JP2019016039A - Method for diagnosing abnormal state of process and abnormal state diagnosis apparatus - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a process abnormal state diagnosis method and an abnormal state diagnosis device capable of estimating the cause of an abnormal state generated in a process such as a manufacturing process. A method for diagnosing an abnormal state of a process uses a plurality of submodels for predicting the state of the process to calculate a deviation index from the normal state of the process, and a deviation index composed of the deviation index calculated for each submodel. Based on the pattern set, the cause of the abnormal state generated in the process is estimated, and in the deviation index pattern set, the pattern of the abnormality shown when there is an abnormality in each of the related variables as the explanatory variable of the submodel is shown. It includes a base pattern creation step of creating as a plurality of base patterns and an abnormality cause estimation step of estimating the cause of an abnormal state generated in a process based on the base pattern. [Selection diagram] Fig. 2

Description

  The present invention relates to an abnormal state diagnosis method and an abnormal state diagnosis apparatus for a process such as a manufacturing process.

  As a method for diagnosing an abnormal state of a process such as a manufacturing process, a power generation process, and a transfer process, there are a model-based approach and a database approach. The model-based approach is an approach for constructing a model that expresses a physical or chemical phenomenon in a process by a mathematical expression and diagnosing an abnormal state of the process using the constructed model. On the other hand, the database approach is an approach in which a statistical analysis model is constructed from operation data obtained in the process, and an abnormal state of the process is diagnosed using the constructed model.

  In a manufacturing process such as a steel process, products of various types and sizes are manufactured on a single manufacturing line, and there are an infinite number of operation patterns. Also, in a manufacturing process such as a blast furnace, natural products such as iron ore and coke are used as raw materials, and thus the manufacturing process varies greatly. For this reason, when diagnosing an abnormal state of a manufacturing process such as a steel process, there is a limit to an approach based only on a model-based approach.

  As a database approach, a diagnosis method for determining the similarity of the operation data at the time of the occurrence of past abnormalities to a database and judging the similarity with the current operation data, or conversely, normal operation data is converted to a database and the difference from the current operation data. There is a diagnostic method for determining the above. However, in manufacturing processes such as the steel process, there are many cases where unprecedented problems occur in the past, especially when there are many facilities used for manufacturing and there are many aging facilities, especially in Japan. . For this reason, in the former diagnostic method based on past trouble cases, there is a limit to the prediction of an abnormal state.

  On the other hand, the latter diagnostic methods are described in Patent Documents 1 and 2. Specifically, Patent Documents 1 and 2 describe a method for predicting or detecting an abnormal state of a manufacturing process based on a prediction based on a model created using normal operation data.

International Publication No. 2013/011745 Japanese Patent No. 4922265

  However, the methods described in Patent Documents 1 and 2 are limited to predicting or detecting an abnormal state of the manufacturing process and cannot estimate the cause of the abnormal state occurring in the manufacturing process. For this reason, provision of the technique which can estimate the cause of the abnormal state which generate | occur | produced in the manufacturing process was anticipated.

  The present invention has been made in view of the above problems, and an object thereof is to provide an abnormal state diagnosis method and an abnormal state diagnosis device for a process capable of estimating the cause of an abnormal state occurring in a process such as a manufacturing process. There is.

  In order to solve the above-described problems and achieve the object, a process abnormal state diagnosis method according to the present invention uses a plurality of submodels for predicting a process state to obtain a deviation index from the normal state of the process. An abnormal condition diagnosis method for a process for estimating a cause of an abnormal condition that has occurred in the process based on a deviation index pattern set comprising deviation indices calculated for each of the submodels, wherein the deviation index pattern set includes: A base pattern creating step for creating a plurality of base patterns of abnormal patterns shown when there is an abnormality for each variable related as an explanatory variable of the sub model, and the process occurred based on the base pattern And an abnormal cause estimating step for estimating the cause of the abnormal state.

  Further, the process abnormal condition diagnosis method according to the present invention is the above-described invention, wherein a plurality of artificial departure index pattern sets are created by combining a plurality of types of the base patterns created in the base pattern creation step, and the artificial departure index pattern A base pattern learning step for learning the type of the base pattern included in the set, and the base pattern included in the deviation index pattern set obtained during actual operation based on a learning result in the base pattern learning step A base pattern estimating step for estimating the type of the abnormal pattern, and the step of estimating the abnormal cause estimates the cause of the abnormal state generated in the process based on the type of the base pattern.

