JP5132746B2 - Operational abnormality detection device and operational abnormality detection method - Google Patents

Description

  The present invention relates to an operation abnormality detection device and an operation abnormality detection method for detecting an abnormality that occurs when equipment is operating, and in particular, collects various physical quantities generated during operation of the equipment, The present invention relates to an operation abnormality detection device and an operation abnormality detection method that can detect a sign of occurrence of an abnormality related to equipment by analyzing the above.

  For example, facility equipment provided in a power generation / transformation facility installed and operated by an electric power company is required to have high availability because it is necessary to supply high-quality power stably and continuously. Such equipment includes various steam turbines, gas turbines and generators for thermal power plants, water turbines and generators for hydropower plants, transformers installed at power stations, and various switches. Things are included.

  In order to guarantee the above-mentioned high availability for these equipment, a wide range of maintenance management is conducted to collect various physical quantities generated during operation of each equipment and analyze them to detect abnormalities in equipment early. In addition, various abnormality detection methods have been studied aiming at detection at a sign of abnormality and detection of abnormality with higher accuracy. The physical quantity includes, for example, sound, vibration, temperature of each part, etc. generated by the operating equipment.

  In order to detect abnormalities in equipment, sensory inspections by the five senses of experts have been carried out classically, but in recent years, with the progress of statistical data analysis techniques, physical quantities such as sound are collected and used. Various methods for realizing early detection of an abnormality by analysis have been proposed.

  Specifically, the physical quantity when the equipment is operating normally and the physical quantity at the time of abnormality are acquired in advance as data, and an analysis mechanism that employs a pattern matching technique using a neural network, support vector machine, etc. ( Specifically, a threshold value for determining normal data and abnormal data is generated by inputting to numerical processing software. Thereafter, it is possible to determine whether or not an abnormality has occurred in the equipment device by inputting the physical quantity data collected during the operation of the equipment device into the analysis mechanism in a timely manner and comparing it with a threshold value.

  An anomaly detection method incorporating such a pattern matching technique is disclosed in, for example, Patent Document 1 and Patent Document 2. Patent document 1 relates to a method and apparatus for detecting an abnormal sign from normal operation data of a mechanical device, and relates to a class of sensor information in a plurality of normal states measured by a plurality of sensors 11 for a detected apparatus in a normal operation state. A configuration is provided in which a combination of exceptional sensor information calculated by a support vector machine is extracted by an exceptional state extraction device 13 and an abnormal sign is detected from a combination pattern of exceptional sensor information. Thereby, it is said that there is an effect that an exceptional situation different from normal can be found only from data during normal operation.

  Further, Patent Document 2 relates to a method for early detection of an abnormality in a plant or facility, and (1) pays attention to temporal data behavior and divides the trajectory into clusters over time. (2) Divided clusters A group is modeled in a partial space, and an outlier is calculated as an abnormal candidate. (3) Utilizing learning data as a reference (comparison / reference, etc.) and grasping state transitions due to changes over time, environmental changes, maintenance (part replacement), and operating conditions. (4) Modeling is performed by sub-space methods such as regression analysis method and projection distance method (N = 0, 1, 2, ...). It is assumed that one is mixed and is modeled except this), or by the local subspace method. Furthermore, (5) integrating the outputs of multiple discriminators, such as the subspace method. Make an abnormality judgment. As a result, even if the learning data is not complete, it is allowed to mix abnormalities, and it is possible to detect abnormalities early and with high accuracy in facilities such as plants.

JP 2005-345154 A JP 2010-92355 A

As described above, depending on the configurations of Patent Document 1 and Patent Document 2, by collecting and analyzing the normal data of the equipment, as a data group deviating from the data group occupying the majority in the normal data A distinguished outlier can be detected and used to detect abnormal operation of the equipment.
However, in the configurations of Patent Literature 1 and Patent Literature 2, data that truly reflects an abnormal state (hereinafter referred to as “abnormal value”) cannot be specified and extracted from the outliers. There was a problem that it was difficult to reliably supplement the signs of abnormality.

  The present invention has been made to solve the above and other problems, and collects various physical quantities generated during operation of the equipment and analyzes them to detect a sign of occurrence of an abnormality related to the equipment. It is an object of the present invention to provide an operation abnormality detection device and an operation abnormality detection method that enable this.

