IDLIFT: Learning Dynamic Fault Trees from Interval-Based Temporal Event Tables

Dynamic Fault Trees (DFTs) augment Static Fault Trees (SFTs) with temporal gates such as Priority AND (PAND) and Sequential Enforcing (SEQ) to capture order-dependent failure logic. Existing data-driven DFT learning methods, such as Dynamic LIFT (DLIFT), rely on Temporal Truth Tables (TTTs), which record only first-event order and cannot represent event duration or co-activity. This paper introduces the Interval-based Temporal Event Table (ITET), which stores event intervals, and Interval-based DLIFT (IDLIFT), which infers OR, AND, PAND, and SEQ gates from interval overlap relations using a statistical association test and a non-greedy search over all candidate subsets. On a benchmark of 18 reference DFT structures, IDLIFT reaches a mean gate recall of 0.58 across nine parameter configurations and 0.70 at the strictest Confidence Level (CL), versus 0.36 and 0.47 for DLIFT. On the NASA Milling Wear dataset, the learned structures differ at one gate, judged plausible for IDLIFT and inconsistent for DLIFT by a domain expert. On held-out runs, the IDLIFT structure predicts tool wear with accuracies of 0.79 at CL=0.90 and 0.78 at CL=0.80, versus 0.76 and 0.52 for DLIFT and 0.55 for a majority baseline. These results show that interval-based IDLIFT improves DFT structure learning.