Adaptive multiscale principal component analysis for on-line monitoring of a sequencing batch reactor

In recent years, multiscale monitoring approaches, which combine principal component analysis (PCA) and multi-resolution analysis (MRA), have received considerable attention. These approaches are potentially very efficient for detecting and analyzing diverse ranges of faults and disturbances in chemical and biochemical processes. In this work, multiscale PCA is proposed for fault detection and diagnosis of batch processes. Using MRA, measurement data are decomposed into approximation and details at different scales. Adaptive multiway PCA (MPCA) models are developed to update the covariance structure at each scale to deal with changing process conditions. Process monitoring by a unifying adaptive multiscale MPCA involves combining only those scales where significant disturbances are detected. This multiscale approach facilitates diagnosis of the detected fault as it hints to the time-scale under which the fault affects the process. The proposed adaptive multiscale method is successfully applied to a pilot-scale sequencing batch reactor for biological wastewater treatment.     

动态MPCA在发酵过程监测与故障诊断中的应用

针对发酵过程非线性和时变特点,提出了一种具有实时性的动态MPCA方法,采用多模型非线性结构代替传统MPCA单模型线性化结构,克服了后者不能处理非线性过程和实时性的问题,并避免了MPCA在线应用时预报未来测量值带来的误差,提高了发酵过程性能监测和故障诊断的准确性。对头孢菌素C发酵过程的拟在线仿真研究,验证了基于动态MPCA的统计过程监测的有效性。     

Enhanced process monitoring of fed-batch penicillin cultivation using time-varying and multivariate statistical analysis

On-line monitoring of penicillin cultivation processes is crucial to the safe production of high-quality products. In the past, multiway principal component analysis (MPCA), a multivariate projection method, has been widely used to monitor batch and fed-batch processes. However, when MPCA is used for on-line batch monitoring, the future behavior of each new batch must be inferred up to the end of the batch operation at each time and the batch lengths must be equalized. This represents a major shortcoming because predicting the future observations without considering the dynamic relationships may distort the data information, leading to false alarms. In this paper, a new statistical batch monitoring approach based on variable-wise unfolding and time-varying score covariance structures is proposed in order to overcome the drawbacks of conventional MPCA and obtain better monitoring performance. The proposed method does not require prediction of the future values while the dynamic relations of data are preserved by using time-varying score covariance structures, and can be used to monitor batch processes in which the batch length varies. The proposed method was used to detect and identify faults in the fed-batch penicillin cultivation process, for four different fault scenarios. The simulation results clearly demonstrate the power and advantages of the proposed method in comparison to MPCA.     

Nonlinear biological batch process monitoring and fault identification based on kernel fisher discriminant analysis

A novel nonlinear biological batch process monitoring and fault identification approach based on kernel Fisher discriminant analysis (kernel FDA) is proposed. This method has a powerful ability to deal with nonlinear data and does not need to predict the future observations of variables. So it is more sensitive to fault detection. In order to improve the monitoring performance, variable trajectories of the batch processes are separated into several blocks. Then data in the original space is mapped into high-dimensional feature space via nonlinear kernel function and the optimal kernel Fisher feature vector and discriminant vector are extracted to perform process monitoring and fault identification. The key to the proposed approach is to calculate the distance of block data which are projected to the optimal kernel Fisher discriminant vector between new batch and reference batch. Through comparing distance with the predefined threshold, it can be considered whether the batch is normal or abnormal. Similar degree between the present discriminant vector and the optimal discriminant vector of fault in historical data set is used to perform fault diagnosis. The proposed method is applied to the process of fed-batch penicillin fermentation simulator benchmark and shows that it can effectively capture nonlinear relationships among process variables and is more efficient than MPCA approach.     

