Apparatus and method for detecting abnormal sign

An abnormal sign detection apparatus comprising:

• a data acquisition portion including a first data acquisition portion configured to acquire first to nth string data each of which is made up of a plurality of monitoring items from an apparatus to be monitored at predetermined or arbitrary time intervals
• a unit space generation portion including
• a first data temporary storage configured to temporarily store acquired first to nth string data;
• a first calculation portion configured to calculate an average and a variation for each of the monitoring items corresponding each of the first to nth string data and calculate correlation information indicating a correlation between the monitoring items for each of the first to nth string data, by using first to nth string data stored in the first data temporary storage;
• and
• a first normalization portion configured to normalize each of the first to nth string data acquired by the first data acquisition portion using the average and the variation of each of the monitoring items corresponding each of the first to nth string data;
• a distance calculation portion including a first distance calculation portion configured to calculate a distance from the correlation information for each of normalized first to nth string data by carrying out a computation using the normalized first to nth string data and the correlation information corresponding to each of the normalized first to nth string data;
• and
• a probability calculation portion configured to calculate probabilities which correspond to respective calculated distances or less from the respective calculated distances and a certain probability distribution;
• wherein
• the data acquisition portion includes a second data acquisition portion configured to acquire a string data having the respective calculated probabilities as monitoring items at predetermined or arbitrary time intervals;
• the unit space generation portion includes
• a second data temporary storage configured to temporarily store each acquired string data;
• a second calculation portion configured to calculate an average and a variation for each of the monitoring items and second correlation information indicating a correlation between the monitoring items by using each string data stored in the second data storage;
• and
• a second normalization portion configured to normalize the string data acquired by the second data acquisition portion using the average and the variation of each of the monitoring items;
• a distance calculation portion includes a second distance calculation portion configured to calculate a second distance from the second correlation information for normalized string data by carrying out a computation using the normalized string data and the second correlation information;
• and
• the abnormal sign detection apparatus further comprises an abnormal sign decision portion configured to decide whether or not there is an abnormal sign in the apparatus to be monitored depending on whether or not calculated second distance falls within a confidence interval set in advance for the certain probability distribution.

An abnormal sign detection method comprising:

• acquiring string data made up of a plurality of monitoring items from an apparatus to be monitored at predetermined or arbitrary time intervals;
• storing each acquired string data in a data temporary storage temporarily;
• calculating an average and variation for each of the monitoring items by using each string data stored in the data temporary storage;
• calculating correlation information indicating a correlation between the monitoring items by using each string data stored in the data temporary storage;
• calculating an average and variation for each of the monitoring items and correlation information indicating a correlation between the monitoring items by using each string data stored in the data temporary storage;
• normalizing the acquired string data by using the average and the variation of each monitoring item;
• calculating a distance from the correlation information for normalized string data by carrying out a computation using the normalized string data and the correlation information;
• and
• deciding whether or not there is an abnormal sign in the apparatus to be monitored depending on whether or not calculated distance falls within a confidence interval set in advance for a certain probability distribution, wherein
• the calculating correlation information includes calculating an inverse matrix of a correlation coefficient matrix among the monitoring items as the correlation information, and
• the calculating a distance includes
• calculating Mahalanobis' distance by carrying out a computation using the normalized string data and the inverse matrix of the correlation coefficient matrix,
• calculating a Euclid distance by carrying out a computation using the normalized string data and a unit matrix corresponding in number to the monitoring items, and
• calculating the distance by dividing the Mahalanobis' distance by the Euclid distance.