Arcface Based Open Set Recognition for Industrial Fault

In industry, fault classification is important to avoid economic losses. Fault type classification is an important because fault detection and classification before equipment shutdown allows accurate maintenance. However, it is difficult to define all fault types in advance. It is impossible to know everything in advance what kind of fault will occur. Therefore, we propose an Arcface-based open set recognition method. We propose an algorithm that can classify a known fault type or an unknown fault type by fusion of a deep learning-based classification model and a distribution model that can estimate whether it is a known type. We apply the proposed method to the AHU dataset. The proposed model shows better performance compared to the existing methods.