EQUIPMENT STATUS MONITORING METHOD, MONITORING SYSTEM, AND MONITORING PROGRAM

• US 20120271587A1
• Filed: 06/16/2010
• Published: 10/25/2012
• Est. Priority Date: 10/09/2009
• Status: Abandoned Application

• Current Assignee: Hitachi, Ltd.
• Original Assignee: Hitachi, Ltd.
• Inventors: Shibuya, Hisae, Maeda, Shunji
• Application Number:
  US13/500,932
• Publication Number:
  US 20120271587A1
• Time in Patent Office: Days
• US Class Current: 702/127
• CPC Class Codes: G05B 23/0229 knowledge based, e.g. expert systems; genetic algorithms

Abstract

Anomaly sign detection methods such as those found in plants cannot detect anomalies when relevant sensor information is not acquired, and while it is possible to detect anomalies through changes in sensor output when manual operations are performed, it is difficult to distinguish between anomalies such as those caused only by the sensor signal and actual anomalies which should be detected. The disclosed method uses event signals, which contain a signal based on the status of a unit unable to acquire sensor information and a signal based on human operations. An event sequence is extracted from an event signal outputted from a piece of equipment and grouped by clustering, then a frequency matrix is created for the alarms generated within a prescribed interval of an event sequence, and a prediction of alarms with a high probability of occurring for an event sequence is output on the basis of the frequency matrix.

Claims

1. An equipment status monitoring method detecting an anomaly on the basis of a time series sensor signal and event signal output from equipment or a device, comprising:

extracting event sequences from the event signal;

grouping the event sequences on the basis of similarity between the event sequences; and

detecting an anomaly using a result of the grouping of the event sequences.

2. An equipment status monitoring method detecting an anomaly on the basis of a time series event signal output from equipment or a device, comprising:

extracting event sequences from the event signal;

grouping the event sequences on the basis of similarity between the event sequences;

extracting an alarm from the event signal;

associating a group of the event sequences with the alarm to calculate frequency matrix;

grouping the event sequence observed on test on the basis of similarity with the learned event sequence; and

predicting occurrence of an alarm strongly associated with the group of the event sequences on the basis of the frequency matrix.

3. The equipment status monitoring method according to claim 2, further comprising:

extracting a feature vector on the basis of the sensor signal output from the equipment or the device to be monitored;

creating a normal state model on the basis of the feature vector, on learning;

calculating an anomaly measure by comparing the normal state model with the feature vector, on detecting of the anomaly; and

an anomaly is identified by comparing the anomaly measure with a preset threshold.

4. The equipment status monitoring method according to claim 2, further comprising:

performing mode division for each operating status on the basis of the event signal;

extracting the feature vector on the basis of the sensor signal output from the equipment or the device to be monitored;

creating a normal state model for each mode on the basis of the feature vector, on learning;

calculating an anomaly measure by comparing the normal state model with the feature vector, on detecting of the anomaly; and

identifying the anomaly by comparing the anomaly measure with a preset threshold.

5. The equipment status monitoring method according to claim 4, wherein the mode division includes:

inputting the event signal;

preliminarily designating start and finish events of a plurality of sequences; and

extracting a period in the sequence or between the sequences while sequentially searching for the start and finish events.

6. The equipment status monitoring method according to claim 2, wherein the creating the frequency matrix includes:

acquiring a result event by adding “

without occurrence”

to the alarm;

regarding the group of the event sequences as a cause event;

setting every element of the matrix to be zero;

examining the alarm generated in an interval until a preliminarily designated time has elapsed, for the event sequence;

counting the element at an intersection of the group to which the event sequence belongs with the generated alarm, if the generated alarm exists; and

counting the element at an intersection of the group to which the event sequence belongs with “

without occurrence”

, if no generated alarm exists.

7. The equipment status monitoring method according to claim 6, wherein the creating the frequency matrix includes:

preliminarily designating a plurality of times as the preset time; and

individually creating the frequency matrices corresponding to the respective times.

8. The equipment status monitoring method according to claim 7, further comprising estimating an alarm occurrence time using the frequency matrices corresponding to the respective plurality of times.

