APPARATUS ANOMALY MONITORING METHOD AND SYSTEM
First Claim
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, whereinthe 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, andthe 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.
1 Assignment
0 Petitions
Accused Products
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.
-
Citations
17 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. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15)
-
-
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. - View Dependent Claims (17)
-
Specification