86 Chapter 4. Fault Tree inference using Multi-Objective Evolutionary Algorithms and Confusion Matrix-based metrics FT-MOEA FT-MOEA-CM-All FT-MOEA-CM-Best 0 20 40 60 Generation 0.0 0.2 0.4 0.6 0.8 Time per gen. (min) ddFT 0 10 20 30 40 Generation 0.0 0.2 0.4 0.6 MPPS 0 25 50 75 100 Generation 0.0 0.8 1.6 2.4 3.2 COVID-19 0 20 40 60 80 Generation 0.0 0.4 0.8 1.2 Time per gen. (min) TS1 0 25 50 75 100 Generation 0 4 8 12 GPT12BE 0 15 30 45 60 Generation 0 4 8 12 GPT15BE Figure 4.7: Computational time per generation: Convergence for all case studies and algorithms based on computational time per generation. Conclusion. The results in terms of convergence speed are less conclusive; in some cases, FT-MOEA-CM outperforms FT-MOEA, but in others, it is rather similar or even slower. However, notice FT-MOEA-CM always achieved global optima. Comparing FT-MOEA-CM-All and FT-MOEA-CM-Best. Both configurations yield the global optima in terms of robustness, yet exhibited less consistency in the TS1 and GPT12BE case studies. In terms of scalability, the two setups are on par. Their convergence speed is also similar, except in the MPPS case study, where FT-MOEA-CM-Best consistently outperformed. Importantly, FT-MOEA-CM-Best used only the top 7 features (identified through PCA in Section 4.5.1), compared to FT-MOEA-CM-All, which utilised all 17 features. This suggests that FT-MOEA-CM-Best provides a more e"cient setup for FT-MOEA-CM. 4.5.3 FT-MOEA-CM’s Features: parallelisation and caching We evaluate the e!ects of parallelisation and caching onFT-MOEA-CM-Best. Caching stores intermediate results to avoid redundant computations, thereby enhancing e"ciency. Parallelisation employs multiple processors to perform tasks concurrently, thus decreasing execution time by distributing the workload. According to Figure 4.8(a), caching benefits all case studies, excluding MPPS, by enabling quicker convergence. From Figure 4.8(b), parallelisation is shown to improve convergence speed by approximately 45% for larger case studies, namely
RkJQdWJsaXNoZXIy MjY0ODMw