204 Appendix B. Appendixes: FT-MOEA Table B.1: (Part I) Data-driven approaches for the automatic inference of FT models. In the columns the references, algorithm name, whether the algorithm is publicly available or not, the key aspects of the methodology, the input data, the benefits and drawbacks. Reference(s) Name Aval? Methodology Input Benefits Drawbacks M. G. Madden and P. J. Nolan, 1994; M. G. Madden, 1970; M. Madden and P. Nolan, 1999 IFT No Based on the ID3 algorithm Quinlan, 1986 to induce Decision Trees (DTs). TS Purely data-driven approach. Provides insights to carry out rules-based diagnostics. It is unclear whether the IFT algorithm guarantees the encoding of cause-e!ect relationships, which is a requirement for FT models. Berikov, 2004 - No They build FT models based on DTs. TS They build the FT model in layers from the DT, this is an interesting way to exploit DTs capabilities. It is di"cult to determine automatically the non-terminal nodes of the FT without expert advice. Mukherjee and Chakraborty, 2007 - No Based on text mining and natural language processing. Text Makes use of data such as maintenance reports. It requires a manually built (partial) FT, which is then refined through their method. Roth, Wolf, and Lindemann, 2015 - No Based on various matrix analysis methods commonly used to model dependencies. SDR The approach helps to identify critical failures or elements from a structural point of view. Also enables handling alternative input data. It requires functional analysis and expert domain support. Nauta, Bucur, and Stoelinga, 2018 LIFT Yes Based on the MantelHaenszel statistical test. BD It works well for small problems and with low noise levels in the data. It requires as input information about the intermediate events. Moreover, the algorithm does an exhaustive search, and thus has exponential time complexity. Waghen and Ouali, 2019 ILTA No Knowledge Discovery in Dataset (KDD) + Interpretable Logic Trees (ILT) TS It does not require human expertise in the construction stage. It generates intuitive logic trees structures that encode hidden system’s causal relations. The ILTA algorithm is insufficient when several subsystems and characteristic variables may have interdependent relationships with each other. Abbreviations: Decision Trees (DTs), Time Series (TS), Binary Data (BD), System Domain and Relations (SDR).
RkJQdWJsaXNoZXIy MjY0ODMw