APPLICATIONS OF ARTIFICIAL NEURAL NETWORKS IN CHEMISTRY AND CHEMICAL ENGINEERING

Abstract

Problems in chemistry and chemical engineering are composed of complex systems. Various chemical processes in chemistry and chemical engineering can be described by different mathematical functions as, for example, linear, quadratic, exponential, hyperbolic et al. There are many of calculated and experimental descriptors/molecular properties to describe the chemical behavior of the substances. It is also possible that many variables can influence the desired response. Usually, chemometrics is widely used as a valuable tool to deal chemical data, and to solve complex problems. In this context, Artificial Neural Networks (ANN) is a chemometric tool that may provide accurate results for complex and non-linear problems that demand high computational costs. The main advantages of ANN techniques include learning and generalization ability of data, fault tolerance and inherent contextual information processing in addition to fast computation capacity. Due to the popularization, there is a substantial interest in ANN techniques, in special in their applications in various fields. The following types of applications are considered: data reduction using neural networks, overlapped signal resolution, experimental design and surface response, modeling, pattern recognition, and multivariate regression.

Headnote

ABSTRACT

Problems in chemistry and chemical engineering are composed of complex systems. Various chemical processes in chemistry and chemical engineering can be described by different mathematical functions as, for example, linear, quadratic, exponential, hyperbolic et al. There are many of calculated and experimental descriptors/molecular properties to describe the chemical behavior of the substances. It is also possible that many variables can influence the desired response. Usually, chemometrics is widely used as a valuable tool to deal chemical data, and to solve complex problems. In this context, Artificial Neural Networks (ANN) is a chemometric tool that may provide accurate results for complex and non-linear problems that demand high computational costs. The main advantages of ANN techniques include learning and generalization ability of data, fault tolerance and inherent contextual information processing in addition to fast computation capacity. Due to the popularization, there is a substantial interest in ANN techniques, in special in their applications in various fields. The following types of applications are considered: data reduction using neural networks, overlapped signal resolution, experimental design and surface response, modeling, pattern recognition, and multivariate regression.

Keywords: artificial neural networks, optimization, modeling, pattern recognition, multivariate regression, chemical engineering

INTRODUCTION

The late 1980s marked the emergence of the field of artificial neural networks (ANNs). This form of non-algorithmic computation is characterized by that system at some level, reminiscent of the structure of the human brain. While not be based on rules, neural computing constitutes an alternative to conventional algorithmic computation.

ANNs are parallel distributed systems consisting of simple processing units (artificial neurons) that calculate certain mathematical functions (typically nonlinear). Such units are arranged in one or more layers and interconnected by a large number of connections unidirectional in general. In most models these connections are associated with weights, which store the knowledge acquired by the model and serve to consider the input received by each neuron network [1].

The solution of problems through ANNs is very attractive since the way they are represented internally by the network and the natural parallelism inherent in the architecture of ANNs create the possibility of a better performance than conventional models. In ANNs, the usual procedure in troubleshooting initially goes through a learning step, in which a set of examples is presented to the network, which extracts the necessary features to represent the information provided. These characteristics are then used to generate responses to the problem [1].

The ability to learn by example and generalize information learned is undoubtedly the main attraction of troubleshooting through ANNs. The generalization that is associated with network capacity to learn through a small set of examples and then give consistent answers to data not known is a demonstration that the ability of ANNs goes far beyond simply mapping input and output relationships. The ANNs can extract information not given explicitly by the examples. Furthermore, the ANNs are capable of acting as universal mappers of multivariate functions with a computational cost that grows only linearly with the number of variables. Another important feature is the ability to self-organization and temporal processing that, allied to those mentioned above, turns ANNs an attractive computational tool for solving complex problems [2].

APPLICATIONS OF ARTIFICIAL NEURAL NETWORKS IN CHEMICAL ENGINEERING

The phenomenological modeling of chemical engineering (CE) systems leads to balances (of material, energy and momentum) and constitutive equations, generally demanding very arduous tasks for determining all the parameters and for implementing and numerically solving the equations obtained. Such problems are highly costly and time-consuming, and often untreatable considering the time and information available. It is even more severe when fast responses are needed from the model, such as in chemical industries for control purposes.

Problems in Chemical Engineering are usually characterized by nonlinearities and a high number of input and output variables. Also, the complexibility of the systems sometimes makes it difficult to develop proper phenomenological models. This way the use of limited empirical models with several adjusted parameters is very common. In these cases, the use of Artificial Neural Networks (ANNs) appears as an excellent alternative, especially when a sufficient and representative set of data is available as it happens in industrial applications. This scenario evidences the appeal of ANNs in CE as very powerful tools that make it possible to correlate the mathematical dependence of variables of interest.

In the neural network modeling, there is no need for a good understanding of the process internal laws and of the equations that describes them. It is especially useful for complex systems. Often, neural models provide better results than phenomenological ones, if a sufficient and representative set of data is available. Both approaches - mechanistic and neural - are nicely complementary, since neural network approach does not clarify the mechanism evolved in a process. It just answers the question "what", but not "why" or "how" [3].

An ANN can be defined as a computing system made up of some simple and highly interconnected processing elements, called neurons, which manage information by their dynamic state response to the external inputs. The advantages of ANNs include its skills of learning from data inputs and capturing patterns usually ignored by common statistical methods [4].

In general, ANNs modeling involves three main stages: (i) data preparation/selection, (ii) topological design, and (iii) selection of training and validation methods along with synaptic weights and biases. Data preparation/selection includes the selection of inputs that significantly affects the output target, such as outliers removal and definition of the operational regions of interest. Further, topological designs deal with the way the neurons are connected, resulting in different ANNs type, such as feed-forward (e.g., Multi-Layer Perceptron - MLP, Radial-Based Networks) and recurrent (e.g., Hopfield Network, Elman Network) architectures. Easy programming associated with general and straightforward application make MLP architecture the most used [3]. For training and validation of ANNs, several methods are available, such as classical backpropagation (BP) and its many variants, as well as numerous evolutionary algorithms.

The flexibility of ANNs allows its use alone as an empirical model or associated to the phenomenological equations in hybrid models. Ensembles of ANNs can be utilized combining several ANNs to capture different aspects of process behavior. ANNs can also be combined with other softcomputing tools, mainly fuzzy systems and evolutionary algorithms [3]. All these applications can be found in Chemical Engineering problems.

