Industrial Fed-batch Fermentation Research Articles

Pseudo label estimation based on label distribution optimization for industrial semi-supervised soft sensor

In process industry, the lack of sufficient labeled data often leads to poor performance of traditional supervised soft sensors. Thus, a pseudo label estimation method based on label distribution optimization (PLELDO) is proposed. PLELDO first converts the pseudo label estimation into an explicit label distribution optimization problem, and then obtains high-confidence pseudo labels. Further, a semi-supervised ensemble soft sensor modeling framework namely EnPLELDO is developed. EnPLELDO first obtains sufficient good pseudo labeled data by repeating small-scale pseudo label estimation. Then, a set of diverse base models are constructed from the extended training sets. Finally, these enhanced base models are combined through stacking strategy. The application results from an industrial fed-batch fermentation process show that, compared with several state-of-the-art methods, PLELDO can obtain better pseudo label estimations and EnPLELDO can deliver more accurate predictions of key quality variables by leveraging the scarce labeled data and large-scale unlabeled data.

Development of dynamic compartment models for industrial aerobic fed-batch fermentation processes

Inhomogeneities in key cultivation variables (e.g., substrate and oxygen concentrations) have been shown to affect key process metrics in large-scale bioreactors. Being able to understand these gradients is hence of key interest from both an industrial and academic perspective. One of the main shortcomings of current modelling approaches is that volume change is not considered. Volume increase is a key feature of fed-batch fermentation processes. Existing models are restricted to simulating snapshots (hundreds of seconds) of industrial processes, which can last several weeks. This study presents a novel methodology that overcomes this limitation by constructing dynamic compartment models for the simulation of fed-batch fermentation processes. This strategy is applied to an industrial aerobic fed-batch fermentation process (40–90 m3) with Saccharomyces cerevisiae. First, it has been validated numerically that the compartmentalization strategy used captures the mixing performance and fluid dynamics. This was done by comparing the mixing times and the local concentration profiles of snapshot fermentation process simulations calculated with both CFD and compartment models. Subsequently, simulations of the entire process have been performed using the dynamic compartment model with kinetics. The simulation allows the spatio-temporal characterization of all process variables (e.g., glucose and DO concentrations), as well as the quantification of the metabolic regimes that the cells experience over time. This strategy enables the rapid characterization and assessment of the impact of gradients on process performance in industrial (aerobic) fed-batch fermentation processes and can be readily generalized to any type of bioreactor and microorganism.

Establishing a Fed-Batch Process for Protease Expression with Bacillus licheniformis in Polymer-Based Controlled-Release Microtiter Plates.

Introducing fed-batch mode in early stages of development projects is crucial for establishing comparable conditions to industrial fed-batch fermentation processes. Therefore, cost efficient and easy to use small-scale fed-batch systems that can be integrated into existing laboratory equipment and workflows are required. Recently, a novel polymer-based controlled-release fed-batch microtiter plate is described. In this work, the polymer-based controlled-release fed-batch microtiter plate is used to investigate fed-batch cultivations of a protease producing Bacillus licheniformis culture. Therefore, the oxygen transfer rate (OTR) is online-monitored within each well of the polymer-based controlled-release fed-batch microtiter plate using a µRAMOS device. Cultivations in five individual polymer-based controlled-release fed-batch microtiter plates of two production lots show good reproducibility with a mean coefficient of variation of 9.2%. Decreasing initial biomass concentrations prolongs batch phase while simultaneously postponing the fed-batch phase. The initial liquid filling volume affects the volumetric release rate, which is directly translated in different OTR levels of the fed-batch phase. An increasing initial osmotic pressure within the mineral medium decreases both glucose release and protease yield. With the volumetric glucose release rate as scale-up criterion, microtiter plate- and shake flask-based fed-batch cultivations are highly comparable. On basis of the small-scale fed-batch cultivations, a mechanistic model is established and validated. Model-based simulations coincide well with the experimentally acquired data.

