Multivariate Statistical Process Control Models Research Articles

In-Line Vis-NIR Spectral Analysis for the Column Chromatographic Processes of the Ginkgo biloba L. Leaves. Part II: Batch-to-Batch Consistency Evaluation of the Elution Process

An in-line monitoring method for the elution process of Ginkgo biloba L. leaves using visible and near-infrared spectroscopy in conjunction with multivariate statistical process control (MSPC) was established. Experiments, including normal operating batches and abnormal ones, were designed and carried out. The MSPC model for the elution process was developed and validated. The abnormalities were detected successfully by the control charts of principal component scores, Hotelling T2, or DModX (distance to the model). The results suggested that the established method can be used for the in-line monitoring and batch-to-batch consistency evaluation of the elution process.

Anomaly detection of the blast furnace smelting process using an improved multivariate statistical process control model

Anomaly detection and early warning of the blast furnace ironmaking process is an important research direction of blast furnace production. In this paper, through the analysis and optimization of the multivariate statistical process control (MSPC) model, MSPC based on TOPSIS and the grey (GT-MSPC) model is established based on the problems of missing alarms, false alarms and untimely prediction in the abnormal detection process of the traditional blast furnace ironmaking production process. First, a composite volatility (CV) that consists of the squared prediction error and T2 is computed to form the MSPC model. The parameters of different training set sizes and different alarm conditions are optimized by using the technique for order preference by similarity to an ideal solution model to optimize the effect of anomaly detection and fault diagnosis. Second, according to the results of the contribution plot, the fault diagnosis of abnormal occurrence is carried out, the main influencing factors of abnormal occurrence are analysed, and the monitoring is strengthened. Finally, the grey model is used to predict CV to realize anomaly early warning and form the GT-MSPC model. By collecting the field data of blast furnace production and performing empirical analysis, the results show that the GT-MSPC model has higher anomaly detectability. The detection rate is 50 % higher than that of the manual observation method, and the GT-MSPC model alarms 16.4861 min earlier than manual detection method. Fault diagnosis can accurately locate the main influencing factors of blast furnace anomalies, such as the furnace body static pressure and differential pressure. In summary, the GT-MSPC model realizes the accurate identification and prediction of anomalies and greatly reduces the occurrence of missing and false alarms and untimely prediction in the blast furnace system, thus reducing the risk of blast furnaces. It can be used to assist field engineers in dealing with abnormal faults and improve the product quality and production safety of blast furnace ironmaking.

Synchronization-Free Multivariate Statistical Process Control for Online Monitoring of Batch Process Evolution

Synchronization of variable trajectories from batch process data is a delicate operation that can induce artifacts in the definition of multivariate statistical process control (MSPC) models for real-time monitoring of batch processes. The current paper introduces a new synchronization-free approach for online batch MSPC. This approach is based on the use of local MSPC models that cover a normal operating conditions (NOC) trajectory defined from principal component analysis (PCA) modeling of non-synchronized historical batches. The rationale behind is that, although non-synchronized NOC batches are used, an overall NOC trajectory with a consistent evolution pattern can be described, even if batch-to-batch natural delays and differences between process starting and end points exist. Afterwards, the local MSPC models are used to monitor the evolution of new batches and derive the related MSPC chart. During the real-time monitoring of a new batch, this strategy allows testing whether every new observation is following or not the NOC trajectory. For a NOC observation, an additional indication of the batch process progress is provided based on the identification of the local MSPC model that provides the lowest residuals. When an observation deviates from the NOC behavior, contribution plots based on the projection of the observation to the best local MSPC model identified in the last NOC observation are used to diagnose the variables related to the fault. This methodology is illustrated using two real examples of NIR-monitored batch processes: a fluidized bed drying process and a batch distillation of gasoline blends with ethanol.

Fusing NIR and Process Sensors Data for Polymer Production Monitoring.

