TITLE OF INVENTION
DIAGNOSIS METHOD AND APPARATUS FOR PLANT
FIELD OF THE INVENTION
The present invention relates to a diagnosis method and apparatus for a plant in a fault diagnosis system for a plant configured with many pieces of equipment, the method and apparatus enabling plant fault diagnosis to be continued without restructuring even if a process involved in fault diagnosis is newly added or deleted in maintenance after commercial operation is started.
BACKGROUND OF THE INVENTION
In power plants typified by thermal plants and nuclear power plants as well as industrial plants typified by medical plants, food plants, and chemical plants, many process signals are monitored to operate those plants in a stable manner.
Specifically, to have an operator grasp the plant status, measuring instruments for measuring pressure, temperature, flow rates, water levels, and the like are disposed at various locations and obtained process signal values are displayed so that the operator can see them. In most plants, the obtained process signal values are stored in a process computer, which is a
dedicated computer, from the viewpoint of provision against faults and problems or of maintenance.
When the plant status has changed, to check whether related process signals have also changed, the operator displays the values of these process signals on a monitoring screen. The operator also displays the previous values of the process signals, which are stored in the process computer, on the monitoring screen, if necessary. Usually, a plant control system and the monitoring system of a plant are combined, so the plant status can be monitored and controlled in real time.
In a conventional practice, when the plant status changes, the operator displays the values of related process signals on the monitoring screen of the monitoring system. The values of process signals to be displayed are updated online according to the retrieving cycle of the monitoring system. The operation status of the plant is grasped from the state of this process value. The plant control system and monitoring system (collectively referred to below as the monitoring and control system) operate in synchronization with alarms corresponding to the states of process values. For example, if a pressure value exceeds a set value, an alarming lamp lights to notify
the operator. Furthermore, control to recover from a faulty state to the normal state is activated to encourage the operator to perform an operation.
Recently, many apparatuses and methods have been studied by which signs of plant faults are predicted before a fault state occurs. There are also support systems and methods by which, after a fault has occurred, its cause is indentified and analyzed.
The system in patent document 1 uses ranked warning according to the level to diagnose a plant.
Patent document 2 describes a network that transmits operation and control information, site information, and equipment information, and also describes a system having an information terminal system that displays these information items and a method.
PRIOR TECHNICAL DOCUMENT PATENT DOCUMENT
Patent document 1: Japanese Patent Laid-open Publication No. 2005-258649
Patent document 2: Japanese Patent Laid-open Publication No. 2011-70334 NON-PATENT DOCUMENT
Non-patent document 1: G. A. Carpenter and S.
Grossberg: "ART2 Self-Organization of stable category recognition codes for analog input patterns", Applied Optics, Vol. 26, No. 23 (1987)
SUMMARY OF THE INVENTION
A plant is configured with many pieces of equipment. Some of them are large and some of them are small. In diagnosing a plant, a statistical model is often used that is created by, for example, using statistical processing on the basis of related process signals and using learning typified by a neural network in accordance with the scale and function of the plant. The statistical model is used to model a correlation of process signals to be input.
The methods described in PTL 1 and the Patent document 1 and 2 are both diagnosis methods for a plant that use the adaptive resonance theory (ART). An ART may be used as a kind of the principle of classifying of the classifier that classifies data that has been input into a plurality of categories.
It is assumed here that process signals of a plant that has no fault have been input to an ART and have been classified into a plurality of categories. When a process signal is input to the ART during diagnosis at the time of fault generation, a new category that does
not belong to existing categories may be generated. In this case, a warning is issued because a plant status that had not been present before has occurred.
When a power plant is operated for a long period of time, new equipment is added and measuring instruments are added or removed at an intermediate point in time. In diagnosis using an ART, if a process signal to be input changes or the number of process signals to be input is increased or decreased as a result of maintenance as described above, the diagnosis cannot be continued, so classification into categories has to be carried out again by using the ART. If the operation has been carried out for many years, a huge amount of time-series data of process signals has been stored, making restructuring difficult.
An object of the present invention is to provide a diagnosis method and an apparatus that solve the above problem.
In the present invention, to solve the above problem, in a diagnosis method for a plant in which a) a model that is used to model a correlation of input variables is retained, b) data that has been input is classified into a plurality of categories according to the correlation of the input variables, and c) a plant fault is determined to have been detected when a
category not belonging to classified normal categories is generated, a new model is created for a model that needs reviewing for addition or deletion of input variables due to plant modification, through a review in which category numbers of the model are used.
When an input variable has been added to an existing model, a diagnosis model is created by using only the added input variable, and a model with a new category number is created from a category number obtained from the existing model and a category number obtained from the added diagnosis model.
When an input variable has been deleted from an existing model, the model is reviewed by using time-series data of input variables for the category number of the model, assuming that the value of an input variable to be deleted, which is one of input variables that have been normalized and have been input, is a fixed value.
When an input variable has been deleted from an existing model, the model is reviewed by using the time-series data of input variables for the category number of the model and deletion processing is performed after the addition processing has been completed, assuming that the value of an input variable to be deleted, which is one of input variables that
have been normalized and have been input, is a fixed value.
In the present invention, to solve the problem described above, in a diagnosis method for a plant in which a) a correlation of input variables obtained from a monitoring target is modeled, b) data that has been input to the resulting model is classified into a plurality of categories, and c) a plant fault is determined to have been detected when a category not belonging to any of the classified normal categories is generated, an integrated model is created by modeling the correlation of a first input variable, retaining the resulting model as an element model, and classifying data that has been input into a plurality of categories according to the correlation of the first input variable and also by modeling the correlation of an added second input variable and classifying data that has been input into a plurality of categories, the integrated model being based on the element model and the modeled second input variable.
