Penicillin Production Process Research Articles

Neural networks-based hybrid beneficial variable selection and modeling for soft sensing

Variable selection plays an important role in soft sensor development. Either including redundant variables or missing important variables can degrade the modeling performance, affecting practical industrial applications. This paper proposes a novel neural networks-based hybrid beneficial variable selection (HBVS) and modeling method for effective soft sensing. First, irrelevant variables are removed through evaluating mutual information (MI) between all candidate variables and the quality variable. Second, proxy variables are introduced and a hidden gain-based evaluation method is employed to temporarily sort variables according to their significance, which facilitates to make use of process knowledge. Then, false discovery rate is employed to identify the model consistency, through which beneficial variables are determined. The proposed soft sensor development method is tested on a penicillin simulation process, and two actual industrial processes, including an oil refining process and an actual penicillin production process. Comparisons to state-of-the-art existing methods verify the effectiveness and superiority of the proposed method.

Automatic segmentation of batch processes into multi-local state-space models for fault detection

Several local models are used to approximate the nonlinear and multiphase characteristics of the batch process. Existing methods using phase partition and modeling can yield a model that does not describe the optimal local behavior. Batch data with uneven length also increases the difficulty of modeling local behavior. This study proposes Auto-Segmentation Subspace IDentification (AS-SID) for modeling a nonlinear batch process using a multi-stage operation over an operating region. It automatically and simultaneously reinforces data that pertains to the corresponding local models and weakens data that does not pertain to other local models. The entire range of the operation is partitioned and multiple state-space models are constructed using the partitioned data. To monitor the running batch using only the online collected data, a residual generator AS-SID is used for online detection. The merits of the proposed AS-SID are demonstrated using a numerical example and a fed-batch penicillin production process.

Data-Driven Fault Detection in Industrial Batch Processes Based on a Stochastic Hybrid Process Model

This paper presents a novel fault detection approach for industrial batch processes. The batch processes under consideration are characterized by the interaction between discrete system modes and non-stationary continuous dynamics. Therefore, a stochastic hybrid process model (SHPM) is introduced, where process variables are modeled as time-variant Gaussian distributions, which depend on hidden system modes. Transitions between the system modes are assumed to be either autonomous or to be triggered by observable events such as on/off signals. The model parameters are determined from training data using expectation-maximization techniques. A new fault detection algorithm is proposed, which assesses the likelihoods of sensor signals on the basis of the stochastic hybrid process model. Evaluation of the proposed fault detection system has been conducted for a penicillin production process, with the results showing a significant improvement over the existing baseline methods.Note to Practitioners—Automatic fault detection makes it possible to limit the effects of faults by taking countermeasures at an early stage. In this work, a data-driven fault detection method for industrial batch processes is proposed, in which the underlying process model is learned from training data. The proposed fault detection system can be used for various industrial batch processes without the need for complex and error-prone manual configuration. In contrast to many other data-driven approaches such as neural networks, only a few process cycles are required to create a robust process model. It should be noted that in data-driven fault detection methods, the training data should cover a large part of the process states that occur during error-free process cycles. The developed method is therefore particularly suitable for cyclical processes, which, however, can have alternative process paths and variability between the process cycles.

Stochastic parcel tracking in an Euler-Lagrange compartment model for fast simulation of fermentation processes.

The compartment model (CM) is a well‐known approach for computationally affordable, spatially resolved hydrodynamic modeling of unit operations. Recent implementations use flow profiles based on Computational Fluid Dynamics (CFD) simulations, and several authors included microbial kinetics to simulate gradients in bioreactors. However, these studies relied on black‐box kinetics that do not account for intracellular changes and cell population dynamics in response to heterogeneous environments. In this paper, we report the implementation of a Lagrangian reaction model, where the microbial phase is tracked as a set of biomass‐parcels, each linked with an intracellular composition vector and a structured reaction model describing their intracellular response to extracellular variations. A stochastic parcel tracking approach is adopted, in contrast to the resolved trajectories used in CFD implementations. A penicillin production process is used as a case study. We show good performance of the model compared with full CFD simulations, both regarding the extracellular gradients and intracellular pool response, using the mixing time as a matching criterion and taking into account that the mixing time is sensitive to the number of compartments. The sensitivity of the model output towards some of the inputs is explored. The coarsest representative CM requires a few minutes to solve 80 h of flow time, compared with approximately 2 weeks for a full Euler–Lagrange CFD simulation of the same case. This alleviates one of the major bottlenecks for the application of such CFD simulations towards the analysis and optimization of industrial fermentation processes.