  In the process abnormal state diagnosis method according to the present invention, in the above invention, the base pattern generation step may be performed on a reference data set including normal state data of the process necessary for the calculation of the submodel. A perturbation data set is created by adding one minute anomaly to each of the variables related as explanatory variables of the submodel, and the basis pattern is obtained by giving the perturbation data set as an input of the submodel. It is characterized by creating.

  In the method for diagnosing an abnormal state of a process according to the present invention, in the above invention, the base pattern learning step selects a plurality of the base patterns created in the base pattern creation step, and appropriately selects each base pattern selected. Creating a plurality of artificial departure index pattern sets by adding and adding various weights, using the artificial departure index pattern set as an input, and performing the machine learning using the type or weight of the selected base pattern as an output It is characterized by.

  In order to solve the above-described problems and achieve the object, an abnormal state diagnosis apparatus for a process according to the present invention uses a plurality of submodels for predicting the state of a process to provide a deviation index from the normal state of the process. An abnormal condition diagnosis apparatus for a process for estimating a cause of an abnormal condition that has occurred in the process based on a deviation index pattern set comprising deviation indices calculated for each of the submodels, wherein the deviation index pattern set includes: The base pattern creating means for creating an abnormality pattern as a plurality of base patterns for each of the variables related as the explanatory variables of the sub-model, and generated in the process based on the base pattern An abnormality cause estimating means for estimating the cause of the abnormal condition.

  Further, the process abnormal condition diagnosis apparatus according to the present invention, in the above invention, creates a plurality of artificial departure index pattern sets by combining a plurality of types of the base patterns created by the base pattern creation means, and the artificial departure index pattern A base pattern learning means for learning the type of the base pattern included in the set; and the base pattern included in the deviation index pattern set obtained during actual operation based on a learning result in the base pattern learning means. Base pattern estimation means for estimating the type of the abnormal pattern, and the abnormality cause estimation means estimates the cause of the abnormal state generated in the process based on the type of the base pattern.

  The process abnormal condition diagnosis apparatus according to the present invention is the above-described invention, wherein the base pattern generation unit is configured to perform a reference data set including normal state data of the process necessary for the calculation of the submodel. A perturbation data set is created by adding one minute anomaly to each of the variables related as explanatory variables of the submodel, and the basis pattern is obtained by giving the perturbation data set as an input of the submodel. It is characterized by creating.

  According to the present invention, even when the same trouble has not occurred in the past, it is possible to quickly identify and deal with the cause of an abnormal state that has occurred in a process such as a manufacturing process.

FIG. 1 is a block diagram showing a configuration of a process abnormal state diagnosis apparatus according to an embodiment of the present invention. FIG. 2 is a flowchart showing an overall flow of the abnormal state diagnosis method by the process abnormal state diagnosis apparatus according to the embodiment of the present invention. FIG. 3 is a diagram illustrating an example of a deviation index pattern set created by the process abnormal state diagnosis device according to the embodiment of the present invention. FIG. 4 is a diagram illustrating an example of a base pattern created by the process abnormal state diagnosis apparatus according to the embodiment of the present invention. FIG. 5 is a diagram illustrating an example of a result of back-estimating a base pattern from a deviation index pattern set in the process abnormal state diagnosis device according to the embodiment of the present invention.

  Hereinafter, an abnormal state diagnosis method and an abnormal state diagnosis apparatus according to an embodiment of the present invention will be described with reference to the drawings. In addition, this invention is not limited to the following embodiment. In addition, constituent elements in the following embodiments include those that can be easily replaced by those skilled in the art or those that are substantially the same. In the following description, “A and / or B” means “at least one of A and B”.

[Abnormal condition diagnosis device]
The abnormal state diagnosis apparatus 1 is an apparatus for diagnosing abnormal states of various processes such as a manufacturing process of a manufacturing facility such as a steel facility, a power generation process of a power generation facility, a transfer process of a transfer facility, and the like, as shown in FIG. The input unit 10, the output unit 20, the external device 30, the storage unit 40, the model definition unit 50, the learning unit 60, and the control unit 70 are provided as main components. In the following description, an example in which the abnormal state diagnosis apparatus 1 is applied to a manufacturing process such as a steel process will be described.