The integrated model includes a new category that is determined by a combination of the category of the element model and the category of the modeled second input variable.
In the present invention, to solve the problem
described above, in a diagnosis apparatus for a plant that detects a fault in a monitoring target from a control signal obtained from a monitoring and control system that controls the monitoring target and from a measured signal obtained from a process computer that receives a process signal of the monitoring target; the diagnosis apparatus for a plant has an element model part that creates an element model by modeling the correlation of a signal that has been input, classifies data that has been input to the element model into a plurality of categories according to the correlation of the input signal, and determines that a fault has been detected in the plant when a category not belonging to classified normal categories is generated and an integrated model part, having an integrated model including at least one element model, that classifies data that has been input to the integrated model into a plurality of categories according to the correlation of the input signal and determines that a fault has been detected in the plant when a category not belonging to classified normal categories is generated; the diagnosis apparatus creates a diagnosis model, for the element model to which the input signal has been added according to modification of a control target, by using only an added input signal, and a model obtained by
creating a new category number from a category number obtained from an existing element model and a category number obtained from the added diagnosis model is stored in the integrated model part as an integrated model.
When an input variable is deleted from an existing model, the model is reviewed by using the time-series data of the input variables for the category number of the model, assuming that the value of an input variable to be deleted, which is one of input variables that have been normalized and have been input, is a fixed value.
With the diagnosis method and apparatus for a plant in the present invention, even if a process signal to be monitored or checked as an input signal to a diagnosis model is changed, added, or deleted, a diagnosis result and a trend display can be continuously presented to an operator or a maintenance person, so the diagnosis method and apparatus can contribute to stable operation of power plants and industrial plants.
BRIEF DESCRIPTION OF THE DRAWINGS
FIG. 1 is a drawing in which a diagnosis apparatus for a plant in the present invention has been applied
to a power plant.
FIG. 2 is a system block diagram indicating the structure of a thermal power plant that is used as a target to be diagnosed.
FIG. 3 is an enlarged view of pipes and an air preheater in the thermal power plant.
FIG. 4 illustrates a format in which data is stored in a process computer 300.
FIG. 5 illustrates an example of information stored in a model information database 450.
FIG. 6 illustrates another example of information stored in the model information database 450.
FIG. 7 illustrates an example of information to be stored in a diagnosis result database 480.
FIG. 8 is a flowchart indicating processing carried out by an element model creating part 420.
FIG. 9 is a flowchart indicating processing carried out by an integrated model creating part 430.
FIG. 10 is a flowchart indicating processing carried out by an elemental diagnosis part 460 and an integrated diagnosis part 470.
FIG. 11 illustrates an initial screen displayed on an image display unit.
FIG. 12 illustrates a diagnosis model setting screen displayed on the image display unit.
FIG. 13 illustrates trend graphs of process signals displayed on the image display unit.
FIG. 14 illustrates a diagnosis result display setting screen displayed on the image display unit.
FIG. 15 illustrates a display example of a diagnosis result displayed on the image display unit.
DESCRIPTION OF THE PREFERRED EMBODIMENTS
A diagnosis method and apparatus for a plant in the best embodiment will be described below with reference to the drawings.
{Embodiment}
FIG. 1 illustrates an example in which a diagnosis apparatus for a plant in this embodiment has been applied to a power plant 100, which is one of targets. In the power plant 100, many measuring instruments are installed to grasp the plant status. The value of a process signal 10 measured by each measuring instrument is transmitted to a monitoring and control system 200 and to a process computer 300 that stores measured values through dedicated lines or general transmission lines.
The monitoring and control system 200 outputs a control signal 20, which is used to maintain plant operation in a desired state, according to the value of the process signal 10. The output control signal 20 is
input to the power plant 100 and also input to a plant diagnosis apparatus 400.
The process computer 300 accumulates the values of process signals 10 obtained from the power plant 100. The accumulated process signals 10 are output to the plant diagnosis apparatus 400 as process signals 30, according to the purpose. The storage format will be described below in detail with reference to FIG. 4.
The plant diagnosis apparatus 400 retrieves control signals 20 and process signals 30 required for diagnosis through an external input interface 410. The plant diagnosis apparatus 400, which is connected to a support tool 910 through an external output interface 490 and the external input interface 410, receives an operation signal from, for example, an operator who is a user and also displays necessary information on an image display unit 950.
The external input interface 410 of the plant diagnosis apparatus 400 switches between a model creation mode and a diagnosis mode according to a command from the support tool 910. The external input interface 410 outputs the input control signal 20 and process signal 30 to the element model creating part 420 while in the model creation mode and to the elemental diagnosis part 460 while in the diagnosis
mode.
The element model creating part 420 creates an ART model according to a diagnosis target. If there are a plurality of diagnosis targets, the element model creating part 420 creates as many ART models as the number of diagnosis targets. To create an ART model, process signals 30 in the normal state are input to the ART model. The ART model classifies input data according to the correlation of the process signals 30 that has been input. Each classified data item is called a category. The resolution of the categories depends on the size of a precaution parameter. A method of appropriately setting the precaution parameter is proposed in PTL 1 or 2 described above. Detailed operation of the ART model is described in non-patent document 1 in detail, so a detailed explanation will be omitted.
