Neural network predictive control by MIMS monitored 2,3-butanediol fermentation by Klebsiella oxytoca

With the aid of a membrane introduction mass spectrometer (MIMS), the major product 2,3-butanediol (2,3-BDL) as well as the other metabolites from the fermentation carried by Klebsiella oxytoca can be measured on-line simultaneously. A backpropagation neural network (BPN) being recognized with superior mapping ability was applied to this control study. This neural network adaptive control differs from those conventional controls for fermentation systems in which the measurements of cell mass and glucose are not included in the network model. It is only the measured product concentrations from the MIMS that are involved. Oxygen composition was chosen to be the control variable for this fermentation system. Oxygen composition was directly correlated to the measured product concentrations in the controller model. A two-dimensional (number of input nodes by number of data sets) moving window for on-line, dynamic learning of this fermentation system was applied. The input nodes of the network were also properly selected. Number of the training data sets for obtaining better control results was also determined empirically. Two control structures for this 2,3-BDL fermentation are discussed and compared in this work. The effect from adding time delay element to the network controller was also investigated.     

A neural network study on the dynamic identification of a fermentation system

The objective of this paper is to propose neural networks for the study of dynamic identification and prediction of a fermentation system which produces mainly 2,3-butanediol (2,3-BDL). The metabolic products of the fermentation, acetic acid, acetoin, ethanol, and 2,3-BDL were measured on-line via a mass spectrometer modified by the insertion of a dimethylvinylsilicone membrane probe. The measured data at different sampling times were included as the input and output nodes, at different learning batches, of the network. A fermentation system is usually nonlinear and dynamic in nature. Measured fermentation data obtained from the complex metabolic pathways are often difficult to be entirely included in a static process model, therefore, a dynamic model was suggested instead. In this work, neural networks were provided by a dynamic learning and prediction process that moved along the time sequence batchwise. In other words, a scheme of two-dimensional moving window (number of input nodes by the number of training data) was proposed for reading in new data while forgetting part of the old data. Proper size of the network including proper number of input/output nodes were determined by trained with the real-time fermentation data. Different number of hidden nodes under the consideration of both learning performance and computation efficiency were tested. The data size for each learning batch was determined. The performance of the learning factors such as the learning coefficient η and the momentum term coefficient α were also discussed. The effect of different dynamic learning intervals, with different starting points and the same ending point, both on the learning and prediction performance were studied. On the other hand, the effect of different dynamic learning intervals, with the same starting point and different ending points, was also investigated. The size of data sampling interval was also discussed. The performance from four different types of transfer functions, x/(1+|x|), sgn(x)·x ²/(1+x ²), 2/(1+e ^{? x} )?1, and 1/(1+e ^{? x} ) was compared. A scaling factor b was added to the transfer function and the effect of this factor on the learning was also evaluated. The prediction results from the time-delayed neural networks were also studied.     

Biological production of 2,3-butanediol

2,3-Butanediol (2,3-BDL), which is very important for a variety of chemical feedstocks and liquid fuels, can be derived from the bioconversion of natural resources. One of its well known applications is the formation of methyl ethyl ketone, by dehydration, which can be used as a liquid fuel additive. This article briefly reviews the basic properties of 2,3-BDL and the metabolic pathway for the microbial formation of 2,3-BDL. Both the biological production of 2,3-BDL and the variety of strains being used are introduced. Genetically improved strains for BDL production which follow either the original mechanisms or new mechanisms are also described. Studies on fermentation conditions are briefly reviewed. On-line analysis, modeling, and control of BDL fermentation are discussed. In addition, downstream recovery of 2,3-BDL and the integrated process (being important issues of BDL production) are also introduced.     

Comparative evaluation of various control schemes for fed-batch fermentation

The crucial problem associated with control of fed-batch fermentation process is its time-varying characteristics. A successful controller should be able to deal with this feature in addition to the inherent nonlinear characteristics of the process. In this work, various schemes for controlling the glucose feed rate of fed-batch baker's yeast fermentation were evaluated. The controllers evaluated are fixed-gain proportional-integral (PI), scheduled-gain PI, adaptive neural network and hybrid neural network PI. The difference between the specific carbon dioxide evolution rate and oxygen uptake rate (Q_c-Q_o) was used as the controller variable. The evaluation was carried out by observing the performance of the controllers in dealing with setpoint tracking and disturbance rejection. The results confirm the unsatisfactory performance of the conventional controller where significant oscillation and offsets exist. Among the controllers considered, the hybrid neural network PI controller shows good performance.     

