A Bayesian approach to design of adaptive multi-model inferential sensors with application in oil sand industry

In the context of process industries, online monitoring of quality variables is often restricted by inadequacy of measurement techniques or low reliability of measuring devices. Therefore, there has been a growing interest in the development of inferential sensors to provide frequent online estimates of key process variables on the basis of their correlation with real-time process measurements. Representation of multi-modal processes is one of the challenging issues that may arise in the design of inferential sensors. In this paper, Bayesian procedures for the development and implementation of adaptive multi-model inferential sensors are presented. It is shown that the application of a Bayesian scheme allows for accommodating the overlapping operating modes and facilitating the inclusion of prior knowledge. The effectiveness of the proposed procedures are first demonstrated through a simulation case study. The efficacy of the method is further highlighted by a successful industrial application of an adaptive multi-model inferential sensor designed for real-time monitoring of a key quality variable in an oil sands processing unit.     

A probabilistic framework for real-time performance assessment of inferential sensors

A definition for the reliability of inferential sensor predictions is provided. A data-driven Bayesian framework for real-time performance assessment of inferential sensors is proposed. The main focus is on characterizing the effect of operating space on the reliability of inferential sensor predictions. A holistic, quantitative measure of the reliability of the inferential sensor predictions is introduced. A methodology is provided to define objective prior probabilities over plausible classes of reliability based on the total misclassification cost. The real-time performance assessment of multi-model inferential sensors is also discussed. The application of the method does not depend on the identification techniques employed for model development. Furthermore, on-line implementation of the method is computationally efficient. The effectiveness of the method is demonstrated through simulation and industrial case studies.     

Statistical and computational intelligence techniques for inferential model development: a comparative evaluation and a novel proposition for fusion

In industry today many products are sold for their efficacy rather than their chemical composition. Many variables (dependent variables), which characterize the quality of the final product in a manufacturing process, can be difficult to measure in real-time. Measurement difficulties can be due to a variety of reasons, including: (1) Reliability of on-line sensors, (2) Lack of appropriate on-line instrumentation. It is often the case that off-line laboratory tests are the only means of determining product quality measurements. However such laboratory analyses introduce delays in the measurement of key performance indicators. This can result in a significant economic loss if the analysed product fails the quality control test. In such situations an improved monitoring system is therefore required to determine product quality online and minimise commercial wastage. To facilitate this, advanced monitoring and control or optimisation techniques require inferred measurements, generated with correlations from readily available process variables (independent variables). Although inferential models are widely used in industry, only a few techniques for inferential model development are discussed in the open literature. This paper therefore will present a comparative evaluation study of the current inferential measurement techniques. An improved systematic approach for the development of inferential models using intelligent and soft computing systems is also highlighted. The proposed approach is designed to address some of the problems that currently exist in the area of inferential modelling through the fusion of statistical and computational intelligence models. A novel method of fusion is also proposed and an industrial case study is then presented to demonstrate the methodology by inferring the ‘Anchorage’ of polymeric-coated substrates (i.e. Tyvek or paper) in the coating industry. The application on which this methodology is demonstrated is unique. No such work in the literature to date has presented any inferential modelling strategies in the area of the coating industry. This strategy developed through the fusion of statistical and artificial modelling to generate a hybrid inferential measurement system has the potential to significantly improve the quality control monitoring system and reduce the economic loss encountered through the production of off-spec material.     

Development and industrial application of soft sensors with on-line Bayesian model updating strategy

This paper deals with the issues associated with the development of data-driven models as well as model update strategy for soft sensor applications. A practical yet effective solution is proposed. Key process variables that are difficult to measure are commonly encountered in practice due to limitations of measurement techniques. Even with appropriate instruments, some measurements are only available through off-line laboratory analysis with typical sampling intervals of several hours. Soft sensors are inferential models that can provide continuous on-line prediction of hidden variables; such models are capable of combining real-time measurements with off-line lab data. Due to the prevalence of plant-model mismatch, it is important to update the model using the latest reference data. In this paper, parameters of data-driven models are estimated using particle filters under the framework of expectation–maximization (EM) algorithms. A Bayesian methodology for model calibration strategy is formulated. The proposed framework for soft sensor development is applied to an industrial process to provide on-line prediction of a quality variable.     