In order to achieve the above object, one aspect of the present invention is an operation abnormality detection device for detecting an abnormality that occurs in an equipment device, and a characteristic of a physical phenomenon that occurs accompanying the operation of the equipment device. A device operation physical information collection unit that collects device operation physical information, which is information that quantitatively represents the information, and a plurality of the device operation physical information acquired from the device operation physical information collection unit . When outlier detection processing for detecting a certain percentage of outliers using a threshold determined as a discriminant function is performed and there is no change in the number of detected outliers, the detected outliers are excluded and the outliers are excluded. and outlier detection unit iteratively performing detection processing, suitable for the outlier detection unit outlier detection process when the number of outliers detected by the outlier detection unit detects that the increased An abnormality determination threshold value setting unit that sets the threshold value that is used as an abnormality determination threshold value that is a threshold value for detecting device operation abnormality, and the device using the abnormality determination threshold value set by the abnormality determination threshold value setting unit. An abnormality determination execution unit that performs an outlier determination process on the operation physical information, and an operation abnormality detection information when an outlier of the device operation physical information that indicates an operation abnormality of the facility device is detected by the abnormality determination execution unit. And a discrimination result processing unit for outputting the above.

  According to the operation abnormality detection device according to one aspect of the present invention, it is possible to collect various physical quantities generated during operation of the equipment and analyze them to detect a sign of occurrence of an abnormality related to the equipment. .

It is a schematic diagram of the abnormality detection system | strain by the operation | movement abnormality detection apparatus which concerns on one Embodiment of this invention. 1 is a configuration example of a computer 10 that can be used as an operation abnormality detection device 500. It is a figure which shows an example of the software configuration of the operation abnormality detection apparatus. It is a schematic diagram which shows the analysis process of the collected acoustic data. It is a schematic diagram which shows the principle which discriminate | determines an outlier from the collected data. It is a figure which shows an example of the outlier detection result table 540. It is a figure which shows an example of the abnormality detection process flow performed by the operation abnormality detection apparatus 500.

  Hereinafter, the present invention will be described in accordance with an embodiment thereof with reference to the accompanying drawings.

<Operation Abnormality Detection System Configuration>
An example of an abnormality detection system using an operation abnormality detection device according to an embodiment of the present invention is schematically shown in FIG. In the example of FIG. 1, for example, the operating state of a rotating shaft that rotatably supports a rotor driven in a turbine device installed in a thermal power plant is shown when the rotating shaft is driven to rotate. It is intended to discover abnormalities by collecting generated sounds as data.
Of course, the equipment for which the abnormality is detected is not limited to the turbine device. As described above, various types of equipment such as a generator, a water turbine of a hydroelectric power plant, a transformer installed in a power generation / substation, and various switches can be the target of abnormality detection.

  In the example of FIG. 1, specifically, in a thermal power plant, the steam turbine apparatus 100 and the generator 200 are connected by a rotating shaft 120, and the high pressure supplied from the boiler outside the figure in the steam turbine apparatus 100. The internal rotors 110 </ b> A and 110 </ b> B are rotationally driven by the steam, and the driving force rotationally drives the rotor in the generator 200 by the rotating shaft 120. The rotating shaft 120 is rotatably supported by a bearing (not shown) at an appropriate position.

  In FIG. 1, a sound collection device 300 (equipment operation physical information collection unit) that collects operation sound generated when the rotary shaft 120 rotates at a preselected position as device operation physical information is installed. The sound collection device 300 includes, for example, a microphone that converts the operation sound of the rotating shaft into an electric signal and outputs the electric signal, and a signal conversion transmission unit that converts the electric signal output from the microphone into a digital signal and transmits the digital signal to the operation abnormality detection device 500. ing.

  The sound collection device 300 and the operation abnormality detection device 500 are connected by a communication line 400. The communication line 400 can be configured by a part of a communication network that employs an appropriate communication protocol, for example, or may be configured as a dedicated communication line between the sound collection device 300 and the operation abnormality detection device 500. Further, the communication line 400 may be configured by an appropriate mode of wireless communication. Alternatively, a configuration may be adopted in which an analog electrical signal output from a microphone is transmitted as it is to the operation abnormality detection device 500 and converted into a digital signal by the operation abnormality detection device 500.