Process analytical technology and multivariate process monitoring are nowadays the most effective approaches to achieve real-time quality monitoring/control in production. However, their use is not yet a common practice, and industries benefit much less than they could from the outcome of the hundreds of sensors that constantly monitor production in industrial plants. The huge amount of sensor data collected are still mostly used to produce univariate control charts, monitoring one compartment at a time, and the product quality variables are generally used to monitor production, despite their low frequency (offline measurements at analytical laboratory), which is not suitable for real-time monitoring. On the contrary, it would be extremely advantageous to benefit from predictive models that, based on online sensors, will be able to return quality parameters in real time. As a matter of fact, the plant setup influences the product quality, and process sensors (flow meters, thermocouples, etc.) implicitly register process variability, correlation trends, drift, etc. When the available spectroscopic sensors, reflecting chemical composition and structure, consent to monitor the intermediate products, coupling process, and spectroscopic sensor and extracting/fusing information by multivariate analysis from this data would enhance the evaluation of the produced material features allowing production quality to be estimated at a very early stage. The present work, at a pilot plant scale, applied multivariate statistical process control (MSPC) charts, obtained by data fusion of process sensor data and near-infrared (NIR) probes, on a continuous styrene-acrylonitrile (SAN) production process. Furthermore, PLS regression was used for real-time prediction of the Melt Flow Index and percentage of bounded acrylonitrile (%AN). The results show that the MSPC model was able to detect deviations from normal operative conditions, indicating the variables responsible for the deviation, be they spectral or process. Moreover, predictive regression models obtained using the fused data showed better results than models computed using single datasets in terms of both errors of prediction and R 2. Thus, the fusion of spectra and process data improved the real-time monitoring, allowing an easier visualization of the process ongoing, a faster understanding of possible faults, and real-time assessment of the final product quality.

Maximizing returns from enterprise manufacturing intelligence and multivariate statistical process control

AbstractThis paper addresses challenges related to deploying analytics in the manufacturing environment. It discusses how to blend univariate and multivariate analyses into a deployment that can be successfully used by those not trained in Data Science. Enterprise manufacturing intelligence (EMI) has found great value in the chemical industry for aiding in timely decision making for improved plant reliability. It typically involves the use of control charts of multiple variables; that is, a univariate approach to data analysis. Another approach is to consider the data all together in a multivariate model, resulting in multivariate statistical process control (MSPC). These two approaches are complementary. Discussed in this report are guidelines for maximizing returns from EMI and MSPC deployments, including (1) considerations when setting up the MSPC model and (2) examples for how to interpret MSPC alerts, especially aimed at users who are not trained in multivariate data analysis.

Real-time monitoring and fault detection of pulsed-spray fluid-bed granulation using near-infrared spectroscopy and multivariate process trajectories

Pulsed spray is a useful tool for granule size control in fluid bed granulation. To improve the quality control of pulsed-spray fluid bed granulation, a combination of in-line near-infrared (NIR) spectroscopy and principal component analysis was used to develop multivariate statistical process control (MSPC) charts. Different types of MSPC charts were developed, including principal component score charts, Hotelling's T2 control charts, and distance to model X control charts, to monitor the batch evolution throughout the granulation process. Correlation optimized warping was used as an alignment method to deal with the time variation in batches caused by the granulation mechanism in MSPC modeling. The control charts developed in this study were validated on normal batches and tested on four batches that deviated from normal processing conditions to achieve real-time fault analysis. The results indicated that the NIR spectroscopy-based MSPC model included the variability in the sample set constituting the model and could withstand external variability. This research demonstrated the application of synchronized NIR spectra in conjunction with multivariate batch modeling as an attractive tool for process monitoring and a fault diagnosis method for effective process control in pulsed-spray fluid bed granulation.

Data fusion strategies to combine sensor and multivariate model outputs for multivariate statistical process control.

