Predicting Twist Condition by Bayesian Classification and Decision Tree Techniques

Railway infrastructures are among the most important national assets of countries. Most of the annual budget of infrastructure managers are spent on repairing, improving and maintaining railways. The best repair method should consider all economic and technical aspects of the problem. In recent years, data analysis of maintenance records has contributed significantly for minimizing the costs. By studying each of line parameters, we can take steps towards condition based maintenance of that parameter. Chain faults as well as the cost of the track inspection will be reduced too. In this research, data mining techniques are used to investigate the relationship between geometric parameters of track and twist failure. By using other parameters of the track, twist can be predicted.  Moreover, the parameter that has the highest impact on twist can be found.  In this paper, Bayesian classification and decision tree techniques have been used. Finally, after conducting this study, it was found that alignment level (AL) and cross level (XLV) have the highest impact on causing track twist. This is the first research in the literature that have used all the above-mentioned methods to investigate twist failure in order to avoid chain failure and achieve condition-based maintenance.