Abstract
Many multivariate statistical analysis methods and their corresponding probabilistic counterparts have been adopted to develop process monitoring models in recent decades. However, the insightful connections between them have rarely been studied. In this study, a generalized probabilistic monitoring model (GPMM) is developed with both random and sequential data. Since GPMM can be reduced to various probabilistic linear models under specific restrictions, it is adopted to analyze the connections between different monitoring methods. Using expectation maximization (EM) algorithm, the parameters of GPMM are estimated for both random and sequential cases. Based on the obtained model parameters, statistics are designed for monitoring different aspects of the process system. Besides, the distributions of these statistics are rigorously derived and proved, so that the control limits can be calculated accordingly. After that, contribution analysis methods are presented for identifying faulty variables once the process anomalies are detected. Finally, the equivalence between monitoring models based on classical multivariate methods and their corresponding probabilistic graphic models is further investigated. The conclusions of this study are verified using a numerical example and the Tennessee Eastman (TE) process. Experimental results illustrate that the proposed monitoring statistics are subject to their corresponding distributions, and they are equivalent to statistics in classical deterministic models under specific restrictions.
Introduction
In modern industrial process control, monitoring systems are essential for detecting anomalies before they escalate into costly failures. Traditionally, engineers have relied on deterministic multivariate methods such as Principal Component Analysis (PCA) or Canonical Correlation Analysis (CCA). However, these methods suffer from two primary drawbacks: "noise blindness"—a lack of explicit accounting for measurement noise—and "fragmentation," where different models are required for random (independent) versus sequential (time-series) data.
This paper introduces the Generalized Probabilistic Monitoring Model (GPMM), a unified framework that bridges the gap between classical multivariate statistical analysis and probabilistic graphical models. GPMM provides a mathematically rigorous way to monitor complex industrial processes while subsuming popular methods like PCA, CCA, and Slow Feature Analysis (SFA) into a single, flexible architecture.
The GPMM Framework
The core of GPMM is a latent variable framework that models the relationship between system inputs ($X$) and outputs ($Y$). Unlike previous models that often required input and output latent variables to be identical, GPMM introduces distinct latent variables for inputs ($s_t$) and outputs ($z_t$), linked by a transition matrix.
This structure allows the model to capture complex correlations without forcing rigid constraints on the variables. One of the most significant properties of GPMM is its "super-model" nature: by applying specific parameter restrictions, it simplifies exactly into Probabilistic PCA, Probabilistic CCA, or Probabilistic Slow Feature Analysis (PSFA).
Parameter Estimation and Sequential Modeling
Because latent variables are unobserved, the Expectation-Maximization (EM) algorithm is employed to estimate model parameters. The paper provides distinct solutions for two scenarios:
- Random Case: Utilizing current samples for independent data.
- Sequential Case: Incorporating temporal dependencies using the entire data history via a forward-backward algorithm.
By incorporating temporal awareness, GPMM can distinguish between static setpoint changes and dynamic anomalies, a common failure point for standard PCA-based systems.
Monitoring Statistics and Fault Isolation
The authors derive exact probability distributions (Chi-square) for monitoring statistics equivalent to $T^2$ and $Q$. This allows for the calculation of precise control limits (e.g., 95% or 99% confidence thresholds). Once an anomaly is detected, the model utilizes contribution analysis methods (GDC and RBC) to pinpoint the specific variables responsible for the alarm.
Experimental Validation
The model was validated using numerical simulations and the Tennessee Eastman (TE) process, a world-standard benchmark for industrial process monitoring.
- Statistical Accuracy: Monte Carlo simulations confirmed that false alarm rates matched target levels, proving the accuracy of the derived control limits.
- Benchmark Performance: In the TE process, GPMM matched or exceeded specialized models like Dynamic PCA. For instance, in fault scenario IDV(5), GPMM achieved high detection rates and provided clear contribution plots for fault isolation.
Conclusion
GPMM successfully bridges the deterministic-probabilistic divide, proving that under certain conditions (such as zero noise), probabilistic monitoring statistics are numerically equivalent to classical deterministic ones. While currently limited by Gaussian assumptions, future work may integrate Kernel Density Estimation (KDE) to handle non-Gaussian distributions and focus on automated fault-tolerant control systems.
Summary updated