WO2015033603A1 - 情報処理システム、情報処理方法及びプログラム - Google Patents
情報処理システム、情報処理方法及びプログラム Download PDFInfo
- Publication number
- WO2015033603A1 WO2015033603A1 PCT/JP2014/056334 JP2014056334W WO2015033603A1 WO 2015033603 A1 WO2015033603 A1 WO 2015033603A1 JP 2014056334 W JP2014056334 W JP 2014056334W WO 2015033603 A1 WO2015033603 A1 WO 2015033603A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- frequency data
- correlation
- data
- information processing
- model
- Prior art date
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F11/00—Error detection; Error correction; Monitoring
- G06F11/07—Responding to the occurrence of a fault, e.g. fault tolerance
- G06F11/0703—Error or fault processing not based on redundancy, i.e. by taking additional measures to deal with the error or fault not making use of redundancy in operation, in hardware, or in data representation
- G06F11/079—Root cause analysis, i.e. error or fault diagnosis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F11/00—Error detection; Error correction; Monitoring
- G06F11/07—Responding to the occurrence of a fault, e.g. fault tolerance
- G06F11/0703—Error or fault processing not based on redundancy, i.e. by taking additional measures to deal with the error or fault not making use of redundancy in operation, in hardware, or in data representation
- G06F11/0706—Error or fault processing not based on redundancy, i.e. by taking additional measures to deal with the error or fault not making use of redundancy in operation, in hardware, or in data representation the processing taking place on a specific hardware platform or in a specific software environment
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F11/00—Error detection; Error correction; Monitoring
- G06F11/07—Responding to the occurrence of a fault, e.g. fault tolerance
- G06F11/0703—Error or fault processing not based on redundancy, i.e. by taking additional measures to deal with the error or fault not making use of redundancy in operation, in hardware, or in data representation
- G06F11/0751—Error or fault detection not based on redundancy
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F11/00—Error detection; Error correction; Monitoring
- G06F11/07—Responding to the occurrence of a fault, e.g. fault tolerance
- G06F11/0703—Error or fault processing not based on redundancy, i.e. by taking additional measures to deal with the error or fault not making use of redundancy in operation, in hardware, or in data representation
- G06F11/0751—Error or fault detection not based on redundancy
- G06F11/0754—Error or fault detection not based on redundancy by exceeding limits
- G06F11/076—Error or fault detection not based on redundancy by exceeding limits by exceeding a count or rate limit, e.g. word- or bit count limit
Definitions
- Some aspects of the present invention relate to an information processing system, an information processing method, and a program.
- Patent Document 1 acquires a plurality of performance information acquired from a plurality of managed devices constituting a system, and then performs a correlation function between performance series information indicating a time-series change of the performance information acquired at regular intervals.
- a device is disclosed that can derive a coefficient and identify the location where an abnormality occurs in response to a change in the correlation function.
- Patent Document 2 discloses that sound data generated from equipment is collected, a frequency spectrum of the collected data is determined, and the presence or absence of abnormality is determined based on a correlation coefficient of spectrum values.
- Some aspects of the present invention are made in view of the above-mentioned subject, and make it an object to provide an information processing system, an information processing method, and a program which can perform data analysis suitably.
- One information processing system includes conversion means for converting a plurality of time series data obtained respectively by detection by a plurality of sensors into first frequency data, and at least two of the plurality of sensors.
- a first model generation unit that generates a first correlation model using first frequency data related to a sensor, a first operation unit that calculates the strength of the correlation of the first correlation model, and an abnormality based on the strength of the correlation
- determination means for determining
- One information processing system includes conversion means for converting a plurality of time series data obtained by detection by a plurality of sensors into first frequency data, and at least two of the plurality of sensors. Using the first frequency data according to a second model, and applying second frequency data obtained by converting another time series data obtained from a sensor according to the correlation model to the correlation model, And determining means for determining an abnormality based on a difference between a predicted value of the second frequency data obtained by performing the second frequency data and a measured value of the second frequency data.
- One information processing method includes the steps of: converting a plurality of time series data obtained respectively by detection by a plurality of sensors into first frequency data; and at least two of the plurality of sensors Processing the step of generating a first correlation model using the first frequency data relating to the step, calculating the strength of the correlation of the first correlation model, and determining an abnormality based on the strength of the correlation. System does.
- One information processing method includes the steps of: converting a plurality of time series data obtained by detection by a plurality of sensors into first frequency data; and at least two of the plurality of sensors. Step of generating a correlation model using the first frequency data, and applying, to the correlation model, second frequency data obtained by converting another time series data obtained from a sensor related to the correlation model.
- the information processing system performs the step of determining abnormality based on the difference between the obtained predicted value of the second frequency data and the actually measured value of the second frequency data.
- a program according to the present invention relates to processing of converting a plurality of time series data obtained respectively by detection by a plurality of sensors into first frequency data, and to at least two of the plurality of sensors. Allowing a computer to execute a process of generating a first correlation model using first frequency data, a process of calculating the strength of the correlation of the first correlation model, and a process of determining an abnormality based on the strength of the correlation .
- a program according to the present invention includes a process of converting a plurality of time series data obtained by detection by a plurality of sensors into first frequency data, and a process related to at least two of the plurality of sensors. It is obtained by processing of generating a correlation model using one-frequency data, and applying, to the correlation model, second frequency data obtained by converting another time-series data obtained from a sensor related to the correlation model.
- the computer is caused to execute a process of determining abnormality based on the difference between the predicted value of the second frequency data and the measured value of the second frequency data.
- an information processing system an information processing method, and a program capable of preferably performing data analysis.
- FIG. 2 is a diagram showing a specific example of the configuration of the information processing system according to the first embodiment.
- FIG. 3 shows a specific example of the functional configuration of the information processing system. It is a flowchart which shows the flow of a process of the information processing system shown in FIG. It is a flowchart which shows the flow of a process of the information processing system shown in FIG. It is a flowchart which shows the flow of a process of the information processing system shown in FIG. It is a flowchart which shows the flow of a process of the information processing system shown in FIG.
- (1 first embodiment) 1 to 10 are views for explaining the first embodiment.
- the present embodiment will be described along the following flow with reference to these drawings.
- an outline of the data analysis method in the present embodiment will be described in “1.1”.
- the outline of the system configuration of the information processing system in this embodiment will be described in “1.2”
- the outline of the functional configuration of the information processing system in this embodiment will be described in “1.3”.
- the flow of processing will be described using a specific example.
- “1.5” a specific example of a hardware configuration capable of realizing an information processing system will be described.
- the effects and the like according to the present embodiment will be described in “1.6”.
- the information processing system detects an abnormality of the system by acquiring data such as vibration, sound, light and the like by, for example, a sensor and analyzing these data.
- data such as vibration, sound, light and the like
- the information processing system in order to analyze data after performing pre-processing of frequency conversion on time series data obtained from a sensor, using data in which an essential change appears in frequency
- the correlation between different sensors can be modeled.
- the correlation between the data is specified not by the correlation coefficient but by the correlation model. As a result, regardless of whether the frequency data is in the increasing trend or the decreasing trend, it is possible to detect an abnormality according to the change in the correlation.
- the present invention will be described more specifically with reference to FIGS. 1 and 2.
- a correlation model between sensors is generated from frequency data obtained by converting time-series data obtained by detecting sound, light, vibration and the like by the sensors.
- the correlation model can be expressed, for example, by the following equation.
- f 0 to f n and f 0 to f m are respectively predetermined frequencies
- s x (f) and s y (f) are frequency data obtained from sensor x and sensor y, respectively
- the intensity at frequency f of (the time series data converted) is shown.
- Each of a 0 to a n , b 0 to b m and C is a coefficient
- the processing for obtaining the correlation model corresponds to the processing for determining the coefficients a 0 to a n , b 0 to b m and C.
- s y (f m ) ' which is a predicted value of the intensity at the frequency fm.
- the predicted values at the frequencies f 0 to f m-1 can be similarly calculated.
- the strength of the correlation there is various indicators for determining the strength of the correlation (hereinafter, also referred to as the strength of the correlation). For example, with respect to normal data used for model generation, the prediction error between the generated predicted value and observed value It can be a value proportional to the sum. That is, the strength of the correlation can be expressed, for example, by the following equation.
- s y (f) is an actual measurement value of the intensity of frequency data at frequency f
- s y (f) ′ is a predicted value of the intensity of frequency data at frequency f calculated using a correlation model.
- the correlation strength of the correlation model in normal data is compared with the correlation strength of the correlation model in observation data By doing this, it is determined whether or not an abnormality has occurred.
- FIG. 2 shows the case where each sensor is observed individually, and the right side shows the case where the correlation between sensors is observed as in the present embodiment.
- the observed value of the sensor indicates within the normal range (normal value) if the generated abnormality is a minor one, and as shown in the upper left of FIG.
- the anomaly can not be detected.
- the correlation between a certain sensor and another sensor may be broken, so that an abnormality may be detected.
- the correlation between a particular sensor and another sensor it may be possible to detect that the observation target of the particular sensor is abnormal.
- a correlation model indicating correlation is generated using frequency data obtained by converting time series data obtained from each sensor, and the correlation of the correlation model An abnormality is detected according to whether or not the relationship is broken.
- Whether or not the correlation is broken can be determined using an index of the strength of the correlation in the correlation model between the sensors. If a large difference occurs between the strength of correlation in the correlation model generated using normal data and the strength of correlation in the correlation model generated using data of the observation target, the correlation is broken It can detect that it is in the state.
- the correlation model used for anomaly detection can be generated even if the fluctuation trend of the data differs between the two data. Therefore, since the strength of the correlation does not depend on the fluctuation trend between the two data, it is possible to preferably generate a correlation model even between data of different sensors, and calculate the strength of the correlation. By comprehensively carrying out such processing between a plurality of sensors, sensor data in which abnormality occurrences are concentrated can be narrowed down as the root cause of the abnormality.
- the information processing system 100 includes vibration sensors 101a to 101n (hereinafter, may be collectively referred to as vibration sensor 101), a signal conversion module 103, a personal computer, a server, etc. And a storage medium 109 and a display 111.
- the vibration sensor 101 is, for example, a sensor installed at a different position, and observes time series data.
- the vibration sensor 101 detects a signal here, it is not restricted to this, It is also considered to observe time series data of sound or light instead of vibration.
- the signal conversion module 103 converts time-series analog data detected by the vibration sensor 101 into digital data that can be processed by the information processing apparatuses 105 and 107, and then converts the time-series digital data into frequency data. Do.
- the frequency data after conversion is output to the information processing apparatuses 105 and 107.
- the information processing apparatus 105 generates a correlation model from frequency data obtained by converting sensor data converted into a digital signal, and outputs the correlation model to the storage medium 109.
- the storage medium 109 is, for example, a hard disk drive (HDD) or a flash memory, and stores the correlation model as described above.
- the storage medium 109 may be built in the information processing apparatus 105 or the information processing apparatus 107.
- the information processing apparatus 107 receives an input of frequency data obtained by converting sensor data newly detected by each vibration sensor 101, and detects an abnormality using the frequency data and model information stored in the storage medium 109. Do the processing.
- the display 111 displays the abnormality detection result by the information processing apparatus 107.
- the present invention is not limited to this, and mounting by one information processing apparatus or three or more information processing apparatuses Is also conceivable.
- the display 111 may be built in the information processing apparatus 107.
- sensing units 201 a to 201 n (hereinafter collectively referred to as sensing unit 201), noise filtering unit 203, frequency conversion unit 205, and storage It comprises a part 207 and 209, a model construction part 211 and 219, a storage part 212, a correlation strength average / maximum deviation calculation part 217, an abnormality detection part 221, and a notification part 223.
- the sensing unit 201 corresponds to the vibration sensor 101 of FIG.
- the noise filtering unit 203 and the frequency conversion unit 205 correspond to the signal conversion module 103 in FIG.
- the storage unit 212 corresponds to the storage medium 109 in FIG. 1
- the storage unit 209 and the abnormality detection unit 221 correspond to the information processing apparatus 107 in FIG. 1.
- the notification unit 223 corresponds to the display 111.
- the sensing unit 201 detects vibration, light, sound, and the like to generate and output time-series data.
- the sensing unit 201 generates and outputs time-series data at least twice or more of a normal state and a state (a target of abnormality detection) in which it is unclear whether the state is normal or not.
- the number n of the sensing units 201 may be any number as long as it is two or more.
- the noise filtering unit 203 removes noise from the time series data output from the sensing unit 201.
- the frequency conversion unit 205 detects time-series data detected by the sensing units 201a to 201n and from which noise has been removed, respectively, as frequency data 208a to 208n (hereinafter collectively referred to as frequency data 208) and frequency data 210a to Convert to 210 n (hereinafter collectively referred to as frequency data 210).
- the frequency data 208 and 210 are, for example, data with different detection timing, and here, the frequency data 208 is normal data used to generate the correlation model 213, and the frequency data 210 is the correlation model 213. It is observation data for detecting abnormalities using.
- the frequency data 208 and 210 are stored in the storage units 207 and 209, respectively.
- the model construction unit 211 generates correlation models 213a to 213m (hereinafter collectively referred to as a correlation model 213) from two combinations of frequency data 208a to 208n which are normal data.
- the generated correlation model 213 is stored in the storage unit 212.
- the model construction unit 211 also causes the storage unit 212 to store the correlation strengths 214a to 214m of the correlation models 213a to 213m.
