Predicting rail defect frequency: An integrated approach using fatigue modeling and data analytics

In maintenance planning of rail track, it is imperative to assess the potential and frequency of rail defects. Although this problem has been mainly studied in the literature by either data-driven or mechanic-based models, in the present study a new method is proposed to account for the strengths of both approaches in a single model. The envisaged model incorporates fatigue crack growth model, through Finite Element Modeling (FEM), into Approximate Bayesian Computation (ABC) framework. The method is applied to the prediction of rail defect frequency for transverse defects obtained from a US Class I Railroad. The results of the proposed model show that inducing the mechanics of rail defects into a data-driven model outperforms the traditional pure data-driven models by over 20%. The outcome of this study, along with necessary future developments to broaden the scope of applicability of the method, will benefit railroad existing practice in capital and maintenance planning.