Industrial Distillation Column Research Articles

Bayesian transfer filter for industrial processes with additional delayed slow-rate measurements

In industrial processes, fast-rate online measurements (FOMs) are commonly associated with delayed slow-rate measurements (DSMs), provided by off-line laboratory analysis. In this paper, we propose a Bayesian transfer filter to transfer the information/knowledge contained in the DSMs to the FOMs. Different from the existing fusion approaches that weight the FOMs and DSMs equally, the proposed method signs the FOMs as the target domain and the DSMs as the source domain. The predicted distribution of states in the target domain is modified by minimizing the Kullback-Leibler divergence by fixing the distribution of measurements. In this way, the resulting algorithm can make a good trade-off between the information transferred from the source domain and the information contained in the target domain. The efficiency of the proposed method is demonstrated by a two-dimensional moving target example and an industrial distillation column case. Furthermore, by comparing it to other existing methods, such as the Kalman Filter and modified delayed track-to-track fusion approach, this method makes a particular accuracy improvement in estimating states.

Surrogate Modelling of an Industrial Distillation Column Obtained from Statistical Techniques and Machine Learning

Purpose: To study a case of modeling an industrial distillation system, using the Aspen Plus simulator as the mathematical model and subsequently generating surrogate models using Machine Learning techniques in Matlab. Theoretical Framework: Metamodels are reduced models obtained from data generated by rigorous models, replacing them entirely or partially when the computational codes originating from them require excessively large computational effort to be used feasibly. Method/Design/Approach: The simulation of the process was performed in Aspen Plus. Subsequently, the most important variables were selected, and an experimental design was created using Latin Hypercube Sampling in Matlab, generating points to be used in sensitivity analysis in Aspen Plus. The next step was the construction of metamodels using the Statistics and Machine Learning toolbox in Matlab, employing linear regression and Gaussian process regression models. Finally, a statistical analysis was conducted. Results and Conclusion: The distillation system simulation converged, and the obtained metamodels had good regression indicators, especially the Gaussian process regression models, making them the most suitable for representing the rigorous Aspen Plus model. Research Implications: Understanding the multicomponent distillation process integrated with computational tools and data regression models, leading to a reduction in computational effort. Originality/Value: Development of a tool that enables the simulation and evaluation of a distillation process without the need to acquire software with rigorous equation databases.

A data-driven soft-sensing approach using probabilistic latent variable model with functional data framework

Functional principal component analysis (FPCA) and functional partial least squares (FPLS) are two mainstream functional data analysis (FDA) methods, which have been commonly used to extract deep information hidden in the original data space. However, the process data always contain random noise, which affects the performance of FDA models. To overcome this issue, two functional probabilistic latent variable models (FPLVMs), including functional probabilistic principal component analysis (FPPCA) and functional probabilistic partial least squares (FPPLS) are proposed in this work. First, the process data are converted into functional data using the FDA. Subsequently, a log-likelihood function considering the noise factor and functional latent variables is designed. Finally, the regression model parameters are estimated using an expectation–maximisation algorithm. In contrast to FPPCA, FPPLS decomposes the process data and the key variable with constrained latent variables, which is similar to the partial least squares (PLS) and the principal component analysis (PCA). Moreover, the degeneration mechanism from FPLVMs into probabilistic latent variable models and latent variable models is discussed. An adaptive strategy with functional covariance is used to satisfy the online predictive capabilities of the model. Finally, the proposed approach is validated using a numerical case, the Tennessee Eastman process and an industrial o-xylene distillation column for evaluation.

Control of an Industrial Distillation Column Using a Hybrid Model with Adaptation of the Range of Validity and an ANN‐based Soft Sensor

AbstractAdvanced control schemes such as model predictive control can be used to minimize the use of resources while guaranteeing the specified product quality. In this paper, we consider an industrial mother liquor distillation column varying flow rate and composition of the feed. There are specifications of the composition for all product streams. To address this challenging control problem, we employ a nonlinear model‐predictive controller using a hybrid model, which consists of a simple phenomenological model augmented by a data‐based component to compensate the plant‐model mismatch. The trustworthiness of the data‐based model is addressed using a domain of validity of the data‐based model, which is estimated using a one‐class support vector machine. During operation, it may turn out that the model is also reliable in a wider range, therefore, data of recently visited operating points is recorded and the domain of validity is extended if the model is sufficiently accurate. To improve the performance of the controller, an artificial neural network model is used to estimate the product composition from available measurements.

