PSI - Issue 57

440 Lorenzo Bercelli et al. / Procedia Structural Integrity 57 (2024) 437–444 4 L. Bercelli, C. Guellec, B. Levieil, C. Doudard, F. Bridier, S. Calloch / Structural Integrity Procedia 00 (2019) 000 – 000 Fig. 3.a). Following this observation, the estimation of the amplitude of the second harmonic 2 appears as a better suited post-processing technique of the infrared data to allow for the monitoring of cracks. If we are to look at the field 2 for the welded joint loaded at =10 , at the same stage as shown in Fig. 2.d, it is possible to assess the size and position of cracks, where the magnitude of 2 is the highest (Fig. 3.b).

(a)

(b)

(c) (d) Fig. 2. Fields of the temperature’s amplitude of the first harmonic 1 for as-welded T-joints: loaded at =0.1 with Δ / =0.29 before (a) and after (b) initiation of cracks, and loaded at =10 with Δ / =0.86 before (c) and after (d) initiation of cracks.

(a) (b) Fig. 3. Illustration of temperaturesignals taken from pixels in a crack and in the uncracked weld toe (a) and field of the second ha rmonic’s amplitude 2 (b) for a T-joint loaded at =10 with Δ / =0.86 . Following this post-processing, it is found that the number of cycles to detection of a crack represents in average only 30% of the number of cycles to a through-width crack . In other words, as cracks are only detected at a size of about 1 , the phase of crack initiation is negligible relative to the totalfatigue life of the welded T-joints. This motivates the use of a predictive fatigue model based on the crack propagation kinetics as assessed by TSA. From this observation, the propagation of cracks throughout the fatigue test is monitored for all loading configurations, through image processing of 1 field maps when there is no crack closure (i.e. =0.1 ) or of 2 field maps when crack closure occurs (i.e. =10 ). From this monitoring, it is possible to link the fatigue cracks size 2 at the surface to the number of cycles (Fig. 4.a). It is observed that for millimetric cracks, before coalescence (for ∈ [1:10] , approximatively), the propagation speed is constant. As a result, it is possible to plot the propagation speed with respect to the normalized nominal stress range Δ (Fig. 4.b): two distinct tendencies are revealed in this graph, where cracks at = 10 seem to propagate slower than at = 0.1 for the same nominal