A quality relevant non‐Gaussian latent subspace projection method for chemical process monitoring and fault detection

Partial least‐squares (PLS) method has been widely used in multivariate statistical process monitoring field. The goal of traditional PLS is to find the multidimensional directions in the measurement‐variable and quality‐variable spaces that have the maximum covariances. Therefore, PLS method relies on the second‐order statistics of covariance only but does not takes into account the higher‐order statistics that may involve certain key features of non‐Gaussian processes. Moreover, the derivations of control limits for T² and squared prediction error (SPE) indices in PLS‐based monitoring method are based on the assumption that the process data follow a multivariate Gaussian distribution approximately. Meanwhile, independent component analysis (ICA) approach has recently been developed for process monitoring, where the goal is to find the independent components (ICs) that are assumed to be non‐Gaussian and mutually independent by means of maximizing the high‐order statistics such as negentropy instead of the second‐order statistics including variance and covariance. Nevertheless, the IC directions do not take into account the contributions from quality variables and, thus, ICA may not work well for process monitoring in the situations when the quality variables have strong influence on process operations. To capture the non‐Gaussian relationships between process measurement and quality variables, a novel projection‐based monitoring method termed as quality relevant non‐Gaussian latent subspace projection (QNGLSP) approach is proposed in this article. This new technique searches for the feature directions within the measurement‐variable and quality‐variable spaces concurrently so that the two sets of feature directions or subspaces have the maximized multidimensional mutual information. Further, the new monitoring indices including I² and SPE statistics are developed for quality relevant fault detection of non‐Gaussian processes. The proposed QNGLSP approach is applied to the Tennessee Eastman Chemical process and the process monitoring results of the present method are demonstrated to be superior to those of the PLS‐based monitoring method. © 2013 American Institute of Chemical Engineers AIChE J 60: 485–499, 2014     

Real time monitoring of bioreactor mAb IgG3 cell culture process dynamics via Fourier transform infrared spectroscopy: Implications for enabling cell culture process analytical technologyŽ 

Compared to small molecule process analytical technology (PAT) applications, biotechnology product PAT applications have certain unique challenges and opportunities. Understanding process dynamics of bioreactor cell culture process is essential to establish an appropriate process control strategy for biotechnology product PAT applications. Inline spectroscopic techniques for real time monitoring of bioreactor cell culture process have the distinct potential to develop PAT approaches in manufacturing biotechnology drug products. However, the use of inline Fourier transform infrared (FTIR) spectroscopic techniques for bioreactor cell culture process monitoring has not been reported. In this work, real time inline FTIR Spectroscopy was applied to a lab scale bioreactor mAb IgG3 cell culture fluid biomolecular dynamic model. The technical feasibility of using FTIR Spectroscopy for real time tracking and monitoring four key cell culture metabolites (including glucose, glutamine, lactate, and ammonia) and protein yield at increasing levels of complexity (simple binary system, fully formulated media, actual bioreactor cell culture process) was evaluated via a stepwise approach. The FTIR fingerprints of the key metabolites were identified. The multivariate partial least squares (PLS) calibration models were established to correlate the process FTIR spectra with the concentrations of key metabolites and protein yield of in-process samples, either individually for each metabolite and protein or globally for all four metabolites simultaneously. Applying the 2^nd derivative pre-processing algorithm to the FTIR spectra helps to reduce the number of PLS latent variables needed significantly and thus simplify the interpretation of the PLS models. The validated PLS models show promise in predicting the concentration profiles of glucose, glutamine, lactate, and ammonia and protein yield over the course of the bioreactor cell culture process. Therefore, this work demonstrated the technical feasibility of real time monitoring of the bioreactor cell culture process via FTIR spectroscopy. Its implications for enabling cell culture PAT were discussed.     

