High Temperatures-High Pressures, 1994, volume 26, pages 163-175 Influence of the boundary conditions on the estimation of thermophysical properties from tran_sient experiments Federico Scarpa, Guido Milano Dipartimento di lngegneria Energetica, Universita di Genova, Via AII'Opera Pia 15a, 116145 Genova, Italy Presented at the 13th European Conference on Thermophysical Properties, Lisboa, Portugal, 30 August-3 September 1993 · Abstract. The aim of this work is to investigate, by means of simulated experiments, the influence of the boundary conditions on the quality of the estimation of thermophysical properties of materials from transient data. In particular, attention is devoted to the problem of simultaneous reconstruction of the temperature-dependent thermal conductivity k(T) and the volumetric heat capacity C( T) from a single experiment of one-dimensional transient conduction. The Jinearised Kalman ftlter is selected as the estimation algorithm and several excitation functions are tested and compared, eg linear increase of heat flux, step variation of heat flux, and step or square wave variation of temperature. The numerical results show that the best performance for the reconstruction process is obtained with a temperature square wave so that the sample is heated and cooled during the c::-xperiment. t. Introduction The design of optimal experiments for nonlinear problems was investigated by Box and Hunter (1964) and Fedorov (1972). Beck and Arnold (1977) provide an overview of various cases with particular regard to the problem of the optimal sensor locations and of the optimal boundary conditions for thermal models with unknown parameters. In this paper we contribute to the search for the best boundary conditions suitable for simultaneously identifying the temperature-dependent thermal conductivity k(T) and the volumetric heat capacity C(T) from experiments of transient conduction. Temperature, T, and heat tlux, q, are assumed known (measured) at the two boundaries of the sample and no information coming from the interior is utilised. This kind of experimental technique, already used in Gamier et al (1991) on the assumption of constant thermophysical properties, appears promising because it reduces the total test time and does not require any intrusion of sensors inside the specimen. Several exdtation functions are tested and compared: linear increase of heat flux, step variation of heat flux, step and square wave variation of temperature etc. As criterion to compare the results, the temperature dependent confidence regions of the unknown functions k(T) and C(T) are selected along with the temperature-dependent correlation function between the two above parameters. The discrete Kalman filter, in the so called 'linearised' version (LKF), is used as an estimation algorithm for its flexibility and because it is able to take into account some quantities partially or completely neglected in other techniques, such as the quality (variance) of the initial temperature distribution and the quality of the input (control) functions. Moreover, it has been shown (Scarpa et al1993) that the Kalman algorithm, for this kind of application, provides results that are more accurate than those given by standard semideterministic methods, such as OLS (ordinary least squares) or MAP (maximum a posteriori). for peer-review only