US 12,460,994 B2
Abnormality detection device
Atsushi Moribe, Kariya (JP)
Assigned to DENSO CORPORATION, Kariya (JP)
Filed by DENSO CORPORATION, Kariya (JP)
Filed on Jul. 29, 2021, as Appl. No. 17/388,505.
Application 17/388,505 is a continuation of application No. PCT/JP2019/050704, filed on Dec. 24, 2019.
Claims priority of application No. 2019-021084 (JP), filed on Feb. 7, 2019.
Prior Publication US 2021/0357727 A1, Nov. 18, 2021
Int. Cl. G06N 3/044 (2023.01); G01M 17/007 (2006.01); G05B 23/02 (2006.01); G06F 18/2133 (2023.01); G06F 18/214 (2023.01); G06N 3/045 (2023.01); G06N 3/08 (2023.01)
CPC G01M 17/007 (2013.01) [G05B 23/02 (2013.01); G06F 18/2133 (2023.01); G06F 18/214 (2023.01); G06N 3/044 (2023.01); G06N 3/045 (2023.01); G06N 3/08 (2013.01)] 17 Claims
OG exemplary drawing
 
1. An abnormality detection device comprising:
a signal acquisition portion configured to acquire learning target data and monitoring target data, the learning target data is data used for machine learning and the monitoring target data is data of a monitoring target;
a state observer generation portion configured to generate a state observer by using a variable in an input variable configuration;
a normal model generation portion configured to generate a threshold by combining a first state observation value obtained by input of the learning target data to the state observer and the learning target data and inputting a combined result to a competitive neural network;
an abnormality degree calculation portion configured to calculate an abnormality degree by combining a second state observation value obtained by input of the monitoring target data to the state observer and the monitoring target data and inputting a combined result to the competitive neural network; and
a determination portion configured to calculate a determination result by comparing the threshold with the abnormality degree, wherein
the abnormality degree is a difference between the learning target data and the second state observation value, which are input to the competitive neural network, and neuron weight data of a winning unit of the competitive neural network.