US 11,703,424 B2
System and method for training anomaly detection analytics to automatically remove outlier data
Dayu Huang, Rexford, NY (US); Frederick Wilson Wheeler, Niskayuna, NY (US); John Joseph Mihok, Schenectady, NY (US); and David C. Korim, Albany, NY (US)
Assigned to General Electric Company, Schenectady, NY (US)
Filed by General Electric Company, Schenectady, NY (US)
Filed on May 21, 2020, as Appl. No. 16/879,828.
Prior Publication US 2021/0364392 A1, Nov. 25, 2021
Int. Cl. G01M 99/00 (2011.01); G06N 20/00 (2019.01); G06N 5/04 (2023.01); G06F 11/07 (2006.01); G06F 17/10 (2006.01); H04L 9/40 (2022.01)
CPC G01M 99/005 (2013.01) [G06F 11/0748 (2013.01); G06F 17/10 (2013.01); G06N 5/04 (2013.01); G06N 20/00 (2019.01); H04L 63/1425 (2013.01)] 16 Claims
OG exemplary drawing
 
1. A method for detecting anomalies during operation of an asset to improve performance of the asset, the method comprising:
collecting, via a server, data relating to operation of the asset or a group of assets containing the asset, the data comprising normal and abnormal asset behavior of the asset or the group of assets containing the asset;
automatically removing, via an iterative algorithm programmed in the server that utilizes one or more inputs or outputs of an anomaly detection analytic, portions of the data containing the abnormal asset behavior to form a dataset containing only the normal asset behavior, wherein the iterative algorithm comprises a random sample consensus algorithm, and wherein the automatically removing comprises:
(a) randomly sampling the data relating to operation of the asset or the group of assets containing the asset;
(b) training a model of the anomaly detection analytic using the sampled data;
(c) evaluating the model using unsampled portions of the data;
(d) counting a number of inliers within the evaluated model, the inliers corresponding to data points confirming to the model;
(e) repeating (a) through (d) until the inliers within the model exceed a certain threshold; and
(f) outputting a trained dataset to the at least one anomaly detection analytic when the inliers within the model exceed the certain threshold, the trained dataset containing only data representative of the normal asset behavior;
training, via a computer-based model programmed in the server, the anomaly detection analytic using, at least, the dataset containing only the normal asset behavior; and,
applying, via the server, the anomaly detection analytic to the asset so as to monitor for anomalies during operation thereof.