Virtual Sensor Development for Multioutput Nonlinear Processes Based on Bilinear Neighborhood Preserving Regression Model With Localized Construction

The traditional data-driven virtual sensors mainly define the outer shape of the data, but they cannot provide any insight into the micro perspective of manifold proximity indicating the local relationships among the data samples. In this article, a regression model with localized construction named bilinear neighborhood preserving regression (BNPR) model is proposed by synchronously exploring the local manifold geometry among both the process variables and primary variables and developing the regression relationship for the estimates of the primary variables. The model is constructed under multi-output to discover the inherent relationships among the primary variables instead of building independent models for each primary variable. The effectiveness of the proposed algorithm is demonstrated by case studies carried out on a simulated penicillin production process and a real-life semiconductor process.

Virtual sensors of nonlinear industrial processes based on neighborhood preserving regression model

The micro perspective of manifold proximity would indicate local relationships with their unique spatial geometric distribution characteristics among the data samples, which are usually neglected by traditional data-driven virtual sensors. This would not guarantee a good prediction performance. In this paper, a regression model with localized construction named neighborhood preserving regression (NPR) model is proposed. It extends the unsupervised neighborhood preserving embedding (NPE) to the supervised form. The projection vector is learned from input process variables and the output quality variable, synchronously exploring the manifold structure of the input process variables for the dimension reduction and developing the regression relationship between the projected input process variables and the output quality variable. The model is developed as a novelly designed optimization whose analytical solution would be compactly and directly calculated without any iterative procedures. The effectiveness of the proposed algorithm is demonstrated by case studies carried out on a simulated penicillin production process.

An algebra approach for nonlinear multivariable fed-batch bioprocess control

In this paper, a linear algebra-based controller design is proposed. This technique allows tracking, with minimum error, predefined optimal profiles in nonlinear and multivariable systems. To achieve this, control actions are obtained by solving a linear equation system. The controller parameters are selected with a Monte Carlo algorithm. The methodology is applied in a fed-batch penicillin production process, where the control action is the feed flow rate and the tracked profiles are the concentration of biomass, product and subtract inside the reactor. Different tests are shown to prove the good performance of the controller adding: parametric uncertainty and perturbations in the control action and in the initial conditions

An algebra approach for nonlinear multivariable fed-batch bioprocess control

In this paper, a linear algebra-based controller design is proposed. This technique allows tracking, with minimum error, predefined optimal profiles in nonlinear and multivariable systems. To achieve this, control actions are obtained by solving a linear equation system. The controller parameters are selected with a Monte Carlo algorithm. The methodology is applied in a fed-batch penicillin production process, where the control action is the feed flow rate and the tracked profiles are the concentration of biomass, product and subtract inside the reactor. Different tests are shown to prove the good performance of the controller adding: parametric uncertainty and perturbations in the control action and in the initial conditions

A New Approach for Nonlinear Multivariable Fed-Batch Bioprocess Trajectory Tracking Control

This paper proposes a new control law based on linear algebra. This technique allows nonlinear path tracking in multivariable and complex systems. This new methodology consists in finding the control action to make the system follow predefined concentration profiles solving a system of linear equations. The controller parameters are selected with a Monte Carlo algorithm so as to minimize a previously defined cost index. The control scheme is applied to a fed-batch penicillin production process. Different tests are shown to prove the controller effectiveness, such as adding parametric uncertainty, perturbations in the control action and in the initial conditions. Moreover, a comparison with other controllers from the literature is made, showing the better performance of the present approach.