E+H组合式过程测控装置及实时监控系统设计

Endress+Hauser(E+H)工业现场仪表广泛应用于石油、冶金、电力、生命科学等多个核心行业,具有可靠集成、精准测量、交互式数据管理等特点.为提高实验装置工业适用性,培养学生熟练运用工业级传感器和过程监控的操作技能,将这类传感器引入实验室,设计了一套E+H组合式过程测控装置与实时监控系统.系统以过程控制涉及的典...     

Semi-supervised variational Bayesian Student’s t mixture regression and robust inferential sensor application

Data-driven inferential sensor has been widely adopted to estimate key quality relevant variables. However, industrial dataset usually presents many characteristics such as nonlinearity, non-Gaussianity, insufficiency of labeled samples, contamination of outliers, etc. These intractable characteristics have rendered significant difficulties in developing high-performance inferential sensor. This paper deals with these issues in the probabilistic way by proposing a robust semi-supervised variational Bayesian Student’s t mixture regression (referred to as the ‘SSVBSMR’). Specifically, in the SSVBSMR, the nonlinear and non-Gaussian characteristics are handled by Bayesian finite mixture models (FMM), and the Student’s t distribution is employed to constitute the components of FMM, which makes the SSVBSMR robust against outliers. In addition, the SSVBSMR exploits unlabeled samples to remedy the insufficiency of labeled samples. Furthermore, the SSVBSMR treats all model parameters as stochastic rather than deterministic such that the model selection can be automatically and efficiently completed and some limitations of the maximum likelihood method (such as overfitting and singular covariance) can be alleviated. A variational Bayesian expectation–maximization-based learning algorithm is also developed to train the SSVBSMR. Two cases are carried out to investigate the performance of the SSVBSMR, and the results demonstrate its effectiveness and feasibility compared to several state-of-the-art methods.     

Comparison of variable selection methods for PLS-based soft sensor modeling

Data-driven soft sensors have been widely used in both academic research and industrial applications for predicting hard-to-measure variables or replacing physical sensors to reduce cost. It has been shown that the performance of these data-driven soft sensors could be greatly improved by selecting only the vital variables that strongly affect the primary variables, rather than using all the available process variables. In this work, a comprehensive evaluation of different variable selection methods for PLS-based soft sensor development is presented, and a new metric is proposed to assess the performance of different variable selection methods. The following seven variable selection methods are compared: stepwise regression (SR), partial least squares with regression coefficients (PLS-BETA), PLS with variable importance in projection (PLS-VIP), uninformative variable elimination with PLS (UVE-PLS), genetic algorithm with PLS (GA-PLS), least absolute shrinkage and selection operator (Lasso), and competitive adaptive reweighted sampling with PLS (CARS-PLS). Their strengths and limitations for soft sensor development are demonstrated by a simulated case study and an industrial case study.     

A novel variable selection approach based on co-linearity index to discover optimal process settings by analysing mixed data

In the last two decades the application of statistical techniques to process control has gained popularity due to the widespread adoption of quality management systems such as ISO9001. Demonstration of continual process improvement by monitoring process effectiveness has become an integral part of satisfying the requirements of clause 8 of the ISO9001:2008 standard. The process effectiveness is measured in terms of one or more process responses. Data driven approaches are often used to associate the variability in process responses with one or more process variables. However, traditional techniques become unpractical in the presence of large number of variables and noisy data sets. This paper extends the co-linearity index and penalty matrix approach (Ransing et al., 2013) for discovering noise free correlations between heterogeneous process variables and responses. Noise is removed by reducing the dimensionality of the variable space and using robust data pre-treatment methods which are more suitable in the presence of outliers and skewed distributions for process variables. Scaling factors have been proposed to balance variance contributions from response variables, quantitative and categorical variables. The proposed method allows process variables with skewed distribution to contribute more to the variance than Gaussian distributed variables so that these variables can be investigated further, if necessary. Correlations are visualised in a single plot and can be used in real industrial settings to assist process engineers in manufacturing diagnosis and root cause analysis. The applicability and validity of this novel method has been demonstrated through two industrial case studies.     