Ensemble just-in-time learning framework through evolutionary multi-objective optimization for soft sensor development of nonlinear industrial processes

Just-in-time learning (JIT) has recently gained growing popularity for soft sensor development of nonlinear processes. However, traditional JIT methods aim to pursue a globally optimal learning configuration while ignoring the diversity of JIT learning. Since strong JIT learners are desirable yet difficult to get while weak JIT learners are easy to obtain in real practice, a novel soft sensor modeling framework, referred to as evolutional multi-objective optimization (EMO) based ensemble JIT learning (EMO-EJIT), is proposed. Within this framework, the diversity of JIT learning is created by perturbing the weighted Euclidean distance (WED) based similarity measure. Then diverse and accurate JIT learners equipped with diverse WED similarity measures are generated using an EMO approach. Finally, a stacking strategy is used to achieve ensemble prediction. By employing the proposed modeling framework, EMO based ensemble locally weighted partial least squares (EMO-ELWPLS) and EMO based ensemble JIT extreme learning machine (EMO-EJITELM) soft sensors are developed. The application results from a simulated fed-batch penicillin fermentation process and an industrial fed-batch chlortetracycline fermentation process have demonstrated the effectiveness and superiority of the proposed method.

Research and application of biological potency soft sensor modeling method in the industrial fed-batch chlortetracycline fermentation process

The potency of fermentation broth is one of the key parameters reflect the yield and quality of fermentation products of chlortetracycline (CTC). But so far, there is no instrument available on-line detection of CTC potency. All the tests are done by manual off-line testing, it takes several hours to sample and analyze the results from a fermentation tank (the production process usually has dozens of small, large fermenters). The use of analytical results to control the amount of operation will lead to severe lag. This paper combines self-organizing feature map (SOM) neural network with accurate classification of data and least squares support vector machine (LSSVM) algorithm with strong described the nonlinear characteristics. The SOM–LSSVM global modeling method of forecasting CTC fermentation potency is establish in this paper. According to the characteristics of nonlinear CTC fermentation process, just-in-time learning-recursive least squares support vector regression (JITL–RLSSVR) is used to perform local real-time modeling and 10-folding cross validation, and a hybrid soft sensor modeling method (JITL–RLSSVR + SOM–LSSVM) for online prediction of CTC fermentation potency is proposed in this paper. Field experiments show that this method can obtain more accurate potency prediction value, and it can meet the requirements of the production process.

Multi-model adaptive soft sensor modeling method using local learning and online support vector regression for nonlinear time-variant batch processes

Batch processes are often characterized by inherent nonlinearity, multiplicity of operating phases, and batch-to-batch variations, which poses great challenges for accurate and reliable online prediction of soft sensor. Especially, the soft sensor built with old data may encounter performance deterioration due to a failure of capturing the time-variant behaviors of batch processes, thus adaptive strategies are necessary. Unfortunately, conventional adaptive soft sensors cannot efficiently account for the within-batch as well as between-batch time-variant changes in batch process characteristics, which results in poor prediction accuracy. Therefore, a novel multi-model adaptive soft sensor modeling method is proposed based on the local learning framework and online support vector regression (OSVR) for nonlinear time-variant batch processes. First, a batch process is identified with a set of local domains and then the localized OSVR models are built for all isolated domains. Further, the estimation for a query data is obtained by adaptively combining multiple local models that perform best on the similar samples to the query point. The proposed multi-model OSVR (MOSVR) method provides four types of adaptation strategies: (i) adaptive combination based on Bayesian ensemble learning; (ii) online offset compensation; (iii) incremental updating of local models; and (iv) database updating. The effectiveness of the MOSVR approach and its superiority over traditional adaptive soft sensors in dealing with the within-batch and between-batch shifting dynamics is demonstrated through a simulated fed-batch penicillin fermentation process as well as an industrial fed-batch chlortetracycline fermentation process.

