PredNet Based Sequence Image Disturbance Processing of Fused Magnesium Furnaces

Abstract

Disturbance processing is necessary for image-based deep learning of abnormal diagnosis for fused processes, e.g., fused magnesium furnace (FMF), since the disturbance of water mist, furnace body, and environment will inevitably affect the visual image relevant to the identification of working conditions. To address this issue, this paper proposes a new predictive neural network (PredNet)-based unsupervised learning method for sequence images processing of fused magnesium furnace. This method consists of a residual extraction of the original sequence images, a feature learning of disturbance via PredNets, and a single frame de-mean operation. Finally, the proposed method is compared to the one using original data and the one using residual extraction method using the collected sequence images from the furnace shell of a real FMF. The application results demonstrate the effectiveness of the proposed method.