  The input unit 10 is a device that receives actual operation data to be diagnosed for performing prediction and cause estimation based on a sub model, which will be described later, via an information / control network. The input unit 10 inputs actual operation data to be diagnosed received from, for example, a process computer (not shown) to the control unit 70.

  The output unit 20 is configured by an output device such as a display device or a printing device, and outputs various processing information of the control unit 70.

  The external device 30 is connected to the model definition unit 50 and the control unit 70 in a form capable of information communication via an electric communication line. The external device 30 includes an operation database (hereinafter referred to as “operation DB”) 31. In the operation DB 31, actual values of a plurality of types of variables acquired during the past operation of the manufacturing process, that is, time series data of a plurality of types of variables (hereinafter referred to as “variable data”), are obtained during normal operation. A plurality of types of variable data are stored in a form that can be read through the telecommunication line.

  The storage unit 40 is configured by a storage device such as a hard disk device, and is connected to the model definition unit 50, the learning unit 60, and the control unit 70. The storage unit 40 includes a sub model database (hereinafter referred to as “sub model DB”) 41, a base pattern database (hereinafter referred to as “base pattern DB”) 42, and a learning result database (hereinafter referred to as “learning result DB”) 43. Is remembered.

  The sub model DB 41 stores, as a sub model, a mathematical formula for calculating a time series predicted value indicating a manufacturing process and / or a state of a product being manufactured in the manufacturing process. The base pattern DB 42 stores a base pattern for each submodel. The learning result DB 43 stores the learning result of the base pattern.

  The model definition unit 50 functions as a submodel creation unit 51 and a base pattern creation unit 52 when the arithmetic processing unit executes a computer program. The functions of these parts will be described later.

  The learning unit 60 functions as the base pattern learning unit 61 when the arithmetic processing unit executes a computer program. The function of the base pattern learning unit 61 will be described later.

  The control unit 70 is configured by an arithmetic processing device such as a CPU (Central Processing Unit) and controls the operation of the entire abnormal state diagnosis device 1. The control unit 70 functions as a deviation index calculation unit 71, an abnormality detection unit 72, a base pattern estimation unit 73, and an abnormality cause estimation unit 74 when the arithmetic processing unit executes a computer program. The functions of these parts will be described later.

(Sub model)
In the present embodiment, the sub-model indicates, for example, the relationship between the state of the material before manufacturing, the setting of equipment before manufacturing, the state of equipment during manufacturing, the state of products during and / or after manufacturing. Means a mathematical expression. As the sub model, for example, a mathematical expression as a forward model for predicting the state of a product during and / or after manufacture from the state of the material before manufacture, the setting state of the equipment before manufacture, the state of the equipment under manufacture, etc. In addition, as an inverse model that reversely estimates whether the setting of equipment before production was appropriate from the state of materials before production, the state of equipment during production, the state of products during and / or after production, etc. There are various mathematical models that can be estimated with each other. Moreover, the model which estimates the state quantity of the electric power generation equipment and conveyance equipment contained in a manufacturing process from various sensors, another state quantity, and a setting value may be sufficient. In this way, by constructing a plurality of types of submodels, it is easier to detect an abnormal state early and to estimate the cause than to construct a single model for the entire manufacturing process.

  In the manufacturing process, various models have been built in order to create products with the quality and dimensions as intended, and the state of the manufacturing process and the state of the product being manufactured are predicted. An existing model may be used as a sub model. Further, when there are a shortage of submodels, a new submodel can be added by statistical processing. For example, a regression equation can be obtained using a plurality of variables other than itself acquired during normal operation of the manufacturing process and used as a sub model. Each submodel is given a reliability (a value that increases as the prediction error decreases) according to the prediction error of the submodel during a predetermined evaluation period. If the calculated reliability is low, it is desirable to review the configuration of the submodel. However, since it is not possible to review it freely at any time, for example, within a certain period, it must be used without reviewing even with low reliability. Sometimes it doesn't happen.

(Deviation index)
In the present embodiment, the deviation index means, for example, a difference value or a ratio between a predicted value calculated from a sub model and an actual value of a manufacturing process corresponding to the predicted value, or a value calculated based on these values. ing. The departure index is more preferably a value calculated by combining the above-described reliability. As a combination method in this case, for example, “deviation index with reliability taken into account” = “deviation index without taking reliability into consideration × reliability” can be considered. The deviation index is a value at the timing to be monitored, whereas the reliability is a value evaluated in a period before the timing to be monitored, and the timings of both are different.