Simultaneous determination of molecular weight and crystallinity of recycled HDPE by infrared spectroscopy and multivariate calibration

An attempt of correlating molecular weight (M_n) of recycled high‐density polyethylene (HDPE) as measured by size‐exclusion chromatography (SEC) with diffuse reflectance near and mid‐infrared spectroscopy (NIR/MIR) was made by means of multivariate calibration. The spectral data obtained was also used to extract information about the degree of crystallinity of the recycled resin. Differential scanning calorimetry (DSC) was used as the reference method. Partial least‐squares (PLS) calibration was performed on the MIR and NIR spectral data for prediction of M_n. Four PC factors described fully the PLS models. The root‐mean‐square error of prediction (RMSEP) obtained with MIR data was 360, whereas a RMSEP of 470 was achieved when calibration was carried out on the diffuse reflectance NIR data. A PLS calibration for prediction of degree of crystallinity was performed on the NIR data in the 1100–1900‐nm region, but the ability of prediction of this model was poor. However a PLS calibration in the region 2000–2500 nm yield better results. Four PC factors explained the most of the variance in the spectra and the RMSEP was 0.4 wt %. © 2002 Wiley Periodicals, Inc. J Appl Polym Sci 85: 321–327, 2002     

Multivariate statistical analysis of mid‐infrared spectra for the online monitoring of 2‐ethylhexyl acrylate/styrene emulsion copolymerization

This article describes the application of multivariate statistical process control techniques to a series of mid‐infrared spectra collected online from a styrene/2‐ethylhexyl acrylate emulsion copolymerization process. Principal component analysis of the mid‐infrared spectral data indicated that in situ monitoring of the complex copolymerization process was feasible in the spectral region of interest. It was also observed that projection to latent structures or partial least squares (PLS) could be used for the effective indirect online prediction of individual monomer conversions and copolymer compositions over a substantial range of process operating conditions. A combination of the developed PLS methodology with a mid‐infrared attenuated total reflection probe proved to be an effective tool for the efficient online characterization of polymer quality, thereby overcoming the lack of robust online conversion and composition measuring devices. © 2001 John Wiley & Sons, Inc. J Appl Polym Sci 82: 1776–1787, 2001     

Multivariate statistical monitoring of process operating performance

Process computers routinely collect hundreds to thousands of pieces of data from a multitude of plant sensors every few seconds. This has caused a “data overload” and due to the lack of appropriate analyses very little is currently being done to utilize this wealth of information. Operating personnel typically use only a few variables to monitor the plant's performance. However, multivariate statistical methods such as PLS (Partial Least Squares or Projection to Latent Structures) and PCA (Principal Component Analysis) are capable of compressing the information down into low dimensional spaces which retain most of the information. Using this method of statistical data compression a multivariate monitoring procedure analogous to the univariate Shewart Chart has been developed to efficiently monitor the performance of large processes, and to rapidly detect and identify important process changes. This procedure is demonstrated using simulations of two processes, a fluidized bed reactor and an extractive distillation column.     

Improvement of Filter Tow Quality through Statistical Analysis

Filter tow production is controlled by individual supervision of the main variables based on the determination of the C_pk, which indicates instability when individual values are below 1. This paper describes a process improvement methodology, which uses statistical tools to correlate the production variables, and thus control the variability of the dependent variables. The analysis was performed over several sets of data, identifying dependent variables – yield (g) and tensile strength (daN) – and establishing relationships of cause and effect with the production variables, indicating the possibility of achieving a new form of control through adjusted multiple regression models, resulting in the C_pk of the dependent variables being greater than 1. These results were improved through an industrial test, using DOE 2^k, which led to the conclusion that the process can be supervised by means of the fitted models, ensuring better control and greater stability.     

化工过程故障诊断中的多元统计方法

综述了多元统计方法在化工过程故障诊断领域的理论进展和应用现状,介绍了一些主要的多元统计方法,其中包括主元分析、部分最小二乘、独立成分分析和Fisher判别分析,展望了多元统计方法在化工过程故障诊断领域的发展与应用前景。     

Process control utilizing data based multivariate statistical models

A process control approach using steady state multivariate statistical models is presented. The goal of this control approach is to improve product quality when the quality measurements are not available on line, or they have long time delays. Principal Component Analysis (PCA) is used to compress information from the process measurements down to a lower dimensional score space, where a control goal is specified using the approach of Piovoso and Kosanovich (1992). A new statistical controller is designed to control the equivalent score space representation of the process. The issue of how to account for the correlation structure of input variables when closing a feedback loop around the PCA model is specifically addressed. A binary distillation column and the Tennessee Eastman process are used for demonstrating the new control approach.     

