Predicting the Remaining Useful Life (RUL) of railway wheels is challenging because wheel–rail degradation is cumulative, stochastic, and influenced by operating conditions. This study evaluates a multi-indicator prognostic framework using real in-service measurements acquired with a CALIPRI C42 optical profilometer (NextSense GmbH, Graz, Austria). The database comprises 80 wheels from ten vehicles of the same rolling-stock type, monitored during five monthly measurement campaigns, and includes flange width (Fw), flange height (Fh), and the flange-gradient dimension (qR). The Gamma process and first-passage formulation are established tools; the contribution of this work is their common application to all three indicators on the same in-service fleet and the benchmarking of long-horizon probabilistic results against an AR(1) short-term predictor embedded in the First-Passage Auto-Regressive (FP-AR) framework using the same dataset. Median Gamma-based RUL values were 21.2–22.0 months for Fw, 13.7–17.6 months for Fh, and 5.7–9.5 months for qR, with qR showing the largest relative percentile dispersion. For one-step prediction, the FP-AR benchmark achieved global MAE/RMSE values of approximately 0.368/0.502 mm for Fw and 0.0187/0.0216 mm for Fh; qR was more difficult to predict, with global MAE/RMSE values of approximately 0.575/0.991 mm. Under the adopted intervention thresholds, these results identify qR as the most variable and operationally constraining indicator under the studied Fès–Marrakech service conditions. The proposed dual-model analysis therefore provides a position-specific, uncertainty-aware basis for comparing wheel-profile degradation indicators, while its maintenance implications remain fleet- and route-specific pending validation in additional operating contexts.
Predictive maintenance of railway wheelsets requires remaining useful life (RUL) estimates and treatment of model limitations. This study develops a Gamma-process-informed XGBoost surrogate for an ONCF fleet of 444 vehicles, 3552 candidate wheel positions, and 14 inspection campaigns. Flange width (Fw), flange height (Fh), and flange steepness (qR) were measured using a CALIPRI C42 gauge. After date normalisation and series-level IQR filtering, 3296 candidate series remained per indicator; Gamma eligibility retained 2366 Fw, 1628 Fh, and 2445 qR series. RUL quantiles were evaluated from the discrete Gamma first-passage distribution, with Monte Carlo used only for convergence checking. Vehicle-level median RUL was 13 months for Fw, 3 months for Fh, and 7 months for qR. RUL below six months occurred for 25/402, 283/344, and 100/404 vehicles, respectively. Under vehicle-grouped five-fold cross-validation, surrogate-label reconstruction yielded Pearson r values of 0.968, 0.759, and 0.964, with MAEs of 4.16, 1.67, and 2.82 months. Ablation confirmed strong target-feature coupling, while calibration and bootstrap diagnostics revealed substantial uncertainty. The framework supports model-conditional fleet screening, not validated real-world RUL forecasting or automatic maintenance decisions, until prospective temporal validation links scores to maintenance outcomes.