Process analytical technologies (PAT) applied to process monitoring and control generally provide multiple outputs that can come from different sensors or from different model outputs generated from a single multivariate sensor. This paper provides a contribution to current data fusion strategies for the combination of sensor and/or model outputs in the development of multivariate statistical process control (MSPC) models. Data fusion is explored through three real process examples combining output from multivariate models coming from the same sensor uniquely (in the near-infrared (NIR)-based end point detection of a two-stage polyester production process) or the combination of these outputs with other process variable sensors (using NIR-based model outputs and temperature values in the end point detection of a fluidized bed drying process and in the on-line control of a distillation process). The three examples studied show clearly the flexibility in the choice of model outputs (e.g. key properties prediction by multivariate calibration, process profiles issued from a multivariate resolution method) and the benefit of using MSPC models based on fused information including model outputs towards those based on raw single sensor outputs for both process control and diagnostic and interpretation of abnormal process situations. The data fusion strategy proposed is of general applicability for any analytical or bioanalytical process that produces several sensor and/or model outputs. Graphical abstract.

Potential of multivariate statistical process monitoring based on the biogas composition to detect free ammonia intoxication in anaerobic reactors

Three anaerobic digestion reactors were initially maintained at steady-state with low nitrogen input, then progressively exposed to increasing nitrogen input until extreme ammonia intoxication, leading to interruption of biogas production. Such interruption occurred for a free ammonia nitrogen (FAN) concentration of about 5 g Lslurry−1. A fourth reactor used as a control was maintained at low nitrogen steady-state. The biogas composition (CH4, CO2, H2 and H2S) was recorded over time and data were fed into a multivariate statistical process control model based on principal components analysis (PCA-MSPC). Static PCA-MSPC failed to distinguish between the actual reactor performance decrease and the drift of the in-control process baseline. However, recursively updated PCA-MSPC (effective memory length of 14 days, i.e. 0.25 hydraulic retention time) allowed detection of the early signs of FAN induced process dysfunction, which were simultaneous with drastic changes in the microbiome of the replicate reactors. However, the process control model was not able to adapt to a drastic modification of the microbiome. Propionate continuously accumulated since the detection of the first signs of process dysfunction.

Real-time monitoring of a coffee roasting process with near infrared spectroscopy using multivariate statistical analysis: A feasibility study

This work proposes the use of near infrared (NIR) spectroscopy in diffuse reflectance mode and multivariate statistical process control (MSPC) based on principal component analysis (PCA) for real-time monitoring of the coffee roasting process. The main objective was the development of a MSPC methodology able to early detect disturbances to the roasting process resourcing to real-time acquisition of NIR spectra. A total of fifteen roasting batches were defined according to an experimental design to develop the MSPC models. This methodology was tested on a set of five batches where disturbances of different nature were imposed to simulate real faulty situations. Some of these batches were used to optimize the model while the remaining was used to test the methodology. A modelling strategy based on a time sliding window provided the best results in terms of distinguishing batches with and without disturbances, resourcing to typical MSPC charts: Hotelling's T2 and squared predicted error statistics. A PCA model encompassing a time window of four minutes with three principal components was able to efficiently detect all disturbances assayed. NIR spectroscopy combined with the MSPC approach proved to be an adequate auxiliary tool for coffee roasters to detect faults in a conventional roasting process in real-time.

Monitoring method of extraction process for Schisandrae Chinensis Fructus based on near infrared spectroscopy and multivariate statistical process control

To establish an on-line monitoring method for extraction process of Schisandrae Chinensis Fructus, the formula medicinal material of Yiqi Fumai lyophilized injection by combining near infrared spectroscopy with multi-variable data analysis technology. The multivariate statistical process control (MSPC) model was established based on 5 normal batches in production and 2 test batches were monitored by PC scores, DModX and Hotelling T2 control charts. The results showed that MSPC model had a good monitoring ability for the extraction process. The application of the MSPC model to actual production process could effectively achieve on-line monitoring for extraction process of Schisandrae Chinensis Fructus, and can reflect the change of material properties in the production process in real time. This established process monitoring method could provide reference for the application of process analysis technology in the process quality control of traditional Chinese medicine injections.