[Abnormal condition diagnosis method]
Hereinafter, an abnormal state diagnosis method using the above-described abnormal state diagnosis apparatus 1 will be described with reference to FIGS. The abnormal state diagnosis method according to the present embodiment creates M submodels that predict the state of a process from actual values of a plurality of types of variables obtained during normal operation, and processes the process based on the prediction error of the submodel. The deviation index from the normal state is calculated, and the cause of the abnormal state generated in the process is estimated based on the deviation index pattern set composed of the M deviation indices calculated for each sub model. Here, it is not always necessary to create a new submodel. If there is an existing model, the existing model can be used.

  Specifically, as shown in FIG. 2, the abnormal state diagnosis method performs a reading step, a deviation index calculating step, an abnormality detecting step, a base pattern estimating step, and an abnormality cause estimating step. In the abnormal state diagnosis method, a sub-model creation step, a base pattern creation step, and a base pattern learning step are performed as necessary. In the following description, an example in which the abnormal state diagnosis method is applied to a manufacturing process such as a steel process will be described.

<Reading step>
In the reading step, the deviation index calculating unit 71 reads a plurality of types of variable data acquired from the manufacturing process at the processing target time from the operation DB 31 (step S1).

<Deviation index calculation step>
Subsequently, in the deviation index calculation step, the deviation index calculation unit 71 uses the plurality of types of variable data read in the reading step to change the manufacturing state of the manufacturing process at the processing target time to the manufacturing state of the manufacturing process at the time of normal operation. A value indicating how much is different is calculated for each sub-model as a deviation index (step S2).

  Specifically, the deviation index calculation unit 71 first acquires the data of the submodel from the submodel DB 41, and substitutes the variable data read from the operation DB 31 into the corresponding submodel, thereby processing time for each variable. The predicted value at is calculated. Next, the deviation index calculation unit 71 normalizes the data of the actual values and the predicted values of a plurality of types of variables in order to normalize differences in absolute quantities and units between variables. Next, the deviation index calculation unit 71 sets the difference value between the normalized predicted value of the variable at the processing target time and the normalized actual value as a deviation index from the normal state of the manufacturing process for each sub model. calculate.

  FIG. 3 shows the size of the deviation index for each sub model when an abnormal condition occurs in the manufacturing process. The vertical axis of the graph indicates the size of the deviation index, and the horizontal axis indicates the number of the sub model. Yes. In FIG. 1 to No. An example using 56 submodels up to 56 is shown. Further, in the present embodiment, as shown in the figure, the size of the deviation index for each of the plurality of submodels is collectively defined as a “deviation index pattern set”.

<Sub model creation step>
Here, in the present embodiment, the submodel creation step is performed before the timing at which the deviation index calculation step is performed. In the sub-model creation step, the sub-model creation unit 51 acquires a plurality of types of variable data obtained during normal operation from the operation DB 31, and M sub-models for predicting the manufacturing state of the manufacturing process from the variable data (this 56 in the embodiment). Then, the sub model creation unit 51 stores the created M sub models in the sub model DB 41. Further, as described above, the sub model is not necessarily created by the model creation unit 51. If there is an existing model, the model creation unit 51 may be stored in advance in the sub model DB 41 without causing the model creation unit 51 to function. Is possible.

  In the present embodiment, for example, as shown in the following formula (1), it is assumed that a sub model is configured by a regression model obtained by modeling the correlation between 56 variables. In the following formula (1), “XOO” is a variable corresponding to the sub model, and “AOO” is a weight of each variable.

<Abnormality detection step>
Subsequently, in the abnormality detection step, the abnormality detection unit 72 detects an abnormal state of the manufacturing process based on the deviation index calculated in the deviation index calculation step (step S3). The abnormality detection unit 72 detects that the manufacturing process is in an abnormal state when, for example, the deviation index calculated in the deviation index calculation step exceeds a predetermined threshold.

  For example, when considering the reliability of the submodel for the deviation index, the abnormality detection unit 72 weights each deviation index according to the magnitude of the reliability of the submodel, and further, the deviation index after weighting. When the total value exceeds a predetermined threshold value, it may be detected that the manufacturing process is in an abnormal state.

<Base pattern estimation step>
Subsequently, in the base pattern estimation step, the base pattern estimation unit 73 estimates a base pattern from the departure index pattern set calculated in the departure index calculation step (step S4).