Historically, ANNs have been widely used in the Chemical Engineering. Hoskins and Himmelblau published one of the first papers on the subject of ANN with its application in chemical engineering [5]. In that article, the authors described the basis of ANNs and demonstrated how an artificial neural network can learn and discriminate successfully among faults in a CE process. So, that work will be considered here as the starting point of the application of ANNs in chemical engineering. Figure 1 presents the number of publications along years reported in the literature using ANNs inthe main areas of CE, according to SCOPUS database. One can see that the number of publications grown linearly from 1994 to 2006 and then remained approximately constant.

Regarding the number of results of this search in Scopus database, we define ????,?? as the fraction of appearance of field j in the results of a field i, as follows

(ProQuest: ... denotes formula omitted.)

From ???? ,?? it is possible to observe the correlation between groups. Table 1 presents the values of ????,?? for the considered fields. It can be observed that f is not a symmetric matrix and that control and safety presents higher correlation with other areas.

Although some applications can be found in systems of costly experiments, the most common applications of ANN in CE can be found when data obtainment is get-at-able, since ANN requires a high amount of data. One of the most profitable fields of ANN applications in CE is chemical industries, where sensors scan the entire plant in a few seconds. Thus, one has the perfect scenario for ANN application: a significant amount of data, a difficult phenomenological modeling and a demand for a fast response.

In fact, industrially the ANNs models provide a good result. For these reasons, such tools are widely employed on processes monitoring, control and optimization strategies.

Generally, the literature reports the applications of ANNs in CE according to different industrial fields: petrochemicals, oil and gas industry; fuel and energy; biotechnology and pharmaceutical industry; environmental, health and safety; food industry; polymer industry; cellular industry; mineral industry, and nanotechnology [3]. However, here we classify these works based on the most common reasons to apply ANNs in chemical engineering:

. Lengthy response or unavailability of physical sensors: in these cases, ANNs are generated to replace the physical sensor in real time. Examples: composition, cell products and rheology, organoleptic properties, food industries, cosmetics, pharmaceuticals and biopharmaceuticals.

. Multipurpose processes: networks are generated to complement the phenomenological modeling (operating in regions in which it diverges) for monitoring and operation of multipurpose plants. Sometimes adaptive groups are created, and the behavior of each operating region is provided by a network model.

. Advanced control, faults detection and diagnosis, and safety: ANNs are generated to replace phenomenological models of complex units, whose resolution can be timeconsuming and even impractical, as well as to model the controller itself. They are also used for mapping the behavior and most common patterns of the plant to recognize improper operation conditions and even safety problems.

The following sections describe some common applications based on the classification described previously.

Lengthy Response or Unavailability of Physical Sensors

Fast response is an important attribute of sensors in process operation. Usually, monitoring systems of industrial plants involve measuring key variables everyone second with storage at every one minute. ANN models can work as virtual sensors - also known as softsensors - predicting key properties that may not be measured in real time, from other variables easily measured in the process. Also, in several processes, some variables cannot be directed measured (also known as intangible parameters).

Pioneer, the oil and gas industry concentrates significant applications of ANNs as softsensors. One of the most common applications consists in the inference of the quality of products in distillation columns [6-8]. Distillation is considered the main unit operation in oil and gas treating, and aims to achieve desired purification degrees; however, direct composition measurements seldom can be made. On-line chromatography measurements typically take around 10-20 minutes, with the high cost of acquisition and intensive maintenance. The use of ANNs is then appealing. For example, to overcome the considerable time delay introduced by the corresponding gas chromatograph, Fortuna et al. developed neural based softsensors to infer, in real time, the stabilized gasoline concentration in the top flow and the butanes concentration in the bottom flow of a debutanizer column, based on available data [9].

The biotechnology industry presents an extensive literature showing how neural networks can be used for classification, estimation and prediction, in which the physical sensors are not available for several variables [10]. Despite the innumerous ANNs applications in bioprocesses few are for on-line implementations,due to the nonlinear and complex behavior of biological systems [11-16]. ANNs can play a significant role in facilitating both qualitative identification and quantitative characterization of biomolecule compositions [17-19]. Bioseparation of proteins in aqueous two-phase systems was also addressed using ANNs [20].

The pulp and paper industry has large applications of ANNs for predicting critical properties [21-23], such as the kappa number (indicating the residual lignin), viscosity, yield, among others [23-25]. Other paper properties extensively modeled by ANNs are tensile index and tear index [23-24]. ANN can also be used in hybrid models to describe the steps of production; for example, Costa et al. proposed a hybrid approach using ANN to model black liquor burning process in an industrial recovery boiler furnace [26].

Polymer industry also deals with the absence of on-line physical sensors of essential properties, and many applications of virtual sensors for polymerization processes based in ANNs can be found [27-28]. As an example of application, Wagner et al. developed a model of an industrial reactive extruder based on ANN, allowing prediction of extrudate viscosity, an excellent measure of product quality [29]. BICC Cables Ltd developed an ANN inferential estimator capable of predicting the tensile strength of polymer coating [30]. Since the reliable method for determining the quality of the polymer coating is through destructive testing 24 h after production, large quantities of sub-standard material may be produced before it is ever detected. In other applications, ANNs were combined with equations of state for predicting the PVT behavior of different types of the polymer melts [31-32].

In polymer composites, measurements and mechanistic description of tribological properties (such as friction and wear) impose a very complicated problem [27, 33]. Jones et al. (1997) simulated the tribological properties of different test rigs with various materials [33]. Several authors introduced ANNs to predict the fatigue life [34-37]. Velten et al. were among the first to explore ANNs for inference in wear of polymer composites, such as the wear volume of short-fiber/particle reinforced thermoplastics [38]. Zhang et al. developed ANNs for prediction of specific wear rate or frictional coefficient from the material compositions, mechanical properties, and testing conditions (temperature, normal force and sliding speed) [39].

Especially the pharmaceutical, food and cosmetics industries deal with intangible parameters such as organoleptic properties (e.g., creaminess, suavity, odor, flavor, color, brightness), which govern the quality of most of their products.

Food industry challenges include multi-criteria decision-making, arose from the lack of objective information regarding the desired qualitative standards of final products [40-42]. Since appropriate mathematical models cannot be derived, ANNs modeling and other methodologies based on artificial intelligence have allowed the construction of flexible and robust automatic decision-making systems for product evaluation [40-42]. For processing olive oil, the ANNs have been used to classify the fruit and to test its ripeness, to detect frauds or adulterations in the final product or to predict its characteristics [42]. In wine production, ANNs were developed for the evaluation of aged wine distillates with emphasis on the properties of aroma and taste [43]. ANN modeling has also been used to calibrate the "electronic nose" arrangement for milk recognition, because the heterogeneous nature of milk makes the analysis of the aroma especially complex, even for classical analytical methods (e.g., gas chromatography, mass spectrometry, gas chromatography, olfactometry) [44].

