Facility state monitoring method and device for same

• US 9,659,250 B2
• Filed: 08/31/2011
• Issued: 05/23/2017
• Est. Priority Date: 08/31/2011
• Status: Active Grant

• Current Assignee: Hitachi Power Solutions Co., Ltd. (Hitachi, Ltd.)
• Original Assignee: Hitachi Power Solutions Co., Ltd. (Hitachi, Ltd.)
• Inventors: Shibuya, Hisae, Maeda, Shunji
• Primary Examiner(s): Hill, Stanley K
• Assistant Examiner(s): Bharadwaj, Kalpana
• Application Number:
  US14/241,780
• Publication Number:
  US 20140279795A1
• Time in Patent Office: 2,092 Days
• CPC Class Codes: G05B 2219/32201 Build statistical model of past normal proces, compare with actual process
• CPC Class Codes: G05B 23/0221 Preprocessing measurements, e.g. data collection rate adjustment; Standardization of measurements; Time series or signal analysis, e.g. frequency analysis or wavelets; Trustworthiness of measurements; Indexes therefor; Measurements using easily measured parameters to estimate parameters difficult to measure; Virtual sensor creation; De-noising; Sensor fusion; Unconventional preprocessing inherently present in specific fault detection methods like PCA-based methods
• CPC Class Codes: G05B 23/0254 based on a quantitative model, e.g. mathematical relationships between inputs and outputs; functions: observer, Kalman filter, residual calculation, Neural Networks
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 5/02 Knowledge representation; Symbolic representation
• CPC Class Codes: Y02P 90/02 Total factory control, e.g. smart factories, flexible manufacturing systems [FMS] or integrated manufacturing systems [IMS]

Abstract

In case-based anomaly indication detection in a facility, there are problems such as error generation due to insufficient learning data or execution difficulty due to increased memory capacity and calculation time when the learning data period has been increased to obtain the learning data sufficiently. Provided is a method for monitoring facility state on the basis of a time series signal outputted from the facility, wherein an operation pattern label for each fixed interval is assigned on the basis of the time series signal, learning data is selected on the basis of the operation pattern label for each fixed interval, a normal model is created on the basis of the selected learning data, an anomaly measure is calculated on the basis of the time series signal and the normal model, and the facility state is determined to be anomaly or normal on the basis of the calculated anomaly measure.

Claims

1. A facility state monitoring method of monitoring a state of a facility on the basis of a time series signal that the facility outputs, the method including the steps of:

assigning an operation pattern label for each fixed interval on the basis of the time series signal;

selecting learning data on the basis of the operation pattern label for each fixed interval;

creating a normal model on the basis of the selected learning data;

calculating an anomaly measure on the basis of the time series signal and the normal model; and

discriminating whether the state of the facility is anomaly or normal on the basis of the calculated anomaly measure;

wherein as the operation pattern label, different operation pattern labels are assigned to a steady off state of the facility, a steady on state of the facility, a start-up operation state of the facility and a shut down operation state of the facility.

2. The facility state monitoring method according to claim 1, wherein in the step of creating, the normal model is created for the operation pattern label for each fixed interval.

3. A facility state monitoring method of monitoring a state of a facility on the basis of a time series signal that the facility outputs, including the steps of:

assigning an operation pattern label which has been categorized to a finite number to the time series signal for each fixed interval;

accumulating the time series signal to which the operation pattern label has been assigned in the operation pattern label assigning step as data;

selecting a predetermined number of pieces of data from within the data accumulated in the data accumulating step on the basis of the operation pattern label assigned to the time series signal and setting them as learning data,

creating a normal model by using the learning data selected in the learning data selecting step;

calculating an anomaly measure of the time series signal on the basis of comparison with the normal model created in the normal model creating step; and

discriminating anomaly on the basis of the anomaly measure calculated in the anomaly measure calculating step.

4. The facility state monitoring method according to claim 3, wherein in the step of selecting, the predetermined number of pieces of data of operation pattern labels which are the same as or close in state to the operation pattern label assigned to the timed sequence signal is selected and set as the learning data.

5. The facility state monitoring method according to claim 3, wherein the step of assigning includes a sub-step of calculating a plurality of operation pattern features which include a steady off state of the facility, a steady on state of the facility, an start-up operation state of the facility and a shut down operation state of the facility for each fixed interval, and a sub-step of assigning the operation pattern label on the basis of a combination of the plurality of operation pattern features.

