PSI - Issue 84

612 Sergio Belluco et al. / Procedia Structural Integrity 84 (2026) 607–614 where the only new coefficient is α 3 , taken as α 3 = 1.5 in case of fatigue verifications (and α 3 = 1.00 otherwise).

Fig. 4. Idealized tendon stress profile at release and member failure according to fib MC20 and 2 nd gen EC2. Notes: the two codes provide two values (distinguished by the apex 1 and 2) to be used in different design situations. 3.2. Uncertainty assessment of code formulations Similarly to the transmission length, also for the anchorage length an extensive database has been collected. Currently, direct measures of anchorage length are not experimentally feasible, and the most common approach to indirectly measure is to perform multiple three- or four-point bending tests on simply supported members having the same characteristics. This type of test does not ensure the exact calculation of the anchorage length, since it is estimated iteratively beam by beam. In detail, if the test results in a combined flexural/bond failure of the specimen (that is, the member exhibits some end-slippage of the tendons when it has reached its maximum flexural capacity), the embedment length of the tendons is identified as a fully-observed experimental realization of the anchorage length , while if a bond (or a flexural) failure occurs, the tested length represents a right-censored (or left-censored) realization of the actual anchorage length, which means a lower (or upper) bound of . Further details on this test method can be found in Belluco and Faleschini 2025. After examining 22 experimental campaigns, 405 experimental observations were collected, out of which a total of 303 observations were classified as censored (44 right- censored, 222 left-censored, and 37 interval-censored). Then, two filtering operations were performed to obtain an homogeneous database (Belluco and Faleschini 2025), consisting of of 224 experimental observations. Fig. 5 shows the distribution of the main variables in their sample space.

Fig. 5. Distribution of the main variables in the database of anchorage lengths (adapted from Belluco and Faleschini 2025). A sample of the model uncertainty was calculated as the ratio of each of the 224 experimental embedment lengths le and their corresponding anchorage length prediction. Then, depending on the registered failure mode, each has been classified as fully-observed (37 observations), right- (17 obs.), left- (153 obs.), or interval-censored (17 obs.). After verifying the independence of Θ against each variable, the distributions parameters are computed with the Maximum Likelihood Estimation Methos for considering all the information provided by the censored realizations of Θ. Results presented in Tabl e 3 show that for both types of release (and for both codes), the coefficient of variation is quite similar, while comparing [ Θ ] indicates that significantly longer experimental lengths are obtained when the strands are suddenly released. Moreover, it is stressed that for ULS, longer lengths imply more onerous verifications, and therefore, [ Θ ] below unity denote model conservativeness.