Search

In this paper, a novel and effective stiction detection method is proposed by combining K-means clustering and the moving window approach. As a byproduct, the proposed stiction detection method off...

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 and a large number of measured variables. The correlation among the variables and units is complex and results in the imperative but challenging monitoring of such plant-wide processes. With the rapid advancement of industrial sensing techniques, process data with meaningful process information are collected. Data-driven multivariate statistical plant-wide process monitoring (DMSPPM) has become popular. The key idea of DMSPPM is first decomposing a plant-wide process into multiple subprocesses and then establishing a data-driven model for monitoring the process, in which process variable decomposition is important for guaranteeing the monitoring performance. In the current review, we first introduce the basics of multivariate statistical process monitoring and highlight the necessity of des...

Flare events in industrial processes are undesirable because of their negative economic and environmental impacts. Early warnings of flare events are needed in preventing occurrences of flaring, as...

Control of a steam-assisted gravity drainage (SAGD) process is a challenging task, because of the presence of various uncertainties, such as geological uncertainty and steam quality uncertainty. They often lead to constraint violations and performance degradation. In this work, a chance-constrained model predictive control (CCMPC) method is presented to generate a safe and optimal control strategy, considering the presence of uncertainties. A novel robust optimization method is applied to solve the chance-constrained optimization problem under general distribution of uncertainties. Two case studies are presented to demonstrate the proposed approach. Furthermore, the modeling of SAGD process is discussed, and the proposed robust optimization-based CCMPC is tested using a reservoir simulator (Petroleum Experts) of the SAGD process. The proposed approach reduces constraint violations that are due to uncertainties and achieves satisfactory performance.

For a nonstationary process which has a time-varying mean, a time-varying variance, or both, it can be difficult to detect incipient disturbances which may be hidden by the time-varying process variations. Besides, stationary and nonstationary characteristics may coexist in complex industrial processes which, however, have not been studied for process monitoring. In the present work, a triple subspace decomposition based dissimilarity analysis algorithm is developed to detect incipient abnormal behaviors in complex industrial processes with both stationary and nonstationary hybrid characteristics. The novelty is how to comprehensively separate the stationary and nonstationary process characteristics and describe them, respectively. First, a stationarity evaluation and separation strategy is proposed to decompose the data space into three subspaces, revealing the linear stationary process characteristics, the nonlinear stationary process characteristics, and the final nonstationary process characteristics....

Molecular reconstruction of petroleum fractions is for determining the detailed molecular compositions in the mixture from a few measurable bulk properties, e.g., density, Reid vapor pressure (RVP), molecular weight, and ASTM boiling point curves etc., which is a great challenge because the number of hydrocarbon compounds is much larger than that of the bulk properties. In this paper, a novel molecular reconstruction method is developed which includes two Bayesian regression models for bulk properties’ prediction and molecular reconstruction. By defining a characteristic function of bulk property and then establishing its general mixing rule with respect to compositions, the bulk property is predicted from a linear regression model with sigmoidal basis functions whose parameters can be estimated by maximizing a posterior distribution from a well-determined database containing bulk properties and molecular information on petroleum fraction samples. Furthermore, by developing a prior distribution of the mol...

Process measurements play a significant role in process identification, control, and optimization. However, they are often corrupted by two types of errors, random and gross errors. The presence of gross errors in the measurements affects the reliability of optimization and control solutions. Therefore, in this work, we characterize the measurement noise model using a Gaussian mixture distribution, where each mixture component denotes the error distribution corresponding to random error and gross error, respectively. On the basis of this assumption, we propose a maximum likelihood framework for simultaneous steady-state data reconciliation and gross error detection. Since the proposed framework involves the noise mode as a hidden variable denoting the existence of gross errors in the data, it can be solved using the expectation maximization (EM) algorithm. This approach does not require the parameters of the error distribution model to be preset, rather they are determined as part of the solution. Several...