The precaution parameter and the number of categories of each ART model are stored in a model information database 450. Information is extracted from the model information database 450 as necessary. Operation in the element model creating part 420 and the structure of the model information database 450 will be described later in detail. FIGs. 5 and 6 illustrate examples of the structure of the model
information database 450.
An integrated model creating part 430 creates a new ART model in which a category number, which is an output of the ART model created in the element model creating part 420, is used as input data. As with the element model creating part 420, the integrated model creating part 430 outputs the precaution parameter, the category number, and the like of the ART model to the model information database 450 and external output interface 490. The integrated model creating part 430 extracts information from the model information database 450 as necessary. Operation in the integrated model creating part 430 and the structure of the model information database 450 will be described later in detail.
If a command in the diagnosis mode is received at the external input interface 410, data is input to the elemental diagnosis part 460. The elemental diagnosis part 460 loads information about a diagnosis model from the model information database 450. Input data is input to the loaded diagnosis model and diagnosis is carried out. According to the data that has been input to the diagnosis model, the data may be classified into a category that has been created in advance or a new category may be created according to the correlation of
the new input data.
When a new category is created, which indicates that a state other than the normal state has been detected, it is determined that there is a fault sign. The output category number at that time and the like are stored in a diagnosis result database 480. They are also output to an integrated diagnosis part 470. Operation in the elemental diagnosis part 460 and the structure of the diagnosis result database 480 will be described later in detail. FIG. 7 illustrates an example of the structure of the diagnosis result database 480.
The integrated diagnosis part 470 loads information about the diagnosis model from the model information database 450, inputs an output from the elemental diagnosis part 460 to the diagnosis model, and carries out diagnosis similarly. The diagnosis result is stored in the external output interface 490 and diagnosis result database 480. Operation in the integrated diagnosis part 470 will be described later in detail.
The external output interface 490 outputs an output result obtained in the integrated model creating part 430 or integrated diagnosis part 470 to the support tool 910.
An operator, who is an example of a user involved
in the power plant 100, can see various types of information about the power plant 100 by using an input unit 900 configured with a keyboard 901 and a mouse 902 and the support tool 910, which is connected to the image display unit 950. The operator can also access the control signal 20 from the monitoring and control system 200, the control signal 30 from the process computer 300, the diagnosis result obtained in the plant diagnosis apparatus 400, and information in the model information database 450 and diagnosis result database 480.
The support tool 910 is configured with an external input interface 920, a data transmission and reception part 930, and an external output interface 940.
An input signal 91 generated in the input unit 900 is input the support tool 910 through the external input interface 920. Similarly, the control signal 20 from the monitoring and control system 200, the control signal 30 from the process computer 300, the diagnosis result 40 from the plant diagnosis apparatus 400, and information in the model information database 450 and diagnosis result database 480 are also input to the support tool 910 through the external input interface 920. The data transmission and reception part 930 processes an input signal 92 according to information
about the input signal 91 supplied from the user, and outputs the processed signal to the external output interface 940 as an output signal 93. An output signal 94 is displayed on the image display unit 950.
Information stored in the databases and signal processing functions will be described below by using, as an example, a case in which the data processing apparatus in the present invention is applied to a thermal power plant.
FIG. 2 is a system block diagram indicating the structure of a thermal power plant that is used as a target to be diagnosed. In this example, a power generation mechanism in a coal-fired thermal power plant will be described.
When coal is used as a fuel, coal is supplied from a coal bunker 111, in which coal is stored, through a coal feeder 112 to a mill 110. The mill 110 finely grinds the coal into pulverized coal by using an internal roller. This pulverized coal, primary air, and secondary air used for combustion adjustment are supplied to a boiler 101 through a burner 102. The pulverized coal and primary air are guided from a pipe 134 to the boiler 101, and the secondary air is guided from a pipe 141 to the boiler 101. After air used for two-stage combustion is supplied from a pipe 142
through an after air port 103 to the boiler 101.
A high-temperature gas generated due to coal combustion flows along the path of the boiler 101 and then passes through an air preheater 104. After that, the gas undergoes exhaust gas treatment, after which the treated gas is released to the atmosphere through a stack.
Feed water, which circulates in the boiler 101, is guided to the boiler 101 through a feed water pump 105, and is heated by a gas in a heat exchanger 106, resulting in a high-temperature, high-pressure steam. Although, in this embodiment, one heat exchanger is used, a plurality of heat exchangers may be provided.
The high-temperature, high-pressure steam, which has passed through the heat exchanger 106, is guided to a steam turbine 108 through a turbine governor valve 107. Energy of the steam is used to drive the steam turbine 108, and electric power is generated by a generator 109. The generated electric power is supplied to an electric power system.
The gas exhausted from the steam turbine 108 is cooled in a condenser 113 and the resulting water is supplied to the feed water pump 105 again. A unit that uses steam extracted from the turbine to heat the feed water is provided at an intermediate point to improve
heat efficiency.
Various measuring instruments are disposed in the thermal power plant generally configured as described above. Information obtained from these measuring instruments is transmitted to the monitoring and control system 200 and the like in FIG. 1 as measurement information 10 (process signal 10). For example, FIG. 2 illustrates a flow rate measuring instrument 150, a temperature measuring instrument 151, a pressure measuring instrument 152, a generated power output measuring instrument 153, and a density measuring instrument 154.