  Here, in the present embodiment, the base pattern creation step and the base pattern learning step are performed before the base pattern estimation step is performed. Below, the content of these steps is demonstrated first.

<Base pattern creation step>
In the base pattern creation step, the base pattern creation unit 52 creates a base pattern. FIG. 4 shows an example of the base pattern, in which the vertical axis of the graph indicates the size of the departure index, and the horizontal axis indicates the submodel number. In the figure, only the base pattern No. 56 on the horizontal axis of the graph is shown only in the bottom “base pattern No. 56”. 01, No. 02, no. 27, no. In FIG. 28, illustration is omitted (the same applies to FIG. 5 described later). The base pattern is an abnormality pattern indicated when there is an abnormality in each of the variables related as the explanatory variables of the submodel. The base pattern creation unit 52 creates M base patterns for each explanatory variable serving as an input signal.

  Specifically, the base pattern creation unit 52 acquires normal state data of the manufacturing process necessary for calculation of M (56 in the present embodiment) submodels from the operation DB 31 and uses this as a reference data set. To do. Next, the base pattern creation unit 52 creates M perturbation data sets by adding one minute abnormality to each of the variables related as sub-model explanatory variables to the reference data set.

  For example, the base pattern creation unit 52 uses the sub model No. By adding minute abnormalities one by one to the explanatory variables other than the explanatory variable “X1” for the items of the reference data set corresponding to the 01 prediction target (“X1” in the above formula (1)) Sub model No. Create a perturbation data set for 01. In addition, the base pattern creation unit 52 includes a sub model No. By adding minute abnormalities one by one to the explanatory variables other than the explanatory variable “X2” for the items in the reference data set corresponding to the prediction target 02 (“X2” in the above formula (1)) Sub model No. Create a perturbation data set for 02. Similarly, the base pattern creation unit 52 creates a total of M × (M−1) perturbation data sets (56 × 55 in this embodiment).

  Next, the base pattern creation unit 52 creates a total of M (56 in this embodiment) departure index pattern sets by giving the created perturbation data sets as submodel inputs. The perturbation data set created above can be classified into M types of data including minute anomalies for each explanatory variable to obtain M data sets.

  Specifically, data of a model that predicts other than “X1” obtained by adding a minute abnormality to only the explanatory variable “X1” is aggregated into one and given to models other than X1. Next, the data of the model that predicts other than “X2” in which the minute abnormality is added only to the explanatory variable “X2” is collected into one and given to the models other than X2. By repeating this, a base pattern can be obtained. Note that it is assumed that there is no submodel that predicts the value of a specific explanatory variable. In that case, there is no submodel, and it is excluded from the target of the base pattern (explanatory variables of other submodels). Only used as). These are the base patterns shown in FIG. Then, the base pattern creation unit 52 stores the created base pattern in the base pattern DB 42.

  When artificially adding a minute abnormality to the above-described reference data set, a certain value of a minute abnormality may be added to each item corresponding to the prediction target of the submodel. In the base pattern, a minute abnormality is added so that the maximum value of the deviation index is 10 times the normal value, that is, the maximum value of the deviation index in each base pattern is normalized to some extent. Also good.

<Base pattern learning step>
Here, as shown in FIG. 28, the sub model no. 28 is given a minor abnormality, sub model no. Of course, the deviation index of 28 shows a high value. 13, no. The deviation index of 42 also shows a somewhat high value. That is, the sub model No. Even if 28 indicates a true abnormality, some other sub-models may show an apparently high deviation index. For this reason, in the deviation index pattern set calculated from the actual operation data, it is not possible to determine that all the deviation indices are high.

  On the other hand, the deviation index pattern set calculated from the actual operation data can be considered as a linear combination of the base patterns as shown in the following equation (2).

  Therefore, if the base pattern included in the deviation index pattern can be back-estimated from the deviation index pattern calculated from the actual operation data, it is possible to identify a true abnormality. However, in reverse estimation, it is usually difficult to uniquely determine the above base pattern.

  Therefore, in the base pattern learning step, the base pattern learning unit 61 creates a plurality of artificial departure index pattern sets by combining a plurality of types of base patterns created in the base pattern creation step, and is included in the artificial departure index pattern set. The type of base pattern is learned by machine learning.