On-line batch process monitoring using dynamic PCA and dynamic PLS models

Producing value-added products of high-quality is the common objective in industries. This objective is more difficult to achieve in batch processes whose key quality measurements are not available on-line. In order to reduce the variations of the product quality, an on-line batch monitoring scheme is developed based on the multivariate statistical process control. It suggests using the past measured process variables without real-time quality measurement at the end of the batch run. The method, referred to as BDPCA and BDPLS, integrates the time-lagged windows of process dynamic behavior with the principal component analysis and partial least square respectively for on-line batch monitoring. Like traditional MPCA and MPLS approaches, the only information needed to set up the control chart is the historical data collected from the past successful batches. This leads to simple monitoring charts, easy tracking of the progress in each batch run and monitoring the occurrence of the observable upsets. BDPCA and BDPLS models only collect the previous data during the batch run without expensive computations to anticipate the future measurements. Three examples are used to investigate the potential application of the proposed method and make a comparison with some traditional on-line MPCA and MPLS algorithms.     

Estimating Glucan,Xylan, and Methylglucuronic Acids in Kraft Pulps of Eucalyptus globulus Using FT-NIR Spectroscopy and Multivariate Analysis

Abstract

Fourier transform near infrared (FT-NIR) associated with multivariate analysis was used to estimate glucan, xylan, 4-O-Methyl-α -D-glucuronic acid (MeGlcA) content, and pulp yield in kraft pulps of Eucalyptus globulus Labill. Several models were applied to correlate chemical composition in samples with the NIR spectral data by means of principal components regression (PCR) or partial least square (PLS) algorithms. Calibration models were built and validated by using all the spectral data and cross-validation methodology. The r_c ² values for the best calibration models for quantification of glucan, xylan, MeGlcA contents and pulp yield were between 0.71–0.92. The model was validated using a set of external samples. The amount of glucan (64–77%), xylan (12–18%), and MeGlcA (204–363 mmol kg pulp^–1) in pulps were predicted with a root mean square error of prediction (RMSEP) of 0.91%, 0.46%, and 15.21% for glucan, xylan, and MeGlcA, respectively. Pulp yield (in the range of 46–70%) was also predicted with good accuracy with a RMSEP of 1.63%. These results suggest that glucan, xylan, MeGlcA composition, and pulp yield in kraft pulps of E. globulus can be adequately estimated by NIR spectroscopy for laboratory or industrial applications. NIR predictions can also provide useful and cost-effective tools for the rapid screening of large numbers of samples.     

Commercial spread optimization using experimental design and sensory data

The cost and quality of food products are issues that concern both the consumer and producer. In this research, the process used for the production of a commercial spread was subjected to a statistical experimental design for the purpose of reducing the cost of production while maintaining or improving the sensory quality. Three factors—the amount of oil added (x ₁), the speed of puddling (x ₂), and the temperature treatment (A or B; x ₃)—were varied according to a full-factorial design at two levels. The experiments were performed over 2 d, and the factorial design was complemented with three replicates for temperature treatments A and B, which were performed on different days. The products were evaluated with both sensory and physicochemical measurements. Special attention was paid to the hardness of the product since it was permissible to reduce it slightly. In contrast, sensory quality aspects of the product including butter-aroma and off-flavor, as well as other quality properties such as spreadability, shine, and meltability, had to be maintained at the present level or improved. Statistical evaluation of the data showed that it was possible to add high amounts of oil (x ₁) without impairing the sensory quality of the product and, hence, reduce the cost of production. The hardness of the product was also slightly reduced when using the high level of oil. In maintaining other sensory qualities such as shine and spreadability at the present levels, the choice of temperature treatment (x ₃) was important.     