Performance evaluation of switched Ethernet for real-time industrial communications

The real-time industrial network, often referred to as fieldbus, is an important element for building automated manufacturing systems. Thus, in order to satisfy the real-time requirements of field devices such as sensors, actuators, and controllers, numerous standard organizations and vendors have developed various fieldbus protocols. As a result, the IEC 61158 standard, including Profibus, WorldFIP, and Foundation Fieldbus, was recently announced as an international standard. These fieldbus protocols have an important advantage over the widely used Ethernet (IEEE 802.3) in terms of the deterministic characteristics. However, the application of fieldbus has been limited due to the high cost of hardware and the difficulty in interfacing with multivendor products. In order to solve these problems, the computer network technology, especially Ethernet, is being adopted by the industrial automation field. The key technical obstacle for Ethernet for industrial applications is that its nondeterministic behavior makes it inadequate for real-time applications, where the frames containing real-time information, such as control command and alarm signal, have to be delivered within a certain time limit. Recently, the development of switched Ethernet shows a very promising prospect for industrial applications due to the elimination of uncertainties in the network operation that leads to the dramatically improved performance. This paper focuses on the application of the switched Ethernet for industrial communications. More specifically, this paper presents the performance evaluation of the switched Ethernet on an experimental network testbed along with an implementation method for using the switched Ethernet for industrial automation.     

Soft sensors for online steam quality measurements of OTSGs

This paper presents a successful application of soft sensor technologies in process industry. The hybrid modeling technique, online prediction update, and robust implementation procedure are presented in detail. Once-through steam generators (OTSGs) have wide applications in In Situ oil sands industry for producing high pressure steam. Real-time control of steam qualities is essential to ensure optimal performance of the OTSGs and ultimately to reduce the production cost and emissions. However, neither existing online measurement nor off-line lab analysis of steam quality can meet this control purpose due to their own limitations. To resolve this problem, soft sensors for steam quality measurements of OTSGs are designed in this work based on a hybrid modeling technique, where online bias update with optimized weighting factor is incorporated to compensate the model error. Furthermore, online outlier detection is considered to ensure the robustness and reliability of the developed soft sensors. The successful applications to an industrial OTSG demonstrate the effectiveness and advantages of the developed soft sensors.     

一种时空协同的图卷积长短期记忆网络及其工业软测量应用

软测量技术的发展有效解决了工业过程中对于难以直接测量的质量变量的感知困难,为过程的控制与优化提供了有力保障.通常在含有多个质量变量的过程中,样本间的时序关系和多个质量变量间相互影响的空间关系能够反映过程本身的特性,这种时空特性的挖掘有益于软测量模型性能的提升,而传统软测量方法往往局限于对时序关系的学习而并未考虑对质量变量间的空间关系进行有效利用.对此,提出一种时空协同的图卷积长短期记忆网络(graph convolution long short-term memory networks, GC-LSTM),并应用于工业软测量场景.采用多通道网络结构将图卷积网络的空间关系挖掘能力与长短期记忆网络的时序关系学习能力相结合,对过程进行时空协同学习以实现软测量应用.具体而言,每条通道用于对每种质量变量进行独立学习;对于过程的时序特性,利用各通道内的长短期记忆网络提取针对不同质量变量的时序特征;对于过程的空间特性,构建质量变量间空间关系的图结构,采用跨通道的图卷积运算将不同通道内不同质量变量的时序特征基于空间关系进行融合,得到兼具过程时空特性的特征,从而在软测量建模中实现过程时空协同学习与融合...     