Online local learning based adaptive soft sensor and its application to an industrial fed-batch chlortetracycline fermentation process

This work presents a new method for adaptive soft sensor development by further exploiting just-in-time modeling framework. In the presented method, referred to as online local learning based adaptive soft sensor (OLLASS), the samples used for local modeling are selected based on the mutual information (MI) weighted or neighbor sample based similarity measure. Then, two adaptive methods, namely self-validation and neighbor-validation, are developed to adaptively select the optimal local modeling size for scenarios without and with the neighbor output information, respectively. Further, a real-time performance improvement strategy is used to enhance the online modeling efficiency. Moreover, an online dual updating strategy is proposed to activate infrequent local model updating and model output offset updating in turn, which allows significantly reducing the online computational load by avoiding unnecessary local model reconstruction while at the same time maintaining high estimation accuracy by performing offset compensation. A maximal similarity replacement rule using MI weighted similarity measure is used for database updating. The superiority of the proposed OLLASS method over traditional soft sensors in terms of the estimation accuracy, adaptive capability and real-time performance is demonstrated through an industrial fed-batch chlortetracycline fermentation process.

An approximate probability graphical modelling method for complex industrial fermentation processes

Fed-batch fermentation process is an effective method for production. Due to the various feeding streams andoperational conditions in different fed-batches, usually it is difficult to formulate a kinetics-based ordinary differential equations model for industrial fed-batch fermentation process. On the other hand, there are plenty of historical data collected during the fermentation process. In this paper, we firstly applied the graphical modeling method to model the fed-batch fermentation process. In this proposed method, the missing data within records are imputed, and then, the correlations between variables are determined by the low order conditional independence method, after that, the parameters of these related variables are learned by the multivariate auto regressive method. The calculation of L-lysine fed-batch fermentation process demonstrates the effectiveness of the proposed approximate model method.

An Approach to Continuous Approximation of Pareto Front Using Geometric Support Vector Regression for Multi-objective Optimization of Fermentation Process

The approaches to discrete approximation of Pareto front using multi-objective evolutionary algorithms have the problems of heavy computation burden, long running time and missing Pareto optimal points. In order to overcome these problems, an approach to continuous approximation of Pareto front using geometric support vector regression is presented. The regression model of the small size approximate discrete Pareto front is constructed by geometric support vector regression modeling and is described as the approximate continuous Pareto front. In the process of geometric support vector regression modeling, considering the distribution characteristic of Pareto optimal points, the separable augmented training sample sets are constructed by shifting original training sample points along multiple coordinated axes. Besides, an interactive decision-making (DM) procedure, in which the continuous approximation of Pareto front and decision-making is performed interactively, is designed for improving the accuracy of the preferred Pareto optimal point. The correctness of the continuous approximation of Pareto front is demonstrated with a typical multi-objective optimization problem. In addition, combined with the interactive decision-making procedure, the continuous approximation of Pareto front is applied in the multi-objective optimization for an industrial fed-batch yeast fermentation process. The experimental results show that the generated approximate continuous Pareto front has good accuracy and completeness. Compared with the multi-objective evolutionary algorithm with large size population, a more accurate preferred Pareto optimal point can be obtained from the approximate continuous Pareto front with less computation and shorter running time. The operation strategy corresponding to the final preferred Pareto optimal point generated by the interactive DM procedure can improve the production indexes of the fermentation process effectively.

Hybrid intelligent control of substrate feeding for industrial fed-batch chlortetracycline fermentation process

The lack of accurate process models and reliable online sensors for substrate measurements poses significant challenges for controlling substrate feeding accurately, automatically and optimally in fed-batch fermentation industries. It is still a common practice to regulate the feeding rate based upon manual operations. To address this issue, a hybrid intelligent control method is proposed to enable automatic substrate feeding. The resulting control system consists of three modules: a presetting module for providing initial set-points; a predictive module for estimating substrate concentration online based on a new time interval-varying soft sensing algorithm; and a feedback compensator using expert rules. The effectiveness of the proposed approach is demonstrated through its successful applications to the industrial fed-batch chlortetracycline fermentation process.