- the model construction unit 211 When the frequency data 208 for a plurality of times is observed for each sensing unit 201, the model construction unit 211 performs the correlation model 213 related to the combination of each sensing unit 201 and the strength 214 of the correlation as the number of times. Only minutes can be generated.
- the correlation strength average / maximum deviation calculation unit 217 calculates the maximum deviation while obtaining the average value of the correlation strengths 214 for a plurality of times related to each sensing unit 201.
- the maximum deviation corresponds to the maximum value of the difference between the average value of the correlation strength 214 and the correlation strength 214 of each correlation model 213 used to calculate the average.
- the said largest deviation can be memorize
- the correlation strength average / maximum deviation calculation unit 217 is unnecessary.
- the model construction unit 219 Similar to the model construction unit 211, the model construction unit 219 also generates correlation models from two combinations of frequency data 210a to 210n, which are observation data, and calculates the strength of the correlation of each correlation model.
- the abnormality detection unit 221 determines the correlation strength 214 of the correlation model 213 generated based on the frequency data 208 which is normal data, and the correlation strength of the correlation model generated based on the frequency data 210 which is observation data.
- the abnormality is detected by comparing More specifically, for example, the difference between the correlation strength 214 of the correlation model 213 generated based on the frequency data 208 and the correlation strength of the correlation model generated based on the frequency data 210 is determined as abnormal.
- the threshold 215 is exceeded, it may be determined that the correlation is broken (an abnormality has occurred).
- the strength of the correlation to be compared is the correlation models according to the combination of the same sensing unit 201.
- the notification unit 223 notifies the user of the result of the abnormality detection by the abnormality detection unit 221.
- a method of notification by the notification unit 223 for example, a method of displaying a message or the like on the display 111 can be considered.
- FIGS. 5 to 9. 5 to 9 are flowcharts showing the process flow of the information processing system 100.
- each processing step described below can be arbitrarily changed in order or executed in parallel as long as no contradiction occurs in the processing content, and even if another step is added between each processing step good. Furthermore, for convenience, the steps described as one step can be divided into a plurality of steps and executed, or the steps described separately as a plurality can be performed as one step. This point is the same as in the second and subsequent embodiments.
- Normal time-series data detected (sensing) by the sensing unit 201 is converted into frequency data 208 by the frequency conversion unit 205 (S 501), and stored in the storage unit 207.
- the model construction unit 211 generates the correlation model 213 for each combination of the frequency data 208a to 208n, and calculates the strength 214 of the correlation related to each correlation model 213 (S503).
- the correlation strength average / maximum deviation calculation unit 217 calculates the average value and the maximum deviation of the correlation strengths 214 for the plurality of times
- the abnormality determination threshold 215 may be calculated.
- time-series data of the observation target newly detected by the sensing unit 201 is converted into frequency data 210 by the frequency conversion unit 205 (S505), and stored in the storage unit 209.
- the model construction unit 219 obtains a correlation model for each combination of the frequency data 210a to 210n, and calculates the strength of the correlation related to each correlation model (S507).
- the abnormality detection unit 221 performs abnormality detection processing by comparing the correlation model 213 generated from the frequency data 208 as normal data with the correlation model generated from the frequency data 210 as observation data (S509). The details of the processing of each of S501, S503, S507, and S509 will be described below with reference to FIGS.
- the sensing units 201a to 201n perform state detection on a measurement target for a predetermined period (for example, 10 seconds) (S601).
- the noise filtering unit 203 and the frequency conversion unit 205 sequentially process the time series data 1 to n obtained from the respective sensing units 201a to 201n.
- the noise filtering unit 203 performs the i-th time-series data (i-th sensing unit After extracting the data 201) (S605), noise is removed from the time-series data (S607). At this time, it is conceivable that the noise filtering unit 203 removes noise by using a Butterworth filter on time-series data. If attention is paid to the frequency band around 1000 Hz, it is conceivable to apply a Butterworth filter so that a component of 10 to 10000 Hz remains.
- the frequency conversion unit 205 converts the time-series data from which noise has been removed by the noise filtering unit 203 into frequency data 208 (S609), and stores the data in the storage unit 207 (S611).
- a conversion method to the frequency data 208 by the frequency conversion unit 205 for example, a method using an AR (Auto-regressive) model (autocorrelation model) can be considered.
- an autocorrelation model of the model order set with reference to the AIC is identified with respect to the filtered time series data, and an impulse response of the identified autocorrelation model is acquired, and then the impulse response A method of frequency conversion is conceivable.
- the information processing system 100 repeats the processing of S603 to S611 until i> n (until there is no unprocessed data).
- the above-described process is the same as in the generation of the frequency data 210 (target of observation data) for abnormality detection (corresponding to the process of S505 in FIG. 5).
- FIG. 7 corresponds to the process of S503 in FIG.
- time series data which is normal data is detected plural times (here, M times) by the sensing units 201a to 201n, and the correlation model 213 is detected with respect to the frequency data 208 corresponding to each time series data.
- M times plural times
- the model construction unit 211 calculates the correlation model 213 for each combination of the frequency data 208 with respect to the normal data detected at the k-th time, and calculates the strength 214 of the correlation of each correlation model 213 (S701). This process will be described in detail later with reference to FIG.
- the sensing units 201i and 201j (0 ⁇ i, j ⁇ n) respectively M From the M pieces of frequency data 208i and 208j generated from the time-series data detected a number of times, M correlation strengths 213 of the correlation model 213 and the correlation model 213 are generated respectively. Therefore, the correlation strength average / maximum deviation calculation unit 217 obtains the average value of the M correlation strengths 214 for each combination of (i, j) (S705 to S711). Thereby, the average value of the strength of the correlation which concerns on each combination of the sensing part 201 can be calculated, respectively. Subsequently, the details of the process according to S701 will be described with reference to FIG.
- the model construction unit 211 sets i and j to 1 and then sets frequency data 208i and 208j (frequency data 208 generated from time-series data detected by the i-th and j-th sensing units 201, respectively). Are extracted from the storage unit 207 (S801 and S803). If i and j are equal (Yes in S805), the value of j is incremented and then the frequency data 208j is extracted again (S803).
- the correlation model 213 is generated using the frequency data 208i and the frequency data 208j (S807).
- Specific examples of the correlation model 213 include, for example, an ARX (Auto-regressive exogeneous) model.
- the model construction unit 211 applies the frequency data 208i and 208j used to generate the model to the generated correlation model 213 to calculate the prediction value at each frequency, and then calculates the prediction value and the observation value (measured
- the strength 214 of the correlation related to the correlation model 213 is calculated from the difference with the value (S809). Further, the model construction unit 211 stores the calculated correlation model 213 and the correlation strength 214 in the storage unit 212 (S811).
- the correlation model 213 used for abnormality detection it is also conceivable to limit the correlation model 213 used for abnormality detection to only one with high prediction accuracy. In this case, only the correlation strength 214 exceeding the threshold value is used for abnormality detection by the abnormality detection unit 221.
- the model construction unit 211 performs the above-described process on all the combinations of frequency data 208 while appropriately incrementing i and j until the value becomes n.
- the correlation model 213 is generated for two combinations of the combination of (frequency data 208i and frequency data 208j) and the combination of (frequency data 208j and frequency data 208i).
- the model construction unit 211 adopts the one with higher prediction accuracy of the correlation model 213 (the one where the correlation strength 214 is larger).
- the model construction unit 219 and the abnormality detection unit 221 perform processing on the frequency data 210 that is a target (observation data) of abnormality detection.
- the model construction unit 219 sets i and j to 1 and then sets the frequency data 210i and 210j (frequency data generated from time-series data extracted by the i-th and j-th sensing units 201, respectively). 210) is extracted from the storage unit 209 (S901 and S903). If i and j are equal (Yes in S905), the value of j is incremented and then the frequency data 210j is extracted again (S903).
- the model construction unit 219 When the values of i and j are different (No in S905), the model construction unit 219 generates a correlation model from the frequency data 210i and 210j related to the sensing units 201i and 201j (S907), and also correlates in the correlation model. The strength of is calculated (S909).
- the abnormality detection unit 221 receives from the model construction unit 219 the strength of the correlation of the correlation model related to the sensing units 201i and 201j, and from the storage unit 212, frequency data 208i and 208j that are normal data And the correlation strength 214 of the correlation model 213 generated from the frequency data 208 generated from the time-series data extracted by the j-th sensing unit 201 (S911). If the correlation strength of the correlation model generated from the observation data and the correlation strength 214 of the correlation model 213 generated from the normal data approximate (for example, the difference is within the abnormality determination threshold 215) ) (Yes in S913), it can be determined that the observation data is normal.
- the notification unit 223 notifies information indicating an abnormality together with information related to the sensing unit 201 in which the correlation is broken (S915).
- the abnormality detection unit 221 notifies that there is an indication that an abnormality occurs in the frequency to be processed.
- the model construction unit 219 and the abnormality detection unit 221 perform the above-described process on all the combinations of the frequency data 210 while appropriately incrementing i and j until the value becomes n.
- the abnormality detection unit 221 lists up the sensing units 201 included in many combinations that are expected to be broken due to the occurrence of an abnormality after processing for all the combinations of frequency data 210 is completed. You may. As a result, it is possible to narrow down the sensing units 201 which may have an abnormality.
- the average value of the correlation strength may be compared with the correlation strength of the correlation model generated from the frequency data 210 which is observation data related to the combination of the same sensing unit 201.
- the computer 1000 includes a processor 1001, a memory 1003, a storage device 1005, an input interface (I / F) 1007, a data I / F 1009, a communication I / F 1011 and a display device 1013.
- a processor 1001 a memory 1003, a storage device 1005, an input interface (I / F) 1007, a data I / F 1009, a communication I / F 1011 and a display device 1013.
- the processor 1001 controls various processes in the computer 1000 by executing programs stored in the memory 1003. For example, the processing relating to the noise filtering unit 203, the frequency conversion unit 205, the model construction unit 211, the correlation strength average / maximum deviation calculation unit 217, the model construction unit 219, and the abnormality detection unit 221 described in FIG. It can be realized as a program which is temporarily stored and mainly operates on the processor 1001.
- the memory 1003 is, for example, a storage medium such as a random access memory (RAM).
- the memory 1003 temporarily stores program code of a program to be executed by the processor 1001 and data necessary for executing the program. For example, in a storage area of the memory 1003, a stack area required for program execution is secured.
- the storage device 1005 is, for example, a non-volatile storage medium such as a hard disk or a flash memory.
- the storage device 1005 includes an operating system, noise filtering unit 203, frequency conversion unit 205, model constructing unit 211, correlation strength average / maximum deviation calculating unit 217, model constructing unit 219, various types for realizing abnormality detection unit 221.
- a program, various data including frequency data 208 and 210, correlation model 213, correlation strength 214, abnormality determination threshold 215, etc. are stored.
- the program and data stored in the storage device 1005 are referred to by the processor 1001 by being loaded into the memory 1003 as necessary.
- the input I / F 1007 is a device for receiving an input from a user.
- a keyboard, a mouse, a touch panel etc. are mentioned as a specific example of input I / F1007.
- the input I / F 1007 may be connected to the computer 1000 via an interface such as USB (Universal Serial Bus).
- USB Universal Serial Bus
- the data I / F 1009 is a device for inputting data from the outside of the computer 1000.
- a specific example of the data I / F 1009 is a drive device for reading data stored in various storage media. It is also conceivable that the data I / F 1009 is provided outside the computer 1000. In that case, the data I / F 1009 is connected to the computer 1000 via an interface such as USB.
- the communication I / F 1011 is a device for performing data communication with an external device of the computer 1000, for example, with the sensing unit 201 or the like in a wired or wireless manner.
- the communication I / F 1011 may be provided outside the computer 1000. In that case, the communication I / F 1011 is connected to the computer 1000 via an interface such as USB.
- the display device 1013 is a device for displaying various information.
- the display 111 shown in FIG. 3 may be implemented as a display device 1013.
- a display device 1013 for example, a liquid crystal display, an organic EL (Electro-Luminescence) display, etc. may be mentioned.
- the display device 1013 may be provided outside the computer 1000. In that case, the display device 1013 is connected to the computer 1000 via, for example, a display cable or the like.
- the information processing system 100 uses the frequency data 208 for the time-series data in which the feature of the change appears in the frequency detected by the sensing unit 201, and The correlation is modeled as a correlation model 213.
- the correlation is specified as the correlation model 213 instead of the correlation coefficient, it is possible to determine whether the sensor data is normal or abnormal from the change in the correlation. . That is, in the information processing system 100 according to the present embodiment, data analysis can be suitably performed.
- the entire configuration of the information processing system 100 whose specific example is shown in FIG. 2 and the configuration of the computer 1000 capable of realizing the signal conversion module 103, the information processing device 105, and the information processing device 107 are the same as those in the first embodiment. Therefore, the explanation is omitted.
- the correlation model is generated from the frequency data 208 and 210 related to each sensing unit 201.
- the average value of the frequency data 208 which is normal data detected by each sensing unit 201
- This process will be described with reference to the specific example of FIG. 11. First, with respect to the frequency data 208 relating to the sensors A to D shown by the graph, average frequency data of broken lines shown at the center in the left part is generated. Then, a correlation model is generated between the average frequency data and the frequency data 208 of the sensors A to D, and the correlation strengths F A-Ave to F D-AVE are calculated.
- the sensing unit 201 for acquiring normal data and the sensing unit 201 for acquiring observation data are the same, and the difference between normal data and observation data is time series data.
- the sensing unit 201 for acquiring normal data and the sensing unit 201 for acquiring observation data are different.