Experimental studies of network parameters and operational variables on recurrent backpropagation neural network adaptive control of penicillin acylase fermentation by Arthrobacter viscosus

This work is to investigate the on-line control of the fermentation by Arthrobacter viscosus. This species of bacteria can secrete penicillin acylase which is a key enzyme in pharmaceutical industry. The growth of more cells during the fermentation will obtain more enzyme. Both the enzyme activity and the cell growth are rather sensitive to the change of pH. Once the pH during a fermentation is not properly controlled, the decay of cells' activity will irreversibly occur. Two peristaltic pumps for supplying acidic and basic solutions, respectively, were connected for the regulation of pH. With superior ability in identification and prediction of dynamic time series, recurrent backpropagation network (RBPN), instead of conventional controllers, was used as the adaptive controller model for the fermentation with dynamic characteristics. Based on a 1-3-1 BPN, a corresponding 4-4-1 RBPN was determined. The deviation of the pH measured at current time from the set point of 7.0, denoted as ( pH(t), was chosen as the input node of the network controller. The output node of this network controller was the predicted flow rate of the peristaltic pump for next control time interval. Such a model was operated by two phases. During the first phase, the network was set as the process model and trained by a fixed set of on-line acquired data. During the second phase, the network was stopped learning and switched to become a predictor, the predicted control action was hence obtained. The optimum sampling time was determined experimentally. To enhance the effective computation of this network, the number of training data was limited. A moving-window type of supplying training data to the network was applied for the on-line learning. The window size was also determined for each learning. With properly chosen network parameters as well as operation conditions, pH of the fermentation was thus well controlled by the RBPN controller.     

An artificial neural network for biomass estimation from automatic pH control signal

This study developed an artificial neural network (ANN) to estimate the growth of microorganisms during a fermentation process. The ANN relies solely on the cumulative consumption of alkali and the buffer capacity, which were measured on-line from the on/off control signal and pH values through automatic pH control. The two input variables were monitored on-line from a series of different batch cultivations and used to train the ANN to estimate biomass. The ANN was refined by optimizing the network structure and by adopting various algorithms for its training. The software estimator successfully generated growth profiles that showed good agreement with the measured biomass of separate batch cultures carried out between at 25 and 35_C.     

A prototype neural network supervised control system for Bacillus thuringiensis fermentations

This article discusses the development of a prototype neural network-based supervisory control system for Bacillus thuringiensis fermentations. The input pattern to the neural network included the type of inoculum, operation temperature, pH value, accumulated process time, optical density in fermentation medium, and change in optical density. The output from the neural network was the predicted optical density for the next sampling time. The control system has been implemented in both a computer simulation and a laboratory fermentation experiment with promising results. (c) 1994 John Wiley & Sons, Inc.     

An on-line adaptive glucose feeding system incorporating patterns recognition for glucose concentration control in glutamate fermentations

In glutamate fermentation, intermittent feeding is the most widely used glucose feed strategy. This feeding strategy causes severe fluctuations of glucose concentration and osmotic pressure in fermentation broth, which deteriorates the viability of the cell and reduces glutamate production in turn. In order to maintain glucose concentration at stable and constant levels, an on-line prediction and feedback control system based an empiric mass balance model was developed. However, the control system did not work properly and sometimes glucose concentration could even decline to 0 level (glucose exhaustion), as the model parameter varies in different runs. As a result, a novel model-based adaptive feedback control system incoporating with an artificial neural network (ANN) based pattern reconition unit for on-line diagnosizing the fault of glucose exhaustion was proposed and applied for glutamate fermentation. This adaptive control system could accurately detect glucose exhaustion when it occurs, and then immediately updates the control parameter based on some pre-defined rule. With the proposed control system, glucose was automatically fed, and its concentration could be maintained at desired levels constantly. As a result, glutamate concentration was 17 ~ 30% higher than that of the traditional fermentations using the intermittent glucose feed strategy.     

