Zero-shot Concurrent Control Chart Pattern Recognition based on Ordinal Pattern Feature Analysis

As the manufacturing process becomes increasingly complex, the occurrence of concurrent abnormal control chart pattern (CCP) is becoming more common. However, collecting and labeling abnormal concurrent CCP samples in real industrial settings is challenging. Due to the lack of training samples of concurrent CCP, the existing CCPR methods are difficult to accurately identify concurrent CCP. In this paper, the idea of zero-shot learning (ZSL) is introduced into process control, and an intelligent recognition method for zero-shot concurrent CCP is proposed. An attribute description space (ADS) of CCP is established to realize the reasoning from single CCP to concurrent CCP. Additionally, the ordinal pattern (OP) feature considering the sequential relationship between CCP data is extracted. The ADS is embedded between the OP features and the sample labels. The attribute classifier associates OP features with CCP attributes. The results show that the proposed method can achieve zerosample concurrent CCPR without their samples.