- normal data is frequency data obtained by converting time series data obtained from sensors A to D
- observation data is frequency data obtained by converting time series data obtained from sensors E and F. is there.
- a correlation model is generated with the average frequency data of the normal data, and the strength of the correlation is calculated (F E-AVE and F F-AVE in FIG. 11).
- the correlation strengths F A-AVE to F D-AVE of the correlation model between the average frequency data and the frequency data relating to the normal data substantially match (for example, the difference is less than or equal to the threshold).
- the correlation strength F E-AVE or F F-AVE of the correlation model between the average frequency data of normal data and the frequency data relating to observation data is largely different (for example, the difference exceeds a threshold)
- the information processing system 100 determines that the sensor E or the sensor F is abnormal.
- the system configuration can be the same as that of the first embodiment, so the description is omitted here.
- the functional configuration of the information processing system 100 according to the present embodiment will be described with reference to FIG.
- the information processing system 100 according to the present embodiment includes an average frequency data calculation unit 225 and average frequency data 228 in addition to the components included in the information processing system 100 according to the first embodiment.
- the sensing unit 201 detects vibration, light, sound, and the like to generate and output time-series data.
- the sensing units 201a to 201k are for detecting normal data
- the sensing units 201k + 1 to 201n are for detecting observation data.
- the number of sensing units 201 for detecting normal data may be any number as long as it is two or more
- the number of sensing units 201 for detecting observation data may be any number as long as it is one or more. .
- the noise filtering unit 203 removes noise from the time series data output from the sensing unit 201 as in the first embodiment.
- the frequency conversion unit 205 converts the time-series data detected by the sensing units 201a to 201k and from which noises have been removed, respectively to frequency data 208a to 208k (collectively referred to as frequency data 208), and a storage unit. Output to 207. Further, the frequency conversion unit 205 converts time-series data, which are respectively detected by the sensing units 201k + 1 to 201n and from which noises have been removed, to frequency data 210a to 201n-k (generally referred to as frequency data 210). Output to the storage unit 209.
- frequency data 208 is normal data indicating a normal state
- frequency data 210 is observation data to be detected.
- the average frequency data calculation unit 225 generates average frequency data 228 which is average data of normal data by calculating an average of the frequency data 208a to 208k for each frequency.
- the calculated average frequency data 228 is stored in the storage unit 227.
- the model construction unit 211 generates correlation models 213a to 213k (hereinafter collectively referred to as a correlation model 213) between the frequency data 208a to 208k, which are normal data, and the average frequency data 228, respectively. .
- the generated correlation model 213 is stored in the storage unit 212.
- the model construction unit 211 also calculates the correlation strengths 214a to 214k of the respective correlation models 213a to 213k, and also stores them in the storage unit 212.
- the model construction unit 211 determines the correlation model 213 and the correlation strength relating to the combination of each sensing unit 201 and the average frequency data 228. 214 can be generated for the number of times.
- the correlation strength average / maximum deviation calculation unit 217 calculates the maximum deviation while obtaining the average value of the correlation strengths 214 generated for a plurality of times.
- the maximum deviation corresponds to the maximum value of the difference between the average value of the correlation strength 214 and the correlation strength 214 of each correlation model 213 used to calculate the average.
- the said largest deviation can be memorize
- the correlation strength average / maximum deviation calculation unit 217 is unnecessary.
- the model construction unit 219 generates a correlation model between each of the frequency data 210a to 210n-k, which is observation data, and the average frequency data 228, and calculates the strength of the correlation of each correlation model.
- the abnormality detection unit 221 determines the correlation strength 214 of the correlation model 213 generated based on the frequency data 208 which is normal data, and the correlation strength of the correlation model generated based on the frequency data 210 which is observation data. The abnormality is detected by comparing More specifically, for example, although the correlation strength 214 of each correlation of the correlation model 213 generated based on the frequency data 208 is similar (for example, within the threshold range), the frequency data 210 If the correlation strength of the correlation model generated based on is not within the approximation range (for example, the threshold range is exceeded), the abnormality detection unit 221 has broken the correlation (an abnormality has occurred) It can be considered to be determined.
- the notification unit 223 notifies the user of the result of the abnormality detection by the abnormality detection unit 221.
- a method of notification by the notification unit 223 for example, a method of displaying a message on the display 111 can be considered.
- Normal time-series data detected (sensing) by the sensing unit 201 is converted into frequency data 208 by the frequency conversion unit 205 (S 1301), and the frequency data 208 is stored in the storage unit 207.
- the average frequency data calculation unit 225 generates average frequency data 228 by averaging the frequency data 208 stored in the storage unit 207 for each frequency (S1303).
- the model construction unit 211 generates the correlation model 213 between each frequency data 208 and the average frequency data 228, and calculates the strength 214 of the correlation related to each correlation model 213 (S1305).
- the correlation strength average / maximum deviation calculating unit 217 determines the strength of the correlation for the plural times.
- the average value of 214 and the maximum deviation (corresponding to the abnormality determination threshold 215) may be calculated.
- time-series data of the observation target detected by the sensing units 201k + 1 to 201n is converted into frequency data 210 by the frequency conversion unit 205 (S1307), and stored in the storage unit 209.
- the model construction unit 219 obtains a correlation model between each of the frequency data 210a to 219n-k and the average frequency data 228, and calculates the strength of the correlation of each correlation model (S1309).
- the abnormality detection unit 221 performs abnormality detection processing by comparing the correlation model 213 generated from the frequency data 208 as normal data with the correlation model generated from the frequency data 210 as observation data (S1311).
- each process of S1301 and S1307 is the same as S501 and S505 described with reference to FIG. 5 in the first embodiment.
- the processes of S1305, S1309 and S1311 will be described below with reference to FIG.
- time series data which is normal data is detected plural times (here, M times) by the sensing units 201a to 201k, and the correlation model 213 is detected with respect to the frequency data 208 corresponding to each time series data.
- M times plural times
- the model construction unit 211 generates the correlation model 213 described with reference to the equation 1 with the average frequency data 228 for the i-th detected normal data.
- the model construction unit 211 uses the correlation model 213 to calculate the predicted value of the intensity at each frequency using the frequency data used for model generation, and obtains the difference between the predicted value and the actual measurement value. It stores as an abnormality determination threshold 215 used for detection.
- the strength 214 of the correlation of the correlation model 213 is calculated based on Equation 2 (S1401). This process will be described in detail later with reference to FIG.
- the model construction unit 211 reads the average frequency data 228 (1501). Further, the model construction unit 211 sets i to 1 and then extracts from the storage unit 207 frequency data 208 i (frequency data 208 generated from time-series data extracted from the i-th sensing unit 201) ( S1503).
- the model construction unit 211 generates the correlation model 213 using the frequency data 208i and the average frequency data 228 (S1505).
- Examples of the correlation model 213 include, for example, an ARX model.
- the model construction unit 211 applies the generated correlation model 213 to the frequency data 208 used for model generation to calculate the prediction value of the frequency data at each frequency, and then the prediction value and the frequency data. From the difference between the measured value (observed value) of 208, the strength of the correlation related to the correlation model 213 is calculated (S1507). Further, the model construction unit 211 stores the calculated correlation model 213 and the correlation strength 214 in the storage unit 212 (S1509).
- the correlation model 213 used for abnormality detection it is also conceivable to limit the correlation model 213 used for abnormality detection to only one with high prediction accuracy. In this case, only the correlation strength 214 exceeding the threshold value is used for abnormality detection by the abnormality detection unit 221.
- the model construction unit 211 performs this on all the frequency data 208 of the above processing while appropriately incrementing the value of i until it becomes k which is the number of sensing units 201 that detect observation data.
- the correlation model 214 can be generated for two combinations of the frequency data 208 and the average frequency data 228 and the combination of the average frequency data 228 and the frequency data 208, among which the correlation model 214 can be generated.
- the model construction unit 211 can adopt one in which the strength 214 is larger. The same applies to the strength of the correlation of the correlation model between the later frequency data 210 and the average frequency data 228.
- the model construction unit 219 performs processing on the frequency data 210 which is a target (observation data) of abnormality detection. First, the model construction unit 219 reads the average frequency data 228 (S 1601), sets i to 1 and then generates frequency data 210 i (frequency generated from time series data extracted by the k + i-th sensing unit 201 The data 210) is extracted from the storage unit 209 (S1603).
- the model construction unit 211 generates the correlation model 213 using the frequency data 210i and the average frequency data 228 (S1605).
- a specific example of the correlation model is the ARX model.
- the model construction unit 211 also calculates the strength of the correlation in the generated correlation model (S1607).
- the abnormality detection unit 221 receives the strength of the correlation of the correlation model according to the sensing unit 201i from the model construction unit 219, and from the storage unit 212, each correlation model 213 generated from the frequency data 208 which is normal data.
- Each correlation strength 214 is read (S1609). If the correlation value 214 of each correlation related to the normal data and the average value of each correlation strength 214 are similar (for example, within the threshold range), and the correlation model of the frequency data 210i When the difference between the correlation strength and the average value of the correlation strength 214 is similar (for example, within the threshold range) (Yes in S1611), it can be determined that the observation data is normal.
- the frequency data 210i When the difference between the correlation strength of the correlation model and the average value of the correlation strength 214 is not approximate (for example, outside the threshold range) (No in S1611), the correlation model relating to the correlation strength is It can be determined that the sensing unit 201i shown is indicating an abnormal value. Therefore, the notification unit 223 notifies the information indicating the above together with the information related to the sensing unit 201i in which the correlation is broken (S1613). For example, at this time, the abnormality detection unit 221 notifies that there is an indication that an abnormality occurs in the frequency to be processed.
- standard of abnormality determination is not restricted to this.
- the magnitude relationship between the average value of the correlation strength 214 of the correlation model 213 generated from the frequency data 208 which is normal data and the absolute value of the maximum value of the difference between the correlation strength 214 of each correlation model 213 good.
- the model construction unit 219 and the abnormality detection unit 221 process all the frequency data 210 while appropriately incrementing i until the value becomes n.
- the abnormality detection unit 221 may list up the sensing units 201 in which an abnormality has occurred after finishing the processing for all the frequency data 210. As a result, it is possible to narrow down the sensing units 201 which may have an abnormality.
- the average value of the correlation strength may be compared with the correlation strength of the correlation model generated from the frequency data 210 which is observation data related to the combination of the same sensing unit 201.
- the information processing system 100 uses the frequency data 208 to average the sensing unit 201 with time series data in which the characteristic of change appears in the frequency detected by the sensing unit 201.
- the correlation with the frequency data 228 is modeled by the model construction unit 211 and the model construction unit 219.
- the correlation since the correlation is specified not as a correlation coefficient but as a correlation model, it is possible to determine normality or abnormality of sensor data from a change in the correlation. That is, in the information processing system 100 according to the present embodiment, data analysis can be suitably performed.
- the entire configuration of the information processing system 100 whose specific example is shown in FIG. 2 and the configuration of the computer 1000 capable of realizing the signal conversion module 103, the information processing device 105, and the information processing device 107 are the same as those in the first embodiment. Therefore, the explanation is omitted.
- normal data is generated from time-series data acquired from the sensing units 201a to 201k
- observation data is generated from time-series data acquired from the sensing units 201k + 1 to 201n.
- normal data and observation data are acquired from the same sensing units 201a to 201n at different timings.
- the correlation between the average value and each frequency data 208 is performed. Calculate the strength of the model and the correlation. This process will be described with reference to FIG. 17. First, for the frequency data relating to the sensors A to D shown by the graph, the average frequency data 228 of the broken line shown in the center of the left part is generated. Then, a correlation model 214 is generated between the average frequency data 228 and the frequency data 208 of the sensors A to D. Also, calculate the correlation strengths F A-Ave to F D-AVE regarding them.
- the frequency data shown in the right part of FIG. 17 and the average frequency data 228 are generated based on the time series data detected by the sensors B and C at timing different from the time series data that is the origin of the frequency data.
- a correlation model 214 is generated between them. Also, the correlation strengths F B '-Ave to F C'- AVE regarding them are calculated.
- the correlation strengths F A-AVE to F D-AVE of the correlation model between the average frequency data and the frequency data relating to the normal data substantially match (for example, the difference is less than or equal to the threshold).
- the correlation strength between the average frequency data 228 of normal data and the frequency data related to the observation data is significantly different (eg, the difference exceeds a threshold) FB'-AVE or F C'-AVE
- the information processing system 100 determines that an abnormality has occurred with respect to the sensor B or the sensor C.
- the system configuration can be the same as that of the first embodiment and the second embodiment, so the description will be omitted here.
- the functional configuration of the information processing system 100 is shown in FIG. As shown in FIG. 18, the functional configuration of the information processing system 100 according to the present embodiment is basically the same as that of the second embodiment, but as described above, it is the frequency data 208 and observation data which are normal data. It differs from the second embodiment in that the frequency data 210 is acquired from the same sensing unit 201.
- the sensing unit 201 detects vibration, light, sound, and the like to generate and output time-series data.
- the sensing unit 201 detects data at least twice or more of a normal state and a state in which it is unclear whether the state is normal or not (error detection target).
- the number of sensing units 201 may be any number as long as it is two or more.
- the noise filtering unit 203 removes noise from the time series data output from the sensing unit 201 as in the first embodiment.
- the frequency conversion unit 205 converts the time-series data, which are respectively detected by the sensing unit 201 and from which noises have been removed, into frequency data 208 and frequency data 210.
- the frequency data 208 and 210 are, for example, data different in detection timing, and the frequency data 208 is normal data converted from time-series data in a normal state.
