Crystallization process optimization using artificial neural networks

This paper presents a new procedure for optimization of continuous mixed suspensionmixed product removal (MSMPR) crystallizing systems. Owing to the difficulties of theoretical modelling, simulation of the MSMPR crystallization process is based on the use of artificial neural networks (ANN). The optimization criterion is a compound objective function corresponding to an intended mean crystal size dimension and a minimal dispersion. The presence of multiple local minima has called for investigation by several optimization techniques. Ultimately, Luus' and Jaakola's random adaptive method proved to be most effective. The results obtained lend support to the general procedure proposed.     

Sintering of WC-Ni nanocomposite powder: Experimental and artificial neural networks modeling study

The sintering behavior of WC-Ni nanocomposite powder was evaluated through experimental and statistical approaches to study the contribution of involving parameters of chemical composition (Ni wt. %) and sintering temperature on sinterability of system by assessing the resulted densification and microhardness. The experimental process was designed based on factorial experimental design for independent effective parameters of Ni percentage (12, 18 and 23 wt %), and sintering temperature (8 different values within 1350–1485 °C). The resulted products of experimental testing after compaction and sintering were analyzed by FESEM and EDX to image the microstructure and evaluate the chemical composition and elemental distribution. The density and microhardness were measured as well. An artificial neural network (ANN) was applied to describe the corresponding individual and mutual impacts on sintering. The ANN model was developed by feed-forward back propagation network including topology 2:5:2 and trainlm algorithm to model and predict density and microhardness. A great agreement was observed between the predicted values by the ANN model and the experimental data for density and microhardness (regression coefficients (R²) of 0.9983 and 0.9924 for target functions of relative density and microhardness, respectively). Results showed that the relative importance of operating parameters on target functions (relative density and microhardness) was found to be 62% and 38% for sintering temperature and Ni percentage, respectively. Also, ANN model exhibited relatively high predictive ability and accuracy in describing nonlinear behavior of the sintering of WC-Ni nanocomposite powder. The experimental results confirmed that the appropriate sintering temperature was influenced by Ni content. The optimum parameters were found to be 12 wt % Ni sintered at 1460 °C with the highest microhardness and relative density.     

Media‐dependent color appearance modeling based on artificial neural networks

In this study, we tried to consider various color appearance factors and device characterization together by visual experiment to simplify the across‐media color appearance reproduction. Two media, CRT display (soft‐copy) and NCS color atlas (hard‐copy), were used in our study. A total of 506 sample pairs of RGB and HVC, which are the attributes of NCS color chips, were obtained according to psychophysical experiments by matching soft copy and hard copy by a panel of nine observers. In addition, a set of error back‐propagation neural networks was used to realize experimental data generalization. In order to get a more perfect generalizing effect, the whole samples were divided into four parts according to different hues and the conversion between HVC and R_HVCG_HVCB_HVC color space was implemented. The current results show that the displays on the CRT and the color chips can match well. In this way, a CRT‐dependent reproduction modeling based on neural networks was formed, which has strong practicability and can be applied in many aspects. © 2006 Wiley Periodicals, Inc. Col Res Appl, 31, 218–228, 2006; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/col.20209     

Control of the Fenton process for textile wastewater treatment using artificial neural networks

BACKGROUND: The Fenton process is a popular advanced oxidation process (AOP) for treating textile wastewater. However, high consumption of chemical reagents and high production of sludge are typical problems when using this process and in addition, textile wastewater has wide‐ranging characteristics. Therefore, dynamically regulating the Fenton process is critical to reducing operation costs and enhancing process performance. The artificial neural network (ANN) model has been adopted extensively to optimize wastewater treatment. This study presents a novel Fenton process control strategy using ANN models and oxygen reduction potential (ORP) monitoring to treat two synthetic textile wastewaters containing two common dyes. RESULTS: Experimental results indicated that the ANN models can predict precisely the colour and chemical oxygen demand (COD) removal efficiencies for synthetic textile wastewaters with correlation coefficients (R²) of 0.91–0.99. The proposed control strategy based on these ANN models effectively controls the Fenton process for various effluent colour targets. For treating the RB49 synthetic wastewater to meet the effluent colour targets of 550 and 1500 ADMI units, the required Fe⁺² doses were 13.0–84.3 and 5.5–34.6 mg L^?1 (Fe⁺²/H₂O₂ = 3.0), resulting in average effluent colour values of 520 and 1494 units. On the other hand, an effluent colour target of 550 ADMI units was achieved for RBB synthetic wastewater. The required Fe⁺² doses were 14.6–128.0 mg L^?1; the average effluent colour values were 520 units. CONCLUSION: The Fenton process for textile wastewater treatment was effectively controlled using a control strategy applying the ANN models and ORP monitoring, giving the benefit of chemical cost savings. Copyright © 2009 Society of Chemical Industry     

Reduced dynamic modeling approach for rectification columns based on compartmentalization and artificial neural networks

