203 Appendix B Appendixes: FT-MOEA B.1 Data-driven methods to infer FTs from data Table B.1 (divided in two parts) contain references associated to data-driven algorithms to infer FTs, the name of the algorithm (if any), whether it is publicly available (if ‘yes’ the table provides a hyperlink redirecting to the respective online repository), the key aspects of the methodology, the input data, the benefits, and drawbacks. B.2 Applying NSGA-II and Crowding-Distance to infer FTs B.2.1 Applying NSGA-II to infer FTs We provide a conceptual visualisation in Figure B.1 that explains our implementation of the NSGA-II and Crowding-Distance in the context of the automatic inference of FTs. To ease the visualisation, we consider the bi-dimensional case where the multi-objective function is sd (see Section 2.6.5, Table 2.4). After computing the metrics for a population of FTs within a given generation, one can depict the FTs with circles as in Figure B.1(a). The output of the NSGA-II algorithm is a set of Pareto fronts, represented in Figure B.1(b) with di!erent colours (red, blue, green and purple). Figure B.1(c) shows some details related to the FTs in the first front (red). Note that these FTs have di!erent structures. Here the top FTs has a higher error based on the failure dataset (ωd) compared to the others, but it is the smallest FTs in the first front. On the contrary, the bottom FTs has a smaller error in ωd, but with the trade-o! of having a larger size.
RkJQdWJsaXNoZXIy MjY0ODMw