A new inverse method for real time eddy current testing is introduced. It is based on particle swarm optimization coupled with a metamodel. This metamodel is generated by interpolating data from an adaptive database. Linear and radial basis function interpolations are compared as means to generate the metamodel. The inverse method gives a result combined with an approximation of the likelihood function to help for decision analysis. It is finally compared to a state-of-art inverse method using support vector machine on a 3D case on simulated data with very promising results.