<< Operation Abnormality Detection Device 500 >>
Next, the configuration of the operation abnormality detection device 500 will be described. FIG. 2 shows a configuration example of the computer 10 that can be applied as the operation abnormality detection apparatus 500 in the present embodiment.

  The computer 10 includes a central processing unit 11 (for example, CPU (Central Processing Unit) or MPU (Micro Processing Unit), hereinafter referred to as “CPU” for simplicity), a main storage device 12 (for example, RAM (Random Access Memory) or ROM ( Read Only Memory), auxiliary storage device 13 (for example, hard disk drive (HDD)), input device 14 (for example, keyboard or mouse) for accepting user operation input, output device 15 (for example, liquid crystal monitor), and others The communication interface 16 (for example, NIC (Network Interface Card)) which implement | achieves communication between apparatuses is provided.

  An operating system (OS) operating on the computer 10 is not particularly limited to a specific system. For example, Windows (registered trademark) and UNIX (registered trademark) operating systems such as Linux (registered trademark) are preferably used as the OS.

  Next, the software configuration of the operation abnormality detection apparatus 500 will be described. FIG. 3 shows an example of the software configuration of the operation abnormality detection device 500. The operation abnormality detection device 500 includes a data I / O unit 510, an OS 520, and an abnormality detection processing unit 530. The data I / O unit 510 receives input of acoustic data from the sound collecting device 300, data input from the input device 14, etc. under the OS 520, and transfers them to the abnormality detection processing unit 530, and an abnormality detection processing unit Output data processing for delivering output data from 530 to the output device 15 or the like is executed. In addition, an outlier detection result table 540 used by the abnormality detection processing unit 530 is set in the operation abnormality detection device 500.

  The OS 520 is basic software that provides a basis for data input / output processing by the data I / O unit 510 and operation of the abnormality detection processing unit 530. As the OS 520, software similar to that running on the computer 10 can be employed.

  The abnormality detection processing unit 530 has a function of realizing the abnormality presence / absence determination of the turbine apparatus 100 executed according to the present embodiment by analyzing the operation sound data from the sound collector 300. The abnormality detection processing unit 530 is a functional block that is realized by the CPU 11 reading and executing a program that realizes this stored in the auxiliary storage device 13.

  The abnormality detection processing unit 530 further includes functional blocks of a feature extraction unit 5310, an outlier detection unit 5320, an abnormality determination threshold setting unit 5330, an abnormality determination execution unit 5340, and a determination result processing unit 5350.

  The feature extraction unit 5310 is a functional block that executes a process of dividing the operation sound data input from the sound collection device 300 into a plurality of frequency bands as a pre-process of a frequency variation pattern matching process described later. FIG. 4 schematically shows frequency band division processing of the operation sound data in the operation abnormality detection device 500 of the present embodiment. In the band division processing, the original operation sound data collected in the range of 0 to 25 kHz is converted into 10 frequency bands (0.02 to 0.03 kHz, 0.03 to 0.05 kHz, 0 using the discrete wavelet transform. 0.05 to 0.2 kHz, 0.2 to 0.39 kHz, 0.39 to 0.78 kHz, 0.78 to 1.56 kHz, 1.56 to 3.13 kHz, 3.13 to 6.25 kHz, 6.25 10 frequency bands of ˜12.5 kHz and 12.5 to 25 kHz). In the example of FIG. 4, in order to collect frequency fluctuations in a low frequency band of 0.02 to 0.03 kHz (20 to 30 Hz), the time length of operation sound collection is approximately 50 ms. In the case of a generator turbine device, the turbine rotation speed for a two-pole generator is 3600 rpm at a power supply frequency of 60 Hz. Therefore, when collecting operation sound for one rotation of the turbine shaft, there should be at least about 16.7 ms. It is enough.

  The frequency variation pattern matching process is executed for each frequency band as will be described later. Note that the number of divisions of the frequency band is not limited to 10, and the frequency boundary value of each band can be arbitrarily determined.