In the pharmaceutical industry, scientists have used ANN models associated with concepts of quantitative structure-property relationships (QSPR) and quantitative structure- activity relationships (QSAR) for the prognosis of the behavior of new molecules, even before they are synthesized [45]. Such studies can infer the organoleptic, physicochemical and therapeutic properties of pharmaceutical compounds. For example, ANNs modeling was utilized to relate the structure of 332 diverse pharmaceuticals compounds to their aqueous solubility [46]. Also, the structure and molecular topology were used by Huuskonen et al. in the prediction of the solubility of the drug molecules [47].

Petrovic et al. demonstrated the possible use of dynamic neural networks to model diclofenac sodium release from polyethylene oxide hydrophilic matrix tablets. Fractions of polymer and compression force have been selected as the most influential factors. The ability of ANN to model drug release has been assessed by the determination of the correlation between predicted and experimentally obtained data [48].

In cosmetic industry, system based ANN is an appropriate technique to assess customer satisfaction on fragrance notes, reducing the needs of perfumery experts to choose the smell for their products [49-50]; with a few experts worldwide, their time availability is limited [51]. In another application in cosmetic industry, Marengo et al. used ANNs to determine the simultaneous separation of 20 typical antimicrobial agents; since such preservatives belong to different classes of chemical species, a high-quality control analysis was needed [45].

The application of ANNs in nanotechnology is often concerned with the prediction of processing parameters and morphologic characteristics of nanoparticles samples in the experimental environments. Some examples of these applications are: modeling and simulation of current-voltage characteristics in carbon nanotube-based gas sensors [52]; prediction of the heat transfer of a silver/water nanofluid in a two-phase closed thermosiphon that is thermally enhanced by magnetic field [53]; estimation of particle size using diffuse reflectance spectra in near-infrared region [54]; analysis and prediction of correlations between processing parameters and the morphologic characteristics of nanocomposites using back-propagation neural network technique [55].

In heating and cooling systems, the applications of ANNs include modeling systems of ventilating and air-conditioning, solar radiation, power-generation, load-forecasting, refrigeration, among others [56]. In fact, estimating the flow of energy and evaluating the performance of the renewable energy systems usually involve the solution of complex mathematical equations. Particularly, in the field of renewable energy systems, data are inherently noisy; thus, such problems can be handled with ANNs [57].

Multipurpose Processes

Multipurpose units are developed to produce different products, according to a schedule. Thus, production must be changed from one process or product to another one periodically. In this case, flexible tools for modeling, monitoring, and control must face pattern change suitably, and again ANN can be used. ANNs were employed to estimate the plant-model mismatch at each sampling instant and to correct the predictions from a process model [58-60].

The polymer industry has used ANN modeling to obtain a specific grade of products, determining operational conditions and inference properties, such as molecular weight, solid content, Mooney viscosity, and polydispersity.

Zhou et al. proposed an ANN approach to achieve an on-line estimation of dry kerosene point in refineries varying crude oils [61]. Megan and Cooper focused on making model adaptations following a load disturbance to a reactor under concentration control and also applied ANN within a DMC algorithm for multivariable composition control of a distillation column [62].

Ruiz et al. presented a strategy for the development and implementation of a fault diagnosis system that interacts with a schedule optimizer in batch chemical plants, using ANN. The information needed to implement the FDS included a historical database of past batches, a Hazard and Operability (HAZOP) analysis and a model of the plant. One of the motivating examples presented corresponded to a multipurpose batch plant [63].

Advanced Control, Faults Detection and Diagnosis, and Safety

ANNs are useful tools for control, safety, fault detection, and diagnosis since a welltrained ANN can appropriately recognize and handle with nonlinear behavior and may be less sensible to noises within their tolerances.

In recent years, with the upsurge of the research in the field of nonlinear control, ANNs have become a popular tool for three major control schemes: predictive control, inverse-model-based control, and adaptive control methods. Typical chemical process units using ANN tools for advanced control and optimization strategies are distillation columns and reactor systems (continuous stirred tank reactors, bioreactors, and the neutralizing reactors) because such systems are multivariable, nonlinear and typically difficult to simulate [7].

For distillation columns, several applications of ANNs can be found, especially to control the top and bottom composition. Macmurray and Himmelblau used an external recurrent ANN within the model predictive control strategy [6]; Basualdo and Ceccato used ANNs and multiloop IMC (Internal Model Control) structures [64]; Ramchandran and Rhinehart used an ANN inverse model incorporated in the GMC (generic model control) strategy to estimate the reflux and holdup rate [65]. Montlaghi et al. constructed an expert system from ANN and genetic algorithm to provide the optimal operating condition and predict the output quality of the crude distillation column [66]. Savafiand Romagnoli used wavelet-based ANNs, also named wave-nets, to the modeling and optimization of an experimental distillation column [67].

Applications of ANNs in the control of reactors systems have also been widely studied [58-59, 68-69]. For example, in an industrial polypropylene plant, a nonlinear predictive control technique employing ANNs was implemented to control the melt flow rate in the polymerization reactor [70]; on-line ANN model was used to control the substrate concentration and pH of an anaerobic digestion system [71], among several others applications.

ANNs has also demonstrated the higher robustness and good performance in detecting and diagnosing process malfunctions, distinguishing between abnormal and normal operations conditions, with quick response, ensuring operation process safety.

Hoskins et al. applied ANNs approach for detection and diagnostic of faults in a large complex chemical plant [72]. Diagnose of multiple faults in a chemical process at steady state operation was addressed by Fan et al. by using ANNs [73]. Dufour et al. developed an ANN tool to detect faults on unmeasured feedstock properties of an industrial pulp digester, since variations of its properties have a significant impact on the kappa number, even under advanced process control [74].

In nuclear engineering, the application of ANNs to fault detection and diagnosis have received more and more attention in the last decades [75]. ANNs were applied for modeling and estimating several nuclear reactor safety parameters [76-78], for sensor fault detection and diagnosis [79-82], for monitoring signs of imminent failure on the melter vessel [30] and nuclear reactor [75], among others. Such applications avoid disasters; besides reducing downtime costs, as well as extra expenses incurred in the destination of the radioactive units and their components.

