How?

The SIRENA Project requires the development of inversion tools operating in real‐time over large data sets describing measurements take on the Structures Under Test (SUTs). As a consequence, suitable meta‐models (or surrogate models) that describe the status/behaviour of the SUTs with a reduced set of descriptors are necessary.

Towards this end, LBE inversion techniques (Fig. 1) will be investigated and adapted to address the NDT‐NDE problem defined within the activities of the SIRENA project. 


Fig. 1 - Example of prediction made by a Kriging surrogate model (1-D case)

Such a choice is motivated by the following features of LBE methods:



Fig. 2 - CIVA model of a metallic cylindrical structure inspected by eddy current testing (ECT)

  1. they can be used to solve problems when closed‐form relationships between the input/data and the output/solution spaces are either not available or very difficult/expensive to determine;
  2. they are suitable tools for both identifying the "nature" of the solution (i.e., LBE classification) and/or predicting the set of solution descriptors (i.e., LBE regression) given a reduced number of available I/O samples, thanks to their well‐known generalization capabilities;
  3. they enable accurate and almost real‐time prediction performances during the (on‐line) test phase, once representative I/O training databases have been provided during the (off‐line) training phase.

As for the generation of the training databases for the LBE techniques, the cooperation with CEA‐LIST will allow to fruitfully exploit the potentialities of the CIVA platform in modelling and solving forward problems concerned with various configurations of the SUTs (Fig. 2).

The use of efficient numerical meta‐models to replace accurate but time‐consuming numerical forward solvers (e.g., CIVA simulator) will be considered, as well.

On the other hand, NDT‐NDE problems will be addressed by analyzing and developing suitable inversion techniques based on Compressive Sensing (CS) (Fig. 3), whose main advantages are:

  1. a low‐complexity and computationally‐efficient algorithmic implementation;
  2. an enhanced inversion robustness against noisy input data;
  3. the possibility to provide an estimate of the confidence level/reliability of the retrieved quantities.


Fig. 3 - NDT-NDE reconstruction obtained by means of compressive sensing


Fig. 4 - Possible experimental setup of interest
for the SIRENA validation

The SIRENA project will be then aimed at adapting the CS‐based approaches developed by the ELEDIA Research Center for solving inverse problems in a wide range of heterogeneous applications to the NDT-NDE framework. In order to achieve such a goal, suitable re‐formulations of the considered NDT-NDE problems will be considered.

The performances, potentialities, as well as the limitations of both LBE and CS‐based techniques developed within the activities of the SIRENA project will be accurately assessed through numerical and experimental (laboratory-ontrolled) validations (Fig. 4).