US 12,229,638 B2
Learning assistance device, processing system, learning assistance method, and storage medium
Yuki Ueyama, Kyoto (JP); Nobuyuki Sakatani, Otsu (JP); Yasuaki Abe, Takatsuki (JP); Kazuhiko Imatake, Osaka (JP); and Takashi Fujii, Kyoto (JP)
Assigned to OMRON Corporation, Kyoto (JP)
Appl. No. 16/977,469
Filed by OMRON Corporation, Kyoto (JP)
PCT Filed Feb. 20, 2019, PCT No. PCT/JP2019/006185
§ 371(c)(1), (2) Date Sep. 2, 2020,
PCT Pub. No. WO2019/176480, PCT Pub. Date Sep. 19, 2019.
Claims priority of application No. 2018-047258 (JP), filed on Mar. 14, 2018.
Prior Publication US 2021/0049506 A1, Feb. 18, 2021
Int. Cl. G06N 20/00 (2019.01)
CPC G06N 20/00 (2019.01) 10 Claims
OG exemplary drawing
 
1. A learning assistance device for performing relearning on a processing part having a learned learning device which has undergone learning for generating a predetermined output from a predetermined input, the learning assistance device comprising one or more processors configured to:
assess an anomaly of the input based on a predetermined reference, an anomaly in the input comprising an input falling outside an input range which has been learned in the learning device; and
in the case where the one or more processors assess that the input is an anomalous input:
generate an output with respect to the anomalous input based on a predetermined physical model capable of simulating servo motor responses to position control commands;
output, by the processing part, the generated output, the generated output comprising a position control command;
generate an ideal output with respect to the anomalous input by performing feedback control on the generated output outputted by the processing part from the anomalous input, the ideal output having an accuracy greater than an accuracy of the generated output outputted by the processing part, wherein the ideal output is generated with respect to the anomalous input based on the predetermined physical model; and
perform relearning of the learning device by taking the anomalous input and the ideal output with respect to the anomalous input as additional learning data under a predetermined condition.