Analysis of SATIR test for the qualification of high heat flux components: defect detection and classification by signal-to-noise ratio maximization

Abstract

Plasma facing components (PFC) in Tore Supra and W7X adopt the flat tile concept using carbon fibre composite (CFC) material for the plasma facing material. As the cooling structure is made of a copper alloy material (CuCrZr), the bonding technique between CFC tiles and CuCrZr is critical. Currently, a soft metallic compliant layer is interposed between the two; in such a way the significant thermal expansion mismatch between carbon and copper can be accomodated. The development of a reliable non-destructive inspection technique (NDT) for the bond, to be performed during the manufacturing process, is obviously of great importance. The SATIR (infrared thermography) test bed operating at Commisariat à l'Energie Atomique (CEA) Cadarache performs this function using transient infrared thermography: the thermal excitation is realized in the cooling channel and the presence of a faulty tile is detected in the form of a delayed thermal response. With this technique, the evolution of the surface temperature of an inspected element was compared to that of a defined free-defect element, using the so-called DTref criterion (maximum of the transient temperature difference). The defect detection capability of the SATIR test bed can be improved using signal processing methods. A first treatment based on spatial image autocorrelation allows a better localization of the bond defect. Moreover, the problem of detection and classification of random signals (like the thin defect signature) can be solved maximizing the signal-to-noise ratio (SNR). Two filters maximizing this ratio were optimized: the stochastic matched filter (SMF) aims at defect detection, while the constrained SMF aims at defect classification. These methods assume that the second-order properties of the process at play are known, through covariance matrices. All these methods process the SATIR signal utilizing any free-defect element as reference signal. The tile temperature signal is either processed by itself or compared to a numerical temperature evolution evaluated by finite element calculations. In view of ITER PFCs qualification, the possibility to extend all these methods from the flat tile concept to the monobloc target element concept is proved.