Process modeling and control applied to real-time monitoring of distillation processes by near-infrared spectroscopy

A distillation device that acquires continuous and synchronized measurements of temperature, percentage of distilled fraction and NIR spectra has been designed for real-time monitoring of distillation processes. As a process model, synthetic commercial gasoline batches produced in Brazil, which contain mixtures of pure gasoline blended with ethanol have been analyzed. The information provided by this device, i.e., distillation curves and NIR spectra, has served as initial information for the proposal of new strategies of process modeling and multivariate statistical process control (MSPC). Process modeling based on PCA batch analysis provided global distillation trajectories, whereas multiset MCR-ALS analysis is proposed to obtain a component-wise characterization of the distillation evolution and distilled fractions. Distillation curves, NIR spectra or compressed NIR information under the form of PCA scores and MCR-ALS concentration profiles were tested as the seed information to build MSPC models. New on-line PCA-based MSPC approaches, some inspired on local rank exploratory methods for process analysis, are proposed and work as follows: a) MSPC based on individual process observation models, where multiple local PCA models are built considering the sole information in each observation point; b) Fixed Size Moving Window – MSPC, in which local PCA models are built considering a moving window of the current and few past observation points; and c) Evolving MSPC, where local PCA models are built with an increasing window of observations covering all points since the beginning of the process until the current observation. Performance of different approaches has been assessed in terms of sensitivity to fault detection and number of false alarms. The outcome of this work will be of general use to define strategies for on-line process monitoring and control and, in a more specific way, to improve quality control of petroleum derived fuels and other substances submitted to automatic distillation processes monitored by NIRS.

Multivariate statistical process control (MSPC) using Raman spectroscopy for in-line culture cell monitoring considering time-varying batches synchronized with correlation optimized warping (COW)

Multivariate statistical process control (MSPC) is increasingly popular as the challenge provided by large multivariate datasets from analytical instruments such as Raman spectroscopy for the monitoring of complex cell cultures in the biopharmaceutical industry. However, Raman spectroscopy for in-line monitoring often produces unsynchronized data sets, resulting in time-varying batches. Moreover, unsynchronized data sets are common for cell culture monitoring because spectroscopic measurements are generally recorded in an alternate way, with more than one optical probe parallelly connecting to the same spectrometer. Synchronized batches are prerequisite for the application of multivariate analysis such as multi-way principal component analysis (MPCA) for the MSPC monitoring. Correlation optimized warping (COW) is a popular method for data alignment with satisfactory performance; however, it has never been applied to synchronize acquisition time of spectroscopic datasets in MSPC application before. In this paper we propose, for the first time, to use the method of COW to synchronize batches with varying durations analyzed with Raman spectroscopy. In a second step, we developed MPCA models at different time intervals based on the normal operation condition (NOC) batches synchronized by COW. New batches are finally projected considering the corresponding MPCA model. We monitored the evolution of the batches using two multivariate control charts based on Hotelling's T2 and Q. As illustrated with results, the MSPC model was able to identify abnormal operation condition including contaminated batches which is of prime importance in cell culture monitoring We proved that Raman-based MSPC monitoring can be used to diagnose batches deviating from the normal condition, with higher efficacy than traditional diagnosis, which would save time and money in the biopharmaceutical industry.

