Dynamic Fault Tree Construction Through Minimal Cut Sequences And Causal Discovery Algorithm

Ensuring system reliability and availability is critical for maintaining operational continuity and supply chain stability in industrial systems, where unexpected downtimes can lead to significant production losses and cascading disruptions. Fault tree analysis is a well-established reliability engineering method for identifying system vulnerabilities and evaluating failure risks. While traditional fault trees are static, Dynamic Fault Trees (DFTs) extend this framework by modeling failure dependencies and sequences, which are essential for representing complex systems with redundancy and interdependent components. However, the manual construction of DFTs remains a major limitation due to its complexity, time consumption, and susceptibility to human error.
This paper proposes a data-driven framework for the automatic construction of DFTs based on the learning of Minimal Cut Sequences (MCSQs) from failure-time data. The approach extracts MCSQs and groups them according to their constituent events. To characterize temporal dependencies, the concepts of Positions of Elements and Relative Positions are introduced to qualitatively capture sequence relationships among failures. These constructs enable the inference of both static and dynamic gates that collectively represent the complete set of MCSQs in a DFT. In addition, a causal discovery algorithm is incorporated to identify causal relationships that cannot be revealed through qualitative analysis alone. The proposed framework facilitates scalable and automated DFT construction, thereby enhancing the applicability of fault tree analysis to complex, data-rich industrial systems.

Keywords: Dynamic Fault Tree Construction, Minimal Cut Sequences, Causal Discovery Algorithm, Data-Driven Approach.