Neural network approach to modelling the behaviour of quantum tunnelling composites asmultifunctional sensors

Abstract

Quantum tunnelling composites, or ‘QTCs’, are composites with an elastomeric polymermatrix and a metal particle filling (usually nickel). At rest, these metal particles do nottouch each other and the polymer acts as an insulator. When the material is suitablydeformed, however, the particles come together (without actually touching) and thequantum tunnelling effect is promoted, which causes the electrical resistance to falldrastically. This paper contains a detailed description of neural networks for a faster,simpler and more accurate modelling and simulation of QTC behaviour thatis based on properly training these neural models with the help of data fromcharacterization tests. Instead of using analytical equations that integrate differentquantum and thermomechanical effects, neural networks are used here due to the notablenonlinearity of the aforementioned effects, which involve developing analytical modelsthat are too complex to be of practical use. By conducting tests under differentpressures and temperatures that encompass a wide range of operating conditionsfor these materials, different neural networks are trained and compared as thenumber of neurons is increased. The results of these tests have also enabled certainpreviously described phenomena to be simulated with more accuracy, especiallythose involving the response of QTCs to changes in pressure and temperature.