  The outlier detection unit 5320 discriminates data considered to be normal values and data considered to be abnormal values from among a plurality of collected operation sound data groups for each frequency band obtained by band division by the feature extraction unit 5310. Then, a process of extracting outliers as abnormal values is executed.

  The outlier detection unit 5320 employs 1000 sample data for each frequency band generated by the feature extraction unit 5310, and detects an outlier as an abnormal value from a plurality of operation sound data groups. FIG. 5 shows a conceptual diagram of outlier extraction processing for motion sound data. In this embodiment, outlier extraction processing is performed using a one-class ν-support vector machine, which is a method of multivariate analysis. In this embodiment, ν indicating the outlier ratio is set to 10% as an example. The 1 class ν-support vector machine itself is a well-known calculation method. For example, Shotaro Akaho, “Kernel Multivariate Analysis”, Iwanami Shoten, November 2008, p.4-10, p.86-105, p. Since it is detailed in .106-112, here is a brief description.

Function consisting of inner product of feature vector and parameter
, The target sample for which an outlier is to be determined is one-dimensional data as f (x ⁽¹⁾ ),..., F (x ⁽ⁿ⁾ ). These data are classified by a positive threshold ρ. Data satisfying ρ ≦ f (x ⁽ⁱ⁾ ) is determined as a normal value, and data satisfying ρ <f (x ⁽ⁱ⁾ ) is determined as an outlier.
Here, in order to set an appropriate threshold value ρ, the following loss function is created.
Considering a criterion for increasing the threshold ρ while suppressing the loss determined by this loss function, the process of discriminating outliers in the sample data can be reduced to solving the following optimization problem.
The discriminant function f (x) for discriminating the outlier shown in FIG.

  Referring back to FIG. 3, the description returns to the software configuration of the operation abnormality detection device 500 again. The abnormality determination threshold value setting unit 5330 uses the discriminant function f (x) calculated by the outlier detection unit 5320 as a threshold value for discriminating between the normal value and the outlier value of the operation sound data to another operation sound data group (identical It has a function of setting as an abnormality determination threshold so that it can be applied to data of other frequency bands collected at the timing of (1) or operating sound data groups collected at different timings.

  The anomaly discrimination execution unit 5340 uses the discriminant function f (x) as a threshold set by the anomaly discrimination threshold setting unit 5330 and relates to the operation sound data that is actually frequency-band divided by the discrete wavelet transform in the feature extraction unit 5310. Then, a process of discriminating and extracting the outlier as an abnormal value in the operation sound data is executed.

  The discrimination result processing unit 5350 receives the discrimination execution result in the anomaly discrimination execution unit 5340, calculates data such as an event where an abnormal value is detected, an abnormal value detection time, and a detection frequency. A process of transmitting to the output device 15 or the like is executed.

  Next, the outlier detection result table 540 will be described. FIG. 6 shows an example of the outlier detection result table 540. The outlier detection result table 540 is a table that records cumulatively the results of outlier detection performed by the set discriminant function f (x) for each operation sound data divided in frequency band.

  As shown in FIG. 6, the outlier detection result table 540 records the number of discriminations 541 and the outlier detection number 542 for each frequency band. The number of times of discrimination 541 stores the number of times outlier discrimination processing has been performed by the discriminant function f (x) as a serial number from 1. The outlier detection number 542 records how many outliers are detected as a result of the outlier discrimination processing in each frequency band, according to the number of outliers detected. In the example of FIG. 6, in the frequency band of 0.03 to 0.05 kHz, the number of detected outliers monotonically increases from 2 to 5 to 10 every time the discrimination process is executed. In this case, it can be determined that some malfunction that appears characteristically in the frequency band of 0.03 to 0.05 kHz occurs in the measurement target, that is, some sign of abnormality appears.

<< Explanation of abnormal operation detection process flow >>
Next, processing of the operation abnormality detection device 500 realized by the above configuration will be described. FIG. 7 shows an example of a processing flow executed by the operation abnormality detection device 500. In FIG. 7, the letter “S” of the reference numerals given to the respective processing steps represents “Step”.

