JP5949032B2 - Pre-processing method and abnormality diagnosis device - Google Patents

Description

The present invention relates to a preprocessing method and an abnormality diagnosis apparatus that are performed as preprocessing for abnormality diagnosis using the MT method.

  A system having a plurality of devices such as a plant is operated by setting conditions for the plurality of devices. Further, in such a system, the state is measured by a plurality of sensors for grasping the operation state.

  Such a system abnormality can be detected by using the plurality of setting values, the plurality of measurement values, and the like. For example, there is a method of diagnosing an abnormality when the measured value of a sensor corresponding to each device exceeds the upper and lower limit values obtained for each measured value of each sensor. However, when setting the upper and lower limit values for each measurement value of each sensor, the number of upper and lower limit values increases as the number of devices used increases, and the management of abnormality detection values and abnormality detection processing become complicated. There's a problem.

  On the other hand, abnormality diagnosis using the MT method (Mahalanobis Taguchi method) has also been proposed (see, for example, Patent Document 1). When the MT method is used, an abnormality can be easily diagnosed by comparing a value (setting value or measurement value) specifying the current state with a value of a unit space as a reference.

  In the MT method, as a method for improving the detection accuracy of abnormality, it is essentially unnecessary for diagnosis of abnormality, and it is considered to exclude values of items (parameters) that lower the diagnosis accuracy. For example, the value of an item (cumulative value item) having a characteristic that the value continues to increase with the passage of time, such as a counter or a timer, is essentially an unnecessary item for abnormality diagnosis. In determining the items to be excluded, a method is generally used in which a contribution (SN ratio) in abnormality diagnosis of each item is obtained and expressed as an orthogonal table, and an item with a low contribution is determined to be unnecessary.

  However, the method using this orthogonal table includes a specific error. Therefore, the current situation is that the operator to be diagnosed selects items to be excluded with reference to the orthogonal table. At this time, it depends on the experience and knowledge of the operator, and if the item is selected incorrectly, there is a problem that the detection accuracy of the abnormality is poor. In addition, trial and error may be repeated to select items to be excluded that are not used for abnormality diagnosis, and there is a problem that it takes time to ensure the accuracy of abnormality diagnosis.

JP 2009-76772 A

  As described above, in the case of abnormality diagnosis using the conventional MT method, it is difficult to improve abnormality detection accuracy.

In view of the above problems, an object of the present invention is to provide a preprocessing method and an abnormality diagnosis device that improve the accuracy of abnormality detection in abnormality diagnosis using the MT method.

In order to achieve the above object, the present invention provides a preprocessing method for selecting candidate parameters to be excluded from an abnormality diagnosis from a plurality of parameters as preprocessing for abnormality diagnosis by the MT method. The method includes a step of generating a histogram using values, and a step of setting a parameter that does not approximate a predetermined normal distribution in the generated histogram as a candidate for an exclusion parameter.
In addition, the abnormality diagnosis apparatus according to the present invention includes an accumulated data storage unit that accumulates values of parameters including set values and measurement values of equipment that identifies the state of a plant that is an object of abnormality diagnosis, and a predetermined that identifies Mahalanobis space. In the case of conditions, the unit space data storage unit that stores data including parameter values representing the state of the past plant as unit space data, and the accumulated data is read out from the accumulated data storage unit at the timing of preprocessing For each parameter of the accumulated data, a generation unit that generates a histogram using a value of a predetermined period, and whether or not the histogram for each parameter generated by the generation unit approximates a normal distribution For the histogram parameters that do not approximate the normal distribution, And at a predetermined timing for diagnosing abnormality, unit data having only parameters used for analysis is extracted from accumulated data and stored in a unit space data storage unit according to the determination of the determination unit. Mahalanobis in a state where the parameter values included in the unit space data read from the unit space data storage unit and the new parameter values input from the plant are excluded, and the parameters excluded by the determination unit are excluded. It is characterized by comprising a calculation unit for obtaining a distance, and a diagnosis unit for determining the presence or absence of abnormality of the plant using the Mahalanobis distance obtained by the calculation unit.

