Rapid, Nonlinear System Identification for NDT, Using Sensor Response Databases

For many systems, accurately estimating multiple parameters or variables of interest from multiple sensor responses requires models that account for nonlinear system behavior. Such multivariate estimation is typically accomplished using iterative methods or local linearity assumptions in the vicinity of a nominal operating point. Here, an alternative approach is described in the context of nondestructive testing (NDT) where databases of precomputed sensor responses are used with error minimization methods to efficiently perform multiple unknown parameter estimations (Goldfine and Melcher, 1995; Goldfine and Melcher, 2002; Melcher, 1991). Databases are generated from "forward" models for the sensor response, requiring the sensor to be accurately modeled over the operating range of interest. The entire nonlinear response over a wide range of material or geometric properties as well as sensor design/operating conditions is captured and used in the "inverse" method. The sensor must be designed so that difficult-to-model (and other unmodeled) contributions to the sensor response are minimized. Specific applications for properly designed meandering winding magnetometer eddy current sensors and interdigitated electrode dielectrometer sensors are described. These sensors were specifically designed to minimize unmodeled contributions.