Host: The Japan Society of Mechanical Engineers
Name : [in Japanese]
Date : June 01, 2022 - June 04, 2022
We have proposed methods to improve the discrimination accuracy without changing the current anomaly discrimination system. The effect of this method was confirmed for the discrimination system by the MT method used in the actual mass production processing machine. We found that the correct rate is improved by preprocessing the raw acceleration data (time series) that is converted into feature vectors and used for learning and discrimination with an optimized multi-band digital filter. Furthermore, optimizing the cutout range of acceleration data at the same time as the filter parameters greatly contributed to the improvement of the correct rate.