Diagnosing the root cause of a propagated oscillation in the operation requires detection of all process variables that are oscillating with similar frequencies followed by application of an appropriate root cause diagnosis procedure. Oscillations in chemical processes are usually caused by controller tuning, valve problems, or external oscillatory disturbances. There are several methods proposed in literature for root cause diagnosis of oscillations within the system. However, most of the methodologies can only work for a specific type of oscillation. For example, the methodologies based on quantifying the nonlinearity of variables can help with root cause diagnosis of a valve-induced oscillation but cannot help if the oscillation actually has occurred due to aggressive controller tuning or due to an external oscillatory disturbance. Therefore, before trying to find out which loop within the system has caused the oscillation, it is important to categorize the oscillation meaning to learn if the oscillati...

Most existing methods for the state estimation in batch processes are similar to those for continuous processes, and these methods usually only consider the state dynamics within a single batch and...

In this paper, a novel dissimilarity-analysis-based method is proposed to monitor the control performance of multi-input–multi-output systems. The proposed approach detects changes in the orientation and volume of hyper-ellipsoids formed by the covariance matrices via analyzing the eigenvalues of transformed covariance matrices. Furthermore, a new performance index is used to quantify performance change of control systems. Simulation results from a numerical example, the Wood Berry distillation column example, and pilot-scale experiment results all demonstrate the effectiveness of the proposed method.

This paper is concerned with the identification problems of the linear parameter varying (LPV) system with missing output in the presence of the time-delay. A multiple-model approach is adopted. Local models varying from one operating point to another are first described by finite impulse response (FIR) models. To handle missing output and time-delay, the expectation-maximization (EM) algorithm is utilized to estimate the unknown parameters and the time-delay simultaneously. Output Error (OE) models are widely used in controller design. Therefore, the auxiliary model principle is employed to recover the OE models based on the initially identified FIR models. The EM algorithm is then used again to refine the unknown parameters of the OE models with the complete data set to obtain the final global model. Simulation examples are presented to demonstrate the performance of the proposed method.

This paper studies the performance assessment of linear controllers in univariate feedback control loops, where the processes to be controlled can be approximated by first-order plus dead time models, and the set points are subject to ramp or step changes. The lower bound of the total variation (TV) of control signals is established. Taking the lower bound of TV and that of integrated absolute error (IAE) as benchmarks, an IAE-TV-based performance index is proposed. Four industrial linear control schemes are investigated to find out the conditions simultaneously achieving satisfactory performances in terms of input load disturbance rejection and set point tracking subject to a constraint on the robustness. A novel performance assessment method is presented to calculate the proposed performance index from measurements. Numerical and experimental examples are provided to validate the performance benchmark and the effectiveness of the proposed performance assessment method.

Automatic detection of oscillatory variables in the presence of multiple oscillations is still a challenging problem in the literature, despite the fact that there are several methods for detection and estimation of single-frequency oscillation. A method is proposed that utilizes the autocorrelation function (ACF) to detect the oscillatory variables and estimate the oscillation periods in the presence of multiple oscillations. The advantage of the developed method is that it requires no or little human interference in the detection process. It is also capable of estimation of the decay rate for decaying oscillations and is advantageous over methods based on analyzing the power spectrum for oscillation detection in case of nonsinusoidal oscillations. The proposed method is verified through a case study.

Wavelength selection is widely accepted as an important step in near-infrared (NIR) spectroscopic model development. In quantitative online applications, the robustness of the established NIR model...

The difficulty in measuring certain types of process variables rapidly has encouraged the use of soft sensors, which can determine the values of difficult to measure process variables based on easily available secondary process variables. A bias update term that allows the system to take into consideration disturbances in the system is often included in such soft sensor systems. The first part of this two-part series considers the bias update for the open-loop situation, including the ideal case, the case where there is measurement delay, the case with multirate sampling, and the case where there is a combination of measurement delay and multirate sampling. Proposed tuning rules are provided for all cases in order to obtain optimal open-loop tracking of the quality variable, especially in the presence of slow or drifting disturbances. Simulation and experimental validation of the results is presented.