- the frequency data 210 is observation data (data of an abnormality detection target) converted from time-series data in a state where it is unclear whether or not it is normal.
- the frequency data 208 and 210 are stored in the storage units 207 and 209, respectively.
- the average frequency data calculation unit 225 generates average frequency data 228 which is average data of normal data by calculating the average of the frequency data 208 for each frequency.
- the generated average frequency data 228 is stored in the storage unit 227.
- the model construction unit 211 generates correlation models 213a to 213n between the frequency data 208a to 208n which is normal data and the average frequency data 228, respectively.
- the generated correlation model 213 is stored in the storage unit 212.
- the model construction unit 211 also calculates the correlation strengths 214 a to 214 n of the respective correlation models 213, and also stores them in the storage unit 212.
- the model construction unit 211 determines the correlation model 213 and the correlation strength relating to the combination of each sensing unit 201 and the average frequency data 228. 214 can be generated for the number of times.
- the correlation strength average / maximum deviation calculation unit 217 calculates the maximum deviation while obtaining the average value of the correlation strengths 214 generated for a plurality of times.
- the maximum deviation corresponds to the maximum value of the difference between the average value of the correlation strength 214 and the correlation strength of each correlation model 213 used to calculate the average.
- the said largest deviation can be memorize
- the correlation strength average / maximum deviation calculation unit 217 is unnecessary.
- the model construction unit 219 generates a correlation model between each of the frequency data 210 to 210 n as observation data and the average frequency data 228, and calculates the strength of the correlation of each correlation model.
- the abnormality detection unit 221 determines the correlation strength 214 of the correlation model 213 generated based on the frequency data 208 which is normal data, and the correlation strength of the correlation model generated based on the frequency data 210 which is observation data. To detect abnormalities by comparing More specifically, for example, although the correlation strength 214 of each correlation model 213 generated based on the frequency data 208 is similar, the correlation of the correlation model generated based on the frequency data 210 In the case where the strength of is not in the approximate range, the abnormality detection unit 221 may determine that an abnormality has occurred in the sensing unit 201 related to the correlation model.
- the notification unit 223 notifies the user of the result of the abnormality detection by the abnormality detection unit 221.
- a method of notification by the notification unit 223 for example, a method of displaying a message on the display 111 can be considered.
- the frequency data 208 and 210 to be processed by the model construction unit 211 and the model construction unit 219 are data acquired from the same sensing unit 201 at different timings.
- the point is different from the second embodiment. However, since the other points are almost the same as those of the second embodiment, the description will be omitted here.
- the information processing system 100 uses the frequency data 208 to average the sensing unit 201 with time series data in which the characteristic of change appears in the frequency detected by the sensing unit 201.
- the correlation with the frequency data 228 is modeled by the model construction unit 211 and the model construction unit 219.
- the correlation since the correlation is specified not as a correlation coefficient but as a correlation model, it is possible to determine normality or abnormality of sensor data from a change in the correlation. That is, in the information processing system 100 according to the present embodiment, data analysis can be suitably performed.
- the entire configuration of the information processing system 100 whose specific example is shown in FIG. 2 and the configuration of the computer 1000 capable of realizing the signal conversion module 103, the information processing device 105, and the information processing device 107 are the same as those in the first embodiment. Therefore, the explanation is omitted.
- the abnormality is detected by comparing the strength of the correlation of the correlation model according to the normal data with the strength of the correlation of the correlation model according to the observation data.
- No correlation model is generated for observed data.
- a correlation model between each sensing unit 201 is generated from frequency data 208 which is normal data, and then observation values are applied to the correlation model to generate prediction values, and the prediction values and actual values of observation data It is detected whether it is abnormal or not by comparing. That is, an abnormality is detected using a prediction error.
- a prediction error As in the information processing system according to the present embodiment, in the case of a method of detecting abnormality using a prediction error, not only the difference of the whole frequency data but also the prediction error is large by displaying the prediction error of each frequency band. It is also possible to specify a frequency band.
- sensor data in which abnormality occurrences are concentrated can be narrowed down as the root cause of the abnormality.
- the method of determining the abnormality determination by the prediction error it is possible to set the value of the prediction error used for the abnormality determination as a rule, with reference to the prediction error in the normal data.
- the method of detecting abnormality with prediction errors related to multiple data can detect minute signs of abnormality as data with a larger range of increase and decrease in the normal state as compared to the case where abnormality is determined with a single data threshold. It is.
- the information processing system 100 includes sensing units 201 a to 201 n, a noise filtering unit 203, a frequency conversion unit 205, storage units 207 and 209, a model construction unit 211, and a storage unit.
- 212 includes an abnormality detection unit 221 and a notification unit 223.
- the sensing unit 201 detects vibration, light, sound, and the like to generate and output time-series data.
- the sensing unit 201 detects data at least twice or more of the normal state and the unknown state (state of abnormality detection target) as to whether the normal state or not.
- the number of sensing units 201 may be any number as long as it is two or more.
- the noise filtering unit 203 removes noise from time-series data output from the sensing unit 201 as in the first to third embodiments.
- the frequency conversion unit 205 converts the time-series data, which are respectively detected by the sensing unit 201 and from which noises have been removed, to the frequency data 208 and the frequency data 210, respectively.
- the frequency data 208 and 210 are, for example, data different in detection timing, and the frequency data 208 is normal data converted from time-series data in a normal state.
- the frequency data 210 is observation data (data of an abnormality detection target) converted from time-series data in a state where it is unclear whether or not it is normal.
- the frequency data 208 and 210 are stored in the storage units 207 and 209, respectively.
- the model construction unit 211 generates correlation models 213a to 213m from two combinations of frequency data 208a to 208n which are normal data.
- the generated correlation model 213 is stored in the storage unit 212.
- the model construction unit 211 also causes the storage unit 212 to store the correlation strength 214 of each of the correlation models 213 as well.
- the abnormality detection unit 221 is different from the frequency data 210 stored in the storage unit 209 (the frequency data 208 which is normal data has a different detection timing by the sensing unit 201 that detected the original data, and is an observation target for abnormality detection).
- An abnormality is detected by applying the correlation model 213 stored in the storage unit 212 to data (corresponding to data). More specifically, for example, with respect to the correlation model 213 generated from the frequency data 208i and the frequency data 208j, the frequency data 210i, the value excluding the frequency f 0 and the value of each frequency of the frequency data 208j are input. Thus, the predicted value of the frequency data 208i at the frequency f 0 can be obtained.
- the abnormality detection unit 221 can detect an abnormality according to whether the predicted value and the measured value at the frequency f 0 of the frequency data 208 i exceed the abnormality determination threshold 215 or the like. In addition, the said process can be performed with respect to all the frequencies of the frequency data 210, and the combination of all the frequency data 210. FIG.
- the notification unit 223 notifies the user of the result of the abnormality detection by the abnormality detection unit 221.
- a method of notification by the notification unit 223 for example, a method of displaying a message or the like on the display 111 can be considered.
- FIGS. 20 and 21 are flowcharts showing the flow of processing of the information processing system 100.
- the normal time-series data detected by the sensing unit 201 is converted into frequency data 208 by the frequency conversion unit 205 (S 2001), and the frequency data 208 is stored in the storage unit 207.
- the model construction unit 211 generates a correlation model 213 for each combination of frequency data 208a to 208n (S2003).
- the time-series data of the detection target newly detected by the sensing unit 201 is converted into frequency data 210 by the frequency conversion unit 205 (S2005), and stored in the storage unit 209.
- the abnormality detection unit 221 calculates a prediction value by applying the correlation model 213 to the frequency data 210, and performs abnormality detection processing from the difference between the prediction value and the actual measurement value (frequency data 210) (S2007). .
- each processing of S2001, S2003 and S2005 is similar to the processing of S501, S503 and S505 described with reference to FIG. 5 in the first embodiment.
- the process of S2007 will be described with reference to FIG.
- the abnormality detection unit 221 processes the frequency data 210 that is the target of abnormality detection. First, the abnormality detection unit 221 sets i and j to 1 and then sets the frequency data 210i and 210j (frequency data generated from time-series data extracted by the i-th and j-th sensing units 201, respectively). 210 is extracted from the storage unit 209 (S1201 and S2103) If i and j are equal (Yes in S2105), the value of j is incremented and the frequency data 210j is extracted again (S2103).
- the correlation model 213 generated from the frequency data 208i and 208j related to the sensing units 201i and 201j is read from the storage unit 212 (S1207), and the correlation model 213 is used. Then, predicted values at each frequency are calculated (S2109). Further, the abnormality detection unit 221 calculates the difference R between the predicted value and the actual measurement value for each frequency (S2111). If the difference R exceeds the abnormality determination threshold 215 (Yes in S2113), the abnormality detection unit 221 notifies the notification unit 223 that there is an indication that an abnormality occurs in the frequency or a prediction error in each frequency band ( S2115).
- the notification unit 223 displays a numerical value on the display 111 may be cited, or a histogram of an actual measurement value and a prediction value or a difference between them (that is, prediction error) It is also conceivable to display on the display 111 as a graph.
- the abnormality detection unit 221 performs the above-described process on all the combinations of frequency data 210 while appropriately incrementing i and j until the value becomes n.
- the abnormality detection unit 221 After the abnormality detection unit 221 has processed the combination of all the frequency data 210, it is assumed that the correlation where the difference R exceeds the abnormality determination threshold 215 (that is, the correlation is broken due to the occurrence of an abnormality). A large number of sensing units 201 included in the expected combination) may be listed. As a result, it is possible to narrow down the sensing units 201 which may have an abnormality.
- the abnormality detection unit 221 may calculate, as an abnormality score, the sum total of the number of correlation models 213 determined to be abnormal for each frequency, and may notify the notification unit 223. With this function, the degree of abnormality determination for each frequency is clearly indicated in the entire sensing unit 201 used, so that it is possible to give the user information for checking the frequency and the abnormal phenomenon.
- the information processing system 100 uses the frequency data 208 for the time-series data in which the feature of the change appears in the frequency detected by the sensing unit 201, and Model correlations.
- the correlation is specified as the correlation model 213 instead of the correlation coefficient, it is possible to determine whether the sensor data is normal or abnormal from the change in the correlation. . That is, in the information processing system 100 according to the present embodiment, data analysis can be suitably performed.
- the abnormality is detected by applying the frequency data 210 to the correlation model, but in the present embodiment, the frequency data is detected.
- a correlation model 213 is generated between the average frequency data 228 and the frequency data 208.
- the average frequency data 230 of the frequency data 210 which is observation data is generated, and then the average frequency data 230 and the frequency data 210 are applied to the correlation model 213 to generate a predicted value of the frequency data 210.
- the abnormality detection unit 221 detects whether or not there is an abnormality based on the difference between the predicted value and the frequency data 210.
- the information processing system 100 includes an average frequency data calculation unit 225 and an average frequency data calculation unit 229 in addition to the components included in the information processing system 100 according to the fourth embodiment. Furthermore, the storage units 207 and 209 store the average frequency data 228 and 229 calculated by the average frequency data calculation units 225 and 229, respectively.
- the operations of the sensing unit 201, the noise filtering unit 203, the frequency conversion unit 205, and the notification unit 223 are the same as in the fourth embodiment, and thus the description thereof is omitted here.
- the average frequency data calculation unit 225 generates average frequency data 228 which is average data of normal data by calculating the average of the frequency data 208 for each frequency.
- the calculated average frequency data 228 is stored in the storage unit 207.
- the model construction unit 211 generates correlation models 213a to 213n between the frequency data 208a to 208n which are normal data and the average frequency data 228, respectively.
- the generated correlation model 213 is stored in the storage unit 212.
- the average frequency data calculation unit 229 generates the average frequency data 230 which is the average data of the normal data by calculating the average of the frequency data 208 for each frequency in the same manner as the average frequency data 228.
- the calculated average frequency data 230 is stored in the storage unit 209.
- the abnormality detection unit 221 detects an abnormality by applying the frequency data 210 and the average frequency data 230 stored in the storage unit 209 to the correlation model 213. More specifically, for example, for the correlation model 213i generated from the frequency data 208i and the average frequency data 228, the value of each frequency of the average frequency data 230 and the value of the frequency data 208j excluding the frequency f 0 by entering, it is possible to find the prediction value of the frequency data 210i in the frequency f 0.
- the abnormality detection unit 221 can detect an abnormality according to whether the predicted value and the measured value at the frequency f 0 of the frequency data 210i exceed the abnormality determination threshold 215 or the like.
- the processing can be performed on all frequencies of the frequency data 210 and all frequency data 210.
- the normal time series data detected by the sensing unit 201 is converted into frequency data 208 by the frequency conversion unit 205 (S 2301), and the frequency data 208 is stored in the storage unit 207.
- the average frequency data calculation unit 225 calculates average frequency data 228 which is the average data of each frequency data 208 stored in the storage unit 207 (S2303).
- the model construction unit 211 generates a correlation model 213 between each frequency data 208 and the average frequency data 228 (S2305).
- the time-series data of the detection target newly detected by the sensing unit 201 is converted into frequency data 210 by the frequency conversion unit 205 (S2305), and stored in the storage unit 209.
- the average frequency data calculation unit 229 calculates average frequency data 230 which is the average data of each frequency data 210 stored in the storage unit 209 (S2309).
- the abnormality detection unit 221 applies the correlation model 213 to the combination of the frequency data 210 and the average frequency data 230 to calculate a predicted value, and from the difference between the predicted value and the actual measured value (frequency data 210), An abnormality detection process is performed (S2311).