Morphological analysis of the filamentous fungus Penicillium chrysogenum using flow cytometry\u2014the fast alternative to microscopic image analysis

An important parameter in filamentous bioreactor cultivations is the morphology of the fungi, due to its interlink to productivity and its dependency on process conditions. Filamentous fungi show a large variety of morphological forms in submerged cultures. These range from dispersed hyphae, to interwoven mycelial aggregates, to denser hyphal aggregates, the so-called pellets. Depending on the objective function of the bioprocess, different characteristics of the morphology are favorable and need to be quantified accurately. The most common method to quantitatively characterize morphology is image analysis based on microscopy. This method is work intensive and time consuming. Therefore, we developed a faster, at-line applicable, alternative method based on flow cytometry. Within this contribution, this novel method is compared to microscopy for a penicillin production process. Both methods yielded in comparable distinction of morphological sub-populations and described their morphology in more detail. In addition to the appropriate quantification of size parameters and the description of the hyphal region around pellets, the flow cytometry method even revealed a novel compactness parameter for fungal pellets which is not accessible via light microscopy. Hence, the here presented flow cytometry method for morphological analysis is a fast and reliable alternative to common tools with some new insights in the pellet morphology, enabling at-line use in production environments.

Controlling the specific growth rate via biomass trend regulation in filamentous fungi bioprocesses

Increasing pressure on product quality and quantity pushes solutions of process control to be a central issue in pharmaceutical bioprocesses. Especially online biomass estimation, and further control of the specific growth rate are of central importance because they describe the catalyst of the reaction. For penicillin producing bioprocesses with filamentous microorganisms, this was underlined by recent findings describing the influence of the specific growth rate on the specific production rate. Hence, the specific growth rate needs to be controlled on a certain level to achieve high productivity.In this study, we developed a control strategy for the specific growth rate based on online estimation of viable biomass via dielectric spectroscopy. The method was verified using an at-line staining method for viability measurement. The online viable biomass estimation is applicable in the growth and decline phase, coping with physiological and morphological changes of filamentous fungi. Furthermore, the control strategy adapts to changing biomass yields, which is a big issue in the bioprocess for penicillin production applied in this study.Two application runs were conducted, yielding in proper online viable biomass estimation and control of the specific growth rate at a constant level of 0.012h−1. We achieved biomass predictions with an average error of 1.5g/l over the whole fed-batch process. In the decline phase, the control of specific growth rates was not possible due to physiological constrains. However, in the growth phase, a total specific growth rate of 0.013h−1 was achieved, which met the pre-defined acceptance criterion for this method.The method is thus ready for viable biomass estimation in the growth and in the decline phase of the penicillin production process. Furthermore, the method is applicable to control the specific growth rate during the growth phase.

Application and Evaluation of Random Forest Classifier Technique for Fault Detection in Bioreactor Operation

Bioreactors and associated bioprocesses are quite complex and nonlinear in nature. A small change in initial condition can greatly alter the output product quality. It is pretty difficult at times to model the system mathematically. In this work, the fault detection problem is studied in the context of bioreactors, mainly, a reactor from the penicillin production process. It is very important to identify the faults in a live process to avoid product quality deterioration. We have focused on the process history-based methods to identify the faults in a bioreactor. We want to introduce random forest (RF), a powerful machine learning algorithm, to identify several types of faults in a bioreactor. The algorithm is simple, easy to use, shows very good generalization ability without compromising much on the classification accuracies, and also has an ability to give variable importance as a part of the algorithm output. We compared its performance with two popular methods, namely support vector machines (SVM) and artificial neural networks (ANN), and found that the overall performance is superior in terms of classification accuracies and generalization ability.

Robust Algorithms for Simultaneous Model Identification and Optimization in the Presence of Model-Plant Mismatch

In the presence of model-plant mismatch, the set of parameter estimates obtained using standard model identification procedures cannot accurately predict the gradients of the optimization problem. Hence, a “two-step” approach involving an identification followed by an optimization step, performed repeatedly, often fails to converge to the true process optimum. In our recent work,1 we proposed a new identification procedure that progressively corrects the model for structural error such that the updated set of parameter estimates simultaneously satisfies both identification and optimization objectives with a guaranteed convergence to the true optimum. In this paper, we expanded our previous methodology in two directions: first, a new model correction based on a higher order approximation of model is proposed, and second, the effect of model uncertainties is considered explicitly by formulating a robust optimization problem at each iteration. The resulting improvements are then illustrated using a simulated study of fed-batch penicillin production process.

