A data-driven FMEA framework integrating median aggregation, entropy weighting and jenks classification for marine gas turbine risk assessment

This study develops a data-driven FMEA framework for marine gas turbine risk assessment using 48 failure modes grouped under four subsystems. Expert ratings are collected on bounded 1–10 occurrence (O), severity (S), and detection (D) scales and aggregated by the median as a robust central-tendency operator; Shapiro–Wilk results are retained as within-expert distribution diagnostics rather than formal tests for individual failure modes. Shannon entropy quantifies criterion-specific information dispersion, and risk is calculated using a weighted-geometric WRPN formulation. A common four-class Jenks Natural Breaks structure is selected as the smallest k for which all transformed components and WRPN achieve GVF ≥ 0.90. Detection receives the largest entropy-derived weight (wD = 0.456), interpreted as greater discriminatory information rather than intrinsic safety importance. WRPN values range from 1.682 to 5.936, with thresholds at 2.931, 3.901, and 4.889. Compared with classical RPN, WRPN remains strongly associated but differs in rank order (Spearman ρ = 0.957); ties decrease from eight groups involving 18 failure modes to three groups involving six. Weight perturbations retain high rank correlations (ρ = 0.989–1.000), while class changes concentrate near data-driven boundaries. The framework supports risk prioritisation, inspection planning, detection improvement, and maintenance management in marine gas turbine systems.