  Specifically, the base pattern learning unit 61 selects and acquires a plurality of base patterns created in the base pattern creation step from the base pattern DB 42. Next, the base pattern learning unit 61 creates a plurality of artificial departure index pattern sets (hereinafter referred to as “artificial departure index pattern sets”) by adding and adding appropriate weights to each selected base pattern. To do. Next, the base pattern learning unit 61 performs machine learning with the artificial departure index pattern set as an input and the type or weight of the selected base pattern as an output. Then, the base pattern learning unit 61 stores the base pattern learning result in the learning result DB 43.

  As machine learning performed in the base pattern learning step, deep learning is effective. Deep learning is a learning method that requires a multi-layered net structure, and is characterized by learning for each layer. Moreover, deep learning has achieved high results in artificial intelligence and image recognition in Go. By using such a learning method, it is possible to estimate which base pattern is included probabilistically in the deviation index pattern set.

  Here, as the learning result of the base pattern, for example, information on the probability that M (56 in the present embodiment) base patterns are included for a certain deviation index pattern set. In this case, the learning result of the base pattern indicates that “probability of including base pattern No. 01 is XX%” and “probability of including base pattern No. 02 is XX%” for a certain deviation index pattern set. ,..., “The probability that base pattern No. 56 is included is OO%” is included.

  Hereinafter, the specific description of the base pattern estimation step will be returned. In the base pattern estimation step, a base pattern included in the deviation index pattern set is estimated based on the base pattern learning result in the base pattern learning step. Specifically, the base pattern estimation unit 73 acquires a base pattern learning result from the learning result DB 43, and calculates based on the learning result in a deviation index pattern set obtained during actual operation, that is, a deviation index calculation step. The type of base pattern included in the set deviation index pattern set is estimated.

  FIG. 5 shows an example of the result of estimating the types of base patterns included in the deviation index pattern set in the base pattern estimation step. As shown in the figure, the base pattern estimation unit 73 includes, for example, two base patterns (base patterns No. 30 and No. 24 in the figure) in the deviation index pattern set calculated in the deviation index calculation step. Presumed to be included. Note that the base pattern estimation unit 73 estimates, for example, a predetermined number (two in the figure) of M base patterns that have a possibility of being included in the deviation index pattern set of 80% or more. Enumerate as

<Abnormality cause estimation step>
Finally, in the abnormality cause estimation step, the abnormality cause estimation unit 74 estimates the cause of the abnormal state from the base pattern (step S5). Specifically, the abnormality cause estimation unit 74 estimates the cause of the abnormal state that has occurred in the manufacturing process, based on the type of the base pattern estimated in the base pattern estimation step and the combination of the base patterns.

  For example, as shown in FIG. 5, the abnormality cause estimation unit 74 includes a base pattern No. in the deviation index pattern set in the base pattern estimation step. 30, no. 24 is estimated to be included, the base pattern No. 30, no. The cause of the abnormal state corresponding to the combination of 24 (for example, deterioration of machine accuracy of a certain facility, etc.) is estimated.

  Note that the correspondence relationship between the type of base pattern or a combination of a plurality of base patterns and the cause of the abnormal state is obtained, for example, experimentally in advance and stored as a table. Then, the abnormality cause estimation unit 74 estimates the cause of the abnormal state corresponding to the type of the base pattern or the combination of the base patterns by referring to this table.

  According to the abnormal condition diagnosis method according to the present embodiment that performs the processing as described above, even if the same trouble has not occurred in the past, the cause of the abnormal condition that occurred in the process such as the manufacturing process can be quickly identified. Can be dealt with.

  As mentioned above, although the abnormal state diagnosis method and the abnormal state diagnosis device of the process according to the present invention have been specifically described in the form for carrying out the invention, the gist of the present invention is not limited to these descriptions. It should be construed broadly based on the claims. Needless to say, various changes and modifications based on these descriptions are also included in the spirit of the present invention.

  For example, the abnormal state diagnosis apparatus 1 can also execute the abnormal state diagnosis method using the submodel, the base pattern, and the learning result of the base pattern that are created in advance and stored in each DB of the storage unit 40. In this case, the abnormal condition diagnosis apparatus 1 may not include the submodel creation unit 51, the base pattern creation unit 52, and the base pattern learning unit 61 illustrated in FIG.

  Moreover, in the abnormal condition diagnosis apparatus 1, the example in which the submodel is configured by the regression model as shown in the above formula (1) has been shown, but the specific configuration of the submodel is not particularly limited, and is configured by, for example, the physical model May be.

