aspg-1-4529-1793910220 20261105202340 American Scientific Publishing Group admin@americaspg.com American Scientific Publishing Group American Journal of Business and Operations Research AJBOR 2770-0216 2692-2967 2027 15 1 Samir Aboul Fotouh Saleh Mansoura University, Mansoura, Egypt A. S. Tolba Mansoura University, Mansoura, Egypt Shimaa Sami Selim Obour University for Science & Technology (OUST), Egypt MohamedElhoseny College of Computing and Informatics, University of Sharjah, United Arab Emirates Maintenance-economic studies often convert condition indicators into apparent costs without verifying whether the resulting targets are independent of the model inputs. This can produce exceptionally high but operationally misleading accuracy. We formulate component-level estimation for a COmbined Diesel-eLectric And Gas (CODLAG) propulsion plant as a leakage-resistant digital-shadow problem. The primary dataset is the public UCI naval propulsion dataset: 11,934 steady-state observations generated by a calibrated numerical frigate simulator across nine speed settings, 51 compressor-health states, and 26 turbine-health states. Because it contains neither monetary maintenance records nor live vessel telemetry, the outputs are explicitly defined as dimensionless compressor and turbine cost-risk indices rather than currency costs. Health coefficients and every target-derived quantity are excluded from predictors. Models are tuned by three-fold grouped inner validation and assessed by five-fold outer validation in which all observations sharing a degradation pair remain in the same fold. Extra Trees achieved an aggregate cost-risk RMSE of 0.0222 (95% CI half-width 0.0049) and R2 of 0.9886 (95% CI half-width 0.0049) under nominal interpolation. However, combined noise, missingness, calibration bias, and channel loss reduced R2 to 0.6335 even after corruption-augmented training, while leave-one-speed-regime-out testing produced mean R2 = −0.379. Scenario-matched 90% split-conformal intervals retained 89.5–90.4% empirical coverage by widening from 0.032 nominally to 0.406 under severe corruption. An external NASA C-MAPSS FD001 benchmark yielded RMSE 16.97 cycles and R2 = 0.821 for remaining-useful-life prediction on 100 unseen engines. The study therefore supports robust condition-risk ranking within the simulated envelope, not direct monetary forecasting or autonomous maintenance scheduling. The principal contribution is a transparent validation protocol that exposes target leakage, simulator determinism, sensor-quality sensitivity, and the boundary between a digital shadow and a deployable digital twin. 2027 50 61 4529 https://creativecommons.org/licenses/by/4.0/ 10.54216/AJBOR.150107 https://www.americaspg.com/journal/1/article/4529 https://www.americaspg.com/storage/articles/manuscripts/01M5TEGE5GW3KNB5AP96VF476S.pdf