A Framework for Thermographic Material Characterization Using Multichannel Neural Network

Research on material characterization has received an increasing amount of attention recently. In several application scenarios, it is essential to effectively estimate the physical properties of objects without coming into contact with them (e.g., tele-operated or autonomous robotics). This article presents the Haptic Eye: a framework using active thermography that uses a custom multichannel neural network approach to perform classification between samples and regression toward their thermal properties. This neural network structure is uniquely suited for effective processing of thermographic data. The framework is realized, implemented, and evaluated with a set of ten samples with diverse thermal/physical properties. Experimental results on a realization of the framework validate this approach, with a 92.20% classification accuracy using multichannel neural network with majority vote, as well as more than 99.6% R2-fit with respect to three different thermal properties, namely thermal conductivity, thermal diffusivity, and thermal effusivity (which is very useful to define thermal exchange during physical interaction).