Imprecision of sensors is one of the main causes of poor control and process performance. Often, instrument measurement bias and variance change over the time and online calibration/re-estimation is necessary. Originated from a real industrial application problem, this paper proposed a Bayesian approach to determine the inconsistency of sensors, based on mass-balance principles. A mass-balance factor model is then introduced, where the factor analysis method is used to determine initial values for estimating instrument noise and process disturbance variance. Because of the structural constraint of mass-balance equations, a gray-box estimation procedure must be adopted for which Bayesian network estimation via the expectation-maximization (EM) algorithm is a very suitable method. Therefore, this paper uses factor analysis to determine the initial values, and, afterward, estimates process and sensor variance by means of Bayesian estimation. After estimating the process and instrument variance, the process s...

We investigate direct estimation of step-response models from closed-loop data using subspace identification. Necessary information concerning impulse-response coefficients is embedded in subspace matrices. Therefore, the step-response coefficients can be directly obtained from this matrix by integration without the need of extracting state space models first, as the conventional subspace identification does. Since the estimated subspace matrix contains more than one set of impulse-response coefficients, a question arises about how to best synthesize them to obtain an optimal estimate of the impulse-response coefficients and subsequently the step-response coefficients. For this purpose, a reformulation of the subspace identification problem is required for which the variance of all elements in the related subspace matrix can be evaluated. The calculated variances are then used to perform a weighted averaging on the estimated impulse-response coefficients to attenuate the noise influence on the final step-response model estimation. Monte Carlo simulations and pilot-scale experiments are provided to illustrate the proposed method.

In this article, first, a hidden Markov model is built to address the temporal mode dependency problem in control loop diagnosis. A data-driven algorithm is developed to estimate the mode transition probability. The new solution to mode dependency is then further synthesized with the solution to evidence dependency to develop a recursive autoregressive hidden Markov model for online control loop diagnosis. When both the mode and evidence transition information sets are considered, the temporal information is effectively synthesized under the Bayesian framework. A simulated distillation column example and a pilot-scale experiment example are investigated to demonstrate the ability of the proposed diagnosis approach.

Model predictive control (MPC) is one of the most studied modem control technologies. Among the various subjects investigated, controller performance assessment of MPC has received considerable attention in recent time. Various approaches and algorithms have been proposed for the assessment of MPCs. In this work, we propose a novel approach to MPC constraint analysis by considering the economic objective function as a continuous-valued function within a Bayesian probabilistic framework. The analysis involves inference of the effect of a decision to adjust the limits of the constrained variables with regards to the achievable profits (decision evaluation) as well as inference of constraint limits that should be adjusted so as to achieve a specified profit value (decision making). The benefits of this approach include a more generalized definition of quality variables, the development of a more rigorous formulation of the problem to address linear and quadratic objective functions and thereby obtaining closed form solutions, and maximum-likelihood location determination of the quality variables in the decision making process. The approach is illustrated with the use of simulations and a pilot-scale experiment.

The main challenge in developing soft sensors in process industry is the existence of irregularity of data, such as measurement noises, outliers, and missing data. This paper is concerned with a comparative study among various data-driven soft sensor algorithms and the Bayesian methods. The algorithms to be considered for a comparative study in this paper include ordinary least-squares, robust regression, error-in-variable methods, partial least-squares, and the Bayesian inference algorithms. Methods for handling irregular data are reviewed. An iterative Bayesian algorithm for handling measurement noise and outliers is proposed. Performance of the Bayesian methods is compared with other existing methods through simulations, a pilot-scale experiment, and an industrial application.

Advanced process control (APC)-in particular, model predictive control (MPC)-has emerged as the most effective control strategy in process industry, and numerous applications have been reported. Nevertheless, there are several factors that limit the achievable performance of MPC. One of the limiting factors considered in this paper is the presence of constraints. To exploit optimal control performance, continuous performance assessment, with respect to the constraints of MPC, is necessary. MPC performance assessment has received increasing interest, both in academia and in industry. This paper is concerned with a practical aspect of performance assessment of industrial MPC by investigating the relationship among process variability, constraints, and probabilistic economic performance of MPC. The proposed approach considers the uncertainties induced by process variability and evaluates the economic performance through probabilistic calculations. It also provides a guideline for the constraint tuning, to improve MPC performance.

