Analysis of Thermal Transient Processes by Means of Neural Network Technique

Abstract

Particularly in the past decade, a very large effort has been expended in developing numerical methods for solving complex multidimensional problems in area of engineering processes. In the last few years the complex behaviour of biological, chemical and industrial systems has been explained in terms of dynamic analysis and many techniques to obtain predictions have been developed. The dynamic investigations of a various processes have focused attention on the problem of the mathematical description. In principle, this knowledge may be obtained by many computational modelling. As an easier alternative, the experimental data may be used to find out a black-box model or an empirical correlation defining the system behaviour. The limitation of this approach is that it requires assumption of the functional form of the proposed correlation. The popular approach to analyse the unsteady and steady heat transfer problems is associated with the availability of non-linear empirical modelling methodologies, such as neural networks, inspired by the biological network of neurons in the brain (Hussain, 1999; Ou & Achenie, 2005). Authors (Liau & Chen, 2006) proposed this methodology to model optimal concentrations of reactants for preparing sub-micron silica particles. Different sets of the reactant concentrations were selected within an operating range and were designed to evaluate the PSD data. The relationship between the reactant concentration and resultant PSD can be evaluated by means of the ANN modelling approach. The neural network models can be successfully used to compute PSD of particles with different shapes in highly concentred suspensions from laser diffraction measurements (Nascimento et al., 1997; Guardani et al. 2002). The ANN pattern recognition (ANNPR) approach has also been proposed for fed-batch cultivation processes of Escherichia coli (Duan et al., 2006). A novel data mining macro-kinetic approach based on ANN was proposed to develop the macrokinetic model of oxidation of p-xylene to terephthalic acid in a industrial type of continuous stirred tank reactor (Yan, 2007). Authors (Liu & Kim, 2008) used the purely mathematic and mechanical model with ANN to model membrane filtration process. As a tool of modelling, neural network technique has been used by (Jones et al., 1999) to magnetic inverse problem of determining the anisotropy field distribution from experimental transverse susceptibility data. Approximation models such as artificial neural networks (ANNs) are powerful and reliable in predicting the complex conditions such as nonlinear and time-variant biological