The flow rate measuring instrument 150 measures the flow rate of feed water to be supplied from the feed water pump 105 to the boiler 101. The temperature measuring instrument 151 and pressure measuring instrument 152 respectively measure the temperature and pressure of the steam to be supplied to the steam turbine 108. The amount of electric power generated by the generator 109 is measured by the generated power output measuring instrument 153. Information about the densities of components (CO, NOx, and the like) included in the gas that is passing through the boiler 101 can be measured by the density measuring instrument 154. Although many other measuring instruments besides
those shown in FIG. 2 are provided in the thermal power plant, they are omitted in FIG. 2.
The paths of the primary air and secondary air supplied from the burner 102 and the after air supplied from the after air port 103 will be described with reference to FIG. 2.
The primary air is guided from a fan 120 into a pipe 130, branches at an intermediate point into a pipe 132, in which passes through the air preheater 104, and a pipe 131, which does not pass through the air preheater 104, joins again in a pipe 133, and is guided to the mill 110. The air passing through the air preheater 104 is heated by a gas. This primary air is used to transport the pulverized coal created in the mill 110 to the burner 102 through the pipe 134.
The secondary air and after air are guided from the fan 121 into a pipe 140 and is heated in the air preheater 104, after which the heated air branches into the pipe 141 for the secondary air and the pipe 142 for the after air. The air in the pipe 141 is guided to the burner 102, and the air in the pipe 142 is guided to the after air port 103.
FIG. 3 is an enlarged view of the pipes through which the primary air, secondary air, and after air pass as well as the air preheater 104. As illustrated

in FIG. 3, air dampers 160, 161, 162, and 163 are provided in the pipes. When the air damper is manipulated, the area of the relevant pipe 133, 141, or 142 through which the air passes can be changed, so the amount of air that passes through the pipe can be adjusted by manipulating the air damper.
Information about the process signal 10 stored in the process computer 300, information stored in the model information database 450 and the diagnosis result database 480, and the calculation functions of the element model creating part 420, integrated model creating part 430, elemental diagnosis part 460, and integrated diagnosis part 470 will be described below.
First, information about the process signal 10 to be stored in the process computer 300 will be described. FIG. 4 illustrates an example of information that has been stored in the process computer 300. As illustrated in FIG. 4, information measured in the power plant 100 is stored for each measurement instrument along with measurement time. For example, measurement time is stored as the vertical axis item in FIG. 4, and proportional integral derivative (PID) numbers uniquely-assigned to measurement instruments are stored as horizontal axis items in relation to types of measurement and units.
For example, PID150, which is a PID number, indicates data measured by the flow rate measuring instrument 150 for the flow rate of feed water in FIG. 2, PID151 indicates data measured by the temperature measuring instrument 151 for the main steam, PID152 indicates data measured by the pressure measuring instrument 152 for the main steam, PID153 indicates data measured by the generated power output measuring instrument 153, and PID154 indicates data measured by the density measuring instrument 154. As far as types are concerned, F indicates the flow rate, T indicates temperature, P indicates pressure, E indicates a generated electric power output, and D indicates the density of NOx included in the exhausted gas. Their units are kg/s, °C, Mps, MW, and ppm. Thus, measured values represented by the PID numbers, types, and units are stored along with information about time.
Although, in FIG. 4, data is obtained and stored at one-second intervals, the sampling cycle in data collection can be set to any value. The PID number uniquely set to each measured value can be used as a key to identify a process signal or search for a desired process signal so that data stored in the process computer 300 can be easily used.
Next, information to be stored in the model
information database 450 will be described. FIGs. 5 and 6 illustrate formats of information required by an ART-based diagnosis model. As illustrated in FIGs. 5 and 6, an element model table TB420 and an integrated model table TB430 have been prepared in the model information database 450. The element model table TB420 is a table that prestores information about an element model handled by the element model creating part 420, and the integrated model table TB430 is a table that prestores information about an integrated model handed by the integrated model creating part 430.
The element model table TB420 in FIG. 5 includes a model number, a model name, the PID number of input variables to be input to the model, a precaution parameter, and the maximum number of the categories created during model creation. In this example, the model number is E-001, the model name is the name of a main steam model related to the main steam of the power plant in FIG. 2, the input variables are PID150 (flow rate measuring instrument 150 for the flow rate of feed water) and PID151 (temperature measuring instrument 151 for the main steam), the precaution parameter is 0.82, and the maximum number of the categories created during model creation is 21. In this element model example, the main steam model is represented by the flow rate of
feed water and the main steam temperature, in this way.
The integrated model table TB430 in FIG. 5 includes a model number, a model name, the model number of an element model to be input to the model, a precaution parameter, and the maximum number of the categories created during model creation. In this example, the model number is T-001, the model name is the name of a main steam loop model related to the main steam of the power plant in FIG. 2, the input model is obtained by adding PID152 (pressure measuring instrument 152 for the main steam) to the main steam model E-001 used in the element model table TB420, the precaution parameter is 0.89, and the maximum number of the categories created during model creation is 25. In this integrated model example, the main steam loop model is represented as the model obtained by adding the main steam pressure to the main steam model E-001 used in the element model table TB420 in this way.
As is clear from the above, the integrated model in this example is a complex model configured by adding a new process amount to an element model. Although not described in the illustrated examples, the integrated model may be a combination of element models. Although the element model in this example is a model that simulates only main steam, the integrated model is a
model that simulates a main steam loop. Thus, the integrated model can be said to be a model that represents a wider and a higher-end range, and it achieves a wider range and a higher-end by integrally-using an element model and a process amount.
The element model and integrated model can be achieved by using a supervised learning type of adaptive resonance theory (ART) network, disclosed in the non-patent document 1 and the like, that classifies a plurality input data items into a plurality of categories. If, in this case, the element model is handled as a low-end network, the integrated model can be considered by positioning it as a high-end network.