Meireles et al. prepared a review aiming to help industrial managing and operational personnel decide which kind of ANN topology and training method would be adequate for their specific problems. The paper presents a comprehensive review of the industrial applications of ANNs since 1988. Common questions that arise to practitioners and control engineers while deciding how to use ANNs for specific industrial tasks are answered. ANN industrial applications are grouped and tabulated by their core functions and what they performed on the referenced papers, focusing mainly on pattern recognition and classification, optimization, modeling, identification, and control [83].

APPLICATIONS OF ARTIFICIAL NEURAL NETWORKS IN CHEMICAL AND RELATED AREAS

The same procedure was applied to investigate the evolution the number of publication of ANNs in chemistry and related areas during 1989 still 2016. In this context, Figure 2 presents the number of publications along years reported in the literature using ANNs in the main areas of chemistry and related areas, according to SCOPUS database. One can see that the number of publications grown linearly from 1994 to 2015 and the number of publication is superior when it compared with the area of Chemical Engineering. The number of publications in chemistry and related areas is almost three times, during the last five years, when it was compared with the number of publication in chemical engineering related with ANNs.

Once more, the same procedure was used to investigate the correlation between different areas of chemistry and related sectors that used ANNs. Table 2 presents the values of for the considered fields. It can be observed that f is not a symmetric matrix and that chromatography presents higher correlation with other areas. This fact is evidenced by the wide dissemination of chromatographic systems in research and industrial laboratories.

The application of artificial neural networks (ANNs) in chemistry and related areas are based on data reduction, overlapped signal resolution, experimental design and surface response, modeling, pattern recognition, and multivariate calibration method.

The following sections describe some common applications based on the classification described above.

Data Reduction

Before demonstrating the application of artificial neural networks in data reduction, it is convenient to introduce a particular type of ANN called Kohonen networks. This kind of ANN belongs to the class of self-organizing maps. In opposite to MLF (Multi-Layer Feedforward) and RBF (Radial Basis Function) networks, they are designed for unsupervised pattern recognition tasks. The Kohonen networks consist of one layer of neurons, ordered in a low-dimensional map. Each neuron or unit contains a weight vector of the same dimension as the input pattern. After training, the individual weight vectors are oriented in such way that the structure of the input space is represented as well as possible in the resultant map. The aim of the Kohonen network is to map similar objects on the same neighbouring neurons. More details about this technique can be found in the literature [84]. In this context, the data mining technique called Kohonen self-organizing map algorithm was applied to the data obtained from the analytical method where "dry distillation" was coupled to headspace solid-phase microextraction (HS-SPME) to investigate the chemical composition of the rosemary plant (Rosmarinus officinalis L.). The large data set obtained was then treated with a rarely used chemometric technique based on nonclassical statistics. This method highlighted a strong correlation between the volatile chemical compositions of the samples and their origins, and it allowed the samples to be grouped according to geographical distribution. Moreover, the method allowed to identify the constituents that exerted the most influence in the classification [85].

Overlapped Signal Resolution

There are some chemometric techniques for simultaneous evaluation of overlapped signals, independent of the type of signal, such as deconvolution or semi-differential techniques coupled to curve fitting, multivariate curve resolution, and multivariate calibration. Specifically, in the field of electroanalytical chemistry, many applications based on different regression methods were reported such as multilinear regression (MLR), principal component regression (PCR), partial least squares regression (PLS), and artificial neural networks (ANNs). In this way, the ANN was compared to MLR, PCR, and PLS to resolve overlapped electrochemical signals using different parameters as input data: position, height, half width, derivative, and area of voltammetric peaks. The peak parameters based strategy can involve a reliable and fast alternative to resolve multicomponent system in voltammetry or, even in other analytical techniques such as chromatography or spectroscopy. The average relative errors of the test sets in the case of ANNs were similar than those belonging to PLS and MLR [86].

Experimental Design and Response Surface

Response surface methodology (RSM) seeks the relationships between several explanatory variables and one or more response variables. The principal idea of RSM is to use a set of designed experiments to obtain an optimal response. RSM simplifies the original problem through some polynomial estimation over small sections of the available area, elaborating on optimum provision through a well-known optimization technique, say gradient method. However, the real world problems are usually complicated; polynomial estimation may not carry out well in providing a good representation of the objective function. Also, the primary issue of the gradient method, getting trapped in a local minimum (maximum), makes RSM at a disadvantage, while defining sub-sections of the available area is also a problem faced to the analyst. In this context, artificial neural networks were used to improve the estimation in the RSM context reducing the calculations. Finally, it was proposed another algorithm to optimize the established model called the simulated annealing method, which it is convenient, for maximizing the estimated objective function in reaching a suitable point. Three examples of different complexities were solved to highlight the merits of the proposed method when it properly adjusted to the problem at hand. The proposed methodology overtook the classical method [87]. Once more, the combination of ANN with multivariate optimization approach was more efficient than the single variable approach to predict the optimal solid-phase extraction (SPE) conditions for determination of cis- and transresveratrol in Australian wines by capillary zone electrophoresis [88].

Modeling

From the view of mathematic, an artificial neural network is often seen as a universal model that is based on the primary concept of artificial intelligent (AI) and tries to simulate the process of the human brain and nervous system. The ANNs contain a series of mathematical correlations that is applied to simulate the learning and memorizing operations. The ANNs techniques learn through an example in which an experimentally measured set of input factors or variables and the corresponding outputs are presented to determine the rules that govern the relationship between them. ANNs are considered to be powerful in capturing a non-linear effect and are practically applicable to every situation existing between the independent and dependent factors or variables [89]. ANN was applied for the prediction of percentage adsorption efficiency for the removal of Cu(II) ions from industrial leachate by pumice. The model was developed using a three-layer feedforward backpropagation network with 4, 8, and 4 neurons in first, second, and third layers, respectively. Lately, radial basis function (RBF) network was also proposed and its performance was compared to traditional network type. The RBF network model was able to predict the removal of Cu(II) from industrial leachate more accurately [90]. The same research group used the ANN for modeling the prediction of percentage adsorption efficiency for the removal of Zn(II) ions from industrial leachate by hazelnut shell [91].