  In the above-described embodiment, an example in which the abnormal state diagnosis device 1 and the abnormal state diagnosis method are applied to a manufacturing process such as a steel process has been described. However, the abnormal state diagnosis device 1 and the abnormal state are applied to a power generation process, a transfer process, and the like. It is also possible to apply a diagnostic method.

DESCRIPTION OF SYMBOLS 1 Abnormal state diagnostic apparatus 10 Input part 20 Output part 30 External apparatus 31 Operation database (operation DB)
40 storage unit 41 submodel database (submodel DB)
42 Base pattern database (Base pattern DB)
43 Learning result database (Learning result DB)
DESCRIPTION OF SYMBOLS 50 Model definition part 51 Sub model creation part 52 Base pattern creation part 60 Learning part 61 Base pattern learning part 70 Control part 71 Deviation index calculation part 72 Abnormality detection part 73 Base pattern estimation part 74 Abnormal cause estimation part

Claims (7)

A deviation index from the normal state of the process is calculated using a plurality of submodels for predicting the state of the process, and is generated in the process based on a deviation index pattern set composed of the deviation indices calculated for each submodel. An abnormal condition diagnosis method for a process for estimating the cause of an abnormal condition,
In the deviation index pattern set, for a variable related as an explanatory variable of the submodel, a base pattern creating step for creating a plurality of base patterns as abnormal patterns that are respectively shown when there is an abnormality,
An abnormal cause estimation step for estimating a cause of an abnormal state that has occurred in the process based on the base pattern;
A method for diagnosing an abnormal state of a process, comprising: A base pattern learning step of creating a plurality of artificial departure index pattern sets by combining a plurality of types of the base patterns created in the base pattern creation step, and learning the types of the base patterns included in the artificial departure index pattern set When,
Based on the learning result in the base pattern learning step, a base pattern estimation step for estimating the type of the base pattern included in the deviation index pattern set obtained during actual operation;
Further including
2. The process abnormal state diagnosis method according to claim 1, wherein the abnormality cause estimation step estimates a cause of an abnormal state generated in the process based on a type of the base pattern.   The base pattern creation step includes one minute abnormality for each variable related as an explanatory variable of the sub model with respect to a reference data set including normal state data of the process necessary for the calculation of the sub model. The process according to claim 1 or 2, wherein the base pattern is created by adding a perturbation data set by adding the perturbation data set as input of the submodel. Abnormal state diagnosis method.   The base pattern learning step selects a plurality of the base patterns created in the base pattern creation step, and adds a plurality of the artificial deviation index pattern sets by adding and adding appropriate weights to the selected base patterns. 4. The machine learning is performed according to claim 1, wherein machine learning is performed using the artificial deviation index pattern set as an input and the type or weight of the selected base pattern as an output. 5. Process abnormal condition diagnosis method. A deviation index from the normal state of the process is calculated using a plurality of submodels for predicting the state of the process, and is generated in the process based on a deviation index pattern set composed of the deviation indices calculated for each submodel. An abnormal condition diagnosis device for a process for estimating the cause of an abnormal condition,
In the deviation index pattern set, for a variable related as an explanatory variable of the submodel, a base pattern creating unit that creates a plurality of base patterns as an abnormality pattern when there is an abnormality respectively;
An abnormal cause estimating means for estimating a cause of an abnormal state generated in the process based on the base pattern;
An apparatus for diagnosing an abnormal state of a process, comprising: A base pattern learning unit that creates a plurality of artificial departure index pattern sets by combining a plurality of types of the base patterns created by the base pattern creation unit, and learns the types of the base patterns included in the artificial departure index pattern set When,
Based on the learning result in the base pattern learning means, a base pattern estimation means for estimating the type of the base pattern included in the deviation index pattern set obtained during actual operation;
Further comprising
6. The process abnormal condition diagnosis device according to claim 5, wherein the abnormality cause estimation means estimates a cause of an abnormal condition generated in the process based on a type of the base pattern.   The base pattern creation means has one micro-abnormality for each of the variables related as explanatory variables of the submodel with respect to a reference data set consisting of normal state data of the process necessary for the calculation of the submodel. 7. The process according to claim 5, wherein the base pattern is created by adding a perturbation data set by adding each of them and providing the perturbation data set as an input of the submodel. Abnormal state diagnosis device.

Images