Bayesian multivariate track geometry degradation modeling and its use in condition-based inspection

• • The multivariate Wiener model captures correlation between degradation rates of multiple track geometry indicators, improving prediction accuracy over independent univariate models by leveraging joint information from longitudinal level, alignment, gauge, cant, and twist. • • Hierarchical Bayesian estimation with MCMC quantifies parameter uncertainty and accommodates limited data, enabling robust degradation forecasts even when historical records are sparse—a common industrial constraint. • • The condition-based inspection policy derived from the model reduces the number of track recording vehicle runs while maintaining abnormal detection levels and failure rates, directly lowering inspection costs and operational disruption. • • The model explicitly incorporates imperfect manual and mechanized tamping through random recovery magnitudes, reflecting real-world maintenance effectiveness and preventing over-optimistic degradation projections.