Multi-model statistical process monitoring and diagnosis of a sequencing batch reactor

Biological processes exhibit different behavior depending on the influent loads, temperature, microorganism activity, and so on. It has been shown that a combination of several models can provide a suitable approach to model such processes. In the present study, we developed a multiple statistical model approach for the monitoring of biological batch processes. The proposed method consists of four main components: (1) multiway principal component analysis (MPCA) to reduce the dimensionality of data and to remove collinearity; (2) multiple models with a posterior probability for modeling different operating regions; (3) local batch monitoring by the T(2)- and Q-statistics of the specific local model; and (4) a new discrimination measure (DM) to identify when the system has shifted to a new operating condition. Under this approach, local monitoring by multiple models divides the entire historical data set into separate regions, which are then modeled separately. Then, these local regions can be supervised separately, leading to more effective batch monitoring. The proposed method is applied to a pilot-scale 80-L sequencing batch reactor (SBR) for biological wastewater treatment. This SBR is characterized by nonstationary, batchwise, and multiple operation modes. The results obtained for the pilot-scale SBR indicate that the proposed method has the ability to model multiple operating conditions, to identify various operating regions, and also to determine whether the biosystem has shifted to a new operating condition. Our findings show that the local monitoring approach can give more reliable and higher resolution monitoring results than the global model.     

Fault diagnosis of a benchmark fermentation process: a comparative study of feature extraction and classification techniques

This paper investigates fault diagnosis in batch processes and presents a comparative study of feature extraction and classification techniques applied to a specific biotechnological case study: the fermentation process model by Birol et al. (Comput Chem Eng 26:1553–1565, 2002), which is a benchmark for advanced batch processes monitoring, diagnosis and control. Fault diagnosis is achieved using four approaches on four different process scenarios based on the different levels of noise so as to evaluate their effects on the performance. Each approach combines a feature extraction method, either multi-way principal component analysis (MPCA) or multi-way independent component analysis (MICA), with a classification method, either artificial neural network (ANN) or support vector machines (SVM). The performance obtained by the different approaches is assessed and discussed for a set of simulated faults under different scenarios. One of the faults (a loss in mixing power) could not be detected due to the minimal effect of mixing on the simulated data. The remaining faults could be easily diagnosed and the subsequent discussion provides practical insight into the selection and use of the available techniques to specific applications. Irrespective of the classification algorithm, MPCA renders better results than MICA, hence the diagnosis performance proves to be more sensitive to the selection of the feature extraction technique.     

Growing multiblock structures: a semi-automated approach to block placement for multiblock hexahedral meshing

Finite element (FE) analysis is a cornerstone of orthopaedic biomechanics research. Three-dimensional medical imaging provides sufficient resolution for the subject-specific FE models to be generated from these data-sets. FE model development requires discretisation of a three-dimensional domain, which can be the most time-consuming component of a FE study. Hexahedral meshing tools based on the multiblock method currently rely on the manual placement of building blocks for mesh generation. We hypothesise that angular analysis of the geometric centreline for a three-dimensional surface could be used to automatically generate building block structures for the multiblock hexahedral mesh generation. Our algorithm uses a set of user-defined points and parameters to automatically generate a multiblock structure based on a surface's geometric centreline. This significantly reduces the time required for model development. We have applied this algorithm to 47 bones of varying geometries and successfully generated a FE mesh in all cases. This work represents significant advancement in automatically generating multiblock structures for a wide range of geometries.     

Growing multiblock structures: a semi-automated approach to block placement for multiblock hexahedral meshing

Finite element (FE) analysis is a cornerstone of orthopaedic biomechanics research. Three-dimensional medical imaging provides sufficient resolution for the subject-specific FE models to be generated from these data-sets. FE model development requires discretisation of a three-dimensional domain, which can be the most time-consuming component of a FE study. Hexahedral meshing tools based on the multiblock method currently rely on the manual placement of building blocks for mesh generation. We hypothesise that angular analysis of the geometric centreline for a three-dimensional surface could be used to automatically generate building block structures for the multiblock hexahedral mesh generation. Our algorithm uses a set of user-defined points and parameters to automatically generate a multiblock structure based on a surface's geometric centreline. This significantly reduces the time required for model development. We have applied this algorithm to 47 bones of varying geometries and successfully generated a FE mesh in all cases. This work represents significant advancement in automatically generating multiblock structures for a wide range of geometries.     