The Statistical Regression Calculation in Plastics Processing – Process Analysis,Optimization and Monitoring

Plastics processing companies can only meet up to present‐day quality requirements if they adopt systematic methods. This holds particularly true for the extremely stringent demands that are now placed on injection molding technology. Working on from a sound experimental basis, it is possible to define cause/effect correlations for two sets of empirical data (the current process conditions and the molded part attributes) for each quality variable by using a statistical process model. The process model enables the processor to calculate the effect of each individual combination of parameters in the experimental area and to perform an optimization. If it proves possible to describe the cause/effect correlation between the fluctuations in the molded part attributes and those in the process parameters by means of a statistical process model, then this can be used for the continuous monitoring of production. The statistical experimentation method and continuous process monitoring are grouped together to form the so‐called CPC concept, permitting traceable, gap‐free documentation of the quality data for a production chain. Three examples are set out to illustrate the possibilities for use of the CPC concept; these are then assessed on the basis of the benefit observed.

Engine‐cooling fan.     

Application of raman spectroscopy to monitor and quantify ethyl esters in soybean oil transesterification

Biodiesel (FA esters) has become very attractive as an alternative diesel fuel owing to its environmental benefits. Transesterification is the most usual and important method to make biodiesel from vegetable oils. This article investigates the potential for using Raman spectroscopy to monitor and quantify the transesterification of soybean oil to yield ethyl esters. The differences observed in the Raman spectra of soybean oil after transesterification were a peak at 2932 cm⁻¹ ( ), the displacement of the v _C=O band from 1748 to 1739 cm⁻¹, and the bands at 861 (v _R-C=O and v _C-C) and 372 cm⁻¹ (δ _CO-O-C). Uni- and multivariate analysis methods were used to build several analytical curves and then applied in known samples, treated as unknowns, to test their ability to predict concentrations. The best results were achieved by Raman/PLS calibration models (where PLS=partial least squares regression) using an internal normalization standard (v _=C-H band). The correlation coefficient (R ²) values so obtained were 0.9985 for calibration and 0.9977 for validation. Univariate regression analysis between biodiesel concentration and the increasing intensity of band or v _C=O displacement showed R ² values of 0.9983 and 0.9742, respectively. Although spectroscopic methods are less sensitive than chromatographic ones, the data obtained by spectroscopy can be correlated with other techniques, allowing biodiesel yield and quality to be quickly assessed.     

Application of partial least squares methods to a terephthalic acid manufacturing process for product quality control

This paper deals with an application of partial least squares (PLS) methods to an industrial terephthalic acid (TPA) manufacturing process to identify and remove the major causes of variability in the product quality. Multivariate statistical analyses were performed to find the major causes of variability in the product quality, using the PLS models built from historical data measured on the process and quality variables. It was found from the PLS analyses that the variations in the catalyst concentrations and the process throughput significantly affect the product quality, and that the quality variations are propagated from the oxidation unit to the digestion units of the TPA process. A simulation-based approach was addressed to roughly estimate the effects of eliminating the major causes on the product quality using the PLS models. Based on the results that considerable amounts of the variations in the product quality could be reduced, we have proposed practical approaches for removing the major causes of product quality variations in the TPA manufacturing process.     

Attempt to correlate the yield processes above and below the glass transition in glassy polymers

This paper is concerned with a model which attempts to describe quantitatively, by the same elementary process, the yield behaviour above and below T_g, as well as the effect of annealing on the yield stress. This model links together theories we have previously proposed and relies on the following main assumptions: the deformation processes imply the cooperation of n activated segments and that the free energy increase of an activated segment depends on the structural state of the polymer. A satisfactory agreement is found with yield stress data on polycarbonate (PC), over a very large range of temperatures and strain rates. The correlation between the yield stress and the annealing treatment is also reasonable.     