Pattern Recognition

Pattern recognition or classification method is defined as the assignation of a sample to one category based on the values of the indices measured on it. When there are Nc possible categories the samples can belong to, classification is usually accomplished by calculating, for each sample, the probability that it belongs to each of these classes given the measurements (posterior probability) and then to assign it to the class corresponding to the maximum value of this probability. Based on this, classification methods are often divided into parametric and nonparametric: the former assume the latter are distribution-free. Artificial neural networks used for classification problems fall in this second group since no assumption is made on the form of the posterior probabilities for each class [92]. A new methodology based on ANNs was proposed to classify food vegetable oils: canola, sunflower, corn, and soybean using fluorescence spectra data. The proposed methodology was able to classify with a 72% a rate of success [93]. The authenticity of grated Protected Denomination of Origin (PDO) Parmigiano Reggiano cheese was investigated using infrared spectroscopy data coupled with softindependent modeling of class analogy (SIMCA) and artificial neural networks (ANN) to classify the cheese samples. ANN was more efficient than SIMCA in the classification of all the cheese classes [94]. The use of ANNs for the independent analysis of GC - MS (gas chromatography - mass spectrometry) profiles of Lucilia sericata was investigated, where ANNs were required to estimate the age of the larvae to aid in the estimation of the postmortem interval (PMI). The ANNs correctly classified the data with accuracy scores of 80.8%, and 87.7% for two independent analysis approaches [95].

Multivariate Calibration Method

Artificial neural networks are increasingly applied in analytical chemistry as a powerful complement to traditional statistical and modeling methods. ANNs represent so-called "soft" modeling without the need to know and establish a mathematical model. The ANN model should have an ability to learn and extract y-x mapping relations from the presentation of a set of training samples. Among the types of ANNs applied in calibration method, backpropagation neural network (BPNN) is the most widely used one. Radial basis function neural network (RBFNN) and generalized regression neural network (GRNN) are ANNs, which both use radial basis functions as transfer functions, offering interesting alternatives to BPNN in the sense that they allow local and fast training [96]. More details about the types of ANNs can be found elsewhere [92]. Near infrared spectroscopy coupled with multivariate regression methods were used to predict four important properties of biodiesel: density (at 150C), kinematic viscosity (at 400C), water content, and methanol content. This study compared the performance of linear and non-linear calibration techniques - namely, multiple linear regression (MLR), principal component regression (PCR), partial least squares regression (PLS), polynomial and Spline-PLS -, and artificial neural networks (ANN) for prediction of biodiesel properties. The artificial neural network (ANN) approach was superior to the linear (MLR, PCR, PLS) and "quasi-nonlinear" (Poly-PLS, Spline-PLS) calibration methods [97].

The determination of the oxidizable amino acids in animal feed samples was investigated using a voltammetric electronic tongue. The quantitative information contained in the voltammograms was extracted employing the discrete wavelet transform (DWT) and then processed using artificial neural networks (ANNs). The ANN was subsequently used to model the system departing from the reduced information, and obtaining the concentrations of the species. The best results were achieved when using two hidden layers in a backpropagation neural network trained with the Bayesian regularization algorithm [98].

CONCLUSION

Artificial neural networks have successfully been applied in several fields related to chemical engineering, chemistry and other relate areas since some years. Their versatility and their ability to deal with highly non-linear trends in data allowed obtaining significantly good results in many applications for which the use of traditional chemometric methods have failed. However, they are usually less easily interpretable, and there are many factors to have into account in the modeling phase so that examples of misuse are commonly encountered in the literature. Due to the robustness and efficiency of ANNs to solve complex problems, these methods have been widely employed in several research fields such as medicinal chemistry, pharmaceutical research, analytical chemistry, biochemistry, food research, etc.

A look to the future would suggest the introduction into the chemometric literature of different network architectures and learning algorithms, for example, recurrent and dynamic networks, or Bayesian learning used until now by the computer scientists or the physicists.

ACKNOWLEDGMENTS

Luna, A. S., and Lima, E.R.A. thank the support of UERJ (Programa Prociência), FAPERJ, and CNPq.

References

REFERENCES

[1] Haykin, Simon. 2009. Neural Networks and Learning Machines. 3th. Edition, New Jersey, USA: Pearson Education, Inc.

[2] Braga, Antonio P., Carvalho, André P. L. F., Ludermir, Teresa B. 2007. Redes Neurais Artificiais. 2a. Edição, Rio de Janeiro, Brazil, Livros Técnicos e Científicos Editora Ltda. [Artificial Neural Networks, 2nd Edition, Rio de Janeiro, Brazil, Livros Técnicos e Científicos Editora Ltda].

[3] Pirdashti, M., Curteanu, S., Kamangar, M. H., Hassim, M. H., Khatami, M. A. "Artificial neural networks: applications in chemical engineering." 2013. Reviews in Chemical Engineering 29 (4): 205 - 239.

[4] Haykin, Simon. 1994. Neural Networks: A Comprehensive Foundation, New York, USA: Macmillan.

[5] Hoskins, J. C. and Himmelblau, D. M. 1998. "Artificial neural network models of knowledge representation in chemical engineering." Computers and Chemical Engineering 12: 881 - 890.

[6] Macmurray, J. C., and Himmelblau, D. M. 1995. "Modeling and control of a packed distillation column using artificial neural networks." Computers and Chemical Engineering 19(10): 1077 - 1088.

[7] Hussain, M. A. 1999. "Review of the applications of neural networks in chemical process control - simulation and online implementation." Artificial Intelligence 13: 55 - 68.

[8] Ramli, N. M., Jan, B. M., Abdullah, B. 2014. "Composition Prediction of a Debutanizer Column using Equation Based Artificial Neural Network Model." Neurocomputing 131: 59 - 76.

[9] Fortuna, F., Graziani, S., Xibilia, M.G. 2005. "Softsensors for product quality monitoring in debutanizer distillation columns." Control Engineering Practice 13: 499 - 508.

[10] Munack, A., and Schönert, K. (Editors). 2014. Computer Applications in Biotechnology, New York, USA: Elsevier.

[11] Bagheri, M., Mirbagheri, S.A., Kamarkhani, A.M. 2016. "Modeling of effluent quality parameters in a submerged membrane bioreactor with simultaneous upward and downward aeration treating municipal wastewater using hybrid models." Desalination and Water Treatment 57 (18): 8068 - 8089.

[12] Han, H., Li, Y., Qiao, J. 2014. "A fuzzy neural network approach for online fault detection in waste water treatment process." Computers and Electrical Engineering 40 (7): 2216 - 2226.

