Predicting strain and stress fields in self-sensing nanocomposites using deep learned electrical tomography

Conductive nanocomposites, enabled by their piezoresistivity, have emerged as a new instrument in structural health monitoring. To this end, studies have recently found that electrical resistance tomography (ERT), a non-destructive conductivity imaging technique, can be utilized with piezoresistive nanocomposites to detect and localize damage. Furthermore, by incorporating complementary optimization protocols, the mechanical state of the nanocomposites can also be determined. In many cases, however, such approaches may be associated with high computational cost. To address this, we develop deep learned frameworks using neural networks to directly predict strain and stress distributions -- thereby bypassing the need to solve the ERT inverse problem or execute an optimization protocol to assess mechanical state. The feasibility of the learned frameworks is validated using simulated and experimental data considering a carbon nanofiber plate in tension. Results show that the learned frameworks are capable of directly and reliably predicting strain and stress distributions based on ERT voltage measurements.