Online monitoring and control of a cyber-physical manufacturing process under uncertainty

Recent technological advancements in computing, sensing and communication have led to the development of cyber-physical manufacturing processes, where a computing subsystem monitors the manufacturing process performance in real-time by analyzing sensor data and implements the necessary control to improve the product quality. This paper develops a predictive control framework where control actions are implemented after predicting the state of the manufacturing process or product quality at a future time using process models. In a cyber-physical manufacturing process, the product quality predictions may be affected by uncertainty sources from the computing subsystem (resource and communication uncertainty), manufacturing process (input uncertainty, process variability and modeling errors), and sensors (measurement uncertainty). In addition, due to the continuous interactions between the computing subsystem and the manufacturing process, these uncertainty sources may aggregate and compound over time. In some cases, some process parameters needed for model predictions may not be precisely known and may need to be derived from real time sensor data. This paper develops a dynamic Bayesian network approach, which enables the aggregation of multiple uncertainty sources, parameter estimation and robust prediction for online control. As the number of process parameters increase, their estimation using sensor data in real-time can be computationally expensive. To facilitate real-time analysis, variance-based global sensitivity analysis is used for dimension reduction. The proposed methodology of online monitoring and control under uncertainty, and dimension reduction, are illustrated for a cyber-physical turning process.

     

Uncertainty-aware soft sensor using Bayesian recurrent neural networks

Data-driven soft sensors have been widely used to measure key variables for industrial processes. Soft sensors using deep learning models have attracted considerable attention and shown superior predictive performance. However, if a soft sensor encounters an unexpected situation in inferring data or if noisy input data is used, the estimated value derived by a standard soft sensor using deep learning may at best be untrustworthy. This problem can be mitigated by expressing a degree of uncertainty about the trustworthiness of the estimated value produced by the soft sensor. To address this issue of uncertainty, we propose using an uncertainty-aware soft sensor that uses Bayesian recurrent neural networks (RNNs). The proposed soft sensor uses a RNN model as a backbone and is then trained using Bayesian techniques. The experimental results demonstrated that such an uncertainty-aware soft sensor increases the reliability of predictive uncertainty. In comparisons with a standard soft sensor, it shows a capability to use uncertainties for interval prediction without compromising predictive performance.     

Development of adaptive soft sensor based on statistical identification of key variables

An adaptive data-driven soft sensor is derived based on a systematic key variables selection of a process system. The key variables are captured using the statistical approach of stepwise linear regression. The online plant measurements can be directly selected as key features to estimate the tardily-detected quality variables. The parameters of the linear inferential model are adapted as the online/offline quality data become available. The problem of multi-collinearity is discussed and the square root filter is used to improve the numerical characteristic of the algorithm. An industrial example, an o-xylene purification column, is implemented to show the capability of the proposed soft sensor. The results show that the inferential model built by the selected key variables not only predicts accurately but it also matches the real plant situation which makes it useful for industrial applications. Rigorous simulation of the distillation column shows that the inferential control is made possible using the derived adaptive model.     

Supervised linear dynamic system model for quality related fault detection in dynamic processes

Dynamic and uncertainty are two main features of industrial processes data which should be paid attentions when carrying out process monitoring and fault diagnosis. As a typical dynamic Bayesian network model, linear dynamic system (LDS) can efficiently deal with both dynamic and uncertain features of the process data. However, the quality information has been ignored by the LDS model, which could serve as a supervised term for information extraction and fault detection. In this paper, a supervised form of the LDS model is developed, which can successfully incorporate the information of quality variables. With this additional data information, the new supervised LDS model can provide a quality related fault detection scheme for dynamic processes. A detailed industrial case study on the Tennessee Eastman benchmark process is carried out for performance evaluation of the developed method.     

Monitoring nitrate concentrations in the denitrifying post-filtration unit of a municipal wastewater treatment plant

Due to stringent environmental regulations, wastewater treatment plants are always challenged to meet new constraints in terms of water pollution prevention. In such an effort, the number of sensors and data available in the plants have increased considerably during the last decades. However, the quality of the collected data and the sensor reliability are often poor mainly due to the hostile environment in which the measurement equipment has to function. In this work, we present the design of an array of soft-sensors to estimate the nitrate concentration in the post-denitrification filter unit of the Viikinmäki wastewater treatment plant in Helsinki (Finland). The developed sensors aim at supporting the existing hardware analyzers by providing a reliable back-up system in case of malfunction. The main stages of the soft-sensors’ design are discussed and the development illustrated in detail, starting from the preliminary preprocessing of the available process measurements where sample and variable selection has been performed, toward the calibration of the regression models and discussion on the performance results. The estimation accuracy together with the light computational cost of the developed soft-sensors demonstrate their potential for an on-line implementation in the plant's control system.     

