An Open-Set Classification Method With Small Samples for Rotating Machinery

With the development of artificial intelligence, deep learning (DL)-based methods for rotating machinery fault diagnosis are emerging. However, these data-driven DL methods rely heavily on a large amount of labeled data, which is often scarce in real industrial scenarios. Additionally, datasets with small samples make it challenging to cover the full range of fault types. With changing operating conditions over time, equipment constantly generates data of unknown fault types. Despite this, most studies are based on the closed-set assumption, where the training and test data label sets are identical. Conversely, the open-set assumption considers data of unknown classes. Small samples exacerbate the challenge of unknown classes. To solve this problem, this article proposes a small sample open-set fault classification (SSO-FC) method. It first makes full use of small samples by data matching. Then, a generic Siamese contrasting (GCS) structure is designed to discriminate whether the data pairs from the same class. Finally, an entropy and confidence based open-set discriminator (ECOD) is proposed to discriminate unknown classes. Experimental results on three datasets show that with small samples, the proposed method can achieve accurate fault classification and effectively discriminate unknown classes.