- processing of S2301 and S2307 is similar to the processing described with reference to FIG. 6 in the first embodiment. Also, the process of S2305 is similar to the process described with reference to FIG. 15 in the second embodiment.
- the abnormality detection unit 221 sets i to 1 and reads the average frequency data 230 (S2401) and reads the frequency data 210i from the storage unit 209 (S2403). Further, the abnormality detection unit 221 reads a correlation model 213i generated from the frequency data 208i and the average frequency data 228 according to the same sensing unit 201i as the frequency data 210i (S2405). The abnormality detection unit 221 calculates the prediction value of the frequency data 210i at each frequency by applying the average frequency data 230 and the frequency data 210i to the read correlation model 213i (S2407). Further, the abnormality detection unit 221 calculates the difference R between the predicted value and the actual measurement value of the frequency data 210i for each frequency (S2409). If the difference R exceeds the abnormality determination threshold 215i (Yes in S2411), the abnormality detection unit 221 notifies the notification unit 223 that there is an indication that an abnormality occurs in the frequency or a prediction error in each frequency band ( S2415).
- various methods can be considered as a notification method by the notification unit 223, for example, it is also conceivable to display a notification on the display 111, or a histogram of an actual measurement value and a prediction value or a difference between them (that is, prediction error). It is also conceivable to display on the display 111 as a graph.
- the abnormality detection unit 221 performs the above-described process on all the frequency data 210 while appropriately incrementing the value of i until it matches n.
- the abnormality detection unit 221 may calculate, as an abnormality score, the sum total of the number of correlation models 213 determined to be abnormal for each frequency, and may notify the notification unit 223. With this function, the degree of abnormality determination for each frequency is clearly indicated in the entire sensing unit 201 used, so that it is possible to give the user information for checking the frequency and the abnormal phenomenon. (5.4 Effects according to the present embodiment)
- the information processing system 100 uses the frequency data 208 to average the sensing unit 201 with time series data in which the characteristic of change appears in the frequency detected by the sensing unit 201.
- the correlation with frequency data 228 is modeled.
- the correlation is specified as the correlation model 213 instead of the correlation coefficient, it is possible to determine whether the sensor data is normal or abnormal from the change in the correlation. . That is, in the information processing system 100 according to the present embodiment, data analysis can be suitably performed.
- the program of the present invention may be a program that causes a computer to execute each operation described in each of the above embodiments.
- Converting means for converting a plurality of time series data respectively obtained by detection by a plurality of sensors into first frequency data, and first frequency data relating to at least two of the plurality of sensors using the first frequency data
- Information processing system comprising: first model generation means for generating a correlation model; first calculation means for calculating the strength of correlation of the first correlation model; and determination means for determining abnormality based on the strength of correlation .
- the first model generation means uses the average frequency data which is an average of a plurality of first frequency data related to the plurality of sensors and the first frequency data of one of the plurality of first frequency data.
- the second model generation unit generates the second correlation model using average frequency data which is an average of a plurality of first frequency data related to the plurality of sensors and one second frequency data. Information processing system as described.
- Appendix 7 The information processing system according to any one of appendices 4 to 6, wherein time series data relating to the first frequency data and time series data relating to the second frequency data are different in detection timing by the plurality of sensors. .
- Appendix 9 The information processing system according to any one of appendices 1 to 8, wherein a threshold value used for abnormality determination is generated by applying frequency data used for generation of the first correlation model to the first correlation model.
- a conversion unit that converts a plurality of time series data obtained by detection by a plurality of sensors into first frequency data, and a correlation model using first frequency data relating to at least two of the plurality of sensors.
- second frequency data obtained by applying, to the correlation model, second frequency data obtained by converting another time-series data obtained from the sensor according to the correlation model.
- An information processing system comprising: determination means for determining abnormality based on a difference between a predicted value and an actual measurement value of the second frequency data.
- model generation unit generates the correlation model using a combination of first frequency data of two sensors among the plurality of sensors.
- the model generation means uses first average frequency data, which is an average of a plurality of first frequency data related to the plurality of sensors, and one first average frequency data of the plurality of first frequency data.
- first average frequency data which is an average of a plurality of first frequency data related to the plurality of sensors, and one first average frequency data of the plurality of first frequency data.
- the determination means is obtained by applying, to the correlation model, second frequency data relating to the correlation model and second average frequency data that is an average of a plurality of second frequency data relating to the plurality of sensors.
- the first correlation model is generated using average frequency data that is an average of a plurality of first frequency data related to the plurality of sensors and first frequency data of one of the plurality of first frequency data.
- Appendix 18 Converting time series data obtained by detection by the plurality of sensors into second frequency data, generating a second correlation model using the second frequency data, and strength of correlation of the second correlation model The method according to any one of Appendices 15 to 17, further comprising the step of: calculating the abnormality based on comparison between the strength of the correlation of the first correlation model and the strength of the correlation of the second correlation model.
- Appendix 19 The information processing method according to appendix 18, wherein the second correlation model is generated using a combination of second frequency data relating to two of the plurality of sensors.
- the correlation model is generated using first average frequency data which is an average of a plurality of first frequency data related to the plurality of sensors, and the first average frequency data of one of the plurality of first frequency data. , The information processing method according to appendix 24.
- Appendix 28 The information processing method according to any one of appendices 24 to 27, wherein the threshold value used for abnormality determination is generated by applying the frequency data used for generating the correlation model to the correlation model.
- the first correlation model is generated using average frequency data that is an average of a plurality of first frequency data related to the plurality of sensors and first frequency data of one of the plurality of first frequency data.
- a process of converting time series data obtained by detection by the plurality of sensors into second frequency data, a process of generating a second correlation model using the second frequency data, and strength of correlation of the second correlation model The method according to any one of appendices 29 to 31, further comprising the step of: calculating the abnormality based on comparison between the strength of the correlation of the first correlation model and the strength of the correlation of the second correlation model. Section program.
- Appendix 34 The program according to Appendix 32, wherein the second correlation model is generated using average frequency data that is an average of a plurality of first frequency data related to the plurality of sensors and one second frequency data.
- Appendix 35 The program according to any one of Appendices 32 to 34, wherein the time series data relating to the first frequency data and the time series data relating to the second frequency data are different in detection timing by the plurality of sensors.
