APPARATUS ANOMALY MONITORING METHOD AND SYSTEM

• US 20110276828A1
• Filed: 01/14/2009
• Published: 11/10/2011
• Est. Priority Date: 01/14/2009
• Status: Active Grant

• Current Assignee: Hitachi, Ltd.
• Original Assignee: Hitachi, Ltd.
• Inventors: Miwa, Toshiharu, Tamaki, Kenji
• Granted Patent: US 8,515,719 B2
• Time in Patent Office: Days
• US Class Current: 714/26
• 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

Abstract

There is provided a technique related to an apparatus anomaly monitoring method and capable of achieving a model which can detect an anomaly with high accuracy, monitoring with using the model, and others. In an apparatus anomaly monitoring system, a model creation module (2) creates an ensemble of models formed of predictive models of an objective variable (Y) for each explanatory variable (X) based on regression analysis with using a plurality of state data items (DS) measured from an apparatus (1) to be a target. With using this, a monitoring execution module (3) monitors a state of the apparatus (1) to detect the anomaly. More particularly, the explanatory variables (X) are categorized into a collinearity item (XA) and an independency item (XB) to create an individual model for each of the collinearity items (XA) with using the collinearity item (XA) and the independency item (XB). With using the ensemble of these models, a predicted value of the objective variable (Y), an error span between the predicted value and a measurement value, an ensemble error span, and others are computed.

Claims

1. An apparatus anomaly monitoring method with using information processing of a computer, the method of performing processes of, based on a plurality of state data items of a target apparatus obtained by measuring a state of the apparatus with a plurality of sensors, monitoring and judging an anomaly of the state of the apparatus, wherein

the method includes:

a first step of performing a model creating process of creating a model for the monitoring and judgment based on the plurality of state data items of the apparatus at a normal time; and

a second step of performing a monitoring execution process of inputting the plurality of state data items of the apparatus at a predetermined time unit, monitoring and judging the anomaly of the state of the apparatus with using the model, and, when the anomaly is detected, outputting detection information,

the first step includes;

a step of categorizing the plurality of state data items into an objective-variable data item and two or more explanatory-variable data items other than the objective-variable data item in regression analysis; and

a step of creating an individual model of predicting the objective-variable data item from the one explanatory-variable data item for each of the explanatory-variable data items as two or more individual models configuring an ensemble of the models to configure the ensemble of the models, and

the second step includes, for inputs of the plurality of state data items;

a step of computing an individual predicted value of the objective-variable data item for each of the individual models configuring the ensemble of the models with taking the explanatory-variable data item as an input;

a step of computing an individual error span between the predicted value and a measurement value of the objective-variable data item for each of the individual predicted values;

a step of computing an ensemble error span in combination with the plurality of error spans obtained for each of the explanatory-variable data items; and

a step of detecting the anomaly of the apparatus by comparing the ensemble error span and a threshold.

2. The apparatus anomaly monitoring method according to claim 1, wherein

the first step includes

a step of categorizing the two or more explanatory-variable data items into a collinearity data item with a large cross-correlation strength and an independency data item other than the collinearity data item, and,

in the step of creating the two or more individual models configuring the ensemble of the models, an individual model of predicting the objective-variable data item from the one collinearity explanatory-variable data item is created for each of the collinearity explanatory-variable data items.

3. The apparatus anomaly monitoring method according to claim 1, wherein

the method includes a third step of performing a process of diagnosing a content of the anomaly of the apparatus based on the detection information obtained when the anomaly is detected in the second step, and,

in the third step, a process is performed in which state data items causing the ensemble error span to exceed the threshold are ranked and specified as candidates, and information indicating the candidates of the cause is outputted.

4. The apparatus anomaly monitoring method according to claim 3, wherein

the method includes a fourth step of performing, based on the information indicating the candidates of the cause obtained in the third step, a process of computing a lifetime of a part of the apparatus corresponding to the data item of the cause and outputting information about the lifetime.

5. The apparatus anomaly monitoring method according to claim 4, wherein

the method includes a fifth step of performing, with using the information about the lifetime of the part obtained in the fourth step, a process of creating maintenance plan information containing a schedule for a maintenance operation for the part of the apparatus performed by a maintenance operator and outputting the maintenance plan information as a maintenance operation instruction.

6. The apparatus anomaly monitoring method according to claim 1, wherein,

in the step of computing the ensemble error span, the ensemble error span is an ensemble maximum value obtained as a maximum value among the plurality of individual error spans.