If the number (order) of process amounts used in the element model is assumed to be N-th (2nd in the example in FIG. 5), the integrated model is a (N + M) -th order model (3rd-order model in the example in FIG. 5). From this, it can be said that creating an integrated model means creating a higher-order model by using an element model.
Therefore, it will be understood that the changing of a model in the diagnosis apparatus due to addition of a new unit or measuring instrument in a plant, which is the technical problem in the present invention, can be achieved by creating an integrated model, to which
an added process amount has been added, with the element model unchanged.
The number of precaution parameters varies depending on the adjustment method to be used. To compensate this, tables in which a plurality of precaution parameters are set are used. FIG. 5 illustrates examples, in each of which only one precaution parameter is set, and FIG. 6 illustrates examples of tables, in each of which a plurality of precaution parameters are set. FIG. 6 differs from FIG. 5 in that one precaution parameter is included for each category number.
FIG. 7 illustrates an example of diagnosis result information to be stored in the diagnosis result database 480. An element model table TB460 and an integrated model table TB470 have been prepared in the model information database 480. The element model table TB460 is a table that prestores diagnosis result information about an element model handled by the elemental diagnosis part 460, and the integrated model table TB470 is a table that prestores information about an integrated model handed by the integrated diagnosis part 470.
The element model table TB460 in FIG. 7 includes a model number, time, a category number, and a diagnosis
result. In this example, the model number is E-001, time is 2010/01/01 00:00:00, and the diagnosis result for category number 1 is recoded as being normal. The integrated model table TB470 in FIG. 7 is configured in a similar format; in this example, the model number is T-001, time is 2010/01/04 00:10:11, and the diagnosis result for category number 3 is recoded as being faulty.
In configuring the elemental diagnosis part 460 and integrated diagnosis part 470 in the present invention, the methods described in PTLs 1 and 2 can be applied. The methods described in PTL 1 and PTL 2 are both diagnosis methods for a plant that use the adaptive resonance theory (ART). The ART may be used as a kind of the principle of classifying of the classifier that classifies data that has been input into a plurality of categories.
It is assumed here that process signals of a plant that has no fault have been input to an ART and have been classified into a plurality of categories. When a process signal is input to the ART during diagnosis at the time of fault generation, a new category that does not belong to existing categories may be generated. In this case, a warning is issued because a plant status that had not been present before has occurred.
In addition to normal and faulty states, a new
category that does not belong to any of the previous states may be created as a diagnosis result. In this case, that information is presented to the user through the support tool 910 as being unknown. If this state is decided to be normal or faulty later, the diagnosis result in the diagnosis result database 480 is replaced. If a similar category is generated later, a diagnosis result is output according to this result.
So far, the structure of the power plant used as a plant to which the apparatus in the present invention is applied has been described. Specific examples of the element model table and integrated model table have also been described in the examples of the process amount used in this case. Assuming that this has been understood, processing by the element model creating part 420 will be described next.
FIG. 8 is a flowchart indicating operation in the element model creating part 420. First, whether a new model is to be created or an existing model is to be modified is decided in step S421. If a new model is to be created, the sequence proceeds to step S422. If not, the sequence proceeds to step S425.
Whether a new model is to be created or an exiting model is to be modified is decided according to a command given by, for example, an operator who is a
user involved in the power plant 100 from the input unit 900. If a command related to the changing of a model has not been given by the user, therefore, the processing in FIG. 8 is not executed. The user gives a model changing command in response to the changing of the power plant 100. A specific method of giving a model changing command will be described later with reference to FIG. 12.
If the model changing command is to create a new model, a new ART model is created in step S422 by using a previous process signal stored in FIG. 4. After that, information about the created ART model is stored in the model information database 450 in step S423. The new resulting ART model that has been created and stored in the model information database 450 is, for example, the main steam model E-001 in FIG. 5, which is an element model determined by the flow rate of feed water and the main steam temperature.
In step S424, whether models for all relevant monitoring items have been created is decided. If models for all relevant monitoring items have been created, the sequence ends. If not, the sequence returns to step S421 and the subsequent steps are repeated until models for all relevant monitoring items have been created.
If the model changing command is to modify an existing model, whether an input variable has been added to the existing model or deleted from the existing model is decided in step S425. If an input variable has been added, the sequence proceeds to step S426. If an input variable has been deleted, the sequence proceeds to step S427.
In step S426 to which the sequence proceeds when an input variable has been added, an ART-based diagnosis model is created by using only the variable added to the existing model. It is assumed here that the existing model is the main steam model E-001 in FIG. 5 and the added variable is PID152 (pressure measuring instrument 152 for main steam). Accordingly, in step S426, an ART-based diagnosis model is created by using only the added variable PID152. Upon completion of the creation, the sequence proceeds to step S423 (in which information about the created ART model is stored in the model information database 450) and then to steps S424 (in which whether all monitoring items have been handled is decided).
In step S423, the existing model number E-001 and the added model number (assumed to be PID152) are stored in the input model field of the integrated model table TB430 in the model information database 450
illustrated in FIG. 5.
In step S427 to which the sequence proceeds when an input variable has been deleted, input variables that have not been changed are left unchanged and a fixed value (0.5 here) is input for deleted variables. In this embodiment, however, it is assumed that all input variables in the ART model have been normalized. That is, the input variables have been grasped as values in a range in which the maximum value is 1 and the minimum value is 0. Therefore, to input 0.5 is to handle the variable in subsequent processing as a value that will remain as an intermediate value and will not change.