[13] Han, H., Li, Y., Guo, Y. N. 2016. "A softcomputing method to predict sludge volume index based on a recurrent self-organizing neural network." Applied SoftComputing Journal 38: 477 - 486.

[14] Baughman, D. R., and Liu, Y. A. 1995. Neural networks in bioprocessing and chemical engineering, San Diego, USA: Academic Press Inc.

[15] Ignova, M., Montague, G. A., Ward, A. C., Glassey, J. 1999. "Fermentation seed quality analysis with self-organizing neural networks." Biotechnology and Bioengineering 64 (1): 82 - 91.

[16] Breusegem, V. V., Thibault, J., Cheruy, A. 1991. "Adaptive Neural Models for On-Line Prediction in Fermentation." Canadian Journal of Chemical Engineering 69: 481 - 487.

[17] Anker, L. S., and Jurs, P. C. 1992. "Prediction of Carbon-13 Nuclear Magnetic Resonance Chemical Shifts by Artificial Neural Networks." Analytical Chemistry 64: 1157 - 1164.

[18] Long, J. R., Mayfield, H. T., Henley, M. V., Kromann, P. K. 1991. "Pattern Recognition of Jet Fuel Chromatography Data by Artificial Neural Networks with Backpropagation of Error." Analytical Chemistry 63: 1256 - 1261.

[19] Meyers, B., Hansen, T., Nute, D. Albersheim, P., Darrill, A. York, W., Sellers, J. 1991. "Identification of the 1H-NMR Spectra of Complex Oligosaccharides with Artificial Neural Networks." Science 251: 542 - 544.

[20] Baughman, D. R., and Liu, Y. A. 1994."An Expert Network for Predictive Modeling and Optimal Design of Extractive Bioseparations in Aqueous Two-Phase Systems." Industrial and Engineering Chemistry Research 33: 2668 - 2687.

[21] Ingman, L.C. 2000. "Utilization of Neural Network Technology for Some Pulp and Paper Industry Applications." TAPPI Process Control, Electrical and Information Conference and ISA-PUPID 39th Annual Symposium 1: 115 - 120.

[22] Rajesh, K., and Ray, A. K. 2006. "Artificial neural network for solving paper industry problem: a review." Journal of Scientific and Industrial Research 65: 565 - 573.

[23] Dahlquist, E. 2008. Use of modeling and simulation in pulp and paper industry. Västerås, Sweden: EU COST and Malardalen University Press.

[24] Zainuddin, Z., Wan Daud, W. R., Pauline, O., Shafie, A. 2011. "Wavelet neural networks applied to pulping of oil palm fronds." Bioresource Technology 102: 10978 - 10986.

[25] Aguiar, H. C., and Filho, R. M. 2001. "Neural network, and hybrid model: a discussion about different modeling techniques to predict pulping degree with industrial data." Chemical Engineering Science 56: 565 - 570.

[26] Costa, A. O. S., Biscaia Jr, E. C., Lima, E. L. 2004. "Mathematical Description of the Kraftrecovery boiler furnace." Computers and Chemical Engineering 28(15): 633 - 641.

[27] Zhang Z., and Friedrich, K. 2003. "Artificial neural networks applied to polymer composites: a review." Composites Science and Technology 63(14): 2029 - 2044.

[28] Duce, C., Micheli, A., Starita, A., Tine, M. R., Solaro, R. 2006. "Prediction of Polymer Properties from their Structure by Recursive Neural Networks." Macromolecular Rapid Communications 27: 711 - 715.

[29] Wagner, M. G., and Montague, G. A., Tham, M.T. 1997. "Neural networks for steady state modeling of an extruder." Artificial Intelligence in Engineering 11(4): 375 - 382.

[30] Lennox, B., Montagueb, G. A., Frithc, A. M., Gentc, C., Bevanc, V. 2001. "Industrial application of neural networks - an investigation." Journal of Process Control 11(5): 497 - 507.

[31] Golzar, K., Amjad-Iranagh, S., Modarress, H. 2016. "Investigation of polymer + solvent mixtures by means of cubic equations of state and artificial neural network." Fluid Phase Equilibria 45: 134 - 143.

[32] Yousefi, F., and Karimi, H. 2013. "Application of the equation of state and artificial neural network to the prediction of volumetric properties of polymer melts." Journal of Industrial and Engineering Chemistry 19(2): 498 - 507.

[33] Jones, S. P., Jansen, R., Fusaro, R.L. 1997. "Preliminary investigation of neural network techniques to predict tribological properties." Tribology Transactions 312: 40 - 320.

[34] Aymerich, F., and Serra, M. 1998. "Prediction of fatigue strength of composite laminates by means of neural networks." Key Engineering Materials 144: 231 - 240.

[35] Lee, J. A., Almond, D. P., Harris, B. 1999. "The use of neural networks for the prediction of fatigue lives of composite materials." Composites Part A: Applied Science and Manufacturing 30(10): 1159 - 1169.

[36] Al-Assaf, Y., and El-Kadi, H. 2001. "Fatigue life prediction of unidirectional glass fiber/epoxy composite laminate using neural networks." Composite Structures 53(1): 65 - 71.

[37] El-Kadi, H., and Al-Assaf, Y. 2002. "Prediction of the fatigue life of unidirectional glass fiber/epoxy composite laminate using different neural network paradigms." Composite Structures 55(2): 239 - 246.

[38] Velten, K., Reinicke, R., Friedrich, K. 2000. "Wear volume prediction with artificial neural networks." Tribology International 33: 731 - 736.

[39] Zhang, Z., Friedrich, K., Velten, K. 2002. "Prediction of tribological properties of short fiber composites using artificial neural networks." Wear 252(7-8): 668 - 675.

[40] Zhang, J., and Chen, Y. 1997. "Food sensory evaluation employing artificial neural networks." Sensor Review 17(2): 150 - 158.

[41] Linko, S. 1998. "Expert systems what can they do for the food industry?" Trends in Food Science and Technology 9(1): 3 - 12.

[42] Funes, E., Allouche, Y., Beltrán, G., Jiménez, A. 2015. "A Review: Artificial Neural Networks as Tool for Control Food Industry Process." Journal of Sensor Technology 5: 28 - 43.

[43] Raptis, C.G., Siettos, C.I., Kiranoudis, C. T., Bafas, G. V. 2000. "Classification of aged wine distillates using fuzzy and neural network systems." Journal of Food Engineering 46(4): 267 - 275.