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.     

Simultaneous control of apparent extract and volatile compounds concentrations in low-malt beer fermentation

Volatile compounds cause undesirable flavor when their concentrations exceed threshold values in beer fermentation. The objective of this study is to develop a system for controlling apparent extract concentration, which indicates the fermentation degree and which should be decreased below a targeted value at a fixed time under a constraint of tolerable amounts of volatile compounds. In beer fermentation, even though the production of volatile compounds is suppressed by maintaining a low fermentation temperature, a low temperature causes a delay in the control of apparent extract concentration. Volatile compound concentration was estimated on-line, and the simulation of apparent extract consumption and volatile compound production was performed. To formulate various beer tastes and conserve energy for attemperation, optimal temperature profiles were determined using a genetic algorithm (GA). The developed feedback control of the brewing temperature profile was successfully applied, and apparent extract and volatile compound concentrations at a fixed time reached their target concentrations. Additionally, the control technique developed in this study enables us to brew a wide variety of beers with different tastes.     

On-line monitoring and control of acetone-butanol fermentation by membrane-sensor mass spectrometry

A mass spectrometry (MS) membrane sensor was developed and applied to on-line product measurement in acetone-butanol fermentation. The sensor facilitated the monitoring of acetone, butanol, ethanol, H₂ and CO₂, and single-compound calibration curves for both acetone and butanol showed a linear relationship between the product concentration and the MS response. However, when an actual fermentation was monitored, the product concentration calculated from the MS response was smaller than the concentration determined by gas chromatography, and the relationship between the response and the product concentration was nonlinear. It was found that large amounts of gases (H₂, CO₂) entering the MS analyzation chamber were causing a ‘space charge effect’, which resulted in an MS response ceiling. The problem could be resolved by reducing the surface area of the sensor membrane. Under some fermentation conditions, a by-product, n-butyl butyrate, was produced, and this interfered with the measurement of butanol due to a peak overlapping effect. However, it was found that this could be compensated for by using an empirical equation. Application of the MS membrane sensor in a fed batch culture of acetone-butanol fermentation resulted in successful control of the butanol concentration.     

Retrospective optimization of time-dependent fermentation control strategies using time-independent historical data

We have previously shown the usefulness of historical data for fermentation process optimization. The methodology developed includes identification of important process inputs, training of an artificial neural network (ANN) process model, and ultimately use of the ANN model with a genetic algorithm to find the optimal values of each critical process input. However, this approach ignores the time-dependent nature of the system, and therefore, does not fully utilize the available information within a database. In this work, we propose a method for incorporating time-dependent optimization into our previously developed three-step optimization routine. This is achieved by an additional step that uses a fermentation model (consisting of coupled ordinary differential equations (ODE)) to interpret important time-course features of the collected data through adjustments in model parameters. Important process variables not explicitly included in the model were then identified for each model parameter using automatic relevance determination (ARD) with Gaussian process (GP) models. The developed GP models were then combined with the fermentation model to form a hybrid neural network model that predicted the time-course activity of the cell and protein concentrations of novel fermentation conditions. A hybrid-genetic algorithm was then used in conjunction with the hybrid model to suggest optimal time-dependent control strategies. The presented method was implemented upon an E. coli fermentation database generated in our laboratory. Optimization of two different criteria (final protein yield and a simplified economic criteria) was attempted. While the overall protein yield was not increased using this methodology, we were successful in increasing a simplified economic criterion by 15% compared to what had been previously observed. These process conditions included using 35% less arabinose (the inducer) and 33% less typtone in the media and reducing the time required to reach the maximum protein concentration by 10% while producing approximately the same level of protein as the previous optimum.     

