Data-derived SEA for Time Domain EMI Sensing of UXO

Electromagnetic induction (EMI) is a prominent technique in Unexploded Ord- nance (UXO) detection and discrimination research. Existing idealized forward models for the EMI response can be defeated by both the material and geometrical heterogeneity of realis- tic UXO. We have developed a new, physically complete modeling system referred to as the Standardized Excitations Approach (SEA). The SEA accounts for all the efiects from these het- erogeneities including their interactions within the object, and is applicable in both the near and far flelds. According to the SEA, the excitation fleld is decomposed into fundamental modes, and the response of a given target to each fundamental mode (denoted as a fundamental solution) is obtained beforehand and saved in a library. In this way, the target response to an arbitrary excitation fleld can be calculated via a simple superposition of these fundamental solutions. The model parameters (i.e., the fundamental solutions) of a given object are extracted from a su-ciently detailed set of measurement data. These parameters will be speciflc to each EMI instrument. The parameter extraction process was developed previously for the frequency domain using the GEM-3 EMI instrument. In this paper, we apply this SEA to time domain using the EM-63 instrument as an example. The receiver coil of the EM63 is a 0.5m by 0.5m square loop and can not be approximated by a point receiver. Therefore, in the model, the data is interpreted as the integration of the secondary fleld over the receiver loop. The objects we consider are all Body of Revolution (BOR) type objects. We exploit the fact that the calculated SEA model parameters also exhibit speciflc behavior because the target is a BOR Thus, the algorithm is improved by enforcing symmetric properties and zero total magnetic charge, which makes the algorithm more robust and more e-cient. Preliminary results show that this approach works well for this time domain EMI instrument. After optimization, this model may be fast enough for implementation in inversion processing algorithms.