- Appendix 37 The program according to any one of appendices 29 to 36, wherein a threshold value used for abnormality determination is generated by applying the frequency data used for generating the first correlation model to the first correlation model.
- (Appendix 38) A process of converting a plurality of time series data obtained by detection by a plurality of sensors into first frequency data and a correlation model using first frequency data relating to at least two of the plurality of sensors. Generation processing, and a predicted value of the second frequency data obtained by applying, to the correlation model, second frequency data obtained by converting another time-series data obtained from a sensor related to the correlation model A program that causes a computer to execute processing for determining an abnormality based on a difference from an actual measurement value of the second frequency data.
- the correlation model is generated using first average frequency data which is an average of a plurality of first frequency data related to the plurality of sensors, and the first average frequency data of one of the plurality of first frequency data. , The program according to appendix 38.
- Appendix 42 The program according to any one of appendices 38 to 41, wherein a threshold value used for abnormality determination is generated by applying the frequency data used to generate the correlation model to the correlation model.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Quality & Reliability (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Testing And Monitoring For Control Systems (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
Description
図1乃至図10は、第1実施形態を説明するための図である。以下、これらの図を参照しながら、以下の流れに沿って本実施形態を説明する。まず、「1.1」で本実施形態におけるデータの解析方法の概要を説明する。その後、「1.2」で本実施形態における情報処理システムのシステム構成等の概要を、「1.3」で本実施形態における情報処理システムの機能構成の概要を説明する。「1.4」では、処理の流れを、具体例を交えながら説明する。「1.5」では、情報処理システムを実現可能なハードウェア構成の具体例を説明する。最後に、「1.6」で、本実施形態に係る効果などを説明する。
(1.1.1 概要)
企業情報システムやデータセンタ等の比較的大規模なシステムにおいては、種々のシステムの監視、制御、運用のための統合運用管理システムが提供されている。ITシステムの大規模化や、地震などの災害時の事故対策としての監視強化に伴い、物理データを扱うプラントや、製造現場における状態監視システムも規模が大きくなってきている。このようなシステムは、従来は管理者が手作業でルールを決めて運用してきたが、このように大規模化、複雑化が進むシステムや装置の運用は、日々困難さを増している。つまり、異常を検知するシステムの必要性が高まっている。
以下、図1や図2を参照しながら、より具体的に説明する。
まず、本実施形態に係る情報処理システム100のシステム構成の具体例を、図3を参照しながら説明する。図3に示す通り、具体例に係る情報処理システム100は、振動センサ101a乃至101n(以下、総称して振動センサ101と呼ぶこともある)と、信号変換モジュール103と、例えばパーソナルコンピュータやサーバ等のコンピュータである情報処理装置105及び107と、記憶媒体109と、ディスプレイ111とから構成される。
情報処理装置105は、デジタル信号となったセンサデータが変換された周波数データから相関モデルを生成し、記憶媒体109へと出力する。
続いて、図4を参照しながら、情報処理システム100の機能構成を説明する。図4に示す通り、本実施形態に係る情報処理システム100は、センシング部201a乃至201n(以下、総称してセンシング部201とも呼ぶ。)と、ノイズフィルタリング部203と、周波数変換部205と、記憶部207及び209と、モデル構築部211及び219と、記憶部212と、相関強さ平均・最大偏差算出部217と、異常検出部221と、通知部223とから構成される。ここで、センシング部201は図3の振動センサ101に相当する。ノイズフィルタリング部203及び周波数変換部205は図1の信号変換モジュール103に対応し、記憶部207及びモデル構築部211、相関強さ平均・最大偏差算出部217は図1の情報処理装置105に対応する。記憶部212は図1の記憶媒体109に相当し、記憶部209、異常検出部221は図1の情報処理装置107に相当する。通知部223は、ディスプレイ111に相当する。
以下、図5乃至図9を参照しながら、本実施形態に係る情報処理システム100の処理の流れを説明する。図5乃至図9は、情報処理システム100の処理の流れを示すフローチャートである。
まず、全体の処理の流れを、図5を参照しながら説明する。
以下、S501、S503、S507及びS509のそれぞれの処理の詳細を、図6乃至図9を参照しながら説明する。
図6を参照しながら、センシング部201が検知した時系列データの変換処理の流れを説明する。なお、図6に係るフローチャートは、図5のS501の処理に対応する。
なお、上述の処理は、異常検知の対象(観察データ)となる周波数データ210の生成時(図5のS505の処理に対応。)においても、同様となる。
続いて、図7及び図8を参照しながら、モデル構築に係る処理の流れを説明する。なお、図7に示すフローチャートは、図5のS503の処理に対応する。
続いて、S701に係る処理の詳細を、図8を参照しながら説明する。
モデル構築部211は、i及びjを、値がnとなるまで適宜インクリメントしつつ、上述の処理を全ての周波数データ208の組み合わせに対して行う。
図9を参照しながら、異常検知処理の流れを説明する。なお、図9に示すフローチャートは、図5のS507及びS509の処理に対応する。
以下、図10を参照しながら、上述してきた信号変換モジュール103や情報処理装置105、情報処理装置107をコンピュータ1000により実現する場合のハードウェア構成の一例を説明する。信号変換モジュール103や情報処理装置105、情報処理装置107の機能は、それぞれ別々のコンピュータとしても実現しても良いし、1台のコンピュータにより実現しても良い。また、4台以上のコンピュータにより実現することも可能である。
以上説明したように、本実施形態に係る情報処理システム100は、センシング部201で検知された周波数に変化の特徴が現れる時系列データに対して、周波数データ208を用いて、センシング部201間の相関関係を相関モデル213としてモデル化する。このように本実施形態に係る情報処理システム100では、相関関係を相関係数ではなく相関モデル213として特定しているため、相関関係の変化から、センサデータの正常又は異常を判定することができる。
つまり、本実施形態に係る情報処理システム100では、好適にデータ解析を行うことができる。
以下、第2実施形態について説明する。以下の説明において、第1実施形態と同一若しくは類似の構成に対しては同一の符号を付与するとともに、必要に応じて説明を省略する。また、第1実施形態と同一若しくは類似する作用効果を得られる場合にも、説明を省略する場合がある。
以下、図11を参照しながら、本実施形態に係る情報処理システム100の処理の概要を簡単に説明する。
システム構成は、第1実施形態と同様とすることができるので、ここでは説明を省略する。以下、図12を参照しながら、本実施形態に係る情報処理システム100の機能構成を説明する。図12に示す通り、本実施形態に係る情報処理システム100は、第1実施形態に係る情報処理システム100が有していた各構成に加えて、平均周波数データ算出部225と、平均周波数データ228を記憶する記憶部227とを更に有する。
なお、センシング部201において複数回の観察を行わない場合には、相関強さ平均・最大偏差算出部217は不要である。
以下、本実施形態に係る情報処理システム100の処理の流れを説明する。
まず、全体の処理の流れを、図5を参照しながら説明する。
センシング部201により検知(センシング)された正常な時系列データは、周波数変換部205によって周波数データ208へと変換され(S1301)、周波数データ208は記憶部207に記憶される。平均周波数データ算出部225は、記憶部207に記憶された各周波数データ208から周波数ごとに平均をとることにより平均周波数データ228を生成する(S1303)。モデル構築部211は、各周波数データ208と平均周波数データ228との間で相関モデル213を生成するとともに、各相関モデル213に係る相関の強さ214を算出する(S1305)。このとき、センシング部201が複数回検出した時系列データからそれぞれ生成された周波数データ208が複数回分ある場合には、相関強さ平均・最大偏差算出部217が、当該複数回分の相関の強さ214の平均値及び最大偏差(異常判定閾値215に相当)を算出してもよい。
図14及び図15を参照しながら、モデル構築に係る処理の流れを参照する。なお、図14に示すフローチャートは、図13のS1305に対応する。
続いて、S1401かかる処理の詳細を、図15を参照しながら説明する。
モデル構築部211は、iの値が、観察データを検出するセンシング部201の数であるkとなるまで適宜インクリメントしつつ、上記処理の全ての周波数データ208に対して行う。
図16を参照しながら、異常検知処理の流れを説明する。なお、図16に示すフローチャートは、図13のS1309及びS1311の処理に対応する。
モデル構築部219及び異常検出部221は、iを値がnとなるまで適宜インクリメントしつつ、全ての周波数データ210に対して処理を行う。
以上説明したように、本実施形態に係る情報処理システム100は、センシング部201で検知された周波数に変化の特徴が現れる時系列データに対して、周波数データ208を用いて、センシング部201と平均周波数データ228との間の相関関係をモデル構築部211やモデル構築部219でモデル化する。このように本実施形態に係る情報処理システム100では、相関関係を相関係数ではなく相関モデルとして特定しているため、相関関係の変化から、センサデータの正常又は異常を判定することができる。
つまり、本実施形態に係る情報処理システム100では、好適にデータ解析を行うことができる。
以下、第3実施形態について説明する。以下の説明において、第1実施形態や第2実施形態と同一若しくは類似の構成に対しては同一の符号を付すとともに、必要に応じて説明を省略する。また、第1実施形態や第2実施形態と同一若しくは類似する作用効果を得られる場合にも、説明を省略する場合がある。
以下、図17を参照しながら、本実施形態に係る情報処理システム100の処理の概要を簡単に説明する。
システム構成は、第1実施形態及び第2実施形態と同様とすることができるので、ここでは説明を省略する。情報処理システム100の機能構成を図18に示す。図18に示す通り、本実施形態に係る情報処理システム100の機能構成は、基本的には第2実施形態と同様であるが、前述の通り、正常データである周波数データ208及び観察データである周波数データ210が、同一のセンシング部201から取得している点で第2実施形態と異なる。
なお、センシング部201において複数回の観察を行わない場合には、相関強さ平均・最大偏差算出部217は不要である。
本実施形態に係る情報処理システム100では上述の通り、モデル構築部211及びモデル構築部219の処理対象となる周波数データ208及び210が、それぞれ同一のセンシング部201から異なるタイミングで取得したデータである点が第2実施形態と異なる。しかしながらその他の点はほぼ第2実施形態と同様であるため、ここでは説明を省略する。
以上説明したように、本実施形態に係る情報処理システム100は、センシング部201で検知された周波数に変化の特徴が現れる時系列データに対して、周波数データ208を用いて、センシング部201と平均周波数データ228との間の相関関係をモデル構築部211やモデル構築部219でモデル化する。このように本実施形態に係る情報処理システム100では、相関関係を相関係数ではなく相関モデルとして特定しているため、相関関係の変化から、センサデータの正常又は異常を判定することができる。
つまり、本実施形態に係る情報処理システム100では、好適にデータ解析を行うことができる。
以下、第4実施形態について説明する。以下の説明において、第1乃至第3実施形態と同一若しくは類似の構成に対しては同一の符号を付すと共に、必要に応じて説明を省略する。また、第1乃至第3実施形態と同一若しくは類似する作用効果を得られる場合にも、説明を省略する場合がある。
第1実施形態乃至第3実施形態では正常データに係る相関モデルの相関の強さと観察データに係る相関モデルの相関の強さとを比較することにより異常を検知していたが、本実施形態では、観察データに対しては相関モデルを生成しない。正常データである周波数データ208から、各センシング部201間の相関モデルを生成した上で、当該相関モデルに観察データを適用することにより予測値を生成し、当該予測値と観察データの実測値とを比較することにより、異常であるか否かを検知する。即ち、予測誤差を用いて異常を検知する。本実施形態に係る情報処理システムのように、予測誤差を用いて異常検知する方法であれば、各周波数帯の予測誤差を表示することにより、周波数データ全体の相違のみならず、予測誤差の大きい周波数帯を明示することも可能である。また、このような処理を複数のセンサ間に対して網羅的に実施することで、異常発生が集中しているセンサデータを、異常の根本原因として絞り込むことができる。
本実施形態に係る情報処理システム100のシステム構成を、図19を参照しながら説明する。図19に示す通り、具体例に係る情報処理システム100は、センシング部201a乃至201nと、ノイズフィルタリング部203と、周波数変換部205と、記憶部207及び209と、モデル構築部211と、記憶部212と、異常検出部221と通知部223とを含む。
以下、図20及び図21を参照しながら、本実施形態に係る情報処理システム100の処理の流れを説明する。図20及び図21は、情報処理システム100の処理の流れを示すフローチャートである。
まず、全体の処理の流れを、図20を参照しながら説明する。
センシング部201により検知された正常な時系列データは、周波数変換部205によって周波数データ208へと変換され(S2001)、周波数データ208は記憶部207に記憶される。モデル構築部211は、周波数データ208a乃至208nの各組み合わせに対して相関モデル213を生成する(S2003)。
図21を参照しながら、異常検知処理の流れを説明する。なお、図21に示すフローチャートは、図20のS2007の処理に対応する。
異常検出部221は、i及びjを値がnとなるまで適宜インクリメントしつつ、上述の処理を全ての周波数データ210の組み合わせに対して行う。
以上説明したように、本実施形態に係る情報処理システム100は、センシング部201で検知された周波数に変化の特徴が現れる時系列データに対して、周波数データ208を用いて、センシング部201間の相関関係をモデル化する。このように本実施形態に係る情報処理システム100では、相関関係を相関係数ではなく相関モデル213として特定しているため、相関関係の変化から、センサデータの正常又は異常を判定することができる。
つまり、本実施形態に係る情報処理システム100では、好適にデータ解析を行うことができる。
以下、第5実施形態について説明する。以下の説明において、第1乃至第4実施形態と同一若しくは類似の構成に対しては同一の符号を付すとともに、必要に応じて説明を省略する。また、第1乃至第4実施形態と同一又は類似する作用効果を得られる場合にも、説明を省略する場合がある。
第4実施形態では、周波数データ208の各組み合わせに対して相関モデルを生成した上で、当該相関モデルに周波数データ210を適用することにより異常を検知していたが、本実施形態では、周波数データ208の平均周波数データ228を生成した上で、当該平均周波数データ228と周波数データ208との間で相関モデル213を生成する。更に、観察データである周波数データ210の平均周波数データ230を生成した上で、当該平均周波数データ230と周波数データ210とを相関モデル213に適用することにより、周波数データ210の予測値を生成する。異常検出部221は、当該予測値と周波数データ210との差異に基づき、異常であるか否かを検知する。
本実施形態に係る情報処理システム100のシステム構成を、図22を参照しながら説明する。図22に示す通り、具体例に係る情報処理システム100は、実施形態4に係る情報処理システム100が有する各構成に加えて、平均周波数データ算出部225及び平均周波数データ算出部229を有する。更に、記憶部207及び209は、それぞれ、平均周波数データ算出部225及び229が算出する平均周波数データ228及び229を記憶する。
以下、図23及び図24を参照しながら、本実施形態に係る情報処理システム100の処理の流れを説明する。図23及び図24は、情報処理システム100の処理の流れを示すフローチャートである。
まず、全体の処理の流れを、図23を参照しながら説明する。
センシング部201により検知された正常な時系列データは、周波数変換部205によって周波数データ208へと変換され(S2301)、周波数データ208は記憶部207に記憶される。平均周波数データ算出部225は、記憶部207に記憶された各周波数データ208の平均データである平均周波数データ228を算出する(S2303)。モデル構築部211は、各周波数データ208と平均周波数データ228との間で相関モデル213を生成する(S2305)。
図24を参照しなら、異常検知処理の流れを説明する。なお、図24に示すフローチャートは、図23のS2311の処理に対応する。
異常検出部221は、iの値をnと一致するまで適宜インクリメントしながら上述の処理を全ての周波数データ210に対して行う。
(5.4 本実施形態に係る効果)
つまり、本実施形態に係る情報処理システム100では、好適にデータ解析を行うことができる。
なお、前述の実施形態の構成は、組み合わせたり或いは一部の構成部分を入れ替えたりしてもよい。また、本発明の構成は前述の実施形態のみに限定されるものではなく、本発明の要旨を逸脱しない範囲内において種々変更を加えてもよい。