Rheological properties of chromosomal and plasmid DNA during alkaline lysis reaction

The experimental apparatus for the simultaneous L-lactic acid fermentation by Rhizopus oryzae immobilized in calcium alginate beads and product separation process was set up in which a three-phase fluidized-bed bioreactor was used as a fermentor and an external electrodialyzer as a separator, and a pump was applied to recycle the fermentation broth between the bioreactor and the separator. The L-lactic acid produced in the fermentor was separated in the separator, product inhibition was alleviated without any addition of alkali or alkali salts and the product purification process could be simplified. The specific productivity and the yield in electrodialysis fermentation (ED-F) process operated in continuous feeding mode were almost the same as that in CaCO₃-buffered fermentation process. A mathematical model of L-lactic acid production in ED-F process was also suggested, in which the model equations for the bioreactor and the electrodialyzer were combined to describe the simultaneous fermentation and product separation. The model predictions were in good agreement with the experimental data.     

Application of neural networks for controlling and predicting quality parameters in beer fermentation

The biochemical pathways involved in the production of ethyl caproate, a secondary product of the beer fermentation process, are not well established. Hence, there are no phenomenological models available to control and predict the production of this particular compound as with other related products. In this work, neural networks have been used to fit experimental results with constant and variable pH, giving a good fit of laboratory and industrial scale data. The results at constant pH were also used to predict results at variable pH. Finally, the application of neural networks obtained from laboratory experiments gave excellent predictions of results in industrial breweries and so could be used in the control of industrial operations. The input pattern to the neural network included the accumulated fermentation time, cell dry weight, consumption of sugars and aminoacids and, in some cases, the pH. The output from the neural network was an estimation of quantity of the ethyl caproate ester.     

On-line monitoring and control of substrate concentrations in biological processes by flow injection analysis systems

Concentrations of substrates, glucose, and ammionia in biological processes have been on-line monitored by using glucose-flow injection (FIA) and ammonia-FIA systems. Based on the on-line monitored data the concentrations of substrates have been controlled by an on-off controller, a PID controller, and a neural network (NN) based controller. A simulation program has been developed to test the control quality of each controller and to estimate the control parameters. The on-off controller often produced high oscillations at the set point due to its low robustness. The control quality of a PID controller could have been improved by a high analysis frequency and by a short residence time of sample in a FIA system. A NN-based controller with 3 layers has been developed, and a 3(input)-2(hidden)-1(output) network structure has been found to be optimal for the NN-based controller. The performance of the three controllers has been tested in a simulated process as well as in a cultivation process ofSaccharomyces cerevisiae, and the performance has also been compared to simulation results. The NN-based controller with the 3-2-1 network structure was robust and stable against some disturbances, such as a sudden injection of distilled water into a biological process.     

An adaptive control algorithm for fed-batch culture

The implementation of adaptive control for a fed-batch culture in order to maximize the output of product based on a self-adjusting model is discussed in the present work. Optimization methods were applied to the generalized mathematical model of a fed-batch fermentation process to determine control algorithms that could be used for on-line process control. The efficiency of the proposed adaptive algorithms was investigated by simulating a model system. The model of amylotytic enzyme fermentation that was proposed by the authors was taken from a real process. Dynamic modelling has shown that the main problem of realization is connected with the on-line identification of the adaptive model's parameters. To avoid this problem, we have introduced special limitations on the parameters' time variations that increased the convergence of the identification algorithm. The results of the investigation have shown the efficiency of the proposed adaptive algorithms, and the results of this work should be investigated for real process control.     

Monitoring of sequencing batch reactor for nitrogen and phosphorus removal using neural networks

The information of nutrient dynamics is essential for the precise control of effluent quality discharged from biological wastewater treatment processes. However, these variables can usually be 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 for the real-time estimation of nutrient concentrations and overcoming the problem of delayed measurements. In order to improve the neural network performance, a split network structure applied separately for anaerobic and aerobic conditions was employed with dynamic modeling methods such as auto-regressive with exogenous inputs. The proposed methodology was applied to a bench-scale sequencing batch reactor (SBR) for biological nutrient removal. The extrapolation problem of neural networks was possible to be partially overcome with the aid of multiway principal component analysis because of its ability of detecting of abnormal situations which could generate extrapolation. Real-time estimation of PO₄³⁻, NO₃⁻ and NH₄⁺ concentrations based on neural network was successfully carried out with the simple on-line information of the SBR system only.     

