Original Article Optimal track geometry maintenance limits using machine learning: A case study Ahmad Kasraei, Jabbar Ali Zakeri and Arash Bakhtiary Abstract The aim of this study has been to determine the optimal maintenance limits for one of the main railway lines in Iran in such a way that the total maintenance costs are minimized. For this purpose, a cost model has been developed by considering costs related to preventive maintenance activities, corrective maintenance activities, inspection, and a penalty costs associated with exceeding corrective maintenance limit. Standard deviation of longitudinal level was used to measure the quality of track geometry. In order to reduce the level of uncertainty in the maintenance model, K-means clustering algorithm was used to classify track sections with most similarity. Then, a linear function was used for each cluster to model the degradation of track sections. Monte Carlo technique was used to simulate track geometry behavior and determine the optimal maintenance limit which minimizes the total maintenance costs. The results of this paper show that setting an optimal limit can affect total annual maintenance cost about 27 to 57 percent. Keywords Maintenance limit, track geometry, degradation, K-means clustering, Monte Carlo Technique Date received: 5 April 2020; accepted: 21 September 2020 Introduction The railway network is known as a safe and reliable means of transportation for transferring passengers and freight. The quality of the track is mainly repre- sented by the track geometry parameters, that is, the cant, alignment, longitudinal level, twist and gauge. 1 Like any other mechanical system, railway compo- nents deteriorate with age, usage and under various environmental conditions. These parameters may result in degradation and failure of railway assets. 1,2 Whenever the quality of track geometry reaches a certain maintenance limit, a proper maintenance action needs to be taken into account to maintain the quality of track in an acceptable level. Maintenance actions like manual intervention, tamp- ing and stone-blowing can be employed to restore the quality of the track geometry to a better condition. 3–5 Tamping is the most applied maintenance action to remedy a degraded track geometry. Therefore, limit action is necessary in order to restore the deteriorated system before failure. 6 Obviously, setting an inappro- priate maintenance limit would negatively affect the track safety and availability and total maintenance cost. On the one hand, setting low maintenance limit increase the down time of the system for fre- quent maintenance activities and as a result of that it negatively affects trains serviceability. On the other hand, high maintenance limit would affect the safety of the trains passing along the track and increase the time that track spend in a bad condition. As a result, it reduces the useful life of track. Determining the optimal track geometry mainte- nance limit had been the main concern of a number of studies in the recent years. Khajehei et al. 1 pro- posed a framework based on which the effective maintenance limit can be determined. These authors consider standard deviation of longitudinal level and the extreme value of isolated defects of longitudinal level as track quality indices. Their aim has been to determine an effective maintenance limit which mini- mizes total maintenance costs. Andrade and Textura 7–9 used the longitudinal leveling indicator to investigate the effect of different limits proposed by EN 13848-1,5. 10,11 Their aim has been to minimize cost and delay. In other study, Araste Khoy et al. 2 tried to determine the optimal maintenance limit by considering standard deviation of longitudinal level and isolated defect of twist as track geometry indica- tor. Andrews et al. 3 have been developed a model to School of Railway Engineering, Iran University of Science and Technology, Tehran, Iran Corresponding author: Jabbar Ali Zakeri, School of Railway Engineering, Iran University of Science and Technology, Tehran, Iran. Email: zakeri@iust.ac.ir Proc IMechE Part F: J Rail and Rapid Transit 0(0) 1–11 ! IMechE 2020 Article reuse guidelines: sagepub.com/journals-permissions DOI: 10.1177/0954409720970096 journals.sagepub.com/home/pif