Total projection to latent structures for process monitoring

Partial least squares or projection to latent structures (PLS) has been used in multivariate statistical process monitoring similar to principal component analysis. Standard PLS often requires many components or latent variables (LVs), which contain variations orthogonal to Y and useless for predicting Y . Further, the X ‐residual of PLS usually has quite large variations, thus is not proper to monitor with the Q‐statistic. To reduce false alarm and missing alarm rates of faults related to Y , a total projection to latent structures (T‐PLS) algorithm is proposed in this article. The new structure divides the X ‐space into four parts instead of two parts in standard PLS. The properties of T‐PLS are studied in detail, including its relationship to the orthogonal PLS. Further study shows the space decomposition on X ‐space induced by T‐PLS. Fault detection policy is developed based on the T‐PLS. Case studies on two simulation examples show the effectiveness of the T‐PLS based fault detection methods. © 2009 American Institute of Chemical Engineers AIChE J, 2010     

基于KPLS模型的间歇过程产品质量控制

针对间歇过程所具有的非线性特性,提出了一种基于核偏最小二乘(KPLS)模型的最终产品质量控制策略。利用初始条件、批次展开后的过程数据以及最终产品质量建立了间歇过程的KPLS模型;采用基于主成分分析(PCA)映射的预估方法对未知的过程数据进行补充,实现了最终产品质量的在线预测。为了解决最终产品质量的控制,利用T2统计量确定KPLS模型的适用范围,并作为约束引入产品质量控制问题,提高控制策略的可行性;采用粒子群优化(PSO)算法实现了优化问题的高效求解。仿真结果表明,与基于偏最小二乘(PLS)模型的控制策略相比,所提出的方法具有更高的预测精度,且能有效解决产品质量控制中出现的各种问题。     

Offset‐Free Inferential Feedback Control of Distillation Compositions Based on PCR and PLS Models

An offset‐free inferential feedback control strategy for distillation composition control using principal component regression (PCR) and partial least squares (PLS) models is presented in this paper. PCR and PLS model based software sensors are developed from process operational data so that the top and bottom product compositions can be estimated from multiple tray temperature measurements. The PCR and PLS software sensors are then used in the feedback control of the top and bottom product compositions. With this strategy the problem of substantial time delay in composition analyzer based control and of substantial bias in single tray temperature control can be overcome. A practically very important issue in software sensor based feedback control is that static control offsets often exist due to a static estimation bias, especially when the process operating condition changes. A technique for eliminating the static estimation bias and the resulting static control offsets through mean updating of process measurements is proposed in this paper. Applications to a simulated methanol‐water separation column demonstrate the effectiveness of this control strategy.     

Inferential sensors for estimation of polymer quality parameters: Industrial application of a PLS-based soft sensor for a LDPE plant

Low-density polyethylene (LDPE) and ethylene vinyl acetate (EVA) copolymers are produced in free radical polymerization using reactors at extremely high pressure. The reactors require constant monitoring and control in order to minimize undesirable process excursions and meet stringent product specifications. In industrial settings, polymer quality is mainly specified in terms of melt flow index (MI) and density. These properties are difficult to measure and usually unavailable in real time, which leads to major difficulty in controlling product quality in polymerization processes. Researchers have attempted first principles modeling of polymerization processes to estimate end use properties. However, development of detailed first principles model for free radical polymerization is not a trivial task. The difficulties involved are the large number of complex and simultaneous reactions and the need to estimate a large number of kinetic parameters. To overcome these difficulties, some researchers considered empirical neural network models as an alternative. However, neural network models provide no physical insight about the underlying process. We consider data-based multivariate regression methods as alternative solution to the problem. In this paper, some recent developments in modeling polymer quality parameters are reviewed, with emphasis given to the free radical polymerization process. We present an application of PLS to build a soft-sensor to predict melt flow index using routinely measured process variables. Issues of data acquisition and preprocessing for real industrial data are discussed. The study was conducted using data collected form an industrial autoclave reactor, which produces LDPE and EVA copolymer using free radical polymerization. The results indicated that melt index (MI) can be successfully predicted using this relatively straightforward statistical tool.