It is assumed here that the existing model is the main steam model E-001 in FIG. 5 and the deleted variable is PID151 (temperature measuring instrument 151 for main steam). In this case, since 0.5 is input for PID151, the main steam model E-001 is practically handled in subsequent processing as a model that is determined only by the input variable PID150, which indicates the flow rate of feed water.
In step S428, since the deleted signal PID151 has been set to a fixed value of 0.5, the existing category is converted to a category used in this model. The main steam model E-001 in FIG. 5 includes 21 categories, so these categories are converted. To perform this
conversion, time-series data of input variables stored in the process computer 300 are retrieved by using times corresponding to category numbers stored in the relevant field in the main steam model E-001 as well as PID information about the variables that have been input to that model, with reference to the element model table TB460 in the diagnosis result database 480 illustrated in FIG. 7.
In this example, time-series data of input variables (PID150 and PID151) stored in the process computer 300 are retrieved by using times 2010/01/01 00:00:00 corresponding to category number 1 as well as PID information (PID150 and PID151) of the variables that have been input to the main steam model E-001, with reference to FIG. 4. The time-series data is input to the ART model. This operation is performed for all the 21 categories. Upon completion of the operation for all the categories, the sequence proceeds to step S423 and then to step S424, as described above.
If both addition and deletion have been carried out in the same model, the flowchart for addition is first decided to be executed in step S425, after which in step S424, the sequence returns to step S421 again and then the flowchart for deletion is decided to be executed in step S425 again.
FIG. 9 is a flowchart indicating operation in the integrated model creating part 430. First, whether the element model has been newly created is decided in step
S431. If the element model has been newly created, the
sequence proceeds to step S436. If not (the element
model has been modified), the sequence proceeds to step
S432. Since, in the case of new creation, there is no
processing to be carried out by the integrated model
creating part 430, the sequence proceeds to step S436,
after which the sequence returns to step S431 to make a
decision for another item or the processing is
terminated.
If the element model has been modified, the state of the input variable is decided in step S432. If the input variable is a one that has been added, the sequence proceeds to step S433. If not (the input variable has been deleted), the sequence proceeds to step S436. Since, in the case of input variable being deleted, there is no processing to be carried out by the integrated model creating part 430, the sequence proceeds to step S436, after which the sequence returns to step S431 to make a decision for another item or the processing is terminated.
In step S433, process signals that have been input to the existing model and newly created additional mode
during element model creation are input again. For example, in the addition processing in FIG. 8 (step S425), PID152 has been added to the existing model (main steam model E-001). As process signals related to them, the flow rate of feed water PID150, the main steam temperature PID151, and the main steam pressure PID152 are input again with reference to FIG. 4. In processing in step S423 in FIG. 8, the existing model number E-001 and added model number (assumed to be PID152) have been stored in the input model field in the integrated model table TB430 in the model information database 450 illustrated in FIG. 5.
In step S434, an ART model that uses category numbers output from the model described above as input variables is created. For example, a combination of category numbers that have been input is classified as a new category and the new category is used as a category of the integrated model.
Specifically, it is assumed that 10 category numbers (A0 to A9) are present for the existing model number E-001 and five category numbers (BO to B4) are present for the added model number PID152. In this case, if A0 (normal) and BO (normal) are combined, a new category is defined to be CO (normal). If A0 (normal) and B1 (faulty) are combined, a new category C1
(faulty) is defined. This operation is executed until the last combination, after which a model having a group of new category numbers is stored in the integrated model table TB430 as an integrated model.
If a measuring instrument or the like has been newly disposed, it is possible to create new models again for all process signals. However in that case, if data for the past year is to be referenced, a huge amount of data has to be processed.
By contrast, in the present invention, when input data for the past year is referenced and modeled to obtain new category numbers, it suffices to handle only the added input variables. Since existing element models focus on category numbers, input data for the past year is compiled into a plurality of category numbers.
Upon completion of model creation, the sequence proceeds to step S436. In step S436, whether all monitoring items have been handled is decided. If all monitoring items have not been handled, the sequence returns to step S431 and processing is repeated until all monitoring items have been handled. If all monitoring items have been handled, the sequence is completed.
Next, FIG. 10 illustrates a flowchart that
indicates the algorithm of the elemental diagnosis part 460 and integrated diagnosis part 470. First, information about each element model is loaded from the element model table TB420 in the model information database 450 in step S461. Next, information about each integrated model is loaded from the integrated model table TB430 in the model information database 450 in step S462. Then, the data in FIGs. 5 and 6 are presented to the elemental diagnosis part 460.
In step S463, whether diagnosis involves only the element model or both the element model and the integrated mode is decided. If diagnosis involves only the element model, the sequence proceeds to step S464. If diagnosis involving both the element model and the integrated model is required, the sequence proceeds to step S466.
Each element model is diagnosed in step S464. This means that the element model table TB420 is diagnosed. In step S465, the diagnosis result of the element model is stored in the element model table TB460 in the diagnosis result database 480.
In step S466, each element model is first diagnosed. This means that each element model in the integrated model table TB430 is executed. In the case of the main steam loop model T-001, for example, the element model
E-001 and the added model PID152 are both diagnosed. The diagnosis result of the element model is stored in the element model table TB460 in the diagnosis result database 480.
In step S467, diagnosis is carried out by using the category number output from each element model as an input, with reference to the element model table TB460 in the diagnosis result database 480.
In step S468, the diagnosis result is stored in the integrated model table TB480 in the diagnosis result database 480.