複数のセンサによる検知によりそれぞれ得られる複数の時系列データを、それぞれ第1周波数データへと変換する変換手段と、前記複数のセンサのうちの少なくとも2つのセンサに係る第1周波数データを用いて第1相関モデルを生成する第1モデル生成手段と、前記第1相関モデルの相関の強さを算出する第1演算手段と、相関の強さに基づき異常を判定する判定手段とを備える情報処理システム。
前記第1モデル生成手段は、前記複数のセンサのうち2つのセンサに係る第1周波数データの組み合わせを用いて前記第1相関モデルを生成する、付記1記載の情報処理システム。
前記第1モデル生成手段は、前記複数のセンサに係る複数の第1周波数データの平均である平均周波数データと、当該複数の第1周波数データのうちの1つの第1周波数データとを用いて前記第1相関モデルを生成する、付記1記載の情報処理システム。
前記複数のセンサによる検知により得られる時系列データを第2周波数データへ変換する変換手段と、第2周波数データを用いて第2相関モデルを生成する第2モデル生成手段と、前記第2相関モデルの相関の強さを算出する第2演算手段と、をさらに備え、前記判定手段は、前記第1相関モデルの相関の強さと、前記第2相関モデルの相関の強さとの比較に基づき、異常を判定する、付記1乃至3のいずれか1項記載の情報処理システム。
前記第2モデル生成手段は、前記複数のセンサのうちの2つのセンサに係る第2周波数データの組み合わせを用いて前記第2相関モデルを生成する、付記4記載の情報処理システム。
前記第2モデル生成手段は、前記複数のセンサに係る複数の第1周波数データの平均である平均周波数データと、1つの第2周波数データとを用いて前記第2相関モデルを生成する、付記4記載の情報処理システム。
前記第1周波数データに係る時系列データと、前記第2周波数データに係る時系列データとは、前記複数のセンサによる検知タイミングが異なる、付記4乃至付記6のいずれか1項記載の情報処理システム。
前記第1周波数データに係る時系列データと、前記第2周波数データに係る時系列データとは、検知するセンサが異なる、付記4乃至付記6のいずれか1項記載の情報処理システム。
前記第1相関モデルに、当該第1相関モデルの生成に用いた周波数データを適用することにより、異常判定に用いる閾値を生成する、付記1乃至付記8のいずれか1項記載の情報処理システム。
複数のセンサによる検知により得られる複数の時系列データを、それぞれ第1周波数データへと変換する変換手段と、前記複数のセンサのうちの少なくとも2つのセンサに係る第1周波数データを用いて相関モデルを生成するモデル生成手段と、前記相関モデルに、当該相関モデルに係るセンサから得られる別の時系列データを変換して得られる第2周波数データを適用することにより得られる当該第2周波数データの予測値と、当該第2周波数データの実測値との差分に基づき異常を判定する判定手段とを備える情報処理システム。
前記モデル生成手段は、前記複数のセンサのうち2つのセンサに係る第1周波数データの組み合わせを用いて前記相関モデルを生成する、付記10記載の情報処理システム。
前記モデル生成手段は、前記複数のセンサに係る複数の第1周波数データの平均である第1平均周波数データと、当該複数の第1周波数データのうちの1つの第1平均周波数データとを用いて前記相関モデルを生成する、付記10記載の情報処理システム。
前記判定手段は、前記相関モデルに、当該相関モデルに係る第2周波数データと、前記複数のセンサに係る複数の第2周波数データの平均である第2平均周波数データとを適用することにより得られる第2周波数データの予測値と、当該第2周波数データの実測値との差分に基づき異常を判定する、付記12記載の情報処理システム。
前記相関モデルに、当該相関モデルの生成に用いた周波数データを適用することにより、異常判定に用いる閾値を生成する、付記10乃至請求項13のいずれか1項記載の情報処理システム。
複数のセンサによる検知によりそれぞれ得られる複数の時系列データを、それぞれ第1周波数データへと変換するステップと、前記複数のセンサのうちの少なくとも2つのセンサに係る第1周波数データを用いて第1相関モデルを生成するステップと、前記第1相関モデルの相関の強さを算出するステップと、相関の強さに基づき異常を判定するステップとを情報処理システムが行う情報処理方法。
前記複数のセンサのうち2つのセンサに係る第1周波数データの組み合わせを用いて前記第1相関モデルを生成する、付記15記載の情報処理方法。
前記複数のセンサに係る複数の第1周波数データの平均である平均周波数データと、当該複数の第1周波数データのうちの1つの第1周波数データとを用いて前記第1相関モデルを生成する、付記15記載の情報処理方法。
前記複数のセンサによる検知により得られる時系列データを第2周波数データへ変換するステップと、第2周波数データを用いて第2相関モデルを生成するステップと、前記第2相関モデルの相関の強さを算出するステップと、をさらに備え、前記第1相関モデルの相関の強さと、前記第2相関モデルの相関の強さとの比較に基づき、異常を判定する、付記15乃至付記17のいずれか1項記載の情報処理方法。
前記複数のセンサのうちの2つのセンサに係る第2周波数データの組み合わせを用いて前記第2相関モデルを生成する、付記18記載の情報処理方法。
前記複数のセンサに係る複数の第1周波数データの平均である平均周波数データと、1つの第2周波数データとを用いて前記第2相関モデルを生成する、付記18記載の情報処理方法。
前記第1周波数データに係る時系列データと、前記第2周波数データに係る時系列データとは、前記複数のセンサによる検知タイミングが異なる、付記18乃至付記20のいずれか1項記載の情報処理方法。
前記第1周波数データに係る時系列データと、前記第2周波数データに係る時系列データとは、検知するセンサが異なる、付記18乃至付記20のいずれか1項記載の情報処理方法。
前記第1相関モデルに、当該第1相関モデルの生成に用いた周波数データを適用することにより、異常判定に用いる閾値を生成する、付記15乃至付記22のいずれか1項記載の情報処理方法。
複数のセンサによる検知により得られる複数の時系列データを、それぞれ第1周波数データへと変換するステップと、前記複数のセンサのうちの少なくとも2つのセンサに係る第1周波数データを用いて相関モデルを生成するステップと、前記相関モデルに、当該相関モデルに係るセンサから得られる別の時系列データを変換して得られる第2周波数データを適用することにより得られる当該第2周波数データの予測値と、当該第2周波数データの実測値との差分に基づき異常を判定するステップとを情報処理システムが行う情報処理方法。
前記複数のセンサのうち2つのセンサに係る第1周波数データの組み合わせを用いて前記相関モデルを生成する、付記24記載の情報処理方法。
前記複数のセンサに係る複数の第1周波数データの平均である第1平均周波数データと、当該複数の第1周波数データのうちの1つの第1平均周波数データとを用いて前記相関モデルを生成する、付記24記載の情報処理方法。
前記相関モデルに、当該相関モデルに係る第2周波数データと、前記複数のセンサに係る複数の第2周波数データの平均である第2平均周波数データとを適用することにより得られる第2周波数データの予測値と、当該第2周波数データの実測値との差分に基づき異常を判定する、付記26記載の情報処理方法。
前記相関モデルに、当該相関モデルの生成に用いた周波数データを適用することにより、異常判定に用いる閾値を生成する、付記24乃至請求項27のいずれか1項記載の情報処理方法。
複数のセンサによる検知によりそれぞれ得られる複数の時系列データを、それぞれ第1周波数データへと変換する処理と、前記複数のセンサのうちの少なくとも2つのセンサに係る第1周波数データを用いて第1相関モデルを生成する処理と、前記第1相関モデルの相関の強さを算出する処理と、相関の強さに基づき異常を判定する処理とをコンピュータに実行させるプログラム。
前記複数のセンサのうち2つのセンサに係る第1周波数データの組み合わせを用いて前記第1相関モデルを生成する、付記29記載のプログラム。
前記複数のセンサに係る複数の第1周波数データの平均である平均周波数データと、当該複数の第1周波数データのうちの1つの第1周波数データとを用いて前記第1相関モデルを生成する、付記29記載の情報処理方法。
前記複数のセンサによる検知により得られる時系列データを第2周波数データへ変換する処理と、第2周波数データを用いて第2相関モデルを生成する処理と、前記第2相関モデルの相関の強さを算出する処理と、をさらに備え、前記第1相関モデルの相関の強さと、前記第2相関モデルの相関の強さとの比較に基づき、異常を判定する、付記29乃至付記31のいずれか1項記載のプログラム。
前記複数のセンサのうちの2つのセンサに係る第2周波数データの組み合わせを用いて前記第2相関モデルを生成する、付記32記載のプログラム。
前記複数のセンサに係る複数の第1周波数データの平均である平均周波数データと、1つの第2周波数データとを用いて前記第2相関モデルを生成する、付記32記載のプログラム。
前記第1周波数データに係る時系列データと、前記第2周波数データに係る時系列データとは、前記複数のセンサによる検知タイミングが異なる、付記32乃至付記34のいずれか1項記載のプログラム。
前記第1周波数データに係る時系列データと、前記第2周波数データに係る時系列データとは、検知するセンサが異なる、付記32乃至付記34のいずれか1項記載のプログラム。
前記第1相関モデルに、当該第1相関モデルの生成に用いた周波数データを適用することにより、異常判定に用いる閾値を生成する、付記29乃至付記36のいずれか1項記載のプログラム。
複数のセンサによる検知により得られる複数の時系列データを、それぞれ第1周波数データへと変換する処理と、前記複数のセンサのうちの少なくとも2つのセンサに係る第1周波数データを用いて相関モデルを生成する処理と、前記相関モデルに、当該相関モデルに係るセンサから得られる別の時系列データを変換して得られる第2周波数データを適用することにより得られる当該第2周波数データの予測値と、当該第2周波数データの実測値との差分に基づき異常を判定する処理とをコンピュータに実行させるプログラム。
前記複数のセンサのうち2つのセンサに係る第1周波数データの組み合わせを用いて前記相関モデルを生成する、付記38記載のプログラム。
前記複数のセンサに係る複数の第1周波数データの平均である第1平均周波数データと、当該複数の第1周波数データのうちの1つの第1平均周波数データとを用いて前記相関モデルを生成する、付記38記載のプログラム。
前記相関モデルに、当該相関モデルに係る第2周波数データと、前記複数のセンサに係る複数の第2周波数データの平均である第2平均周波数データとを適用することにより得られる第2周波数データの予測値と、当該第2周波数データの実測値との差分に基づき異常を判定する、付記40記載のプログラム。
前記相関モデルに、当該相関モデルの生成に用いた周波数データを適用することにより、異常判定に用いる閾値を生成する、付記38乃至請求項41のいずれか1項記載のプログラム。
101 :振動センサ
103 :信号変換モジュール
105 :情報処理装置
107 :情報処理装置
109 :記憶媒体
111 :ディスプレイ
201 :センシング部
203 :ノイズフィルタリング部
205 :周波数変換部
207 :記憶部
208 :周波数データ
209 :記憶部
210 :周波数データ
211 :モデル構築部
212 :記憶部
213 :相関モデル
214 :相関モデル
215 :異常判定閾値
217 :相関強さ平均・最大偏差算出部
219 :モデル構築部
221 :異常検出部
223 :通知部
225 :平均周波数データ算出部
227 :記憶部
228 :平均周波数データ
229 :平均周波数データ算出部
230 :平均周波数データ
1000 :コンピュータ
1001 :プロセッサ
1003 :メモリ
1005 :記憶装置
1007 :入力インタフェース
1009 :データインタフェース
1011 :通信インタフェース
1013 :表示装置
Claims (18)
- 複数のセンサによる検知によりそれぞれ得られる複数の時系列データを、それぞれ第1周波数データへと変換する変換手段と、
前記複数のセンサのうちの少なくとも2つのセンサに係る第1周波数データを用いて第1相関モデルを生成する第1モデル生成手段と、
前記第1相関モデルの相関の強さを算出する第1演算手段と、
相関の強さに基づき異常を判定する判定手段と
を備える情報処理システム。 - 前記第1モデル生成手段は、前記複数のセンサのうち2つのセンサに係る第1周波数データの組み合わせを用いて前記第1相関モデルを生成する、
請求項1記載の情報処理システム。 - 前記第1モデル生成手段は、前記複数のセンサに係る複数の第1周波数データの平均である平均周波数データと、当該複数の第1周波数データのうちの1つの第1周波数データとを用いて前記第1相関モデルを生成する、
請求項1記載の情報処理システム。 - 前記複数のセンサによる検知により得られる時系列データを第2周波数データへ変換する変換手段と、
第2周波数データを用いて第2相関モデルを生成する第2モデル生成手段と、
前記第2相関モデルの相関の強さを算出する第2演算手段と、
を更に備え、
前記判定手段は、前記第1相関モデルの相関の強さと、前記第2相関モデルの相関の強さとの比較に基づき、異常を判定する、
請求項1乃至請求項3のいずれか1項記載の情報処理システム。 - 前記第2モデル生成手段は、前記複数のセンサのうちの2つのセンサに係る第2周波数データの組み合わせを用いて前記第2相関モデルを生成する、
請求項4記載の情報処理システム。 - 前記第2モデル生成手段は、前記複数のセンサに係る複数の第1周波数データの平均である平均周波数データと、1つの第2周波数データとを用いて前記第2相関モデルを生成する、
請求項4記載の情報処理システム。 - 前記第1周波数データに係る時系列データと、前記第2周波数データに係る時系列データとは、前記複数のセンサによる検知タイミングが異なる、
請求項4乃至請求項6のいずれか1項記載の情報処理システム。 - 前記第1周波数データに係る時系列データと、前記第2周波数データに係る時系列データとは、検知するセンサが異なる、
請求項4乃至請求項6のいずれか1項記載の情報処理システム。 - 前記第1相関モデルに、当該第1相関モデルの生成に用いた周波数データを適用することにより、異常判定に用いる閾値を生成する、
請求項1乃至請求項8のいずれか1項記載の情報処理システム。 - 複数のセンサによる検知により得られる複数の時系列データを、それぞれ第1周波数データへと変換する変換手段と、
前記複数のセンサのうちの少なくとも2つのセンサに係る第1周波数データを用いて相関モデルを生成するモデル生成手段と、
前記相関モデルに、当該相関モデルに係るセンサから得られる別の時系列データを変換して得られる第2周波数データを適用することにより得られる当該第2周波数データの予測値と、当該第2周波数データの実測値との差分に基づき異常を判定する判定手段と
を備える情報処理システム。 - 前記モデル生成手段は、前記複数のセンサのうち2つのセンサに係る第1周波数データの組み合わせを用いて前記相関モデルを生成する、
請求項10記載の情報処理システム。 - 前記モデル生成手段は、前記複数のセンサに係る複数の第1周波数データの平均である第1平均周波数データと、当該複数の第1周波数データのうちの1つの第1平均周波数データとを用いて前記相関モデルを生成する、
請求項10記載の情報処理システム。 - 前記判定手段は、前記相関モデルに、当該相関モデルに係る第2周波数データと、前記複数のセンサに係る複数の第2周波数データの平均である第2平均周波数データとを適用することにより得られる第2周波数データの予測値と、当該第2周波数データの実測値との差分に基づき異常を判定する、
請求項12記載の情報処理システム。 - 前記相関モデルに、当該相関モデルの生成に用いた周波数データを適用することにより、異常判定に用いる閾値を生成する、
請求項10乃至請求項13のいずれか1項記載の情報処理システム。 - 複数のセンサによる検知によりそれぞれ得られる複数の時系列データを、それぞれ第1周波数データへと変換するステップと、
前記複数のセンサのうちの少なくとも2つのセンサに係る第1周波数データを用いて第1相関モデルを生成するステップと、
前記第1相関モデルの相関の強さを算出するステップと、
相関の強さに基づき異常を判定するステップと
を情報処理システムが行う情報処理方法。 - 複数のセンサによる検知により得られる複数の時系列データを、それぞれ第1周波数データへと変換するステップと、
前記複数のセンサのうちの少なくとも2つのセンサに係る第1周波数データを用いて相関モデルを生成するステップと、
前記相関モデルに、当該相関モデルに係るセンサから得られる別の時系列データを変換して得られる第2周波数データを適用することにより得られる当該第2周波数データの予測値と、当該第2周波数データの実測値との差分に基づき異常を判定するステップと
を情報処理システムが行う情報処理方法。 - 複数のセンサによる検知によりそれぞれ得られる複数の時系列データを、それぞれ第1周波数データへと変換する処理と、
前記複数のセンサのうちの少なくとも2つのセンサに係る第1周波数データを用いて第1相関モデルを生成する処理と、
前記第1相関モデルの相関の強さを算出する処理と、
相関の強さに基づき異常を判定する処理と
をコンピュータに実行させるプログラム。 - 複数のセンサによる検知により得られる複数の時系列データを、それぞれ第1周波数データへと変換する処理と、
前記複数のセンサのうちの少なくとも2つのセンサに係る第1周波数データを用いて相関モデルを生成する処理と、
前記相関モデルに、当該相関モデルに係るセンサから得られる別の時系列データを変換して得られる第2周波数データを適用することにより得られる当該第2周波数データの予測値と、当該第2周波数データの実測値との差分に基づき異常を判定する処理と
をコンピュータに実行させるプログラム。
Priority Applications (3)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
EP14842340.3A EP3045889B1 (en) | 2013-09-09 | 2014-03-11 | Information processing system, information processing method, and program |
US14/916,007 US10228994B2 (en) | 2013-09-09 | 2014-03-11 | Information processing system, information processing method, and program |
JP2015535335A JP6237774B2 (ja) | 2013-09-09 | 2014-03-11 | 情報処理システム、情報処理方法及びプログラム |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2013-185947 | 2013-09-09 | ||
JP2013185947 | 2013-09-09 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2015033603A1 true WO2015033603A1 (ja) | 2015-03-12 |
Family
ID=52628098
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2014/056334 WO2015033603A1 (ja) | 2013-09-09 | 2014-03-11 | 情報処理システム、情報処理方法及びプログラム |
Country Status (4)
Country | Link |
---|---|
US (1) | US10228994B2 (ja) |
EP (1) | EP3045889B1 (ja) |
JP (1) | JP6237774B2 (ja) |
WO (1) | WO2015033603A1 (ja) |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2017064855A1 (ja) * | 2015-10-13 | 2017-04-20 | 日本電気株式会社 | 構造物異常検知システム、構造物異常検知方法及び記録した記録媒体 |
WO2017064854A1 (ja) * | 2015-10-13 | 2017-04-20 | 日本電気株式会社 | 構造物異常検知装置、構造物異常検知方法、記録媒体及び構造物異常検知システム |
WO2019054433A1 (ja) * | 2017-09-14 | 2019-03-21 | 日本電気株式会社 | 診断装置、診断方法及びプログラム記憶媒体 |