The availability of reduced-dimensional, accurate dynamic models is crucial for the optimal operation of chemical processes in fast-changing environments. Herein, we present a reduced modeling approach for rectification columns. The model combines compartmentalization to reduce the number of differential equations with artificial neural networks to express the nonlinear input–output relations within compartments. We apply the model to the optimal control of an air separation unit. We reduce the size of the differential equation system by 90% while limiting the additional error in product purities to below 1 ppm compared to a full-order stage-by-stage model. We demonstrate that the proposed model enables savings in computational times for optimal control problems by ~95% compared to a full order and ~99% to a standard compartment model. The presented model enables a trade-off between accuracy and computational efficiency, which is superior to what has recently been reported for similar applications using collocation-based reduction approaches.     

人工神经网络在化学工程中的应用

介绍了人工神经网络在化学工程中的应用，讨论了应用中的优点和局限性，预示了其应用的发展前景。     

Modeling the kinetics of a photochemical water treatment process by means of artificial neural networks

We have investigated the kinetics of the degradation of 2,4-dimethyl aniline (2,4-xylidine), chosen as a model pollutant, by the photochemically enhanced Fenton reaction. This process, which may be efficiently applied to the treatment of industrial waste waters, involves a series of complex reactions leading eventually to the mineralization of the organic pollutant. A model based on artificial neural networks has been developed for fitting the experimental data obtained in a laboratory batch reactor. The model can describe the evolution of the pollutant concentration during irradiation time under various conditions. It has been used for simulating the behavior of the reaction system in sensitivity studies aimed at optimizing the amounts of reactants employed in the process — an iron(II) salt and hydrogen peroxide. The results show that the process is much more sensitive to the iron(II) salt concentration than to the hydrogen peroxide concentration, a favorable condition in terms of economic feasibility.     

The use of artificial neural networks in electrostatic force microscopy

ABSTRACT: The use of electrostatic force microscopy (EFM) to characterize and manipulate surfaces at the nanoscale usually faces the problem of dealing with systems where several parameters are not known. Artificial neural networks (ANNs) have demonstrated to be a very useful tool to tackle this type of problems. Here, we show that the use of ANNs allows us to quantitatively estimate magnitudes such as the dielectric constant of thin films. To improve thin film dielectric constant estimations in EFM, we first increase the accuracy of numerical simulations by replacing the standard minimization technique by a method based on ANN learning algorithms. Second, we use the improved numerical results to build a complete training set for a new ANN. The results obtained by the ANN suggest that accurate values for the thin film dielectric constant can only be estimated if the thin film thickness and sample dielectric constant are known.PACS: 07.79.Lh; 07.05.Mh; 61.46.Fg.     

神经网络在动力配煤中的应用

在动力配煤中考虑了混煤这个各煤参数之间的相关关系,建立动力配煤的数学模型,并提出神经网络算法,改进了一般的动力配煤模型。     

Discoloration process modeling by neural network

The photo-oxidation of acid orange 52 dye was performed in the presence of H₂O₂, utilizing UV light, aiming the discoloration process modeling and the process variable influence characterization. The discoloration process was modeled by the use of feedforward neural network. Each sample was characterized by five independent variables (dye concentration, pH, hydrogen peroxide volume, temperature and time of operation) and a dependent variable (absorbance). The neural model has also provided, through Garson Partition coefficients and the Pertubation method, the independent variable influence order determination. The results indicated that the time of operation was the predominant variable and reaction mean temperature was the lesser influent variable. The neural model obtained presented coefficients of correlation on the order 0.98, for sets of trainability, validation and testing, indicating the power of prediction of the model and its character of generalization.     

Applications of artificial neural networks in chemical engineering

A growing literature within the field of chemical engineering describing the use of artificial neural networks (ANN) has evolved for a diverse range of engineering applications such as fault detection, signal processing, process modeling, and control. Because ANN are nets of basis functions, they can provide good empirical models of complex nonlinear processes useful for a wide variety of purposes. This article describes certain types of neural networks that have proved to be effective in practical applications, mentions the advantages and disadvantages of using them, and presents four detailed chemical engineering applications. In the competitive field of modeling, ANN have secured a niche that now, after one decade, seems secure.     

Nonlinear system identification and model reduction using artificial neural networks

We present a technique for nonlinear system identification and model reduction using artificial neural networks (ANNs). The ANN is used to model plant input–output data, with the states of the model being represented by the outputs of an intermediate hidden layer of the ANN. Model reduction is achieved by applying a singular value decomposition (SVD)-based technique to the weight matrices of the ANN. The sequence of state values is used to convert the model to a form that is useful for state and parameter estimation. Examples of chemical systems (batch and continuous reactors and distillation columns) are presented to demonstrate the performance of the ANN-based system identification and model reduction technique.     

