Multivariate statistical process monitoring involves dimension reduction and latent feature extraction in large-scale processes typically incorporates all measured variables. However, involving variables without beneficial information may degrade performance. This study analyzes the effect of variable selection on principal component analysis (PCA) Then, it proposes a fault-relevant Bayesian inference-based distributed method for efficient fault detection isolation. First, optimal subset is...

Process monitoring is crucial for maintaining favorable operating conditions and has received considerable attention in previous decades. Currently, a plant-wide process generally consists of multiple operational units large number measured variables. The correlation among the variables complex results imperative but challenging such processes. With rapid advancement industrial sensing techniques, data with meaningful information are collected. Data-driven multivariate statistical (DMSPPM)...

Abstract Non-Markovian models of stochastic biochemical kinetics often incorporate explicit time delays to effectively model large numbers intermediate processes. Analysis and simulation these models, as well the inference their parameters from data, are fraught with difficulties because dynamics depends on system’s history. Here we use an artificial neural network approximate time-dependent distributions non-Markovian by solutions much simpler time-inhomogeneous Markovian models;...

Synthetic minority oversampling technique (SMOTE) has been widely used in dealing with the imbalance classification problem machine learning field. However, classical SMOTE implements by linear interpolation between adjacent class samples, which may fail to consider uneven distribution of samples. This article proposes a clustering (MC-SMOTE) method that involves samples improve performance. First, from are clustered into several clusters. Second, is performed clusters create new different...

Batch-end quality modeling is used to predict the by using batch measurements and generally involves a large number of predictor variables. However, not all variables are beneficial for prediction. Conventional multiway partial least squares (PLS) may function properly batch-end because many irrelevant This paper proposes an optimized sparse PLS (OSPLS) approach simultaneous prediction relevant-variable selection. The effect on quality-prediction performance analyzed, importance selection...

Large-scale processes have become common, and fault detection for such is imperative. This work studies the data-driven distributed local problem large-scale with interconnected subsystems develops a genetic algorithm (GA)-regularized canonical correlation analysis (CCA)-based scheme. For each subsystem, GA-regularized CCA first performed its all coupled systems, which aims to preserve maximum minimal communication cost. A CCA-based residual then generated, corresponding statistic...

Sensitive principal component analysis (SPCA) is proposed to improve the (PCA) based chemical process monitoring performance, by solving information loss problem and reducing nondetection rates of T2 statistic. Generally, components (PCs) selection in PCA-based subjective, which can lead poor performance. The SPCA method subsequently build a conventional PCA model on normal samples, index PCs reflect dominant variation abnormal observations, use these sensitive (SPCs) monitor process....

Industrial big data and complex process nonlinearity have introduced new challenges in plant-wide monitoring. This article proposes a local-global modeling distributed computing framework to achieve efficient fault detection isolation for nonlinear processes. First, stacked autoencoder is used extract dominant representations of each local unit establish the inner monitor. Second, mutual information (MI) determine neighborhood variables unit. Afterward, joint representation learning then...

Although principal component analysis (PCA) is widely used for chemical process monitoring, improvements in the selection of components (PCs) are still needed. Given that determination complicated and changing fault information not guaranteed using offline‐selected PCs, this study proposes a just‐in‐time reorganized PCA model objectively selects PCs online monitoring. The importance evaluated by kernel density estimation. indicating more varied then selected to reorganize model. most useful...

Industrial processes generally have various operation modes, and fault detection for such is important. This paper proposes a method that integrates variational Bayesian Gaussian mixture model with canonical correlation analysis (VBGMM-CCA) efficient multimode process monitoring. The proposed VBGMM-CCA maximizes the advantage of VBGMM in automatic mode identification superiority CCA local detection. First, applied to unlabeled historical data determine number modes cluster each mode. Second,...

Deep neural network (DNN) extracts hierarchical representations from process data and is promising for nonlinear monitoring. Obtaining meaningful generating efficient fault detection residual are the main challenges in DNN-based This study proposes a regularized deep correlated representation (RDCR) method that incorporates belief networks (DBNs) canonical correlation analysis (CCA) Hierarchical initially extracted using DBN to input output variables. Second, modeled through CCA characterize...

Dynamics and nonlinearity may exist in the time batch directions for processes, thereby complicating monitoring of these processes. In this article, we propose a two-dimensional deep correlated representation learning (2D-DCRL) method to achieve efficient fault detection isolation nonlinear Three-way historical data are first unfolded as two-way time-slice data. Second, stacked autoencoder based neural network is constructed characterize correlation among process variables. Considering that...

A modern batch process can be characterized by a large scale and multiple operation units, local fault detection for the key units of such is imperative. time-slice canonical correlation analysis (CCA)-based multivariate statistical monitoring scheme processes proposed. First, three-way data are unfolded into data. Second, CCA modeling performed at each time instant to explore between entire process. Then, residual generated statistics constructed. The discriminate both status type detected...

Conventional Bayesian fault diagnosis assumes that all measurements are available synchronously; however, this condition does not always hold in practical industry because a process can be characterized by multiple sampling or transmitting rates. This paper introduces system incorporating both historical and online information to address the asynchronous measurement problem. First, Expectation Maximization approach is utilized deal with measurements; second, incomplete handled through...

Successive batch processes generally involve within-batch and batch-to-batch correlations, monitoring of such is imperative. This paper proposes a multiobjective two-dimensional canonical correlation analysis (M2D-CCA)-based fault detection scheme to achieve efficient successive processes. First, three-way historical process data are unfolded into two-way time-slice data. Second, for each measurement, CCA performed between the current measurement previous measurements from both time...

This study develops a novel data-driven latent variable correlation analysis (LVCA) framework to achieve communication efficient distributed monitoring for industrial plant-wide processes. Process data of local unit are first projected into dominant subspace and residual characterize the within unit. Then, least absolute shrinkage selection operator is used determine variables from neighboring units that beneficial Thereafter, canonical performed between units. Finally, monitor established...

Soft sensors provide a means to reliably estimate unmeasurable variables, thereby playing prevalent role in formulating closed-loop control batch processes. In soft sensor development, enhancing quality-relevant information and eliminating quality-irrelevant are important. This study proposes neural network-based deep representation learning approach improve the sensing performance dynamic The structure of network is optimized layer-by-layer manner. First, given generally abundant predictor...