A novel ternary-latent variable structure for monitoring of dynamic processes with multiple sampling rates

Abstract

The monitoring of quality-related aspects in multi-sampling rate dynamic processes has consistently been a research focus in recent years. This paper investigates a novel ternary-latent variable structure, which aims to provide a comprehensive explanation for the correlations among variables observed at different sampling rates. The structure incorporates three types of latent variables formed through a first-order Markov chain. The first type of latent variables is designed to capture dynamic information related to quality (RTQ), while the other two types can offer additional insights into the first type by focusing on information unrelated to quality (UTQ) and unrelated to process (UTP). Furthermore, an adaptive parameter training method, namely expectation maximization algorithm, is employed to obtain posterior estimates of each type of latent variable in an incomplete data collection. The study concludes by proposing a fault detection method based on the multi-sampling rate dynamic ternary-latent variable (MDTLV) model, which demonstrates superior monitoring performance compared to similar approaches in experimental evaluations.