  ADVANTAGE OF THE INVENTION According to this invention, the abnormality detection precision in abnormality diagnosis using MT method can be improved.

It is a figure explaining the structure of the abnormality diagnosis apparatus which performs abnormality diagnosis using the pre-processing method of this invention. It is the orthogonal table and histogram explaining the contribution of each parameter used by abnormality diagnosis. It is a graph which shows the Mahalanobis distance before and behind excluding a parameter with the pre-processing method of this invention.

  A preprocessing method according to the present invention will be described with reference to the drawings. The pretreatment method according to the present invention is implemented as a pretreatment for abnormality diagnosis using the MT method. For example, abnormality diagnosis detects abnormality of a plant having a plurality of devices. The pre-processing method according to the present invention is a method for selecting parameter candidates to be excluded from abnormality diagnosis from parameters for a plurality of devices.

  For example, the plant targeted for abnormality diagnosis is a power plant. The power plant includes a plurality of devices (valves, pumps, etc.), and values for controlling these devices are set as set values. This set value is, for example, a pump pressure, a valve opening / closing, or the like. In addition, the power plant includes a plurality of sensors, and each sensor measures a measured value such as temperature, pressure, and power generation amount. Therefore, in the pre-processing method according to the present invention, parameter candidates to be excluded at the time of abnormality diagnosis are selected from these set values and measured values.

  In the abnormality diagnosis by the MT method, a unit space is generated in advance based on the set values and measurement values that are the basis, and the abnormality diagnosis is performed by comparing the set values and measurement values at the time of diagnosis with the values of the reference unit space. This MT method is used to diagnose whether or not the state is “same as usual”, and the Mahalanobis distance of the object to be diagnosed is determined based on the Mahalanobis space that is a unit space that represents the same state as usual. This is a method for diagnosing an abnormality when it is farther than this Mahalanobis space.

  The preprocessing method according to the present invention is executed in, for example, the abnormality diagnosis apparatus 1 shown in FIG. For example, the abnormality diagnosis apparatus 1 includes a calculation unit 13, a diagnosis unit 14, and an update unit 15 for abnormality diagnosis. In addition, the abnormality diagnosis apparatus 1 includes a generation unit 11 and a determination unit 12 in order to execute the preprocessing method according to the present invention.

  The abnormality diagnosis device 1 is, for example, an information processing device including a central processing unit (CPU) 10 and a storage device 20, and the abnormality diagnosis program P1 stored in the storage device 20 is read and executed. As shown in FIG. 1, a calculation unit 13, a diagnosis unit 14, and an update unit 15 are mounted on the CPU 10. Further, in the abnormality diagnosis apparatus 1, the generation unit 11 and the determination unit 12 are mounted on the CPU 10 as illustrated in FIG. 1 by reading and executing the preprocessing program P2 stored in the storage device 20. .

  The storage device 20 of the abnormality diagnosis device 1 stores an abnormality diagnosis program P1 and a preprocessing program P2, and has an accumulated data storage unit 21 for storing accumulated data and a unit space data storage unit 22 for storing unit space data. ing.

  Further, the abnormality diagnosis device 1 is connected to the input device 2 and the output device 3. The input device 2 is a keyboard, an operation button, or the like, and inputs an operation signal or a selection signal and transmits it to the abnormality diagnosis device 1. The output device 3 is a display or the like, and receives and outputs a diagnosis result and a processing result from the abnormality diagnosis device 1.

  The accumulated data stored in the accumulated data storage unit 21 is data for accumulating parameter values such as set values and measured values of devices that specify the state of the plant that is the target of abnormality diagnosis. For example, it is data that accumulates records including “measured values” measured in the plant in the case of the “set value” of the plant operation that is a parameter representing the state of the past plant at the “time” and the set value. .

  The unit space data stored in the unit space data storage unit 22 includes a history of parameter values representing past plant states in the case of a predetermined condition for specifying the Mahalanobis space, each average value, and standard deviation of each parameter. This is data in which the decision value set in the Mahalanobis space is associated with the inverse matrix of the correlation matrix for each parameter.

