A monitoring method based on stacked AutoEncoder considering sensor fault

Abstract

In the process of online monitoring based on multi-sensor machining, in view of the problem that the accuracy rate of the monitoring model decreases due to the sudden fault of a certain sensor, this paper proposes a method to rapidly train a new monitoring model, and takes tool condition monitoring as an example to verify the effectiveness of the method. In this paper, the stacked autoencoder (SAE) model is used to monitor the tool conditions during the machining process, and the accuracy rate is 99.9%; when a sensor fault occurs suddenly during the machining process, the monitoring model accuracy rate drops to 73.6%. In order to solve the problem of the monitoring model accuracy rate declining and prevent the monitoring process from being interrupted, the method proposed in this paper to quickly train a new monitoring model only takes 5s to train a new stacked autoencoder model. The average accuracy rate of the new monitoring model is 98.6%, and it can still accurately monitor tool conditions.