Developing a soft sensor with online variable reselection for unobserved multi-mode operations

Soft sensors are used to predict response variables, as these variables are difficult to measure, the prediction models use data of predictors that are relatively easier to obtain. Arranging time-lagged data of predictors and applying the partial least squares (PLS) method to the dataset is a popular approach for extracting the correlation between data of the responses and predictors of the process dynamic. Because irrelevant inputs deteriorate the prediction performance of the soft sensor, the selection of variables in the PLS-based model is a critical step for developing a robust and accurate model. Furthermore, it is necessary to reselect the important predictors of a soft sensor when the operating mode is changed. However, a switch in the operating mode may not be measured, directly. In this study, two statistics are proposed to detect a change of operating mode to enable the reselection of the predictors of the soft sensor. This work involved the development of a soft sensor based on operating data from the industrial ethane removal (de-ethane) process. The changeover of crude oil types cannot be observed from the data of process variables; however, the correlation between input and output variables is significantly affected by the different types of crude oil. The result shows that the use of a soft sensor with online variable reselection is capable of maintaining the accuracy and robustness of the inferential model, effectively.     

Inferential estimation of kerosene dry point in refineries with varying crudes

A bootstrap aggregated model approach to the estimation of product quality in refineries with varying crudes is proposed in this paper. The varying crudes cause the relationship between process variables and product quality variables to change, which makes product quality estimation by soft-sensors a difficult problem. The essential idea in this paper is to build an inferential estimation model for each type of feed oil and use an on-line feed oil classifier to determine the feed oil type. Bootstrap aggregated neural networks are used in developing the on-line feed oil classifier and a bootstrap aggregated partial least square regression model is developed for each data group corresponding to each type of feed crude oil. The amount of training data in crude oil distillation is usually small and this brings difficulties for classification and estimation modelling. In order to enhance model reliability and robustness, bootstrap aggregated models are developed. The inferential estimation results of kerosene dry point on both simulated data and industrial data show that the proposed method can significantly improve the overall inferential estimation performance.     

Quality inferential control of an industrial high density polyethylene process

The melt index (MI) is an important quality variable of control in polyolefin processes. However, it is difficult to measure this variable at frequent and regular sample intervals. A practical on-line inferential scheme is proposed in this paper for predicting the MI using secondary on-line measurements. This on-line MI inferential estimation scheme is combined with a quality control system that utilizes a two-degree of freedom cascaded model predictive controller. The proposed system has been successfully evaluated, under regulatory as well as tracking conditions, on an industrial polyolefin operation.     

Data-driven soft sensor development based on deep learning technique

In industrial process control, some product qualities and key variables are always difficult to measure online due to technical or economic limitations. As an effective solution, data-driven soft sensors provide stable and reliable online estimation of these variables based on historical measurements of easy-to-measure process variables. Deep learning, as a novel training strategy for deep neural networks, has recently become a popular data-driven approach in the area of machine learning. In the present study, the deep learning technique is employed to build soft sensors and applied to an industrial case to estimate the heavy diesel 95% cut point of a crude distillation unit (CDU). The comparison of modeling results demonstrates that the deep learning technique is especially suitable for soft sensor modeling because of the following advantages over traditional methods. First, with a complex multi-layer structure, the deep neural network is able to contain richer information and yield improved representation ability compared with traditional data-driven models. Second, deep neural networks are established as latent variable models that help to describe highly correlated process variables. Third, the deep learning is semi-supervised so that all available process data can be utilized. Fourth, the deep learning technique is particularly efficient dealing with massive data in practice.