Adaptive soft sensor modeling framework based on just-in-time learning and kernel partial least squares regression for nonlinear multiphase batch processes

Batch processes are characterized by inherent nonlinearity, multiple phases and time-varying behavior that pose great challenges for accurate state estimation. A multiphase just-in-time (MJIT) learning based kernel partial least squares (KPLS) method is proposed for multiphase batch processes. Gaussian mixture model is estimated to identify different operating phases where various JIT-KPLS frameworks are built. By applying Bayesian inference strategy, the query data is classified into a particular phase with the maximal posterior probability, and thus the corresponding JIT-KPLS framework is chosen for online prediction. To further improve the predictive accuracy of the MJIT-KPLS algorithm, a hybrid similarity measure and an adaptive selection strategy are proposed for selecting local modeling samples. Moreover, maximal similarity replacement rule is proposed to update database. A procedure of input variable selection based on partial mutual information is also presented. The effectiveness of the MJIT-KPLS algorithm is demonstrated through application to industrial fed-batch chlortetracycline fermentation process.

OPTIMIZATION OF MEDIUM COMPOSITION FOR cis,cis-MUCONIC ACID PRODUCTION BY A Pseudomonas sp. MUTANT USING STATISTICAL METHODS

cis,cis-Muconic acid (CCMA) is used as a platform chemical for the production of several high-value compounds. For this article, an optimization strategy has been used to optimize medium composition for CCMA production from fairly cheap benzoate by Pseudomonas sp. 1167. The effect of different concentrations of medium components on CCMA production was studied. CCMA yields obtained from Plackett–Burman design (PBD) showed wide variation (3.95–5.87 g/L), and the first-order model indicated that (NH4)2SO4 (P < 0.01) and K2HPO4 · 3H2O (P < 0.02) were the significant components for CCMA production. Then the optimization was performed by steepest ascent design (SAD) and central composite design (CCD), and a validation experiment was conducted to verify the predicted value. The optimal medium composition was: 12 g/L sodium benzoate, 2.5 g/L sodium succinate, 0.7932 g/L (NH4)2SO4, 1.5612 g/L K2HPO4 · 3H2O, 1.2 g/L MgSO4 · 7H2O, 0.4 g/L yeast extract, 0.08 g/L FeCl3 · 6H2O, and 0.08 g/L ethylenediamine tetraacetic acid (EDTA). Under these conditions, a maximum of 7.18 g/L CCMA was produced per 12 g/L benzoate with a highly efficient process within 11 hr and a molecular conversion yield of 61%. Altogether, our results provide valuable insights into nutritional supplementation of CCMA production by using statistical methods, which may benefit a cost-competitive industrial fed-batch fermentation process using a cheap substrate.

On-line evolutionary optimization of an industrial fed-batch yeast fermentation process

This paper presents two genetic algorithms based on optimization methods to maximize biomass concentration, and to minimize ethanol formation. The objective function is maximized according to the values of feed flow rate, using genetic search approaches. Five case studies were carried out for different initial conditions, which strongly influence the optimal profiles of feed flow rate for the fermentation process. The ethanol and glucose disturbance effects were examined to stress the effectiveness of proposed approaches. The proposed genetic approaches were implemented for an industrial scale baker’s yeast fermentor which produces Saccharomyces cerevisiae known as baker’s yeast. The results show that optimal feed flow rate was obtained in a satisfactory and successful way for fed-batch fermentation process.

Dynamic hybrid neural network model of an industrial fed-batch fermentation process to produce foreign protein

An industrial pharmaceutical company has provided industrial pilot scale fed-batch data from a biological process used to produce a foreign protein from fed-batch fermentation. This process had proven difficult to control due to the complex behavior of the bacteria after induction. Because of the difficulty of modeling the process fundamentally, neural networks are an attractive alternative. To capture dynamic systems a gray box model approach of parameter function neural networks was used. The parameter function neural network approach has been able to capture well this pilot scale fed-batch fermentation process. In order to obtain accurate training data, the data sets were fit and smoothed using smoothing cubic spline functions. Neural networks were found for the five critical parameter functions of growth rate, glucose consumption rate, oxygen consumption rate, acetate production rate, and protein production rate. Relatively simple networks were used in order to capture process behavior and not the significant noise in the industrial scale pilot data. Simulations using the neural network parameters predicted dynamic response data well.