Transfer of a static PCA-MSPC model from a steady-state anaerobic reactor to an independent anaerobic reactor exposed to organic overload

A static multivariate statistical process control model based on principal component analysis (PCA-MSPC) was developed for an anaerobic reactor maintained in steady-state, joining the biogas composition (CH4, CO2, H2) to the total solids (TS), volatile solids (VS), total inorganic carbon (TIC) and total ammonia nitrogen (TAN) contents of the slurry. The principal component analysis (PCA) highlighted a lack of correlation between the individual process variables (IPVs) measured in the slurry and the gas phase. The application of this model to the data set collected for an independent anaerobic reactor (fed with the same substrate) progressively led from steady-state to critical volatile fatty acids (VFA) intoxication did not allow evaluating its process status. A second static PCA-MSPC model was built for the steady-state reactor in excluding the TS and VS content of the slurry and was successfully transferred to the overfed reactor. The TA² and SPE control charts built for the overfed reactor using this second model closely reflected its process status and delivered valuable warning signals approaching the VFA intoxication. The gas phase composition (CH4, CO2, H2) brought the main contribution to these warnings.

Monitoring method for macroporous resin column chromatography process of salvianolic acids based on near infrared spectroscopy

To study and establish a monitoring method for macroporous resin column chromatography process of salvianolic acids by using near infrared spectroscopy (NIR) as a process analytical technology (PAT).The multivariate statistical process control (MSPC) model was developed based on 7 normal operation batches, and 2 test batches (including one normal operation batch and one abnormal operation batch) were used to verify the monitoring performance of this model. The results showed that MSPC model had a good monitoring ability for the column chromatography process. Meanwhile, NIR quantitative calibration model was established for three key quality indexes (rosmarinic acid, lithospermic acid and salvianolic acid B) by using partial least squares (PLS) algorithm. The verification results demonstrated that this model had satisfactory prediction performance. The combined application of the above two models could effectively achieve real-time monitoring for macroporous resin column chromatography process of salvianolic acids, and can be used to conduct on-line analysis of key quality indexes. This established process monitoring method could provide reference for the development of process analytical technology for traditional Chinese medicines manufacturing.

Data density-based fault detection and diagnosis with nonlinearities between variables and multimodal data distributions

Multivariate statistical process control (MSPC) is an important means of monitoring multiple process variables and their interrelationships while controlling chemical and industrial plants efficiently and stably. To consider nonlinearities between process variables and multimodal data distributions, the data density can be used as an index for fault detection. Data domains with a low data density are considered abnormal states. However, after fault detection, faulty process variables cannot be diagnosed with an MSPC model based on the data density. Therefore, we have developed a new index to diagnose the process variables that contribute to process faults using a data density-based MSPC model. The proposed index uses the partial derivative of an MSPC model with respect to each process variable. We demonstrate the effectiveness of the proposed method using numerical simulation data, Tennessee Eastman process data, and real plant data analyses.

Real-Time Process Analytical Technology Assurance for Flow Synthesis of Oligonucleotides

A feasibility study has been conducted using Process Analytical Technology (PAT) monitoring and chemometric modeling techniques to mitigate risks identified for the solid-phase synthesis of a model oligonucleotide compound in a flow reactor manifold. This paper will discuss three of the key risks identified for this automated process that are generally applicable to oligonucleotide manufacturing: (1) connection of an incorrect chemical solution to a designated input port of the synthesizer; (2) incorrect quantitative preparation of a chemical solution; and (3) errors that may come from mechanical or other unknown sources associated with the synthesizer. Mid-infrared and Raman spectroscopy were the key PAT techniques used to monitor the automated synthetic process in real time. Classification, quantitative (partial least-squares), and multivariate statistical process control (MSPC) modeling were then used to analyze the acquired data and detect the occurrence of the aforementioned failure modes. MSPC model...

