Multimode Continuous Processes Monitoring Based on Hidden Semi-Markov Model and Principal Component Analysis

Abstract

Several studies have applied the hidden Markov model (HMM) in multimode process monitoring. However, because the inherent duration probability density of HMM is exponential, which is inappropriate for modeling the multimode process, the performance of these HMM-based approaches is not satisfactory. As a result, the hidden semi-Markov model (HSMM), which integrated the mode duration probability into HMM, is combined with principal component analysis (PCA) to handle the multimode feature, named as HSMM-PCA. PCA is a powerful monitoring algorithm for the unimodal process, and HSMM specializes in mode division and identification. HSMM-PCA inherits the advantages of these two algorithms and hence it performs much better than the existing HMM-based approaches do. In addition, HSMM-PCA can detect the mode disorder fault, which challenges the most multimode approaches.

Soft sensor modelling of acrolein conversion based on hidden Markov model of principle component analysis and fireworks algorithm

“…In particular, these systems can be modelled as the Markov model . The HMM is a statistical model to describe a Markov process with hidden unknown parameters . In this paper, the HMM is applied to describe the soft measurement model of acrolein conversion in the second oxidation reactor.…”

A new monitoring method for the between‐mode transition of multimode processes

“…In order to guarantee the manufacturing safety and the quality of products, process monitoring has become a popular issue. Over the last two decades, data‐driven methods have been extensively studied . In particular, multivariate statistical process control (MSPC) methods such as principal component analysis (PCA) and partial least squares (PLS) are the typical models for online fault detection.…”

“…Over the last two decades, data-driven methods have been extensively studied. [1][2][3][4][5][6] In particular, multivariate statistical process control (MSPC) methods such as principal component analysis (PCA) [7] and partial least squares (PLS) [8,9] are the typical models for online fault detection. In practice, the same production line is often used to manufacture different kinds of products, so the working conditions or control objectives may change frequently.…”

Multimode processes monitoring based on hierarchical mode division and subspace decomposition

“…They can be used for multimode process modelling . The hidden semi‐Markov model (HSMM) was combined with PCA by using the mode affiliation information of the historical data, the mode shifting probability, and the mode duration probability for mode identification to address the multimode problem in process monitoring, which can detect the mode disorder fault . However, nonlinear and multimodal uncertainty characteristics should be considered.…”

“…[28,29] The hidden semi-Markov model (HSMM) was combined with PCA by using the mode affiliation information of the historical data, the mode shifting probability, and the mode duration probability for mode identification to address the multimode problem in process monitoring, which can detect the mode disorder fault. [30] However, nonlinear and multimodal uncertainty characteristics should be considered. Considering the interrelationship of each mode, monitoring according to the common and specific bases of two neighbouring modes was presented to reconstruct the transitional samples, which restricted the specific number of modes and the hard partition.…”