B.5. Varying parent Fault Trees (details) 211 0 10 20 30 40 50 60 Generation 0 0.1 0.2 0.3 0 10 20 30 40 50 60 Generation 0 0.1 0.2 0.3 Setup A Setup B Setup C 0 10 20 30 40 50 60 Generation 0 10 20 30 40 50 0 10 20 30 40 50 60 Generation 0 2000 4000 6000 Conv. time (min) (a) (b) (c) (d) d c s Figure B.5: Influence of varying the parent FTs on (a) εd, (b) εd, (c) εs (green dashed line is the ground truth FT size), (d) convergence time. For the m.o.f sdc, the case study MPPS (ps =400, ng =100, uc =20, ϑ =6). Setup A: as described in Section 2.6.3; Setup B: using as parent FT the disjunctive normal form; Setup C: using as parent FT a sub-optimal FT of ςs =98 obtained with the m.o.f. d in Figure 2.9(d). B.5 Varying parent Fault Trees (details) We defined in Section 2.6.3 what a parent FT is, and here we evaluate the e!ects of varying this parameter. We use the case study MPPS, the m.o.f. sdc with the following parameters ps = 400, ng = 100, uc = 20, ε = 6. We consider three setups. Setup Ais our reference, we have been using it throughout this paper (see Section 2.6.3). Setup B consists of a single parent FT based on the Disjunctive Normal Form. Setup C takes an inferred sub-optimal FT model with ωs =98 previously obtained using the m.o.f. d in Figure 2.9(a). We already discussed in detail the results of Setup A in Figure 2.6, Section 2.7.3. Figure B.5(a) and (b) shows that Setup B has an error of zero since the onset (i.e., ωd =ωd =0.0). This is expected because Setup B considers as parent FT the Disjunctive Normal Form (DNF). However, this does not mean the parent FT has the optimal structure, this is what we observe in Figure B.5(c) where around the 30th generation the FT-MOEA found a smaller structure with the same performance. The latter structure is the same as the one where Setup A converged.
RkJQdWJsaXNoZXIy MjY0ODMw