4 4.5. Results 83 FT-MOEA-CM-All FT-MOEA FT-MOEA-CM-Best MPPS ddFT COVID-19 TS1 GPT12BE GPT15BE MPPS ddFT COVID-19 TS1 GPT12BE GPT15BE 0.00 0.04 0.08 0.12 Accuracy (a) (b) 15 20 25 30 FTSize Figure 4.3: Evaluating robustness for all case studies and algorithms. In (a), based on accuracy. In (b) based on FT size. to the potential existence of multiple optimal FT structures for the same failure dataset. Discussion. Robustness is evaluated by examining the variance in box plots for accuracy and FT size, where a variance close to zero suggests higher robustness. In Figure 4.3(a), accuracy results for all case studies are presented using box plots to compare FT-MOEA-CM-All, FT-MOEA-CM-Best, and FT-MOEA. An accuracy of 0 is optimal. It is observed that FT-MOEA-CM’s configurations consistently achieved an Accuracy of 0.0 with zero variance, indicating they consistently reach the global optima across all case studies. In contrast, FT-MOEA failed to achieve the global optima in the COVID-19 and GPT15BE case studies. Regarding FT size, Figure 4.3(b) shows that results vary significantly among algorithms. For instance, in the TS1 case study, all algorithms yield varying FT sizes. For GPT12BE, FT-MOEA showed greater robustness compared to FT-MOEA-CM-All and FT-MOEA-CM-Best, whereas the opposite holds for COVID-19. Conclusion. The results indicate that in terms of Accuracy, FT-MOEA-CM-All and FT-MOEA-CM-Best are more consistent compared to FT-MOEA, achieving the global optima for all case studies. However, regarding FT Size and the related FT structure, the consistency of the results varies. Scalability. The scalability analysis assesses the algorithms’ e"ciency in managing more complex case studies. The complexity of an FT is linked to its number of elements (i.e., BEs and logic gates) and MCSs, indicating that a more complex FT represents a more challenging inference process. Discussion. Table 4.2 presents the FT Size and the number of MCSs per case study, arranged from the one with the fewest BEs to that with the most. The
RkJQdWJsaXNoZXIy MjY0ODMw