Summary

In this article, it is shown that artificial neural networks and ANFIS-type neuro-fuzzy systems are effective tools for identifying thermophysical properties and, consequently, a relevant aid to non-destructive testing. The measurements used are responses to excitations with a pseudo-random time profile (PRBS).

As a first step, a forward-propagating neural network of the multilayer Perceptron type was used. Once its topology had been defined and its adjustable parameters identified, this network was capable of accurately estimating the thermal diffusivity of a homogeneous material from its response to a forward or reverse PRBS. To achieve this, we first trained it with a series of fictitious materials of varying thermal behavior, derived from simulations. Then, an initial validation was carried out, again using simulated materials, to check the relevance of the...