Performance assessment of model predictive control (MPC) systems has been focusing on evaluation of the variability with, for example, minimum variance or LQG/MPC tradeoff curve as benchmarks. These previous studies are mainly concerned with the dynamic performance of MPC. However, the benefit of MPC is largely attributed to its capability for economic optimization. The economic performance, on the other hand, is also dependent on the variability reduction achieved through dynamic control. There is a need to assess MPC performance by considering economic performance, variability reduction, and their relationships. One of the good indications of this relation is the constraint tuning. In practical MPC applications, the constraint setups are important whenever an MPC is commissioned, and constraint tunings are not uncommon, even when the MPC is already on-line. Thus, the questions to ask are which constraints should be adjusted, and what is the benefit to do so? By investigating the relationship between var...

Multivariate controller performance assessment (MVPA) has been developed over the last several years, but its application in advanced model predictive control (MPC) has been limited mainly due to issues associated with comparability of the variance control objective and that of MPC applications. MPC has been proven as one of the most effective advanced process control (APC) strategies to deal with multivariable constrained control problems with an ultimate objective toward economic optimization. Any attempt to evaluate MPC performance should therefore consider constraints and economic performance. In this work, we show that the variance based performance assessment may be transferred to performance assessment of MPC applications. The MPC economic performance can be evaluated by solving benefit potentials through either variability reduction of quality output variables or tuning of constraints. Algorithms for MPC performance assessment and constraint/variance tuning guidelines are developed through linear matrix inequalities (LMIs) using routine operating process data plus the process steady-state gain matrix. The proposed approach for MPC economic performance evaluation is illustrated and verified via a simulation example of an MPC application as well as a pilot-scale experiment.

This paper is concerned with the feedforward and feedback control performance assessment of linear time-variant processes. The developed algorithms provide a way to calculate the time-varying minimum-variance feedforward and feedback control benchmark from routine operating data. The proposed techniques are illustrated through a simulated stirred-tank heater example and an industrial application.

This paper is concerned with the performance assessment problem for linear time-variant (LTV) processes. It is a continuation of the work by Huang (J. Process Control 2002, 12, 707−719) with further theoretical development and industrial applications. Systematic algorithms and procedures for the performance assessment of LTV feedback control loops are proposed. The developed algorithms are illustrated through a simulated stirred-tank heater example and, as an industrial case study, applied to a sulfur recovery unit that is under adaptive control.

Fault detection and isolation (FDI) has become a critical issue for increasing availability, reliability, and production safety in industrial process monitoring. While model-based FDI methods rely on the mathematical model and input-output data of a process to perform detection, the local approach is a new model-based FDI method that aims to detect slight changes of parametric properties of a system. Robustness with respect to model uncertainties is an important issue for the local approach. To reduce false alarms caused by the estimation error of process parameters, a new algorithm is proposed to recalculate the threshold based on the original threshold and covariance matrix of the estimated parameters. A similar algorithm is also provided to recalculate the threshold to reduce false alarms due to regular parameter fluctuations. The effectiveness of the proposed algorithms is shown by simulation studies on a papermaking cross-direction control system.

A dynamic matrix is a lower triangular matrix containing the step response coefficients of the deterministic input used in the model predictive control schemes such as the dynamic matrix controller. Subspace matrices (defined in subspace state-space identification methods) corresponding to the deterministic input and the stochastic input contain the impulse response coefficients of the deterministic and stochastic models, respectively. This paper proposes a new subspace identification based method for the estimation of the dynamic matrix of the deterministic input(s) directly from the closed-loop data. The noise model is simultaneously obtained from the closed-loop data in the impulse response form. The method is extendable to the case of measured disturbances. All of the results presented in this paper are applicable to the multivariate systems. Guidelines for the practical implementation of the algorithm are also presented in this paper. The proposed method is illustrated through MATLAB simulations and an application on a pilot-scale plant.