Multiblock copolymers of L-lactide and trimethylene carbonate

Sequential copolymerizations of trimethylene carbonate (TMC) and l-lactide (LLA) were performed with 2,2-dibutyl-2-stanna-1,3-oxepane as a bifunctional cyclic initiator. The block lengths were varied via the monomer/initiator and via the TMC/l-lactide ratio. The cyclic triblock copolymers were transformed in situ into multiblock copolymers by ring-opening polycondensation with sebacoyl chloride. The chemical compositions of the block copolymers were determined from (1)H NMR spectra. The formation of multiblock structures and the absence of transesterification were proven by (13)C NMR spectroscopy. Differential scanning calorimetry (DSC), wide-angle X-ray scattering (WAXS), and dynamic mechanical analysis (DMA) measurements confirmed the existence of a microphase-separated structure in the multiblock copolymers consisting of a crystalline phase of poly(LLA) blocks and an amorphous phase formed by the poly(TMC) blocks. Stress-strain measurements showed the elastomeric character of these biodegradable multiblock copolymers, particularly in copolymers having epsilon-caprolactone as comonomer in the poly(TMC) blocks.     

Multiblock PLS analysis of an industrial pharmaceutical process

The performance of an industrial pharmaceutical process (production of an active pharmaceutical ingredient by fermentation, API) was modeled by multiblock partial least squares (MBPLS). The most important process stages are inoculum production and API production fermentation. Thirty batches (runs) were produced according to an experimental planning. Rather than merging all these data into a single block of independent variables (as in ordinary PLS), four data blocks were used separately (manipulated and quality variables for each process stage). With the multiblock approach it was possible to calculate weights and scores for each independent block. It was found that the inoculum quality variables were highly correlated with API production for nominal fermentations. For the nonnominal fermentations, the manipulations of the fermentation stage explained the amount of API obtained (especially the pH and biomass concentration). Based on the above process analysis it was possible to select a smaller set of variables with which a new model was built. The amount of variance predicted of the final API concentration (cross-validation) for this model was 82.4%. The advantage of the multiblock model over the standard PLS model is that the contributions of the two main process stages to the API volumetric productivity were determined.     

Nonlinear modeling and adaptive monitoring with fuzzy and multivariate statistical methods in biological wastewater treatment plants

A new approach to nonlinear modeling and adaptive monitoring using fuzzy principal component regression (FPCR) is proposed and then applied to a real wastewater treatment plant (WWTP) data set. First, principal component analysis (PCA) is used to reduce the dimensionality of data and to remove collinearity. Second, the adaptive credibilistic fuzzy-c-means method is used to appropriately monitor diverse operating conditions based on the PCA score values. Then a new adaptive discrimination monitoring method is proposed to distinguish between a large process change and a simple fault. Third, a FPCR method is proposed, where the Takagi-Sugeno-Kang (TSK) fuzzy model is employed to model the relation between the PCA score values and the target output to avoid the over-fitting problem with original variables. Here, the rule bases, the centers and the widths of TSK fuzzy model are found by heuristic methods. The proposed FPCR method is applied to predict the output variable, the reduction of chemical oxygen demand in the full-scale WWTP. The result shows that it has the ability to model the nonlinear process and multiple operating conditions and is able to identify various operating regions and discriminate between a sustained fault and a simple fault (or abnormalities) occurring within the process data.     

Application of a kNN-based similarity method to biopharmaceutical manufacturing

Machine learning-based similarity analysis is commonly found in many artificial intelligence applications like the one utilized in e-commerce and digital marketing. In this study, a kNN-based (k-nearest neighbors) similarity method is proposed for rapid biopharmaceutical process diagnosis and process performance monitoring. Our proposed application measures the spatial distance between batches, identifies the most similar historical batches, and ranks them in order of similarity. The proposed method considers the similarity in both multivariate and univariate feature spaces and measures batch deviations to a benchmarking batch. The feasibility and effectiveness of the proposed method are tested on a drug manufacturing process at Biogen.     