Physics Informed Piecewise Linear Neural Networks for Process Optimization

Constructing first-principles models is usually a challenging and time-consuming task due to the complexity of real-life processes. On the other hand, data-driven modeling, particularly a neural network model, often suffers from overfitting and lack of useful and high-quality data. At the same time, embedding trained machine learning models directly into the optimization problems has become an effective and state-of-the-art approach for surrogate optimization, whose performance can be improved by physics-informed machine learning. This study proposes using piecewise linear neural network models with physics-informed knowledge for optimization problems with neural network models embedded. In addition to using widely accepted and naturally piecewise linear rectified linear unit (ReLU) activation functions, this study also suggests piecewise linear approximations for the hyperbolic tangent activation function to widen the domain. Optimization of three case studies, a blending process, an industrial distillation column, and a crude oil column are investigated. Physics-informed trained neural network-based optimal results are closer to global optimality for all cases. Finally, associated CPU times for the optimization problems are much shorter than the standard optimization results.

Gamma Photon Column Scanning of Prototype and Industrial Distillation Columns

Gamma Photon Column Scanning of Prototype and Industrial Distillation Columns

Model Predictive Control of an Industrial Distillation Column Based on a Hybrid Model: Adapting the Domain of Validity

In any model predictive control (MPC) implementation, the key factor is the quality of the employed system model in order to guarantee constraint satisfaction and to optimize the performance of the controlled system. However, it is difficult in practice to model all the phenomena of the true system which leads to plant-model mismatch. We propose to augment a simple first principles based model with a machine learning model to capture the plant dynamics which are not represented in the semi-rigorous model. The use of data-based models however can be problematic outside the region where the model was trained. In this work, we estimate the domain of validity of the data-based model using a one-class support vector machine (SVM) which is trained on a low-dimensional projection of the available training data. During the application of the controller, data of newly visited regions is collected and the domain of validity is adapted based on checking a performance criterion. The hybrid model incorporates a weighted contribution of the data-based model based on its predicted validity. We demonstrate the versatility and the benefits of the proposed adaptive MPC based on a hybrid model (AHMPC) approach via an industrial distillation column case study.

A soft-sensing method for product quality monitoring based on particle swarm optimization deep belief networks

A novel soft-sensing method for quality parameters of aviation kerosene in atmospheric distillation column based on least absolute shrinkage and selection operator and particle swarm optimization deep belief network (LASSO-PSO-DBN) is proposed. First, to reduce the dimension of the input variables, the least absolute shrinkage and selection operator (LASSO) algorithm is used to select the input variables that are irrelevant to the soft sensor of aviation kerosene quality parameters. Then, to improve the generalization of soft sensor model, a deep learning algorithm, deep belief network (DBN), is proposed for soft sensing of aviation kerosene quality parameters. Considering that the structure characteristics and parameters of DBN algorithm have a great impact on the learning and prediction results, the parameters of DBN are optimized based on particle swarm optimization (PSO) algorithm. The benchmark data sets and the industrial atmospheric distillation column data are used for simulation analysis and evaluation of the soft-sensing performance. The simulation results show that the novel proposed algorithm can effectively reduce the dimension of the input variables and simplify the structure of the soft sensor model. It also has good generalization ability and the predicted value is in good agreement with the actual measured value.

Data-based design of inferential sensors for petrochemical industry

Inferential (or soft) sensors are used in industry to infer the values of imprecisely and rarely measured (or completely unmeasured) variables from variables measured online (e.g., pressures, temperatures). The main challenge, akin to classical model overfitting, in designing an effective inferential sensor is the selection of a correct structure of the sensor. The sensor structure is represented by the number of inputs to the sensor, which correspond to the variables measured online and their (simple) combinations. This work is focused on the design of inferential sensors for product composition of an industrial distillation column in two oil refinery units, a Fluid Catalytic Cracking unit and a Vacuum Gasoil Hydrogenation unit. As the first design step, we use several well-known data pre-treatment (gross error detection) methods and compare the ability of these approaches to indicate systematic errors and outliers in the available industrial data. We then study effectiveness of various methods for design of the inferential sensors taking into account the complexity and accuracy of the resulting model. The effectiveness analysis indicates that the improvements achieved over the current inferential sensors are up to 19%.

