Validation of a multilayer perceptron for rapid, direct solution of the electrical impedance tomography inverse problem

Electrical impedance tomography is a method of mapping the conductivity distribution of a domain. For decades it has been considered a potential in situ nondestructive evaluation technique for characterization of conductivity changes in aerospace composites. Yet, several challenges must be addressed before this technique can be transitioned from the laboratory to meaningful practice; for example, the expense of the inverse problem that must be solved to estimate conductivity. An alternative is to characterize damage from the measured voltage-current relationship using deep learning. In this work, we develop and test a deep learning algorithm to characterize time-independent damage events in complex geometry.