Optimizing Ultrasonic Inspection Regimes of Railway Rails

Abstract

Ultrasonic inspection of railway rails is considered as an important safety barrier. Fatigue cracks develop due to the cyclic loads of the passing trains, and the objective is to detect the cracks before they develop to rail breakages. It is crucial to understand the speed of crack propagation to define the inspection regime. Trains equipped with ultrasonic instruments can run at a speed of approximately 50 km/h. As the train runs suspects are identified with their position along the track. The ultrasonic scans are investigated and critical suspects are recorded for manual follow up. The manual inspection uses a hand-held trolley also with ultrasonic instruments for a more precise classification. In Norway a classification regime consisting of the categories 2b, 2a, 1 and 0 are used. The main strategy is to monitor 2a and 2b defects, whereas 1 defects have to be fixed within a month, and 0 defects have to be fixed immediately. The inspection-and follow-up regime is now under revision. Markov modelling is the basis for this study. There are several challenges with the Markov model. First of all, since we assume that fatigue is the main failure mechanism, it is not realistic to assume that transition times follow the exponential distribution. Secondly, times between running the inspection car are almost deterministic which requires a special treatment when solving the Chapman-Kolmogorov differential equations. Finally, the follow-up activities are also deterministic, and phasetype models may be used to handle transitions representing the results of the follow-up activities. In this paper we investigate how we can simplify the modelling without compromising the results too much. Statistics from the Norwegian rail network is used for estimating the transition rates.