人工神经网络在环境领域中的研究进展

综述了人工神经网络在环境领域的国内外最新研究现状,总结了人工神经网络在环境质量预测、环境质量评价、环境监测、环境化学和其他环境领域的一些应用进展,分析了目前人工神经网络在环境领域中存在的问题和不同人工神经网络模型的优缺点,最后对人工神经网络在环境领域的应用前景进行了展望。     

Evolutionary polymorphic neural network in chemical process modeling

Evolutionary polymorphic neural network (EPNN) is a novel approach to modeling dynamic process systems. This approach has its basis in artificial neural networks and evolutionary computing. As demonstrated in the studied dynamic CSTR system, EPNN produces less error than a traditional recurrent neural network with a less number of neurons. Furthermore, EPNN performs networked symbolic regressions for input–output data, while it performs multiple step ahead prediction through adaptable feedback structures formed during evolution. In addition, the extracted symbolic formulae from EPNN can be used for further theoretical analysis and process optimization.     

Context-based recognition of process states using neural networks

Patterns whose interpretation varies across contexts are common in many engineering domains. The resulting one-to-many mapping between patterns and their classes cannot be adequately handled by traditional pattern recognition approaches. This important class of pattern recognition problems although common has not received any attention in chemical engineering domain. In this paper, we show that identification of the state of chemical or biological processes is context dependent. Two types of features are important for context-based pattern recognition—primary features, which determine the class of a pattern, and contextual features, which cannot themselves predict the class, but can improve the effectiveness of the primary features. Process measurements can be used as primary features for identifying the current process state, and the previous process state provides the context in which the primary features have to be interpreted. We also propose a dynamic neural network architecture for context-based operating state identification. Three variations of the architecture, each using a different approach to identify change of context, are described. These are illustrated using two case studies for operating state identification—the startup of a simulated fluidized catalytic cracking unit and operation of a lab-scale fermentation process. A comparison with traditional neural networks reveals that the performance of the proposed context-based pattern recognition architecture is superior in all cases.     

Reinforcing the phenomenological consistency in artificial neural network modeling of multiphase reactors

Artificial neural networks (ANN) aided with dimensional analysis have been successfully applied in multiphase reactors modeling when considerable amount of experimental data (or database) is available. An important problem that stemmed from this approach was the ambiguity to select the fittest combination of dimensionless numbers to be used as ANN inputs to predict a variable of interest. A genetic algorithm (GA) based methodology was proposed to optimize the combination of inputs by taking into account the phenomenological consistency (PC) of the resulting ANN models along with their fitting capabilities. PC is a measure of the capability of an ANN model to simulate outputs with specified gradient conditions with respect to the process variables. These conditions are imposed based on a priori knowledge of the system's behavior. PC used to be evaluated in the vicinity of a particular point in the database space. The novelty of the approach was the extension of the PC test around all the points available in the training data set. This technique may be regarded as a robust method to prevent data overfitting when the function to be learned by ANN is characterized by a monotonic behavior with respect to some of the process variables. The new approach was illustrated using as a case study the correlation of two-phase pressure drop in randomly packed beds with countercurrent flow.     

Application of neural networks to meltblown process control

Process modeling is essential for the control of optimization and an on-line prediction is very useful for process monitoring and quality control. Up to now, no satisfactory methods have been found to model an industrial meltblown process since it is of highly dimensional and nonlinear complexity. In this article, back-propagation neural networks (BPNNs) were investigated for modeling the meltblown process and on-line predicting the product specifications such as fiber diameter and web thickness. The feasibility of this application was successfully demonstrated by agreement of the prediction results from the BPNN to the actual measurements of a practical case. The network inputs included extruder temperature, die temperature, melt flow rate, air temperature at die, air pressure at die, and die-to-collector distance (DCD). The output of the fiber diameter was obtained by neural computing. The network training was based on 160 sets of the training samples and the trained network was tested with 70 sets of test samples which were different from the training data. This research is preliminary and of industrial significance and especially valuable for the optimal control of advanced meltblown processes. © 1996 John Wiley & Sons, Inc.     

Predicting compressive strength of different geopolymers by artificial neural networks

In the present study, six different models based on artificial neural networks have been developed to predict the compressive strength of different types of geopolymers. The differences between the models were in the number of neurons in hidden layers and in the method of finalizing the models. Seven independent input parameters that cover the curing time, Ca(OH)₂ content, the amount of superplasticizer, NaOH concentration, mold type, geopolymer type and H₂O/Na₂O molar ratio were considered. For each set of these input variables, the compressive strength of geopolymers was obtained. A total number of 399 input-target pairs were collected from the literature, randomly divided into 279, 60 and 60 data and were trained, validated and tested, respectively. The best performance model was obtained through a network with two hidden layers and absolute fraction of variance of 0.9916, the absolute percentage error of 2.2102 and the root mean square error of 1.4867 in training phase. Additionally, the entire trained, validated and tested network showed a strong potential for predicting the compressive strength of geopolymers with a reasonable performance in the considered range.