Monitoring and control of methanol concentration during polysaccharide fermentation using an on-line methanol sensor

Combining principles of membrane separation and semiconductor gas sensor technology, we constructed a methanol sensor to follow methanol concentrations on-line. A length of silicone tubing allowed for mass transfer of methanol from the fermentation medium to a carrier gas which then flowed over a semiconductor gas sensor for detection. The sterilizable sensor demonstrated excellent ability in following methanol concentrations during the batch production of a polysaccharide by the organism Methylomonas mucosa, even as the fermentation broth became increasingly viscous. During fed-batch control by feeding methanol to the fermentation to maintain setpoint methanol levels, a drift in the sensor signal was noted and quantified. A drift factor was determined which, after it was incorporated into the calibration calculations, improved methanol concentration control greatly. Methanol concentration was held constant over a range of set point concentrations during fedbatch fermentations.     

Medium optimization and proteome analysis of (R,R)-2,3-butanediol production by Paenibacillus polymyxa ATCC 12321

Paenibacillus polymyxa can produce the (R,R)-stereoisomer of 2,3-butanediol (2,3-BDL) which is industrially very useful. Two important factors affecting (R,R)-BDL production by P. polymyxa ATCC 12321, medium composition, and addition of acetic acid to the culture were investigated in this study with accompanying comparative proteomic analysis. For this purpose, a simple control strategy of O₂ supply was applied on the basis of an optimized basal medium: after a short period of batch cultivation with relatively high O₂ supply, the culture is switched into strong O₂ limitation, thereby promoting BDL formation. Three parallel fed-batch cultures starting from the same batch culture in an early stationary phase were then comparatively studied: the first one was running as control with the only change of O₂ supply; the second was, in addition, supplemented with 0.5 g/L yeast extract; and the third one was further added with 6 g/L acetate. Proteomic analyses of the three fed-batch cultures identified more than 86 proteins involved primarily in the central carbon metabolism, amino acid biosynthesis, energy metabolism, and stress responses. The examination of expression patterns of selected proteins, especially combined with fermentation data, gave valuable insights into the metabolic regulation of P. polymyxa under the different given conditions. Based on the proteomic analysis and further medium optimization studies, methionine was identified as one major growth-limiting factor in the basal medium and explains well the effect of yeast extract. Acetic acid was found to trigger the so far less studied acetone biosynthesis pathway in this organism. The latter is suggested in turn to enhance the switch from acidogenesis to solventogenesis. Thus, these findings extended our knowledge about BDL formation in P. polymyxa and provided useful information for further strain and process optimization.     

Enhanced production of L-(+)-lactic acid in chemostat by Lactobacillus casei DSM 20011 using ion-exchange resins and cross-flow filtration in a fully automated pilot plant controlled via NIR

Due to the lack of suitable in-process sensors, on-line monitoring of fermentation processes is restricted almost exclusively to the measurement of physical parameters only indirectly related to key process variables, i.e., substrate, product, and biomass concentration. This obstacle can be overcome by near infrared (NIR) spectroscopy, which allows not only real-time process monitoring, but also automated process control, provided that NIR-generated information is fed to a suitable computerized bioreactor control system. Once the relevant calibrations have been obtained, substrate, biomass and product concentration can be evaluated on-line and used by the bioreactor control system to manage the fermentation. In this work, an NIR-based control system allowed the full automation of a small-scale pilot plant for lactic acid production and provided an excellent tool for process optimization. The growth-inhibiting effect of lactic acid present in the culture broth is enhanced when the growth-limiting substrate, glucose, is also present at relatively high concentrations. Both combined factors can result in a severe reduction of the performance of the lactate production process. A dedicated software enabling on-line NIR data acquisition and reduction, and automated process management through feed addition, culture removal and/or product recovery by microfiltration was developed in order to allow the implementation of continuous fermentation processes with recycling of culture medium and cell recycling. Both operation modes were tested at different dilution rates and the respective cultivation parameters observed were compared with those obtained in a conventional continuous fermentation. Steady states were obtained in both modes with high performance on lactate production. The highest lactate volumetric productivity, 138 g L(-1) h(-1), was obtained in continuous fermentation with cell recycling.