JP2019098515A (ja) * | 2017-11-28 | 2019-06-24 | 先馳精密儀器(東莞)有限公司Techmark Precision Instrument Co.Ltd. | 刃具状態検査システム及び方法 |
WO2020166072A1 (ja) * | 2019-02-15 | 2020-08-20 | 日本電気株式会社 | 時系列データ処理方法 |
JP2021189717A (ja) * | 2020-05-29 | 2021-12-13 | 株式会社東芝 | プラント監視支援装置、方法及びプログラム |
JP7358791B2 (ja) | 2019-06-11 | 2023-10-11 | 中国電力株式会社 | プラント監視システムおよびプラント監視方法 |
Families Citing this family (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP6237774B2 (ja) * | 2013-09-09 | 2017-11-29 | 日本電気株式会社 | 情報処理システム、情報処理方法及びプログラム |
US10419877B2 (en) * | 2015-10-07 | 2019-09-17 | Samsung Electronics Co., Ltd. | Electronic apparatus and IoT device controlling method thereof |
US10191799B2 (en) * | 2016-12-29 | 2019-01-29 | Sandisk Technologies Llc | BER model evaluation |
US10585739B2 (en) * | 2017-04-28 | 2020-03-10 | International Business Machines Corporation | Input data correction |
US10769007B2 (en) * | 2018-06-08 | 2020-09-08 | Microsoft Technology Licensing, Llc | Computing node failure and health prediction for cloud-based data center |
JP7420144B2 (ja) * | 2019-10-15 | 2024-01-23 | 日本電気株式会社 | モデル生成方法、モデル生成装置、プログラム |
JP7481897B2 (ja) * | 2020-05-12 | 2024-05-13 | 株式会社東芝 | 監視装置、監視方法、プログラムおよびモデル訓練装置 |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPH06281547A (ja) * | 1993-03-26 | 1994-10-07 | Hitachi Ltd | 機器の異常監視方法および装置 |
JPH076882B2 (ja) * | 1990-10-31 | 1995-01-30 | 旭化成工業株式会社 | 異常検出方法および装置 |
JP3780299B1 (ja) * | 2005-06-24 | 2006-05-31 | 独立行政法人科学技術振興機構 | 対象設備の診断方法、コンピュータプログラム、及び、対象設備を診断するための装置 |
JP2010066244A (ja) | 2008-09-13 | 2010-03-25 | Chugoku Electric Power Co Inc:The | 設備異常診断方法およびシステム |
JP4872944B2 (ja) | 2008-02-25 | 2012-02-08 | 日本電気株式会社 | 運用管理装置、運用管理システム、情報処理方法、及び運用管理プログラム |
JP2012098149A (ja) * | 2010-11-02 | 2012-05-24 | Jfe Mechanical Co Ltd | 携帯型振動診断装置 |
Family Cites Families (18)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2812347B2 (ja) * | 1992-02-17 | 1998-10-22 | 日本電気株式会社 | 再同期復調装置 |
US5623609A (en) * | 1993-06-14 | 1997-04-22 | Hal Trust, L.L.C. | Computer system and computer-implemented process for phonology-based automatic speech recognition |
JP3321487B2 (ja) | 1993-10-20 | 2002-09-03 | 株式会社日立製作所 | 機器/設備診断方法およびシステム |
JP2000125177A (ja) * | 1998-10-12 | 2000-04-28 | Ricoh Co Ltd | 自動合焦装置 |
JP2001198498A (ja) * | 1999-11-08 | 2001-07-24 | Nissan Motor Co Ltd | 塗装ガンの異常検出装置およびその方法 |
US6594595B2 (en) * | 2001-04-03 | 2003-07-15 | Advantest Corporation | Apparatus for and method of measuring cross-correlation coefficient between signals |
JP3864722B2 (ja) * | 2001-04-26 | 2007-01-10 | 日産自動車株式会社 | 塗装ガンの異常検出装置およびその方法 |
US6909808B2 (en) * | 2002-03-08 | 2005-06-21 | Anzus, Inc. | Image compression to enhance optical correlation |
JP4942353B2 (ja) * | 2006-02-01 | 2012-05-30 | 株式会社ジェイテクト | 音又は振動の解析方法及び音又は振動の解析装置 |
KR100829870B1 (ko) * | 2006-02-03 | 2008-05-19 | 한국전자통신연구원 | 멀티채널 오디오 압축 코덱의 음질 평가 장치 및 그 방법 |
JP2008033532A (ja) * | 2006-07-27 | 2008-02-14 | Denso Corp | 可動部を備えた設備の異常を検出する方法及び異常検出装置 |
US8107631B2 (en) * | 2007-10-04 | 2012-01-31 | Creative Technology Ltd | Correlation-based method for ambience extraction from two-channel audio signals |
JP2009187293A (ja) * | 2008-02-06 | 2009-08-20 | Nec Corp | 時系列データ解析システム、方法およびプログラム |
US20100149073A1 (en) * | 2008-11-02 | 2010-06-17 | David Chaum | Near to Eye Display System and Appliance |
JPWO2011046228A1 (ja) | 2009-10-15 | 2013-03-07 | 日本電気株式会社 | システム運用管理装置、システム運用管理方法、及びプログラム記憶媒体 |
US8909641B2 (en) | 2011-11-16 | 2014-12-09 | Ptc Inc. | Method for analyzing time series activity streams and devices thereof |
JP6237774B2 (ja) * | 2013-09-09 | 2017-11-29 | 日本電気株式会社 | 情報処理システム、情報処理方法及びプログラム |
US20160161339A1 (en) * | 2014-12-05 | 2016-06-09 | Intel Corporation | Human motion detection |
-
2014
- 2014-03-11 JP JP2015535335A patent/JP6237774B2/ja active Active
- 2014-03-11 EP EP14842340.3A patent/EP3045889B1/en active Active
- 2014-03-11 US US14/916,007 patent/US10228994B2/en active Active
- 2014-03-11 WO PCT/JP2014/056334 patent/WO2015033603A1/ja active Application Filing
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPH076882B2 (ja) * | 1990-10-31 | 1995-01-30 | 旭化成工業株式会社 | 異常検出方法および装置 |
JPH06281547A (ja) * | 1993-03-26 | 1994-10-07 | Hitachi Ltd | 機器の異常監視方法および装置 |
JP3780299B1 (ja) * | 2005-06-24 | 2006-05-31 | 独立行政法人科学技術振興機構 | 対象設備の診断方法、コンピュータプログラム、及び、対象設備を診断するための装置 |
JP4872944B2 (ja) | 2008-02-25 | 2012-02-08 | 日本電気株式会社 | 運用管理装置、運用管理システム、情報処理方法、及び運用管理プログラム |
JP2010066244A (ja) | 2008-09-13 | 2010-03-25 | Chugoku Electric Power Co Inc:The | 設備異常診断方法およびシステム |
JP2012098149A (ja) * | 2010-11-02 | 2012-05-24 | Jfe Mechanical Co Ltd | 携帯型振動診断装置 |
Non-Patent Citations (1)
Title |
---|
See also references of EP3045889A4 |
Cited By (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US10641681B2 (en) | 2015-10-13 | 2020-05-05 | Nec Corporation | Structure abnormality detection system, structure abnormality detection method, and storage medium |
WO2017064854A1 (ja) * | 2015-10-13 | 2017-04-20 | 日本電気株式会社 | 構造物異常検知装置、構造物異常検知方法、記録媒体及び構造物異常検知システム |
JPWO2017064855A1 (ja) * | 2015-10-13 | 2018-08-02 | 日本電気株式会社 | 構造物異常検知システム、構造物異常検知方法及び記録した記録媒体 |
JPWO2017064854A1 (ja) * | 2015-10-13 | 2018-08-09 | 日本電気株式会社 | 構造物異常検知装置、構造物異常検知方法、記録媒体及び構造物異常検知システム |
US10697861B2 (en) | 2015-10-13 | 2020-06-30 | Nec Corporation | Structure abnormality detection device, structure abnormality detection method, storage medium, and structure abnormality detection system |
WO2017064855A1 (ja) * | 2015-10-13 | 2017-04-20 | 日本電気株式会社 | 構造物異常検知システム、構造物異常検知方法及び記録した記録媒体 |
WO2019054433A1 (ja) * | 2017-09-14 | 2019-03-21 | 日本電気株式会社 | 診断装置、診断方法及びプログラム記憶媒体 |
JP2019098515A (ja) * | 2017-11-28 | 2019-06-24 | 先馳精密儀器(東莞)有限公司Techmark Precision Instrument Co.Ltd. | 刃具状態検査システム及び方法 |
JPWO2020166072A1 (ja) * | 2019-02-15 | 2021-12-02 | 日本電気株式会社 | 時系列データ処理方法 |
WO2020166072A1 (ja) * | 2019-02-15 | 2020-08-20 | 日本電気株式会社 | 時系列データ処理方法 |
JP7218765B2 (ja) | 2019-02-15 | 2023-02-07 | 日本電気株式会社 | 時系列データ処理方法 |
JP7358791B2 (ja) | 2019-06-11 | 2023-10-11 | 中国電力株式会社 | プラント監視システムおよびプラント監視方法 |
JP2021189717A (ja) * | 2020-05-29 | 2021-12-13 | 株式会社東芝 | プラント監視支援装置、方法及びプログラム |
JP7391765B2 (ja) | 2020-05-29 | 2023-12-05 | 株式会社東芝 | プラント監視支援装置、方法及びプログラム |
Also Published As
Publication number | Publication date |
---|---|
US10228994B2 (en) | 2019-03-12 |
EP3045889B1 (en) | 2021-08-11 |
EP3045889A4 (en) | 2017-08-09 |
US20160196175A1 (en) | 2016-07-07 |
JPWO2015033603A1 (ja) | 2017-03-02 |
JP6237774B2 (ja) | 2017-11-29 |
EP3045889A1 (en) | 2016-07-20 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
JP6237774B2 (ja) | 情報処理システム、情報処理方法及びプログラム | |
JP6835251B2 (ja) | 損傷診断装置、損傷診断方法、及び、損傷診断プログラム | |
US20200259725A1 (en) | Methods and systems for online monitoring using a variable data | |
WO2015045319A1 (ja) | 情報処理装置、及び、分析方法 | |
WO2018008708A1 (ja) | 震央距離推定装置、震央距離推定方法、及びコンピュータ読み取り可能な記録媒体 | |
KR20200043196A (ko) | 진동 신호를 이용하여 설비 또는 부품의 잔여 수명을 예측하는 딥러닝 기반의 분석 장치 및 방법 | |
KR20190025474A (ko) | 플랜트 데이터 예측 장치 및 방법 | |
CN103090895A (zh) | 用于将来自传感器的数据与可视显示相关的检验系统和方法 | |
Tran et al. | Automated and model-free bridge damage indicators with simultaneous multiparameter modal anomaly detection | |
JP5283607B2 (ja) | コンクリート構造物の耐力評価方法及びコンピュータプログラム | |
JP6904418B2 (ja) | 情報処理装置、情報処理システム、情報処理方法、及び、プログラム | |
WO2021127646A1 (en) | Device and method for monitoring a system | |
CN104977922A (zh) | 设备监视装置和方法 | |
JP2010261754A (ja) | 常時微動計測に基づく建物の健全性診断法、診断装置及び診断プログラム | |
Diao et al. | Structural damage identification under variable environmental/operational conditions based on singular spectrum analysis and statistical control chart | |
Pham‐Bao et al. | Energy dissipation‐based material deterioration assessment using random decrement technique and convolutional neural network: A case study of Saigon bridge in Ho Chi Minh City, Vietnam | |
CN111596619B (zh) | 计算机系统及设施的监视方法 | |
US10866163B2 (en) | Anomaly monitoring device and method for producing anomaly signs according to combinations of sensors based on relationship of sensor fluctuations | |
US20230123872A1 (en) | Method for detection of anomolous operation of a system | |
Staffa et al. | Development and Validation of a Low-Cost Device for Real-Time Detection of Fatigue Damage of Structures Subjected to Vibrations | |
JP6247777B2 (ja) | 異常診断装置および異常診断方法 | |
JP2019095370A (ja) | 疲労損傷診断方法、疲労損傷診断システム及び疲労損傷診断プログラム | |
JP2017150887A (ja) | 構造物診断装置、構造物診断方法及びプログラム | |
Lenart et al. | Real-time monitoring and analyses of sensory data integrated into the BIM platform | |
JP6932467B2 (ja) | 状態変動検出装置、状態変動検出システム及び状態変動検出用プログラム |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 14842340 Country of ref document: EP Kind code of ref document: A1 |
|
ENP | Entry into the national phase |
Ref document number: 2015535335 Country of ref document: JP Kind code of ref document: A |
|
WWE | Wipo information: entry into national phase |
Ref document number: 14916007 Country of ref document: US |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
REEP | Request for entry into the european phase |
Ref document number: 2014842340 Country of ref document: EP |
|
WWE | Wipo information: entry into national phase |
Ref document number: 2014842340 Country of ref document: EP |