US 12,014,279 B2
Anomaly detection using a non-mirrored dimensional-reduction model
Sari Andoni, Austin, TX (US); Udaivir Yadav, Austin, TX (US); and Tyler S. McDonnell, Austin, TX (US)
Assigned to SPARKCOGNITION, INC.
Filed by SparkCognition, Inc., Austin, TX (US)
Filed on Mar. 23, 2021, as Appl. No. 17/209,508.
Application 17/209,508 is a continuation of application No. 16/927,300, filed on Jul. 13, 2020.
Application 16/927,300 is a continuation of application No. 16/716,850, filed on Dec. 17, 2019, granted, now 10,733,512, issued on Aug. 4, 2020.
Prior Publication US 2021/0209477 A1, Jul. 8, 2021
Int. Cl. G06N 3/086 (2023.01); G06N 3/045 (2023.01); G06N 3/082 (2023.01); G06N 3/088 (2023.01); G06N 3/084 (2023.01)
CPC G06N 3/086 (2013.01) [G06N 3/045 (2023.01); G06N 3/082 (2013.01); G06N 3/088 (2013.01); G06N 3/084 (2013.01)] 26 Claims
OG exemplary drawing
 
1. A computer system comprising:
an interface to receive sensor data indicative of operation of one or more devices from a sensor distinct from the interface, wherein the sensor data includes status data indicative of uptime for the one or more devices; and
a processor configured to:
provide input data based on the sensor data to a dimensional-reduction model, wherein the dimensional-reduction model has a decoder portion and an encoder portion, wherein the decoder portion and the encoder portion are not mirrored;
obtain output data from the dimensional-reduction model responsive to the input data;
determine a particular reconstruction error indicating a difference between the input data and the output data;
perform a comparison of the particular reconstruction error to an anomaly detection criterion, wherein the dimensional-reduction model is configured to generate a reconstruction error that satisfies the anomaly detection criterion responsive to the input data corresponding to historical data associated with a time period prior to a historical device failure or a historical fault condition, the time period distinct from a particular time corresponding to the historical device failure or historical fault condition; and
generate an anomaly detection output for the one or more devices based on a result of the comparison during operation of the one or more devices, wherein the anomaly detection output selectively predicts a future anomaly based on the result of the comparison, wherein the anomaly detection output is generated responsive to a determination that an average value based on the particular reconstruction error is greater than a threshold specified by the anomaly detection criterion, and wherein the anomaly detection output identifies a corrective action to prevent the future anomaly responsive to predicting the future anomaly.