Towards multivariate statistical process control in the wood pellet industry

Abstract Multivariate statistical process control (MSPC) based on principal component analysis (PCA) and partial least squares (PLS) regression was simulated, based on industrial data collected over a two-year period within a plant producing wood pellets as biofuel. The data used in the simulations consisted of values of five variables of analysed intermediate products (sawdust and powder) and end products (pellets), acquired during processes with seven on-line settings of controls. PCA global modelling revealed an overlap in the data between years and detected three different pellet types. Correlations within the dataset indicated there was a time lag of up to 14 h. Therefore, PLS prediction of current product values was based on observations containing the current process settings and all variable values within a preceding 18 h time interval. Global models showed that predictions of the dryness of sawdust, milled sawdust and pellets had good accuracy, whereas predictions of pellet bulk density and mechanical durability were less accurate. Dynamic and local PLS modelling showed that more accurate predictions of pellet dryness were obtained if all previous observations were included in the calibration set rather than observations in calibration windows of the 10 or 100 preceding observations. The results illustrate the possibilities to implement MSPC in the wood pellet industry, potentially handling huge amounts of data. To develop and implement the next phase of process control more parameters must be included in the MSPC models, e.g. data acquired using on-line instruments to continuously collect information on variations in the stream of material.

Process performance monitoring in the presence of confounding variation

The aim of this paper is to present a two-stage partial least squares (PLS) methodology, for the monitoring of processes that are known to be affected by sources of variation that are an inherent part of routine operations. These sources, termed nuisance or confounding variation, will typically either mask the more subtle process changes that are of particular interest to operational personnel, or make the determination of the source of non-conforming operation more difficult to locate. In the first stage of the two-stage algorithm, the sources of confounding variation are extracted and removed through the application of a PLS based filter. The second stage takes the filtered ‘signal’ and through the application of PLS those latent variables that are uncorrelated with the nuisance source of variation are extracted. These latent variables then form the basis of a multivariate statistical process control model. The algorithm is compared with the ordinary PLS based performance monitoring and a monitoring scheme that is formed from the latent variables of a PLS model developed after the application of Orthogonal Signal Correction (OSC). The methodologies are illustrated and compared by application to a mathematical simulation example and an industrial semi-discrete batch manufacturing process. In both cases it is shown that the two-stage PLS algorithm is more able to detect and locate the sources of subtle process changes.

In situ near Infrared Monitoring of Activated Dairy Sludge Wastewater Treatment Processes

A sequential batch reactor (SBR) for aerobic treatment of dairy residues (activated sludge process) was monitored on-line by near infrared (NIR) spectroscopy. NIR spectra in the wavelength range of 900–1700 nm were acquired in situ with a transflectance probe. Measured in situ NIR spectra allowed the development of multivariate statistical process control models for supervision of the SBR process. The directly acquired spectra permitted characterisation of the different SBR operation stages and the detection of process disturbances. Calibration models based on partial least squares regression were developed for three process parameters: total solids (TS), total suspended solids (TSS) and chemical oxygen demand (COD). Model errors were estimated as root mean squared errors of cross-validation. The cross-validation errors obtained were 0.066 g TS L−1, 0.075 g TSS L−1 and 86.6 mg O2 L−1 respectively. Results demonstrated the viability of NIR spectroscopy as an alternative method for monitoring and control of activated sludge wastewater treatment processes having advantages both in terms of efficient determination of process parameters (fast, in situ and requiring no toxic reagents) and in terms of process monitoring.

Online monitoring of multi-phase batch processes using phase-based multivariate statistical process control

Online monitoring of batch processes using multivariate statistical methods has attracted enormous research interests due to its practical importance. In this paper, we focus on an important issue that continues to confound online batch process monitoring—run-to-run variations that do not confirm to a normal distribution around a reference trajectory. Here, we show that a phase-based decomposition of the trajectory offers a systematic way to overcome this challenge. In our approach, phase changes are detected online using Singular points in key variables. Run-to-run variations among different instances of a phase are synchronized by using time warping. Finally, phased-based multivariate statistical process control models are used to monitor the execution of the batch and detect abnormalities. This phase-based monitoring approach is robust to run-to-run variations arising from changes in initial conditions and event timings as is illustrated using a well-known fermentation process simulation.