Provided is a trained artificial intelligence that is used for detecting non-stationarity of an object, and functions accurately even in the presence of ambient sound. This autoencoder is input with stationary vibration feature quantity data that is generated from stationary vibration data regarding stationary vibration including the vibration generated, when stationary, from a target object to be subjected to detection of the non-stationarity thereof on the basis of sound and that is regarding a feature quantity of the stationary vibration identified by the stationary vibration data, and the autoencoder is caused to output estimated stationary vibration feature quantity data. A loss function between the stationary vibration feature quantity data and the estimated stationary vibration feature quantity data is generated, and the autoencoder is caused to perform learning such that a difference therebetween is minimized. By repeating the above process, the autoencoder becomes a trained autoencoder.
展开▼