[44] Sanzogni, L., and Kerr, D. 2001. "Milk production estimates using feed forward artificial neural networks." Computers and Electronics in Agriculture 32: 21 - 30.

[45] Marengo, E., Gianotti, V., Angioi, S., Gennaro, M. C. 2004. "Optimization by experimental design and artificial neural networks of the ion-interaction reversed-phase liquid chromatographic separation of twenty cosmetic preservatives." Journal of Chromatography A 1029(1-2): 57 - 65.

[46] Mitchell, B. E., and Jurs, P.C. 1998. "Prediction of aqueous solubility of organic compounds from molecular structure." Journal of Chemical Information and Modeling 38: 489 - 496.

[47] Huuskonen, J., Salo, M., Taskinen, J. 1998. "Aqueous solubility prediction of drugs based on molecular topology and neural network modeling." Journal of Chemical Information and Modeling 38: 450 - 456.

[48] Petrovic, J., Ibric, S., Betz, G., Parojc ic, J., Ðuric, Z. 2009. "Application of dynamic neural networks in the modeling of drug release from polyethylene oxide matrix tablets." European Journal of Pharmaceutical Sciences 38: 172 - 180.

[49] Kengpola, A., and Wangananonb, W. 2006. "The expert system for assessing customer satisfaction on fragrance notes: Using artificial neural networks." Computers and Industrial Engineering 51(4): 567 - 584.

[50] Hanafizadeh, P., Ravasan, A. Z., Khaki, H. R. 2010. "An expert system for perfume selection using the artificial neural network." Expert Systems with Application 37: 8879 - 8887.

[51] Dowthwaite, Stephen V. 1998. Training the ABC's of perfumery, vol. 3(1). UK, Perfumer and Favourite Allured Publications.

[52] Akbari, E., Buntat, Z., Enzevaee, A., Ebrahimi, M.; Yazdavar, A. H., Yusof, R. 2014. "Analytical modeling, and simulation of I-V characteristics in carbon nanotube-based gas sensors using ANN and SVR methods." Chemometrics and Intelligent Laboratory Systems 137: 173 - 180.

[53] Salehi, H., Zeinali Heris, S., Koolivand Salooki, M., Noei, S.H. 2011. "Designing a neural network for closed thermosyphon with nanofluid using a genetic algorithm." Brazilian Journal of Chemical Engineering 28: 157 - 168.

[54] Khanmohammadi, M., Garmarudi, A. B., Khoddami, N., Shabani, K., Khanlari, M. A. 2010. "A novel technique based on diffuse reflectance near-infrared spectrometry and back-propagation artificial neural network for estimation of particle size in TiO2 nanoparticle samples." Microchemical Journal 95: 337 - 340.

[55] Ma, J., Zhu, S.G., Wu, C. X., Zhang, M.L. 2009. "Application of back-propagation neural network technique to high-energy planetary ball milling process for synthesizing nanocomposite WC-MgO powders." Materials and Design 30: 2867 - 2874.

[56] Kalogirou, Soteris A. 2000. "Applications of artificial neural networks for energy systems." Applied Energy 67(1-2): 17 - 35.

[57] Kalogirou, Soteris A. 2001. "Artificial neural networks in renewable energy systems applications: a review." Renewable and Sustainable Energy Reviews 5(4): 373 - 401.

[58] Willis, M. J., Montague, G. A., Di Massimo, C., Tham, M.T., Morris, A. J. 1992. "Artificial neural networks in process estimation and control." Automatica 28(6): 1181 - 1187.

[59] Hunt, K. J., and Sbarbaro, D. 1992. "Studies in a neural network based control." IEE Control Engineering Series, London.

[60] Turner, P., Montague, G. A., Morris, A. J. 1995. "Neural networks in dynamic process state estimation and nonlinear predictive control." Fourth International Conference on Artificial Neural Networks. 1: 284 - 289.

[61] Zhou, Y. P., Zhao, B. Q., Wu, D.X. 2000. "Application of genetic algorithms to fault diagnosis in nuclear power plants." Reliability Engineering and System Safety 67(2): 153-160.

[62] Megan, L., and Cooper, D. J. 1995. "A neural network strategy for disturbance pattern classification and multivariable adaptive control." Computers and Chemical Engineering 19 (2): 171 - 186.

[63] Ruiz, D., Cantón, J., Nougués, J. M., Espuña, A., Puigjaner, L. 2001. "On-line fault diagnosis system support for reactive scheduling in multipurpose batch chemical plants." Computers and Chemical Engineering 25: 829 - 837.

[64] Basualdo, M. S., and Ceccatto, H. A. 1995. "Predictive control methods for distillation columns using neural networks." DYCORD International Federation of Automatic Control (IFAC) Symposium 95:171 - 176.

[65] Ramchandran, S., and Rhinehart, R.R. 1995. "A very simple structure for neural network control of distillation." Journal of Process Control 5(2): 115 - 128.

[66] Motlaghi, S., Jalali, F., Ahmadabadi, M. N. 2008. "An expert system design for a crude oil distillation column with the neural networks model and the process optimization using genetic algorithm framework." Expert Systems with Applications 35(4): 1540 - 1545.

[67] Safavi, A., and Romagnoli, J. A. 1997. "Application of wavelet-based neural networks to the modeling and optimization of an experimental distillation column." Engineering Applications of Artificial Intelligence 10(3): 301 - 313.

[68] Hernandez, E., and Arkun, Y. 1990. "Neural network modeling and an extended DMC algorithm to control nonlinear systems." American Control Conference 1: 2454 - 2459.

[69] Psichogios, D. M., and Ungar, L.H. 1991. "Direct and indirect model based control using artificial neural networks." Industrial and Engineering Chemistry Research 30: 2564 - 2573.

[70] Keeler, J., Martin, G. et al. 1997. The Process Perfector - the next step in multivariable control and optimization. Technical Report, Pavilion Technologies Inc., Austin, TX.

[71] Emmanouilides, C., and Petrou, L. 1997. "Identification, and control of anaerobic digesters using adaptive, online trained neural networks." Computers and Chemical Engineering 21(1): 113 - 143.

[72] Hoskins, J. C., Kaliyur, K. M., Himmelblau, D. M. 1991. "Fault diagnosis in complex chemical plants using artificial neural networks." AIChE Journal 37(1): 137 - 141.

[73] Fan, J. Y., Nikolaou, M., White, R. E. 1993. "An approach to fault diagnosis of chemical processes via neural networks." Process System Engineering 39(1): 82 - 88.