  The generation unit 11 reads stored data from the stored data storage unit 21 at the timing of preprocessing. Here, the preprocessing timing is, for example, the timing at which data for a certain period is accumulated in the accumulated data. Further, the generation unit 11 generates a histogram with a value for a predetermined period for each parameter included in the read accumulated data.

  The determination unit 12 determines whether the histogram for each parameter generated by the generation unit 11 is approximated to a normal distribution. Further, the determination unit 12 determines a parameter of a histogram that does not approximate the normal distribution as a candidate for an exclusion parameter to be excluded at the time of abnormality diagnosis and outputs it to the output device 3. After that, when a signal for selecting this exclusion parameter candidate as an exclusion parameter is input via the input device 2 via the input device 2, the determination unit 12 determines this parameter as an “exclusion parameter” and calculates the result. To the unit 13.

  The calculation unit 13 reads unit space data from the unit space data storage unit 22 at a predetermined timing for diagnosing abnormality. Here, the predetermined timing for diagnosing abnormality is, for example, a regular timing.

  In addition, the calculation unit 13 calculates the average value and standard deviation of each parameter value (variable) included in the read unit space data, the inverse matrix of the correlation matrix for each parameter value, and a predetermined timing for diagnosing abnormality. The Mahalanobis distance is obtained by a general Mahalanobis distance calculation method using the new parameter value input from, and the obtained value is output to the diagnosis unit 14. Specifically, the calculation unit 13 first standardizes data by using the value of each parameter, the average value and the standard deviation obtained for each parameter, and obtains a normalized value for each parameter. Subsequently, the calculation unit 13 obtains the Mahalanobis distance using the normalized values of the parameters and the obtained inverse matrix.

  Here, when any parameter is determined as an “exclusion parameter” in the determination unit 12, the calculation unit 13 uses the value of each parameter excluding this “exclusion parameter” to calculate the Mahalanobis distance. Ask.

  When the diagnosis unit 14 inputs the Mahalanobis distance from the calculation unit 13, the diagnosis unit 14 reads the unit space data from the unit space data storage unit 22, and extracts a determination value included in the read unit space data. Further, the diagnosis unit 14 compares the Mahalanobis distance input from the calculation unit 13 with a determination value extracted from the unit space data, determines whether the operation state of the plant is normal or abnormal, and outputs a determination result. Output to device 3.

  When the value of each parameter is newly input from the plant, the update unit 15 generates a new record including the input value of each parameter and the input time of the value, and adds the generated record to the accumulated data storage unit 21. Update stored data to be stored.

  FIG. 2A is an orthogonal table showing the contribution of each parameter when the Mahalanobis distance is obtained. For example, when there are 133 parameters, the Mahalanobis distance is obtained using all the parameters of 133, and the contribution when obtaining the Mahalanobis distance of each parameter is obtained. It is a graph shown by (a). The S / N ratio when obtaining the Mahalanobis distance is the contribution.

  When parameters are selected using an orthogonal table as shown in FIG. 2A, parameters with low contribution are excluded. However, for example, even if the parameter is a higher contribution level (for example, fifth place), there are actually cases where the accuracy of abnormality diagnosis is lowered. Even if the contribution level is a low-order parameter, in reality, when the abnormality is diagnosed except for this parameter, the accuracy of the abnormality diagnosis may be lowered.

  FIG. 2B is a histogram generated by the generation unit 11 with the data of a certain period for the parameter ranked fifth in FIG. 2A. In FIG. 2B, the horizontal axis is the parameter value, and the vertical axis is the number of times this value is obtained. Here, the MT method is a method based on the assumption that parameters of a normal distribution are used. Therefore, when the parameter histogram is represented as shown in FIG. 2B, it is clear that this parameter is not a normal distribution. Therefore, the determination unit 12 determines this parameter as a parameter that is preferably excluded in abnormality diagnosis using the MT method, that is, a candidate for an exclusion parameter.