Industrial fermentation end-product modelling with multilinear PLS

In this paper, a trilinear version of the partial least squares (PLS) algorithm was used to model the performance of an industrial fed-batch fermentation process. Trilinear data obtained from process operation were used to derive a model for the end-process active product ingredient (API) concentration prediction. Obtained mutilinear PLS models were compared with the correspondent bilinear models. A genetic algorithm was used to select appropriate calibration sets (to reduce the influence of nominal batches). A validation coefficient of determination ( Q Y 2) of 91.4% was obtained for the multilinear PLS model after batch selection (prediction intervals were estimated using bootstrapping). Examination of the multilinear PLS model weights led to the delimitation of a small time region (from 50 to 75 processing hours) almost exclusively responsible for the fermentation performance.

LOCAL LINEAR AND NON-LINEAR MULTI-WAY PARTIAL LEAST SQUARES BATCH MODELLING

An industrial fed-batch fermentation process forms the basis of a study that illustrates the development and use of local models for the construction of a performance monitoring scheme. Linear models are initially built for different operating regions within the batch using Dynamic Partial Least Squares (DPLS) to capture the underlying process dynamics. The local models are then pieced together and a global model developed based on the latent variable scores. A comparison with a global non-linear quadratic DPLS model is then made prior to using the latent variable scores from the local models to build a nominal process representation.

Moving Window MSPC and Its Application to Batch Processes

Multivariate statistical process control (MSPC) has been demonstrated to provide assistance in the monitoring offed-batch fermentation systems. The application of MSPC to batch systems is complicated by the non-linear profiles that the systems exhibit. These complications are typically overcome through what is termed unfolding the data. This approach however does pose two problems. The first is that the future profile of batch operation must be assumed in the analysis and the second is that the length of each batch must be the same. In this paper an alternative approach to unfolding the data is presented. This approach, termed moving window MSPC, determines the statistical model on a moving window of data that is continuously updated. The approach is applied to an industrial fed-batch fermentation system and its benefits over traditional data unfolding techniques are presented.

The Monitoring of an Industrial Fed-Batch Fermentation Process

The monitoring of the evolution of batch and fed-batch processes based on the multivariate statistical projection technique of principal component analysis is investigated. Prior to developing a process representation, a sensitivity analysis is carried out to identify the key variables that influence process behaviour. Graphical representations are then used to demonstrate the suitability of techniques for process monitoring where the objective is to identify, at any early stage, the onset of changes in process operation and deviations in product quality. The strategy is illustrated on an industrial biological fed-batch fermentation

Process monitoring of an industrial fed-batch fermentation.

Market demand places great emphasis in industry on product quality. Consequently, process monitoring and control have become important aspects of systems engineering. In this article we detail the results of a 2-year study focusing on the development of a condition monitoring system for a fed-batch fermentation system operated by Biochemie Gmbh in Austria. We also demonstrate the suitability and limitations of current state of the art technologies in this field and suggest novel modifications and configurations to improve their suitability for application to a fed-batch fermentation system.

Supervision of fed-batch fermentations

Process faults may be detected on-line using existing measurements based upon modelling that is entirely data driven. A multivariate statistical model is developed and used for fault diagnosis of an industrial fed-batch fermentation process. Data from several (25) batches are used to develop a model for cultivation behaviour. This model is validated against 13 data sets and demonstrated to explain a significant amount of variation in the data. The multivariate model may directly be used for process monitoring. With this method faults are detected in real time and the responsible measurements are directly identified. The fault detection and identification is enabled through inspection of a few simple plots. Thus, the presented methodology allows the process operator to actively monitor data from several cultivations simultaneously.