[74] Dufour, P., Bhartiya, S., Dhurjati, P. S., Doyle, F. J. 2005. "Neural network-based software sensor: training set design and application to a continuous pulp digester." Control Engineering Practice 13: 135 - 143.

[75] Messai, A., Mellit, A., Abdellani, I., Massi Pavan, A. 2015. "On-line fault detection of a fuel rod temperature measurement sensor in a nuclear reactor core using ANNs." Progress in Nuclear Energy 79: 8 - 21.

[76] Coban R., and Can, B. 2010. "A trajectory tracking genetic fuzzy logic controller for nuclear research reactors." Energy Conversion and Management 51(3): 587 - 593.

[77] Coban, Ramazan. 2013. "A context layered locally recurrent neural network for dynamic system identification." Engineering Applications of Artificial Intelligence 26(1): 241 - 250.

[78] Coban, Ramazan. 2014. "Power level control of the TRIGA Mark-II research reactor using the multi-feedback layer neural network and the particle swarm optimization." Annals of Nuclear Energy 69: 260 - 266.

[79] Jin, Y. L., and Hong, X. 2013. "Application of fuzzy neural network to fault diagnosis of the sensor with mechanical properties in nuclear power plant." Advances in Materials Research 644: 68 - 71.

[80] Elnokity, O., Mahmoud, I. I., Refai, M. K., Farahat, H. M. 2012. "ANN based Sensor Faults Detection, Isolation, and Reading Estimates - SFDIRE: Applied in a nuclear process." Annals of Nuclear Energy 49: 131 - 142.

[81] Sabura Banu, U., and Uma, G. 2011. "ANFIS based sensor fault detection for continuous stirred tank reactor." Applied SoftComputing Journal 11(2): 2618 - 2624.

[82] Kasinathan, M., Saidhar Rao, B., Murali, N., Swaminathan, P. 2009. "An artificial neural network approach for the discordance sensor data validation for SCRAM parameters." IEEE First International Conference on Advancements in Nuclear Instrumentation Measurement Methods and Their Applications (ANIMMA) 1: 1 - 5.

[83] Meireles, M. R. G., Almeida, P. E. M., Simões, M. G. 2003. "A Comprehensive Review for Industrial Applicability of Artificial Neural Networks.". IEEE Transactions on Industrial Electronics 50(3): 585 - 601.

[84] Vandeginste, B. G. M., Massart, D. L., Buydens, L. M. C., de Jong, S., Lewi, P. J., Smeyers-Verbeke, J. 1998. Handbook of Chemometrics and Qualimetry, Part B, Amsterdam, Elsevier.

[85] Tigrine-Kordjani, N., Chemat, F., Meklati, B. Y., Tuduri, L., Giraudel, J. L., Montury, M. 2007. "Relative characterization of rosemary plants according to their geographical origins using microwave-accelerated distillation, solid-phase microextration, and Kohonen self-organizing maps." Analytical and Bioanalytical Chemistry 389: 631 - 641.

[86] Palacios-Santander, J. M., Cubillana-Aguilera, L. M., Naranjo-Rodríguez, J. L., Hidalgo-Hidalgo-de-Cisneros, J. L. 2007. "A chemometric strategy based on peak parameters to resolve overlapped eletochemical signals." Chemometrics and Intelligent Laboratory Systems 85: 131 - 139.

[87] Abbasi, B., and Mahlooji, H. 2012. "Improving response surface methodology by using artificial neural network and simulated annealing." Expert Systems and Applications 39: 3461 - 3468.

[88] Spanilá, M., Pazourek, J., Farková, M., Havel, J. 2005. "Optimization of solid-phase extraction using artificial neural networks in combination with an experimental design for determination of resveratrol by capillary zone electrophoresis in wines." Journal of Chromatography A 1084: 180 - 185.

[89] Esfandian, H., Samadi-Maybodi, A., Parvini, M., Khoshandam, B. 2016. "Development of a novel method for the removal of diazinon pesticide from aqueous solution and modeling by artificial neural networks (ANN)." Journal of Industrial and Engineering Chemistry 35: 295 - 308.

[90] Turan N. G., Mesci, B., Ozgonenel, O. (2011). "The use of artificial neural networks (ANN) for modeling of adsorption of Cu(II) from industrial leachate by pumice." Chemical Engineering Journal 171: 1091 - 1097.

[91] Turan N. G., Mesci, B., Ozgonenel, O. (2011). "Artificial neural networks (ANN) approach for modeling Zn(II) adsorption from leachate using a new biosorbent. Chemical Engineering Journal 173: 98 - 105.

[92] Marini, Frederico. 2009. "Artificial neural networks in foodstuffanalyses: Trends and perspectives." Analytica Chimica Acta 635: 121 - 131.

[93] da Silva, C. E. T., Filardi, V. L., Pepe, I. M., Chaves, M. A., Santos, C. M. S. 2015. "Classification of food vegetable oils by fluorimetry and artificial neural networks." Food Control 47: 86 - 91.

[94] Cevoli, C., Gori, A., Nocetti, M., Cuibus, L., Caboni, M. F., Fabbri, A. 2013. FT-NIR and FT-MIR spectroscopy to discriminate competitors, non-compliance and compliance grated Parmigiano Reggiano chesse. Food Research International 52: 214 - 220.

[95] Butcher, J. B., Moore, H. E., Day, C. R., Adam, C. D., Drijfhout, F. P. 2013. "Artificial neural network analysis of Lucilia sericata for post mortem interval estimation." Forensic Science International 232: 25 - 31.

[96] Chen, Y., He, J., Zhang, J., Yu, Z. 2009. "Extending the working calibration ranges of four hexachlorocyclohexane isomers in gas chromatography-electron capture detector by radial basis function neural network." Talanta 79: 916 - 925.

[97] Balabin, R. M., Lomakina, E. I., Safieva, R. Z. 2011. "Neural network (ANN) approach to biodiesel analysis: Analysis of biodiesel density, kinematic viscosity, methanol and water content using near infrared (NIR) spectroscopy." Fuel 90: 2007 - 2015.

[98] Moreno-Barón, L., Cartas, R., Merkoçi, A., Alegret, S., del Valle, M., Leija, L., Hernandez, P. R.; Muñoz, R. 2006. "Application of the wavelet transform coupled with artificial neural networks for quantification purposes in a voltammetric electronic tongue." Sensors and Actuators B 113: 487 - 499.