QUANTITATIVE RISK ASSESSMENT OF RAIL INFRASTRUCTURE PROJECTS USING MONTE CARLO SIMULATION AND FUZZY LOGIC

Abstract

Rail infrastructure programs frequently face complex, intertwined risks spanning cost overruns, schedule delays, safety concerns, and interface uncertainties. This systematic review critically examines how quantitative risk assessment (QRA) methods—Monte Carlo simulation (MCS), fuzzy logic (FL), and hybrid approaches—have been employed across the rail project lifecycle to manage these multidimensional challenges. Following PRISMA guidelines, we conducted a comprehensive search across Scopus, Web of Science, IEEE Xplore, ASCE Library, ScienceDirect, and TRB databases, applied pre-registered eligibility criteria, implemented double-screening for study inclusion, and rigorously appraised methodological practices encompassing data provenance, dependence modeling, validation, and sensitivity analysis. From an initial pool of studies, 95 peer-reviewed publications met all inclusion standards. Findings indicate MCS dominates (47%) owing to its strength in producing distributional forecasts and governance-ready percentiles; FL supports imprecise or linguistic inputs (33%) often encountered during early-stage planning and safety screening; while hybrid models (20%) bridge probabilistic propagation and evidential uncertainty, particularly in interface-intensive phases. Applications cluster within construction (68%), followed by design (46%), feasibility analysis (39%), testing/commissioning (24%), and operations & maintenance (21%). Methodologically, MCS studies primarily use triangular and PERT/beta distributions, with approximately 42% employing Latin hypercube sampling. However, dependence modeling remains limited—38% of studies assume independence, 23% employ rank or copula methods, and only 31% jointly simulate cost–schedule interactions. FL studies typically apply triangular/trapezoidal membership functions with centroid defuzzification; while two-thirds disclose reproducible rule bases, one-third lack transparency. Hybrid models frequently convert fuzzy assessments into probabilistic inputs or embed fuzzy and evidential nodes into Bayesian structures, enabling richer risk representation at system interfaces. Sensitivity analysis is reported in 64% of studies, but only 21% adopt global approaches and a mere 5% include tail-focused diagnostics, while external validation is rare (18%).