Then, whether diagnosis has been carried out for all monitoring items is decided in step S469. If diagnosis has not been carried out for all monitoring items, the sequence returns to step S461 and processing is repeated until all monitoring items have been handled. Upon completion of the handling of all monitoring items, the sequence ends.
In this step, operations in steps S467 and S468 are executed by the integrated diagnosis part 470 and operations in other steps S are executed by the elemental diagnosis part 460.
The external output interface 490 outputs diagnosis results of the element model and integrated model to the support tool 910 as an output result.
Next, the method by which the user uses the support tool 910 to display information about the control signal 20, process signal 30, diagnosis result 40, model information database 450, and diagnosis result database 480 on the image display unit 950 will be described.
FIGs. 11 to 15 are exemplary screens displayed on the image display unit 950. The user uses the keyboard 901 and mouse 902 to, for example, input parameter values in the blank fields on a screen 90.
FIG. 11 illustrates an initial screen displayed on the image display unit 950. A Diagnosis model creation button 951 and a Diagnosis result display button 952 are displayed on the screen 90, which is displayed as an initial screen. Of these, the user selects one that is required, after which the user moves a cursor 953 with the mouse 902 and clicks the mouse 902, displaying a desired screen.
FIG. 12 illustrates a setting screen for an element model and integrated model, which is displayed when the Diagnosis model creation button 951 is selected on the initial screen. The top field on the screen 90 indicates that the current screen is a screen on which to set display information. Fields described later are displayed at various portions on the screen 90. To set
a model, data is input to these fields or their buttons are selected.
In a process signal display field 961, the user inputs, to the input fields, measurement signals or operation signals that the user wants to input to the diagnosis model along with their ranges (upper limits and lower limits). In the example in the drawing, the flow rate of main steam is selected with its button as the measuring signal, and 300 (kg/s) and 0 (kg/s) are numerically input as its upper limit and lower limit, respectively. The unit (kg/s) is displayed by default when the flow rate of main steam is selected as the measuring signal.
When the flow rate of main steam is added to the diagnosis model, a time zone that the user wants to use and display during model creation is also input to a time input field 962. In the example in the drawing, one day on 2010/01/01 has been set as a start time and an end time.
When a Display button 963 is clicked on the screen 90, trend graphs are displayed on the image display unit 950 as illustrated in FIG. 13. In the example in FIG. 13, changes in a plurality of process amounts with time can be displayed. When a Return button 971 in FIG. 13 is clicked, the screen in FIG. 12 is displayed again.
A model number, a model name, input variables, and other information required in element model creation are displayed in an element model creation display area 964 in FIG. 12. A New button 965 and a Modify button 966 are displayed on the right side of the element model creation display area 964.
When the New button 965 is pressed, the element model creation display area 964 becomes blank, waiting for inputs. A new model can then be created with user's manual inputs.
The element model creation display area 964 in the drawing indicates an example of a screen displayed when the Modify button 966 is selected. In the example in the drawing, the model number is E-001, the model name indicates a main steam model, and the input variables are PID150 and PID151.
A fast way to create a model is to modify an existing model. In this case, it is preferable to display a model to be modified, press the Modify button 966, and modify the model. To achieve this, a search key input field 967 is provided so that an existing model can be searched for. When a search key is input and then a Search button 968 is pressed, information about the target model is displayed in the element model creation display area 964.
Basically, an integrated model creation display area 974 is also configured as is the element model creation display area 964. A model number, a model name, an input model, and other information required in integrated model creation are displayed in the integrated model creation display area 974. When a New button 975 is pressed, the integrated model creation display area 974 waits for inputs. A new model can then be created. To modify an existing model to create a new model, display the model to be modified, press a Modify button 976, then modify the model. A search key input field 977 is provided so that an existing model can be searched for. When a search key is input and a Search button 978 is then pressed, information about the target model is displayed in the integrated model creation display area 974.
When a Create button 992 is pressed after the above setting, relevant models are created. When a Return button 969 in FIG. 12 is pressed, the screen in FIG. 11 is displayed again.
FIG. 14 illustrates a setting screen 90 used to display diagnosis results on the image display unit 950. When the Diagnosis result display button 952 on the initial screen illustrated in FIG. 11 is clicked, the screen illustrated in FIG. 14 is displayed. The top
field on the screen 90 in FIG. 14 indicates that the current screen is a diagnosis result display setting screen. Fields described later are displayed at various portions on the screen 90. To set a model, data is input to these fields or their buttons are selected.
In a process signal selection field 981, the user inputs, to the input fields, measurement signals or operation signals the user want to display on the screen 90 of the image display unit 950 along with their ranges (upper limits and lower limits). The example in the drawing illustrates a screen displayed when a generator output and the flow rate of main steam are input along with their ranges (upper limits and lower limits).
Time that the user wants to display is also input to a time input field 982. After process signals the user want to display have been determined, when their selection columns are clicked and thereby checked, their selection is determined.
As in the case of the display information setting screen in FIG. 12, a model number, a model name, input variables, and other information are displayed for each model in an element model selection area 983. A search key input field 984 is provided so that a model that the user wants to display can be searched for. To
search for a model, input a search key and then press a Search button 985. The search result is displayed in the element model selection area 983. When the selection column is clicked, it is checked and the selection is determined. The display in the example in the drawing is in the case of the model E-001.