Gaussian process functional regression modeling for batch data

A Gaussian process functional regression model is proposed for the analysis of batch data. Covariance structure and mean structure are considered simultaneously, with the covariance structure modeled by a Gaussian process regression model and the mean structure modeled by a functional regression model. The model allows the inclusion of covariates in both the covariance structure and the mean structure. It models the nonlinear relationship between a functional output variable and a set of functional and nonfunctional covariates. Several applications and simulation studies are reported and show that the method provides very good results for curve fitting and prediction.     

Chromatographic peak alignment using derivative dynamic time warping

Chromatogram overlays are frequently used to monitor inter‐batch performance of bioprocess purification steps. However, the objective analysis of chromatograms is difficult due to peak shifts caused by variable phase durations or unexpected process holds. Furthermore, synchronization of batch process data may also be required prior to performing multivariate analysis techniques. Dynamic time warping was originally developed as a method for spoken word recognition, but shows potential in the objective analysis of time variant signals, such as manufacturing data. In this work we will discuss the application of dynamic time warping with a derivative weighting function to align chromatograms to facilitate process monitoring and fault detection. In addition, we will demonstrate the utility of this method as a preprocessing step for multivariate model development. © 2013 American Institute of Chemical Engineers Biotechnol. Prog., 29: 394–402, 2013     

Real time phase detection based online monitoring of batch fermentation processes

Industrial fermentations conducted in a batch or semi-batch mode demonstrate significant batch-to-batch variability. Current batch process monitoring strategies involve manual interpretation of highly informative but low frequency offline measurements such as concentrations of products, biomass and substrates. Fermentors are also fitted with computer interfaced instrumentation, enabling high frequency online measurements of several variables and automated techniques which can utilize this data would be desirable. Evolution of a batch fermentation, which typically uses complex medium, can be conceptualized as a sequence of several distinct metabolic phases. Monitoring of batch processes can then be achieved by detecting the phase change events, also termed as singular points (SP). In this work, we propose a novel moving window based real-time monitoring strategy for SP detection based only on online measurements. The key hypothesis of the strategy is that the statistical properties of the online data undergo a significant change around an SP. The strategy is easily implementable and does not require past data or prior knowledge of the number or time of occurrence of SPs. The efficacy of the proposed approach has been demonstrated to be superior compared to that of reported techniques for industrially relevant model organisms. The proposed approach can be used to decide offline sampling timings in real time.     

Process performance models in the optimization of multiproduct protein production plants

In this work we propose a model that simultaneously optimizes the process variables and the structure of a multiproduct batch plant for the production of recombinant proteins. The complete model includes process performance models for the unit stages and a posynomial representation for the multiproduct batch plant. Although the constant time and size factor models are the most commonly used to model multiproduct batch processes, process performance models describe these time and size factors as functions of the process variables selected for optimization. These process performance models are expressed as algebraic equations obtained from the analytical integration of simplified mass balances and kinetic expressions that describe each unit operation. They are kept as simple as possible while retaining the influence of the process variables selected to optimize the plant. The resulting mixed-integer nonlinear program simultaneously calculates the plant structure (parallel units in or out of phase, and allocation of intermediate storage tanks), the batch plant decision variables (equipment sizes, batch sizes, and operating times of semicontinuous items), and the process decision variables (e.g., final concentration at selected stages, volumetric ratio of phases in the liquid-liquid extraction). A noteworthy feature of the proposed approach is that the mathematical model for the plant is the same as that used in the constant factor model. The process performance models are handled as extra constraints. A plant consisting of eight stages operating in the single product campaign mode (one fermentation, two microfiltrations, two ultrafiltrations, one homogenization, one liquid-liquid extraction, and one chromatography) for producing four different recombinant proteins by the genetically engineered yeast Saccharomyces cerevisiae was modeled and optimized. Using this example, it is shown that the presence of additional degrees of freedom introduced by the process performance models, with respect to a fixed size and time factor model, represents an important development in improving plant design.     

