振動データ解析を目的とした深層ニューラルネットワークおよび その軸受故障識別への適用
概要
Condition-based maintenance (CBM) technology is required to perform maintenance on rotating machinery at the optimal time before deterioration and failure occurs. In this study, we describe a deep neural network (DNN) model and corresponding training method for analyzing vibration data to realize CBM for rotating machinery. To apply DNN models in order to analyze time-series data such as vibration data, we first need to optimize the parameters of the models by training them on a dataset consisting of data as well as annotations about anomalies contained in the data. In addition, when assigning annotations to time-series data, understanding quantitatively where the anomalies are contained in the data and the extent of those anomalies is crucial. In this research, we establish a DNN model and corresponding training method that can detect anomalies from time-series data, which are difficult to annotate. In our experiment, we evaluate the applicability of this method to bearing fault diagnosis using vibration data acquired by acceleration sensors attached to bearings.
