PSI - Issue 41

Mohamed Amine Belyamna et al. / Procedia Structural Integrity 41 (2022) 372–383 Mohamed Amine Belyamna et al. /Structural Integrity Procedia 00 (2022) 000–000

380

9

4.3. Predicted Leak Probability at End of Life versus Operating Parameters

The reliability for a large number of welds and fittings in a piping system can be estimated quickly if the results of detailed MCS and ANN are provided in a structured parametric format. The IG SCC D  presented in equation (2) is used to generalize the results of PFM calculations. 297 values of IG SCC D  for various degrees of sensitization, different levels of applied stress (both the applied service-induced pressure, thermal and residual stresses), conductivity equal to 0.2 µS/cm, different steady state temperature and different O 2 content are presented in Figure 5. The parametric calculations as presented below consisted of many actual M-PRAISE runs using ANN that covered a range of leak probabilities from 1.0E-04 to 4.0E-01. It was believed that IG SCC D  could serve as a suitable parameter to summarize results for calculated failure probabilities of stainless steel piping. Results presented later in this section show a good correlation between 40-year cumulative leak probabilities and IG SCC D  . This parameter does provide a useful basis to generalize results for piping-leak probabilities (Figure 6). High correlation coefficient would indicate a good prediction capacity of the model. Figure 7 shows the results of training, test and validation for 297 data sets. These results would confirm, and therefore, validate the learning and generalization capacity of the ANN model. The models used in the reliability analysis of IG-SCC pipes usually involve many different types of parameters, including geometry, property, and operating parameters. A large number of physical parameters could increase the number of input neurons of ANN for further reliability estimation as well as complex the ANN structure, which could reduce the computational efficiency. To guarantee the efficiency and accuracy of the ANN, the number of input neurons should be limited. However, an arbitrary reduction of the number of the parameters needed in burst pressure prediction could impact the accuracy of ANN prediction. To realize a reliable structural simplification, sensitivity analysis could be employed to determine which parameters could be neglected.

Fig. 5. IG SCC D  Values for various degrees of sensitization, different levels of applied stress, a conductivity equal to 0.2 µS/cm. (a) temperature equal to 288°C O 2 content equal to 8ppm; (b) temperature equal to 288°C, O 2 content equal to 0.2ppm; (c) temperature equal to 288°C, O 2 content equal to 2ppm; (d) temperature equal to 140°C, O 2 content equal to 0.2ppm.