The simultaneous transport of salt and moisture is very complex and coupled, thus rather difficult to model. The main storage parameter which describes the hygral state of any porous material is the degree of saturation. It is usually described by the sorption isotherm, which gives the relation between the degree of saturation (mass of adsorbed water) and the relative humidity of surrounded air.

The sorption isotherm depends on the kind and the density of salt, which occupies the surface of pores. This paper is devoted to experimental and theoretical analysis of the sorption isotherm of cement mortar containing different densities of NaCl. The moisture content by various relative humidity and salt concentration was measured by means of the saturated salt solution method. Then the sorption isotherm and both derivatives were approximated using neural networks.

The mathematical model of coupled heat, moisture and salt transport consists of five governing equations: moisture, air and salt mass balance equations, enthalpy conservation equation, linear momentum conservation equation. A feed-forward neural network (FFNN) with nonlinear, sigmoidal activation functions is used in this paper. This kind of network is known to be a universal approximation tool for arbitrary functions and their partial derivatives. This second property is of special interest, because the trained model is intended to be used directly in the finite element code and the derivatives are required for obtaining numerical solution. The concise introduction to the feed-forward neural networks in the framework of the graph theory is given, together with the formulas needed to calculate the derivatives. The general derivative network concept is proposed. All training was performed by means of FFNET package, written entirely in the Python language. It implements the ideas described, in particular, it provides access to the exact derivatives. After the training stage, the network was exported to a Fortran code and incorporated in the HMTRA_SALT [1] numerical routines. In this way we obtained a very efficient numerical model reflecting the experimental data with very good correlations. It was shown that a surprisingly simple, layered network (2-4-1) fits the experimental data well.

The drying experiment for the cement mortar containing pure water and two salt solutions with different NaCl concentrations were simulated. The material containing pure water dried faster than the materials with salt. The higher salt concentration is the slower drying. Salt affects the hygral state of the material. It also influences the energy transport by changing the effective conductivity and thermal capacity of multiphase domain.

1

M. Koniorczyk, D. Gawin, "Heat and Moisture Transport in Porous Building Materials Containing Salt", Journal of Building Physics, 31(4), 279-300, 2008. doi:10.1177/1744259107088003

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £140 +P&P)