Development of Robust and Physically Interpretable Soft Sensor for Industrial Distillation Column Using Transfer Learning with Small Datasets

In the development of soft sensors for industrial processes, the availability of data for data-driven modeling is usually limited, which led to overfitting and lack of interpretability when conventional deep learning models were used. In this study, the proposed soft sensor development methodology combining first-principle simulations and transfer learning was used to address these problems. Source-domain models were obtained using a large amount of data generated by dynamic simulations. They were then fine-tuned by a limited amount of real plant data to improve their prediction accuracies on the target domain and guaranteed the models with correct domain knowledge. An industrial C4 separation column operating at a refining unit was used as an example to illustrate the effectiveness of this approach. Results showed that fine-tuned networks could obtain better accuracy and improved interpretability compared to a simple feedforward network with or without regularization, especially when the amount of actual data available was small. For some secondary effects, such as interaction gain, its interpretability is mainly based on the interpretability of the corresponding source models.

Hybrid model for fault detection and diagnosis in an industrial distillation column

Abstract The present work describes a method of automatic fault detection and identification based on a hybrid model (HM): First Principles – Neural Network. The FPM can simulate a wide range of situations while the NN corrects the model output using information from the historical data of the process. Operating conditions corresponding to different types of faults were simulated with the HM and saved with their description in a process state library. To detect a fault, the online measured data was compared with that corresponding to the operation under normal conditions. If a significant deviation was detected, the current state was compared with all the states stored in the process state library and it was identified as the one at the shortest distance. The method was tested with real data from a methanol-water industrial distillation column. During the studied period of operation of the plant, two faults were identified and reported. The proposed method was able to identify such failures more effectively than an equivalent model of first principles. The results obtained show that the proposed method has a great potential to be used in the automatic diagnosis of faults in refining and petrochemical processes.

Novel Soft Sensor for Measuring and Controlling Product Recovery in a High-Purity, Multicomponent, Side-Draw Distillation Column

The use of soft sensors for monitoring purposes is an established practice in the process industry. In this study, the focus has been on developing a soft sensor that can be used to monitor the product recovery rate of double-ended, high-purity distillation columns with a side-draw. To this end, this study specifically focuses on developing a cost-effective and accurate soft sensor for an industrial methanol distillation unit where a side-draw is used to meet the parts per million level impurity specifications. The novel soft sensor is based on the unique characteristics of the mass balance in this type of column, and uses the density measurement at the side-draw together with the flow rates of the side-draw and product draw to calculate the product recovery rate. The developed soft sensor was validated against real plant data as well as on a process simulation of an industrial methanol distillation column. The soft sensor demonstrated the ability to predict product recovery to an accuracy of 0.05% and sh...

Managing Trace Components in a Dividing Wall Distillation Column: An Experimental Study

Industrial distillation columns must manage impurities inherently present in feed mixtures. Proving that dividing wall columns (DWCs) can deal with trace components is an important step toward thei...

Advance warning for loss of separation in an industrial distillation column

AbstractThis paper describes the development of a data‐driven advance warning system for the onset of loss of separation in an industrial distillation column. The system would enable preventive actions to avoid several hours of bad operation and subsequent recovery of the process. Data of more than 2 years of process operation were used to identify and validate various monitoring systems based on both static principal component analysis (PCA) and dynamic PCA. Despite the presence of autocorrelation in the data, only minor differences in advance warning were observed between PCA and dynamic PCA.The developed system provides warnings for 35% to 45% of the observed periods of bad column operation, with respective advance warning times of 16 and 6 minutes. It proves a valuable additional tool to monitor the operation of the distillation column and avoid losses of product, with the potential of reducing bad operation (and the associated costs) by up to 45% and substantially improving overall process reliability.

A variable selection method for soft sensor development through mixed integer quadratic programming

Soft sensors are widely employed in industry to predict quality variables, which are difficult to measure online, by using secondary variables. To build an accurate soft sensor, a proper variable selection is critical. In this project, a method of selecting the optimal secondary variables for a soft sensor model is proposed. It is formulated as a nested optimization problem. In each iteration, a mixed integer quadratic programming (MIQP) is conducted with the Bayesian information criterion (BIC) to estimate the prediction error. A warm start (WS) technique is developed to speed up the convergence. The proposed method is evaluated using a number of instances from the UCI Machine Learning Repository. The computational results demonstrate that this method is well suited for finding the best variable subsets. The method is successfully applied to build soft sensors for an industrial distillation column. The results show that the proposed method can effectively select feature variables that will improve the model prediction performance and reduce the model complexity. Comparisons with other methods, including the traditional partial least square technique, are also presented.