6. The facility state monitoring method according to claim 3, further including a step of calculating a feature indicating a macro-fluctuation of the time series signal for each fixed interval, wherein in the step of learning, the predetermined number of pieces of data is selected from within the accumulated data on the basis of the operation pattern label assigned to the time series signal and the calculated macro-feature and is set as the learning data.

7. The facility state monitoring method according to claim 6, wherein in the step of selecting, the predetermined number of pieces of data of operation pattern labels which are the same as or close in state to the operation pattern label assigned to the time series signal and similar in macro-feature relating to the time series signal is selected from within the accumulated data and set as the learning data.

8. The facility state monitoring method according to claim 6, wherein in the step of learning, presence/absence of a noticeable state change is determined on the basis of the time series signal, in a case where the noticeable state change is not present, the predetermined number of pieces of data of the operation pattern labels which are the same as or close in state to the operation pattern label assigned to the time series signal is selected from within the accumulated data and set as the learning data, and in a case where the noticeable stage change is present, the predetermined number of pieces of data of the operation pattern labels which are the same as or close in state to the operation pattern label assigned to the time series signal and similar in macro-feature relating to the time series signal is selected from within the accumulated data and set as the learning data.

9. The facility state monitoring method according to claim 6, wherein the macro-feature includes at least any one of a mean, a variance, a maximum value and a minimum value over a whole period, or a mean and a variance at steady ON times in the period, or a mean, a variance at steady OFF times in the period of sensor signals.

10. The facility state monitoring method according to claim 8, wherein in the step of learning, decision as to presence/absence of the noticeable state change is made on the basis of a change in the macro-feature.

11. A facility state monitoring device, comprising:

a sensor signal analysis unit that inputs and analyzes a time series signal output from a facility;

an anomaly diagnosis unit that receives a result of analysis by the sensor signal analysis unit and the time series signal and diagnoses anomaly of the facility; and

an input/output unit which is connected with the sensor signal analysis unit and the anomaly diagnosis unit to input and output data,

wherein the sensor signal analysis unit has an operation pattern label assigning sub-unit that assigns an operation pattern label for each fixed interval on the basis of the time series signal output from the facility, a learning data creation sub-unit that selects learning data on the basis of the operation pattern label which has been assigned by the operation pattern label assigning sub-unit for each fixed period, a normal model creation sub-unit that creates a normal model on the basis of the learning data created by the learning data creation sub-unit, an anomaly measure calculation sub-unit that calculates an anomaly measure of the time series signal output from the facility on the basis of the normal model created by the normal model creation sub-unit, and an anomaly discrimination sub-unit that performs discrimination as to whether the state of the facility is anomaly or normal on the basis of the anomaly measure calculated by the anomaly measure calculation sub-unit;

further comprising a database sub-unit that stores the time series signal to which the operation pattern label has been assigned by the operation pattern label assigning sub-unit, wherein the learning data creation sub-unit selects pieces of data of the same operation pattern label or operation pattern labels which are close in state from within data accumulated in the database sub-unit by a predetermined number on the basis of the operation pattern label which has been assigned to the time series signal by the operation pattern label assigning sub-unit and sets as the learning data.

12. A facility state monitoring device for monitoring a facility state by inputting a time series signal output from a facility, comprising:

an operation pattern label assigning unit that inputs the time series signal and assigns an operation pattern label categorized to a finite number to the externally input time series signal for each fixed interval;

a data accumulation unit that accumulate the time series signal to which the operation pattern label has been assigned by the operation pattern label assigning unit;

a learning data selection unit that selects pieces of data of the same operation pattern label or operation pattern labels which are close in state by a predetermined number from within the data accumulated in the data accumulation unit on the basis of the label assigned to the time series signal by the operation pattern label assigning unit and sets as learning data;

a normal model creation unit that creates a normal model using the learning data selected by the learning data selection unit;

an anomaly measure calculation unit that compares the time series signal to which the operation pattern label has been assigned by the operation pattern label assigning unit with the normal model created by the normal model creation unit and calculates an anomaly measure of the time series signal; and

an anomaly discrimination unit that performs discrimination of anomaly of the time series signal to which the operation pattern label has been assigned on the basis of the anomaly measure calculated by the anomaly measure calculation unit.

13. The facility state monitoring device according to claim 12, further comprising an input unit that inputs conditions for assigning the operation pattern label to the operation pattern label assigning unit.