Basically, an integrated model selection area 986 is also configured as is the element model selection area 983. A model number, a model name, an input model, and other information are displayed for each model in the integrated model selection area 986. A search key input field 987 is provided so that a model that the user wants to display can be searched for. To search for a model, input a search key and then press a Search button 988. The search result is displayed in the element model selection area 986. When the selection column is clicked, it is checked and the selection is determined. The display in the example in the drawing is in the case of the model T-001.
Upon completion of the above input or selection operation, when a Display button 989 is clicked, trend graphs are displayed on the screen 90 of the image display unit 950, as illustrated in FIG. 15. In the example in FIG. 15, changes in the generator output and the flow rate of main steam, which have been selected
in the process signal selection field 981, with time are displayed in contrast with each other. For the selected element model E-001 or integrated model T-001 and the newly added input variable PID152, trend graphs are displayed in time series for each category number. As illustrated in FIG. 15, when a new category is generated, the outline portion on the trend screen is highlighted in another color, for example, to notify the user. When a Return button 991 in FIG. 15 is clicked, the setting screen in FIG. 14 is displayed again. When a Return button 999 in FIG. 14 is pressed, the initial screen in FIG. 11 is displayed again.
Although FIG. 15 has been shown as an example of a diagnosis result display screen, this is not a limitation; a display method can be considered in which a system diagram of a plant to be diagnosed is displayed, and when a new category is generated, that portion is highlighted and the trend display in FIG. 15 is given by clicking the portion.
Although an embodiment in which the model information database 450 and diagnosis result database 480 are included in the plant diagnosis apparatus 400 has been described, another embodiment in which they are not included in the plant diagnosis apparatus 400 but they are implemented by other hardware components
is also possible.
Next, advantageous effects obtained by applying the inventive diagnosis method and apparatus for a plant to fault and sign diagnosis of the power plant 100 will be described according to results obtained from the diagnosis method and apparatus.
If the inventive diagnosis method and apparatus for a plant is applied to fault and sign diagnosis of a power plant, when a diagnosis model is modified due to replacement or repair of part of facilities in operation, diagnosis can be carried out by using diagnosis data accumulated so far as it is. As a result, the diagnosis model can be easily restructured and trends of faults and signs can be detected by using previous accomplishments.
Since results of both an element model and an integrated model are presented to the operator by displaying trend graphs, visual monitoring becomes easier. Furthermore, when a plurality of diagnosis models are layered as an integrated model, a model that can diagnose a power plant in a wider range can be easily created.

WHAT IS CLAIMED IS:
1. A diagnosis method for a plant in which a model that is used to model a correlation of input variables is retained, data that has been input is classified into a plurality of categories according to the correlation of the input variables, and a plant fault is determined to have been detected when a category not belonging to classified normal categories is generated, wherein a new model is created for a model that needs reviewing for addition or deletion of input variables due to plant modification, through a review in which category numbers of the model are used.
2. The diagnosis method for a plant according to Claim 1, wherein when an input variable has been added to an existing model, a diagnosis model is created by using only the added input variable, and a model with a new category number is created from a category number obtained from the existing model and a category number obtained from the added diagnosis model.
3. The diagnosis method for a plant according to Claim 1, wherein when an input variable has been deleted from an existing model, the model is reviewed by using time-series data of input variables for a category number of the model, assuming that the value of an input variable to be deleted, which is one of
input variables that have been normalized and have been input, is a fixed value.
4. The diagnosis method for a plant according to Claim 2, wherein when an input variable has been deleted from an existing model, the model is reviewed by using the time-series data of input variables for a category number of the model and deletion processing is performed after the addition processing has been completed, assuming that the value of an input variable to be deleted, which is one of input variables that have been normalized and have been input, is a fixed value.
5. A diagnosis method for a plant in which a correlation of an input variable obtained from a monitoring target is modeled, data that has been input to a resulting model is classified into a plurality of categories, and a plant fault or an omen of a sign thereof is determined to have been detected when a category not belonging to any of classified normal categories is generated, wherein an integrated model is created by modeling a correlation of a first input variable, retaining a resulting model as an element model, and classifying data that has been input into a plurality of categories according to the correlation of the first input variable and also by modeling a
correlation of an added second input variable and classifying data that has been input into a plurality of categories, the integrated model being based on the element model and the modeled second input variable.
6. The diagnosis method for a plant according to Claim 5, wherein the integrated model includes a new category that is determined by a combination of a category of the element model and a category of the modeled second input variable.
7. A diagnosis apparatus for a plant that detects a fault in a monitoring target from a control signal obtained from a monitoring and control system that controls the monitoring target and from a measured signal obtained from a process computer that receives a process signal of the monitoring target, wherein the diagnosis comprising: an element model part that creates an element model by modeling a correlation of a signal that has been input, classifies data that has been input to the element model into a plurality of categories according to the correlation of the input signal, and determines that a fault has been detected in the plant when a category not belonging to classified normal categories is generated; and an integrated model part, having an integrated model including at least one element model, that classifies
data that has been input to the integrated model into a plurality of categories according to the correlation of the input signal and determines that a fault has been detected in the plant when a category not belonging to classified normal categories is generated; the diagnosis apparatus creates a diagnosis model, for the element model to which the input signal has been added according to modification of the monitoring target, by using only an added input signal, and a model obtained by creating a new category number from a category number obtained from an existing element model and a category number obtained from the added diagnosis model is stored in the integrated model part as an integrated model.
8. The diagnosis apparatus for a plant according to Claim 7, wherein
when an input variable is deleted from an existing model, the model is reviewed by using time-series data of input variables for a category number of the model, assuming that a value of an input variable to be deleted, which is one of input variables that have been normalized and have been input, is a fixed value.