Detection of phase shifts in batch fermentation via statistical analysis of the online measurements: a case study with rifamycin B fermentation

Industrial production of antibiotics, biopharmaceuticals and enzymes is typically carried out via a batch or fed-batch fermentation process. These processes go through various phases based on sequential substrate uptake, growth and product formation, which require monitoring due to the potential batch-to-batch variability. The phase shifts can be identified directly by measuring the concentrations of substrates and products or by morphological examinations under microscope. However, such measurements are cumbersome to obtain. We present a method to identify phase transitions in batch fermentation using readily available online measurements. Our approach is based on Dynamic Principal Component Analysis (DPCA), a multivariate statistical approach that can model the dynamics of non-stationary processes. Phase-transitions in fermentation produce distinct patterns in the DPCA scores, which can be identified as singular points. We illustrate the application of the method to detect transitions such as the onset of exponential growth phase, substrate exhaustion and substrate switching for rifamycin B fermentation batches. Further, we analyze the loading vectors of DPCA model to illustrate the mechanism by which the statistical model accounts for process dynamics. The approach can be readily applied to other industrially important processes and may have implications in online monitoring of fermentation batches in a production facility.     

Raman spectroscopy and chemometrics for on‐line control of glucose fermentation by Saccharomyces cerevisiae

This work presents the use of Raman spectroscopy and chemometrics for on‐line control of the fermentation process of glucose by Saccharomyces cerevisiae. In a first approach, an on‐line determination of glucose, ethanol, glycerol, and cells was accomplished using multivariate calibration based on partial least squares (PLS). The PLS models presented values of root mean square error of prediction (RMSEP) of 0.53, 0.25, and 0.02% for glucose, ethanol and glycerol, respectively, and RMSEP of 1.02 g L^?1 for cells. In a second approach, multivariate control charts based on multiway principal component analysis (MPCA) were developed for detection of fermentation fault‐batch. Two multivariate control charts were developed, based on the squared prediction error (Q) and Hotelling's T². The use of the Q control chart in on‐line monitoring was efficient for detection of the faults caused by temperature, type of substrate and contamination, but the T² control chart was not able to monitor these faults. On‐line monitoring by Raman spectroscopy in conjunction with chemometric procedures allows control of the fermentative process with advantages in relation to reference methods, which require pretreatment, manipulation of samples and are time consuming. Also, the use of multivariate control charts made possible the detection of faults in a simple way, based only on the spectra of the system. © 2012 American Institute of Chemical Engineers Biotechnol. Prog., 2012     

Hexahedral meshing of subject-specific anatomic structures using mapped building blocks

To extend the use of computational techniques like finite element analysis to clinical settings, it would be beneficial to have the ability to generate a unique model for every subject quickly and efficiently. This work is an extension of two previously developed mapped meshing tools that utilised force and displacement control to map a template mesh to a subject-specific surface. The objective of this study was to map a template block structure, common to multiblock meshing techniques, to a subject-specific surface. The rationale is that the blocks are considerably less refined and may be readily edited after mapping, thereby yielding a mesh of high quality in less time than mapping the mesh itself. In this paper, the versatility and robustness of the method was verified by processing four data-sets. The method was found to be robust enough to cope with the variability of bony surface size, spatial position and geometry, producing building block structures (BBSs) that generated meshes comparable to those produced using BBSs that were created manually.     

Neural network modeling for on-line estimation of nutrient dynamics in a sequentially-operated batch reactor.

In monitoring and controlling wastewater treatment processes, on-line information of nutrient dynamics is very important. However, these variables are determined with a significant time delay. Although the final effluent quality can be analyzed after this delay, it is often too late to make proper adjustments. In this paper, a neural network approach, a software sensor, was proposed to overcome this problem. Software sensor refers to a modeling approach inferring hard-to-measure process variables from other on-line measurable process variables. A bench-scale sequentially-operated batch reactor (SBR) used for advanced wastewater treatment (BOD plus nutrient removal) was employed to develop the neural network model. In order to improve the network performance, the structure of neural network was arranged in such a way of reflecting the change of operational conditions within a cycle. Real-time estimation of PO3-(4), NO-3, and NH+4 concentrations was successfully carried out with the on-line information of the SBR system only.