Evaluation of aerobic co-composting of penicillin fermentation fungi residue with pig manure on penicillin degradation, microbial population dynamics and composting maturity

Improper treatment of penicillin fermentation fungi residue (PFFR), one of the by-products of penicillin production process, may result in environmental pollution due to the high concentration of penicillin. Aerobic co-composting of PFFR with pig manure was determined to degrade penicillin in PFFR. Results showed that co-composting of PFFR with pig manure can significantly reduce the concentration of penicillin in PFFR, make the PFFR-compost safer as organic fertilizer for soil application. More than 99% of penicillin in PFFR were removed after 7-day composting. PFFR did not affect the composting process and even promote the activity of the microorganisms in the compost. Quantitative PCR (qPCR) indicated that the bacteria and actinomycetes number in the AC samples were 40–80% higher than that in the pig-manure compost (CK) samples in the same composting phases. This research indicated that the aerobic co-composting was a feasible PFFR treatment method.

Progressive multi-block modelling for enhanced fault isolation in batch processes

A multi-block progressive modelling approach is proposed for enhanced fault isolation in batch processes. The unfolding of batch data typically leads to matrices with a large number of columns and this complicates contribution analysis. In order to rapidly focus fault isolation in batch processes, it would be desirable to employ multi-block modelling techniques. Multi-block model such as consensus principal component analysis (CPCA) can produce multiple monitoring charts for sub-blocks and block loadings and block scores can be obtained which can represent unique behaviour of each sub-block. CPCA model uses super score which is the same as score from normal principal component analysis (PCA) model and it does not produce enhanced monitoring performance. Multi-block PCA (MBPCA) model using block score for model calculation can represent sub-blocks’ character but block scores are obtained from super loading so it may not be the best way to describe sub-blocks. A new MBPCA model is proposed for better expression of each sub-block. Through progressive modelling and contribution analysis, variables related to or affected by the fault, as well as the associated time information, are gradually identified. This enables a fault propagation path being established. The proposed method is applied to a benchmark simulated penicillin production process, PenSim.

The inducers 1,3-diaminopropane and spermidine cause the reprogramming of metabolism in Penicillium chrysogenum, leading to multiple vesicles and penicillin overproduction

In this article we studied the differential protein abundance of Penicillium chrysogenum in response to either 1,3-diaminopropane (1,3-DAP) or spermidine, which behave as inducers of the penicillin production process. Proteins were resolved in 2-DE gels and identified by tandem MS spectrometry. Both inducers produced largely identical changes in the proteome, suggesting that they may be interconverted and act by the same mechanism. The addition of either 1,3-DAP or spermidine led to the overrepresentation of the last enzyme of the penicillin pathway, isopenicillin N acyltransferase (IAT). A modified form of the IAT protein was newly detected in the polyamine-supplemented cultures. Both inducers produced a rearrangement of the proteome resulting in an overrepresentation of enzymes involved in the biosynthesis of valine and other precursors (e.g. coenzyme A) of penicillin. Interestingly, two enzymes of the homogentisate pathway involved in the degradation of phenylacetic acid (a well-known precursor of benzylpenicillin) were reduced following the addition of either of these two inducers, allowing an increase of the phenylacetic acid availability. Both inducers produced also an increase in the intracellular content of vesicles that derived to vacuoles in late stages and promoted sporulation of P. chrysogenum in solid medium. The analysis of global protein changes produced in response to polyamines 1,3-DAP and spermidine provides a valuable information for the understanding of the molecular mechanisms underlying the production of penicillin. This represents useful information to improve the production of this antibiotic and many other bioactive secondary metabolites not only in P. chrysogenum, but in other filamentous fungi as well.