  First, the feature extraction unit 5310 of the abnormality detection processing unit 530 acquires the operation sound data collected and transmitted by the sound collection device 300, and performs a process of dividing it into a plurality of frequency bands by discrete wavelet transform (S601, S602). Next, the abnormality determination threshold value setting unit 5330 of the abnormality detection processing unit 530 selects one frequency band data from the plurality of frequency band data generated by being divided in S602 (S603), and a threshold value is set for the selected frequency band data. It is determined whether it has been set (S604), and when it is determined that it has not been set (S604, No), the process is transferred to the outlier detection unit 5320, and the outlier detection process is executed (S605). After recording in the outlier detection result table 540 (S606), the process proceeds to the abnormality determination threshold value setting unit 5330. The abnormality determination threshold value setting unit 5330 temporarily sets the discrimination function f (x) applied to the outlier detection by the outlier detection unit 5320 as a threshold (S607). Thereafter, the abnormality determination threshold value setting unit 5330 shifts the processing to S623 and checks whether there is another frequency band to be processed. If it is determined that the processing has not been completed for all frequency bands (S623, No), another frequency band is selected (S603), and the processes after S604 are performed. When it is determined that the processing has been completed for all frequency bands (S623, Yes), the processing is shifted to the feature extraction unit 5310, the next operation sound data is captured (S624), and the next operation sound data group Continue the abnormality detection process for.

On the other hand, if it is determined that the threshold value is set (S604, Yes), it is determined whether the threshold value is the currently set threshold value (S608). If it is determined that the threshold value is not set (S608, No), the processing is performed. Is shifted to the abnormality determination execution unit 5340, and an outlier is detected based on the temporarily set threshold (S609), and the number of outliers detected is cumulatively stored in the outlier detection result table 540 (S610).

  Next, it is determined whether the number of determinations that have detected an outlier has reached a specified number (S611). If it is determined that the number has not been reached (S611, No), the process is transferred to the abnormality determination threshold value setting unit 5330, and It is checked whether there is a frequency band to be processed (S623).

  When it is determined that the number of determinations has reached the specified number or more (S611, YES), it is determined whether the outlier tends to increase (S612), and when it is determined that the outlier tends to increase (S612, YES), The abnormality determination execution unit 5340 shifts the processing to the abnormality determination threshold value setting unit 5330, and uses the identification function f (x) temporarily set as the threshold value by the abnormality determination threshold value setting unit 5330 in S607 for subsequent operation abnormality determination processing. (S616), it is checked whether there is another frequency band to be processed (S623).

  When it is determined that the detected outlier does not tend to increase (S612, No), the extracted outlier is excluded from the population (S613). Further, the temporarily set threshold value is deleted (S614), and all the determination results up to this time stored in the outlier detection result table 540 are also deleted (S615). Thereafter, the process is transferred to the outlier detection unit 5320, the outlier detection process is performed again (S605), and a new threshold value is temporarily set (S607).

  If the outlier is an abnormal value indicating an operational noise abnormality, that is, an abnormal value indicating a sign of abnormal operation due to wear of the turbine shaft bearing portion or the like, it is generally considered that the number of outliers increases as the operation continues. Based on this assumption, in the present embodiment, when the increase in the number of outlier detections is not observed, it is determined that the outliers are not caused by an abnormality in the equipment, and subsequent outlier detection is performed. It is intended to be excluded from the target.

  On the other hand, when it is determined that the threshold value is the preset threshold value (S608, Yes), the abnormality determination threshold value setting unit 5330 shifts the process to the abnormality determination execution unit 5340, and detects an outlier by the set threshold value. (S617) The outlier detection count is cumulatively stored in the outlier detection result table 540 (S618).

  Next, the abnormality determination execution unit 5340 shifts the processing to the determination result processing unit 5350, grasps in detail the increasing tendency of the outlier based on the data of the outlier detection result table 540 (S619), and the operation state of the device The degradation state of the device is evaluated with reference to the data etc. (S620). As a result, when it is determined that an increasing tendency satisfying a predetermined condition is observed, such as the rate of increase in outliers exceeding a predetermined reference value (S621, Yes), the determination result processing unit 5350 detects an abnormality in the equipment. As a result of detecting the sign, the determination result is transmitted to the output device 15 via the data I / O unit 510 (S622). Thereafter, the process proceeds to S623, and it is checked whether there is another frequency band to be processed. The same applies to the case where the predetermined increasing tendency is not observed in S621 (S621, No).