  FIG. 3A is a graph showing changes in Mahalanobis distance obtained using all parameters without excluding the input parameter values. FIG. 3B is a graph showing a change in the Mahalanobis distance obtained by using the histogram of each parameter generated by the generation unit 11 and excluding values of parameters whose histogram is not a normal distribution. In FIG. 3, the horizontal axis represents time, and the vertical axis represents the Mahalanobis distance obtained by using the parameter value at each time.

  In the case of the graph in FIG. 3A, the Mahalanobis distance tends to increase with the passage of time, although the Mahalanobis distance is increased several times. Therefore, as shown in FIG. 3A, when all the parameters are used, it is understood that after a predetermined time has passed, all are diagnosed as abnormal, and the accuracy of abnormality diagnosis is low.

  On the other hand, in the case of the graph of FIG. 3B, the Mahalanobis distance does not increase with time, the period during which the Mahalanobis distance is increased is limited to one time, and the others are stable. . Therefore, it can be seen that an abnormality is diagnosed only during a period in which an abnormality actually occurs and the accuracy of the abnormality diagnosis is high.

  As described above, according to the preprocessing method of the present invention, a histogram is generated for each parameter, and parameters that do not approximate the normal distribution can be determined as candidates for exclusion. Further, when using the above-described preprocessing method, it is possible to determine candidates for exclusion parameters without depending on the experience and knowledge of the operator. Therefore, when performing an abnormality diagnosis after pre-processing by the pre-processing method according to the present invention, it is easy to determine the Mahalanobis distance by determining candidates to be actually excluded from the exclusion parameter candidates determined by the pre-processing. The accuracy of abnormality diagnosis can be improved.

  As mentioned above, although this invention was demonstrated in detail using embodiment, this invention is not limited to embodiment described in this specification. The scope of the present invention is determined by the description of the claims and the scope equivalent to the description of the claims.

1 ... Abnormality diagnosis device 10 ... CPU
DESCRIPTION OF SYMBOLS 11 ... Generating part 12 ... Determining part 13 ... Calculation part 14 ... Diagnosis part 15 ... Update part 20 ... Storage device 21 ... Accumulated data storage part 22 ... Unit space data storage part 2 ... Input device 3 ... Output device

Claims (2)

As a pre-process for abnormality diagnosis by MT method, a pre-processing method for selecting parameter candidates to be excluded from abnormality diagnosis from a plurality of parameters,
Generating a histogram using the value of a predetermined period for each accumulated parameter , with the timing at which the parameter for a certain period is accumulated in the accumulated data storage unit as a timing for preprocessing ;
Of the generated histogram, a parameter that does not approximate a predetermined normal distribution is set as an exclusion parameter candidate; and
A pre-processing method comprising: An accumulated data storage unit that accumulates parameter values including set values and measured values of equipment that identifies the state of the plant that is the target of abnormality diagnosis;
A unit space data storage unit that stores data including values of parameters representing the state of the past plant in the case of a predetermined condition for specifying the Mahalanobis space as unit space data;
The timing at which the parameters for a certain period are accumulated in the accumulated data storage unit is used as the preprocessing timing, the accumulated data is read from the accumulated data storage unit, and the value of the predetermined period is used for each parameter of the accumulated data read. A generator for generating a histogram,
A determination unit that determines whether or not a histogram for each parameter generated by the generation unit approximates a normal distribution, and determines a parameter of a histogram that does not approximate a normal distribution as an excluded parameter to be excluded at the time of abnormality diagnosis When,
At a predetermined timing for diagnosing an abnormality, unit data having only parameters used for analysis is extracted from stored data, stored in the unit space data storage unit, and read from the unit space data storage unit according to the determination of the determination unit The calculation unit for obtaining the Mahalanobis distance in a state where the parameter value excluded from the determination unit is used, using the value of each parameter included in the unit space data and the value of the new parameter input from the plant,
A diagnostic unit for determining the presence or absence of abnormality of the plant using the Mahalanobis distance obtained by the calculation unit;
An abnormality diagnosis apparatus comprising:

Images