Discriminating between critical and non-critical disturbances in (bio-)chemical batch processes using multi-model fault detection and end-quality prediction

This paper proposes a novel multimodel methodology for discriminating between critical and noncritical process disturbances in (bio)chemical batch processes, in addition to providing online prediction of batch-end quality. A multivariate multiway partial least squares (MPLS) or multiway principal component analysis (MPCA) model monitoring all available measurements is coupled with an MPLS or MPCA model monitoring only those measurements influencing the final product quality. Hence, process disturbances are labeled critical or noncritical, depending on whether they impact final quality and require immediate attention. This avoids unnecessary control actions or even early batch terminations for noncritical disturbances. The presented approach is illustrated on a simulated industrial-scale penicillin production process. On the basis of extensive simulation results, it is concluded that the proposed methodology discriminates between critical (according to a hypothesis test with 0.05 significance level) and noncritical disturbances. In addition, accurate online estimations of the batch-end product quality are provided.

Measurement noise influence on statistical properties of batch-end quality predictions

In this paper, an extensive Monte Carlo simulation is performed to investigate the influence of output measurement noise on Multiway Partial Least Squares (MPLS) batch-end quality predictions. MPLS models are well suited for monitoring (bio)chemical batch processes, but the lack of insight in noise influence leaves companies reluctant to accept the technique. Simplified relations between prediction variance and measurement noise exist for spectroscopy calibration problems, but are based on assumptions that do not necessarily hold for batch process modelling. The non-linear properties of the PLS predictor and the lack of knowledge about its statistical distribution make the derivation of an analytical relation extremely difficult.Based on an extensive case study of a penicillin production process, MPLS predictions of final batch quality are shown to outperform offline quality measurements. Even at very high noise levels, the models capture the important information in the measurements and discard most of the noise. Prediction bias and variance are studied and found to behave inversely with respect to the model order. This inverse behaviour has important consequences for model order selection, which becomes a trade-off between bias and variance. In this light, several crossvalidation-based techniques for selection of the optimal number of principal components are compared. An adjusted Wold's R criterion proves to be slightly favorable to the minimum MSE and general Wold's R criterion.

Recursive weighted kernel regression for semi-supervised soft-sensing modeling of fed-batch processes

Soft-sensor techniques have been increasingly applied in chemical industry to establish an online monitor of unmeasured product indices. However, one intrinsic obstacle of soft-sensing modeling is insufficiency of labeled data while unlabeled data is abundant. In this work, a semi-supervised recursive weighted kernel regression (RWKR) method is proposed to model the soft-sensor by leveraging both labeled and unlabeled data. A novel weight strategy is presented to improve the prediction and its recursive algorithm is formulated, which adopts the incremental and decremental learning mechanism to update the soft-sensor model online and control the model complexity. Simulative soft-sensor for penicillin production process indicates that RWKR is superior to both relevance vector machine (RVM) and harmonic functions to model such fed-batch processes. Additionally, it sheds light on more competitive semi-supervised soft-sensing modeling approaches.

Fault Localization in Batch Processes through Progressive Principal Component Analysis Modeling

A technique for fault localization in batch processes using progressive principal component analysis (PCA) modeling is proposed in this paper. A PCA model is developed from normal process operation data and is used for online process monitoring. Once a fault is detected by the PCA model, process variables that are related to the fault are identified using contribution analysis. The time information on when abnormalities occurred in these variables is identified using a time series plot of the squared prediction errors (SPE) on these variables. These variables are then removed and another PCA model is developed using the remaining variables. If the faulty batch cannot be detected by the new PCA model, then the remaining variables are not related to the fault. If the faulty batch can still be detected by the new PCA model, then further variables associated with the fault are identified from SPE contribution analysis. The procedure is repeated until the faulty batch can no longer be detected using the remaining variables. Using the time information on when abnormalities presented in the variables associated with the fault, fault propagation paths can be established and the origin of the fault could be traced. The proposed method is tested on a benchmark simulated fed-batch penicillin production process, PenSim. The results demonstrate that the proposed method is particularly effective in isolating faults that have occurred on measured variables. For more complex faults that have occurred on unmeasured variables, the method can identify variables affected by the fault, and process knowledge is required to determine the fault.