7. The apparatus anomaly monitoring method according to claim 1, wherein,

in the step of computing the ensemble error span, the ensemble error span is an ensemble average value obtained as an average value among the plurality of individual error spans.

8. The apparatus anomaly monitoring method according to claim 1, wherein,

in the step of computing the ensemble error span, the ensemble error span is a shape of distribution or a statistical quantile of the ensemble of the plurality of individual error spans.

9. The apparatus anomaly monitoring method according to claim 1, wherein,

in the step of computing the ensemble error span, the ensemble average value is an effective value obtained as a root mean square of the plurality of individual error spans.

10. The apparatus anomaly monitoring method according to claim 1, wherein,

in the second step, a normalized error span is computed by normalizing a magnitude of a value of each of the individual error spans, and then, the ensemble error span in combination with the plurality of normalized error spans is computed.

11. The apparatus anomaly monitoring method according to claim 1, wherein,

in the second step,

in the step of computing the ensemble error span, an ensemble average value obtained as an average value among the plurality of individual error spans and an ensemble maximum value obtained as a maximum value among the plurality of individual error spans are computed as the ensemble error span,

a differential value between the ensemble average value and the ensemble maximum value is computed,

when the differential value is smaller than a predetermined threshold, it is judged that a large-area anomaly related to a plurality of state data items of the apparatus occurs, and, when the differential value is larger than the predetermined threshold, it is judged that a local-area anomaly restricted to a part of the state data items of the apparatus occurs.

12. The apparatus anomaly monitoring method according to claim 1, wherein

either of a linear regression model or a non-linear model is used for each of the explanatory-variable data items as the plurality of individual models configuring the ensemble of the models, and, both of the linear regression model and the non-linear model are mixed to be used as the ensemble of the models.

13. The apparatus anomaly monitoring method according to claim 3, wherein,

in the third step, a process is performed in which parts of the apparatus corresponded to state data items causing the ensemble error span to exceed the threshold are ranked and specified as candidates, and information indicating the candidates is outputted.

14. The apparatus anomaly monitoring method according to claim 3, wherein,

in the third step, the state data item of the cause is specified by analyzing a causational relation among the plurality of state data items with using a cross-correlation function.

15. The apparatus anomaly monitoring method according to claim 14, wherein,

in the third step, the state data item of the cause is specified by examining a rank of change occurrence of each of the explanatory-variable data items in vicinity of a changing point of the objective-variable data item on a time axis.

16. An apparatus anomaly monitoring system with using information processing of a computer, the system of performing processes of, based on a plurality of state data items of a target apparatus obtained by measuring a state of the apparatus with a plurality of sensors, monitoring and judging an anomaly of the state of the apparatus, wherein

the system includes:

first means of performing a model creating process of creating a model for the monitoring and judgment based on the plurality of state data items of the apparatus at a normal time; and

second means of performing a monitoring execution process of inputting the plurality of state data items of the apparatus at a predetermined time unit, monitoring and judging the anomaly of the state of the apparatus with using the model, and, when the anomaly is detected, outputting detection information,

the first means perform;

a process of categorizing the plurality of state data items into an objective-variable data item and two or more explanatory-variable data items other than the objective-variable data item in regression analysis; and

a process of configuring an ensemble of the models by creating an individual model of predicting the objective-variable data item from the one explanatory-variable data item for each of the explanatory-variable data items as two or more individual models configuring the ensemble of the models, and

the second means perform, for inputs of the plurality of state data items;

a process of computing an individual predicted value of the objective-variable data item for each of the individual models configuring the ensemble of the models with taking the explanatory-variable data item as an input;

a process of computing an individual error span between the predicted value and a measurement value of the objective-variable data item for each of the individual predicted values;

a process of computing an ensemble error span in combination with the plurality of error spans obtained for each of the explanatory-variable data items; and

a process of detecting the anomaly of the apparatus by comparing the ensemble error span and a threshold.

17. The apparatus anomaly monitoring system according to claim 16, wherein

the first means perform

a process of categorizing the two or more explanatory-variable data items into a collinearity data item with a large cross-correlation strength and an independency data item other than the collinearity data item, and,

in the step of creating the two or more individual models configuring the ensemble of the models, an individual model of predicting the objective-variable data item from the one collinearity explanatory-variable data item is created for each of the collinearity explanatory-variable data items.