Online Composition Prediction of a Debutanizer Column Using Artificial Neural Network

The current method for composition measurement of an industrial distillation column includes an offline method, which is slow, tedious and could lead to inaccurate results. Among advantages of using online composition designed are to overcome the long time delay introduced by laboratory sampling and provide better estimation, which is suitable for online monitoring purposes. This paper presents the use of an online dynamic neural network to simultaneously predict n-butane composition of the top and bottom products of an industrial debutanizer columns. Principal component and partial least square analysis are used to determine the important variables surrounding the column prior to implementing the neural network. It is due to the different types of data available for the plant, which requires proper screening in determining the right input variables to the dynamic model. Statistical analysis is used as a model adequacy test for the composition prediction of n-butane in the column. Simulation results demonstrated that the Artificial Neural Network (ANN) can reliably predict the online composition of n-butane of the column. It is further confirmed by the statistical analysis with low Root Mean Square Error (RMSE) value indicating better prediction.

A comparison of a novel robust decentralised control strategy and MPC for industrial high purity, high recovery, multicomponent distillation

In this work we have developed a novel, robust practical control structure to regulate an industrial methanol distillation column. This proposed control scheme is based on a override control framework and can manage a non-key trace ethanol product impurity specification while maintaining high product recovery. For comparison purposes, a MPC with a discrete process model (based on step tests) was also developed and tested. The results from process disturbance testing shows that, both the MPC and the proposed controller were capable of maintaining both the trace level ethanol specification in the distillate (XD) and high product recovery (β). Closer analysis revealed that the MPC controller has a tighter XD control, while the proposed controller was tighter in β control. The tight XD control allowed the MPC to operate at a higher XD set point (closer to the 10ppm AA grade methanol standard), allowing for savings in energy usage. Despite the energy savings of the MPC, the proposed control scheme has lower installation and running costs. An economic analysis revealed a multitude of other external economic and plant design factors, that should be considered when making a decision between the two controllers. In general, we found relatively high energy costs favour MPC.

Efficiency-based nonequilibrium modeling of industrial-scale multicomponent distillation columns

ABSTRACTAn efficiency-based nonequilibrium model is proposed to simulate two industrial distillation columns. In this model, the behavior of trays is studied by considering the mass transfer at interphase. The interphase mass transfer is simply evaluated by the simultaneous solution of the vapor material balance equation and the definition of the Murphree efficiency. A MATLAB code is developed to implement the simulation procedure. For verification, the product compositions and plate temperatures are compared with the reported experimental data. The discrepancy of the experimental and simulation results is about 0.5% and 20% for the temperature and the product compositions for both columns, respectively.

OPERATIONAL RISK ANALYSIS FOR UNDESIRED POLYMERIZATION IN OVERHEAD CONDENSER OF BUTADIENE COLUMN

An accident model based on fault tree analysis was developed and applied to analyze the recurrence of undesired reactions producing popcorn polymers in the overhead condenser of an industrial butadiene distillation column. The modeling framework incorporates reliability data associated with asset integrity and human performance along with data on selected process variables. These variables, i.e., pressure, feed velocity and temperature of the condenser, which were identified from root cause analysis of the incident provided dynamic contributions to the model, and were represented in the form of Weibull distribution functions. The results obtained proved the potentials of the proposed methodology. Based on the case study considered, operating pressure was identified as the most influential process variables that needed closer monitoring. The methodology provides an opportunity for risk management to be implemented dynamically to facilitate maintenance plan and management of change.

Nonsquare multivariable non-minimal state space-proportional integral plus (NMSS-PIP) control for atmospheric crude oil distillation column

The goal of this paper is to show implementation of non-minimal state space (NMSS) representation with proportional-integral-plus (PIP) controller for industrial crude distillation column. The control structure was designed and developed for MIMO nonsquare configuration. Weighting matrices of LQ optimal PIP control structure were achieved based on ITAE criteria for the column. The control performance of PIP control was compared with conventional PI control with different scenarios. The results show that PIP control handles the process interactions truly in order to utilize all variables simultaneously. In feed composition change scenario, the PIP control was stable and behaved robustly while the PI control had some difficulties. Furthermore, the proposed control structure indicates efficient performance for both disturbance rejection and set point tracking.