plant diagnosis device that determines whether there is a delay in the signal to a node or a failure, and accurately displays a predicted message to the operator. An embodiment of the present invention will be described below with reference to the drawings. In Fig. 5, 1, 2, 3, 4, and 6 are of the same type as the conventional one, and 5' is the process stored in 3 and 4 in addition to the conventional cause identification and prediction functions. A diagnostic processing device that has the ability to calculate logical operation values for all nodes using information and CCT. 7 is a storage device that stores the logical operation value of each node obtained in 5'. The CCT execution processing method of the present invention will be explained. The diagnostic processing device 5' inputs the observed value of the primary node in the storage device 3 and inputs it to the storage device 4.
According to CCT gate logic, calculate the logical operation values of all other nodes every execution cycle,
It is stored in the storage device 7. This calculation is performed from the bottom up every time, regardless of the status value or change of the observed value of the primary node. Then, CCT diagnostic processing is performed based on the logical operation results stored in the storage device 7, and in this execution method, in prediction, it is possible to correctly identify detector failures and signal delays and suggest future events. The point is that I did it like this. Since we proceed with the diagnosis assuming that the logical operation result is correct, if the logical operation value and the observed value of a node are different, it can be said that one of the following is the case. Observation results are incorrect due to instrument failure. Although the event has already occurred, it has not yet appeared in the observed values due to the instrument delay. The event will occur in the future, but it has not yet occurred. Since no event has occurred in the condition of the instrument outlet, they can be treated the same. The following will be called a failure and the following will be called a delay. Prediction is performed for each node above the entry node, but since the processing differs depending on the type of node, each item will be described separately. () When the upper node is unobserved In this case, there is only the result of a logical operation and it is correct, so if it is true, it is active, and if it is false, it ends and goes to the next entry node. () When the operation result of the upper node is equal to the observed value In this case, the result is correct because the result of the logical operation and the observation result are equal. Therefore, if it is false, the node is terminated, and if it is true, it is an active node. () When the operation result of the upper node and the observation result are not equal In this case, there are cases where the operation result is false and the observation result is true, and vice versa. Since the calculation result is correct, in the former case, there is an instrument failure. In the latter case, the observed value is a failure or delay. Since it is unknown from this node alone which of these is the case, higher observable points are checked. In other words, the conventional system did not have a function to search further upstream. () When the observed value of the upper node of the upper node is true At this time, the operation result of the upper node is true, the observation result is false, the operation result of the upper node of the upper node is true, and the observation result is true. Since the upper node of the higher order is true, it cannot be considered that at least the upper node of the higher order is behind. Also, assuming that we do not consider that two observation devices are out of order, this means that the upper node is faulty and the result is false, and the upper node is normal and shows True; It is considered as active and advances to the top. () When the observed value of the upper node of the upper node is false In this case, the operation result of the upper node is true and the observation result is false, and the operation result of the upper node is true and the observation result is false. At this time as well, assuming that we do not consider that two observers are out of order in a row, the upper node assumes that it is false due to a delay, outputs a potential active signal, and ends the process. Even if the upper node is false due to a failure and the upper node is false due to a delay, the potential is
This is simply made active, and if the upper node becomes true when the time comes, the failure can be discovered from the logic in () and predictive processing can be executed. FIG. 6 shows the processing flow at the time of the above-mentioned prediction. In the above embodiment, the diagnostic processing device 5' is
However, a configuration in which the present invention is divided into two parts, a logic operation section that calculates the logic operation values of all nodes and a diagnosis section, may be used, and the same effects as in the above embodiment can be obtained. Furthermore, the storage devices may each have different configurations but may have a single hardware configuration, and the same effects as in the above embodiments can be obtained. By configuring the plant diagnosis device as described above, even if there is an abnormality in the detector, it is possible to show the operator the future propagation process of the abnormality in the plant, and as a man-machine system, the operator can receive dense information. This has the effect of increasing diagnostic efficiency and plant operation availability.

[Brief explanation of drawings]

Fig. 1 is a block diagram showing a conventional plant diagnosis device, Fig. 2 is an explanatory diagram of CCT showing an example of plant diagnosis logic, Fig. 8 is a CCT diagram for explaining an example of CCT execution processing, and Fig. 4 is a diagram showing an example of the status of an observable node corresponding to the CCT in Figure 8.
The figure is a block diagram of a plant diagnosis system showing one embodiment of the present invention, and FIG. 6 is a prediction processing flow diagram. In the figure, 1 is a data collection device, 2 is a first arithmetic processing unit, 3 is a first storage device, 4 is a second storage device, 5 is a diagnostic processing device, and 5' is also the logic operation function of all nodes. 6 is a cathode ray tube display device, and 7 is a third storage device for storing logical operation values of each node. In the figures, the same reference numerals indicate the same or corresponding parts.

Claims (1)

[Scope of Claims] 1. A data collection device that collects process data for each node in the plant; an arithmetic processing device that compares each collected data with a reference value and outputs the comparison result as an observation result; Input the cause-and-effect tree in which the propagation sequence of the abnormal event that occurs is described in a logical formula and the observation results, and proceed from the entry node among the nodes to the lower nodes in the cause-and-effect tree to determine the conditions for establishing the logic. is searched to identify the cause of the abnormality, and based on the observation results for the primary node at each node, the logical operation results are sequentially obtained for each logic in the cause-effect tree, and the After determining that the logical operation result for each node in the cause-effect tree is correct, each node higher than the entry node is inspected, and the logical operation result of the node to be inspected indicates that the node is abnormal. Processing is terminated if the result of the logical operation indicates that an abnormality has occurred for that node, and the observation result for that node also indicates that an abnormality has occurred. If the result of the logical operation indicates that an abnormality has occurred in that node, the higher-ranking node is to be inspected, but the observation result for that node does not indicate an abnormality. If the observation results indicate that an abnormality has occurred, check the observation results for the higher-rank nodes, and if the observation results indicate that an abnormality has occurred, check the higher-rank nodes and check whether the observation results indicate that an abnormality has occurred. a diagnostic processing device that terminates the process if the diagnostic processing device indicates that there is no abnormality; and a display that displays a message that is to be output during the processing of the diagnostic processing device as predictive information indicating that the abnormality may spread. A plant diagnostic device equipped with a device.