초록 ▼

A malfunction condition judgment apparatus (MCJA) that judges malfunction condition (MC) of an observed object based on a change of observed values. The MCJA acquires time series data for values of each of a plurality of variables; calculates, with respect to each of the variables, a statistic defin

A malfunction condition judgment apparatus (MCJA) that judges malfunction condition (MC) of an observed object based on a change of observed values. The MCJA acquires time series data for values of each of a plurality of variables; calculates, with respect to each of the variables, a statistic defining a probability density function of that variable at T1, based on the value of that variable at T1 and that statistic at a point of time prior to T1; calculates dissimilarity showing an extent of variation between the statistic calculated for each variable and a statistic of a criterial probability density function predetermined corresponding to that variable; and picks, out of the plurality of variables, a variable for which the calculated dissimilarity is larger than a predetermined reference value, as the variable by which MC of the observation object is detected.

대표청구항 ▼

What is claimed is: 1. A control method of a malfunction condition judgment apparatus that judges a malfunction condition of an observed object based on a change of observed values observed from the observation object, comprising the steps of: acquiring time series data for values of each of a plur

What is claimed is: 1. A control method of a malfunction condition judgment apparatus that judges a malfunction condition of an observed object based on a change of observed values observed from the observation object, comprising the steps of: acquiring time series data for values of each of a plurality of variables that has a plurality of observed values observed from the observation object as a value; calculating, with respect to each of the variables, a statistic defining a probability density function of that variable at a point of time (T1), by adding into a value where the statistic at the prior time to T1 is subtracted by a value obtained by multiplying a forgetting rate set corresponding to the variable at the prior time to said point of time, a value where a value of the variable at T1 is multiplied by said forgetting rate; calculating, with respect to each of the variables, dissimilarity showing an extent of variation between the statistic calculated for that variable and a statistic of a criterial probability density function predetermined corresponding to that variable; pick up, out of the plurality of variables, a variable in which the calculated dissimilarity is larger than a predetermined reference value, as the variable by which a malfunction condition of the observed object is detected; and setting a forgetting rate with respect to each of the variables in a manner that, in a case where an updating frequency at which that variable is updated in a unit time is higher, the forgetting rate is set lower than in a case where the updating frequency is lower.