QUALITY ABNORMALITY DETECTION METHOD, QUALITY ABNORMALITY DETECTION DEVICE, AND QUALITY ABNORMALITY DETECTION PROGRAM

摘要

PROBLEM TO BE SOLVED: To accurately detect quality abnormalities.;SOLUTION: An acquisition process acquires time series data 12a for a sensor from a machine unit for each prescribed production unit of a product. A division process generates a division data group 12b by dividing the time series data 12a into multiple divisions. A setting process associates first additional information indicating there are no failures with a prescribed normal quality included in the division data group, and unconditionally associates second additional information indicating there are failures with a latest production included in the division data group. A grouping process forms two groups in which the normal quality and the latest production are mixed. A generation process generates a prediction model 12e for predicting additional information corresponding to input data on the basis of learning using one of the two groups. A calculation process calculates a probability 12fa for predicting that the additional information is the second additional information in the prediction results using the other of the two groups as input data. A determination process determines quality abnormality for the product on the basis of the probability.;SELECTED DRAWING: Figure 2A;COPYRIGHT: (C)2018,JPO&INPIT