Quality-related Process Monitoring based on Analytic Stationary Subspace Analysis and Kernel Projection To Latent Structures

With the rising demand for high-quality goods, both academia and business are paying close attention to quality-related process monitoring. Most industrial processes exhibit nonstationarity. Traditional quality-related monitoring methods like projection to latent Structures (PLS) are only able to handle stationary processes. In this article, a novel quality-related process monitoring method based on analytic stationary subspace analysis (ASSA) and kernel projection to latent structures (KPLS) is proposed. The method simultaneously takes the nonstationarity and nonlinearity of processes into account. First, The multivariate time series are divided into stationary and nonstationary sources using the ASSA method. Then, the KPLS model is used to establish the monitoring model with stationary sources as input to detect quality-related faults effectively. In order to validate the superiority of the proposed method, the well-known Tennessee Eastman process (TEP) is used. Experimental results demonstrate the exceptional performance of the proposed method.