  The abnormality detection process flow of FIG. 7 continuously monitors signs of operational abnormality during operation of the turbine device, for example, by continuously executing the operation after the turbine device to be monitored in this embodiment is started. Can be found.

  According to the abnormality detection process flow described above with reference to FIG. 7, based on the outliers of the operation sound without collecting the abnormality detection teaching data for the turbine apparatus that is the target of operation sound monitoring in this embodiment. It is possible to detect signs of abnormal operation. Furthermore, since detected outliers that are considered to be less relevant to operational abnormalities are excluded without being used as an indicator of operational abnormalities, the detection of operational abnormalities can be reduced by reducing the detection after operational abnormalities. Accuracy can be improved.

  In addition, this invention is not limited to said embodiment, A various change is possible in the range which does not deviate from the summary.

DESCRIPTION OF SYMBOLS 10 Computer 11 Central processing unit 12 Main storage device 13 Auxiliary storage device 14 Input device 15 Output device 16 Communication interface 100 Steam turbine apparatus 110A, 110B Rotor 120 Rotating shaft 200 Generator 300 Sound collector 400 Communication line 500 Operation abnormality detection device 510 Data I / O unit 520 OS 530 Abnormality detection processing unit 540 Outlier value detection table 5310 Feature extraction unit 5320 Outlier detection unit 5330 Abnormality discrimination threshold setting unit 5340 Abnormality discrimination execution unit 5350 Discrimination result processing unit

Claims (4)

An operation abnormality detection device for detecting an abnormality that occurs in equipment,
A device operation physical information collection unit that collects device operation physical information, which is information that quantitatively represents the characteristics of a physical phenomenon that accompanies when the equipment operates
For a plurality of the device operation physical information acquired from the device operation physical information collection unit, an outlier detection process for detecting a certain percentage of outliers using a threshold determined as an identification function of the 1 class-ν support vector machine is executed. And, if there is no change in the number of detected outliers, an outlier detection unit that repeats the outlier detection process by excluding the detected outliers ;
In order to detect an abnormal operation of the device, the threshold value applied by the outlier detection unit to the outlier detection process when the number of outliers detected by the outlier detection unit is detected to increase. An abnormality determination threshold value setting unit that is set as an abnormality determination threshold value that is a threshold value of;
An abnormality determination execution unit that performs outlier determination processing for the device operation physical information using the abnormality determination threshold set by the abnormality determination threshold setting unit;
A determination result processing unit for outputting operation abnormality detection information when an outlier of the device operation physical information representing an operation abnormality of the equipment is detected by the abnormality determination execution unit;
An operation abnormality detection device comprising: The operation abnormality detection device according to claim 1,
The apparatus operation physical information is frequency characteristic data of sound or vibration generated by the facility apparatus. The operation abnormality detection device according to claim 2,
The operation abnormality detection apparatus, wherein the collected frequency characteristic data as the device operation physical information is divided into a plurality of frequency bands by discrete wavelet transform processing, and the outlier detection processing is executed for each frequency band . An operation abnormality detection method for detecting an abnormality that occurs in equipment,
In a computer with a processor and memory,
A process of collecting equipment operation physical information, which is information quantitatively representing the characteristics of a physical phenomenon that accompanies when the equipment is operating;
For a plurality of the device operation physical information acquired from the device operation physical information collection unit, an outlier detection process for detecting a certain percentage of outliers using a threshold determined as an identification function of the 1 class-ν support vector machine is executed. When the number of detected outliers does not change, a process of excluding the detected outliers and repeatedly executing the outlier detection process ;
When detecting that the number of outliers detected by the outlier detection unit is increasing, the threshold applied by the outlier detection unit to the outlier detection process is used to detect an apparatus operation abnormality. Processing to set as an abnormality determination threshold that is a threshold;
A process of performing an outlier determination process for the device operation physical information using the abnormality determination threshold set by the abnormality determination threshold setting unit;
When an outlier of the device operation physical information representing an operation abnormality of the equipment is detected by the abnormality determination execution unit, a process of outputting operation abnormality detection information;
An abnormal operation detection method characterized in that