9. An equipment status monitoring method detecting an anomaly on the basis of a time series sensor signal and event signal output from equipment or a device, comprising:

performing mode division for each operating status on the basis of the event signal;

extracting the feature vector on the basis of the sensor signal;

creating a first normal state model for each mode on the basis of the feature vector;

calculating a first anomaly measure by comparing the first normal state model with the feature vector;

extracting an event sequences from the event signal;

grouping the event sequences on the basis of similarity between the event sequences;

extracting the event sequence having a significantly high anomaly measure on the basis of correlation between presence or absence of occurrence of the grouped event sequence and the first anomaly measure;

creating learned data by removing data in a prescribed period during which the event sequence having the significantly high anomaly measure has been occurred, from the feature vector;

creating a second normal state model for the each mode using the learned data;

calculating a second anomaly measure by comparing the second normal state model with the feature vector; and

identifying an anomaly by comparing the second anomaly measure with a preset threshold.

10. The equipment status monitoring method according to claim 9, further comprising:

presetting whether to allow the event sequence having the significantly high anomaly measure as an exception or not; and

canceling the anomaly determination by the anomaly identification in a prescribed period during which the event sequence set as the exception has been occurred.

11. An equipment status monitoring system, comprising:

equipment to be monitored that outputs a time series sensor signal and an event signal;

a mode division unit performing mode division for each operating status on the basis of the event signal;

a normal state model creation unit that extracts a feature vector on the basis of the sensor signal to create a normal state model;

an anomaly measure calculation unit calculating an anomaly measure by comparing the normal state model with the feature vector;

an anomaly identification unit identifying an anomaly by comparing the anomaly measure with a preset threshold;

an event sequence grouping unit that groups event sequences from the event signal on the basis of similarity of extraction;

a causality extraction unit that associates a group of the event sequences with an alarm extracted from the event signal, and calculates a frequency matrix; and

an anomaly prediction unit that groups the event sequence to be observed on the basis of similarity with the learned event sequence, and predicts occurrence of an alarm strongly associated with the observed event sequence on the basis of the frequency matrix.

12. An equipment status monitoring system, comprising:

equipment to be monitored that outputs a time series sensor signal and an event signal;

a mode division unit performing mode division for each operating status on the basis of the event signal;

a normal state model creation unit that extracts a feature vector on the basis of the sensor signal to create a normal state model;

an anomaly measure calculation unit calculating an anomaly measure by comparing the normal state model with the feature vector;

an anomaly identification unit identifying an anomaly by comparing the anomaly measure with a preset threshold;

an event sequence grouping unit that extracts event sequences from the event signal, and groups the event sequences on the basis of similarity;

a correlation calculation unit that calculates correlation between presence or absence of occurrence of the grouped event sequence and the anomaly measure, and extracts the event sequence having a significantly high anomaly measure; and

an anomaly identification exception setting unit setting whether to allow the event sequence having the significantly high anomaly measure as an exception of anomaly identification or not.

13. An equipment status analysis program causing a computer to execute:

a step of receiving, as an input, a time series event signal output from equipment or a device;

a step of extracting event sequences from the event signal;

a step of grouping the event sequences on the basis of similarity between the event sequences; and

a step of associating a group of the event sequences with an alarm extracted from the event signal, and calculating a frequency matrix.

14. An equipment status analysis program causing a computer to execute:

a step of receiving, as inputs, a time series sensor signal and event signal output from equipment or a device;

a step of performing mode division for each operating status on the basis of the event signal;

a step of extracting a feature vector on the basis of the sensor signal;

a step of creating a first normal state model for the each mode on the basis of the feature vector;

a step of calculating an anomaly measure by comparing the first normal state model with the feature vector;

a step of extracting an event sequences from the event signal;

a step of grouping the event sequences on the basis of similarity between the event sequences;

a step of calculating correlation between occurrence of a group of the event sequences and the anomaly measure;

a step of extracting a group of event sequences having a significantly high anomaly measure on the basis of the correlation;

a step of creating learned data by removing data in a prescribed period during which the event sequence having the significantly high anomaly measure has been occurred, from the feature vector; and

creating a second normal state model for the each mode using the learned data.