Data-Driven Contact-Based Thermosensation for Enhanced Tactile Recognition

Abstract

Thermal feedback plays an important role in tactile perception, greatly influencing fields such as autonomous robot systems and virtual reality. The further development of intelligent systems demands enhanced thermosensation, such as the measurement of thermal properties of objects to aid in more accurate system perception. However, this continues to present certain challenges in contact-based scenarios. For this reason, this study innovates by using the concept of semi-infinite equivalence to design a thermosensation system. A discrete transient heat transfer model was established. Subsequently, a data-driven method was introduced, integrating the developed model with a back propagation (BP) neural network containing dual hidden layers, to facilitate accurate calculation for contact materials. The network was trained using the thermophysical data of 67 types of materials generated by the heat transfer model. An experimental setup, employing flexible thin-film devices, was constructed to measure three solid materials under various heating conditions. Results indicated that measurement errors stayed within 10% for thermal conductivity and 20% for thermal diffusion. This approach not only enables quick, quantitative calculation and identification of contact materials but also simplifies the measurement process by eliminating the need for initial temperature adjustments, and minimizing errors due to model complexity.