US 12,405,604 B2
Fault prediction device and fault prediction method
Toshiyuki Syo, Tokyo (JP)
Assigned to RENESAS ELECTRONICS CORPORATION, Tokyo (JP)
Filed by RENESAS ELECTRONICS CORPORATION, Tokyo (JP)
Filed on Jun. 10, 2022, as Appl. No. 17/837,440.
Claims priority of application No. 2021-103286 (JP), filed on Jun. 22, 2021.
Prior Publication US 2022/0404822 A1, Dec. 22, 2022
Int. Cl. G05B 23/02 (2006.01)
CPC G05B 23/0254 (2013.01) [G05B 23/0283 (2013.01)] 10 Claims
OG exemplary drawing
 
1. A fault prediction device for predicting a fault of a target device whose deterioration state transitions with elapse of time, the fault prediction device comprising:
a plurality of autoencoders respectively corresponding to deterioration states of the target device, the plurality of autoencoders including i) a first autoencoder trained with state signals associated with a first deterioration state of the target device and ii) a second autoencoder trained with state signals associated with a second deterioration state of the target device, the second deterioration state representing a progressed deterioration state compared to the first deterioration state; and
a processor configured to:
acquire a state signal indicating a state of the target device;
input the state signal into the plurality of autoencoders;
compute a reconstruction loss for each of the plurality of autoencoders, the reconstruction loss being a difference between the state signal input into the respective autoencoder and an output signal reconstructed by the respective autoencoder;
compare the reconstruction loss computed by the first autoencoder to a first threshold corresponding to the first deterioration state;
determine that the target device exists in the first deterioration state based on determining that the reconstruction loss computed by the first autoencoder is within the first threshold;
compare the reconstruction loss computed by the second autoencoder to a second threshold corresponding to the second deterioration state when the reconstruction loss computed by the first autoencoder exceeds the first threshold; and
determine that the target device exists in the second deterioration state based on determining that the reconstruction loss computed by the second autoencoder is within the second threshold,
wherein the plurality of autoencoders are configured to evaluate deterioration states of the target device in a sequential manner such that each of the plurality of autoencoders determines whether the target device exists in a corresponding deterioration state based on variations in reconstruction loss values.