A case study on multivariate statistical modeling for maintenance of freight railway wheelsets

This paper undertakes a linear mixed model analysis framework to explore the impact of maintenance decisions on the turning (re-profiling) of freight railway wheelsets. Univariate and multivariate linear models analyzed all the different variables (including fixed and random factors) that mainly impact maintenance and life cycle. The impact of maintenance is based on response variables such as changes in wheel diameter, in flange thickness, in flange height, and in the flange slope. All response variables are considered individually and combined to produce different statistical model specifications. The evaluation and comparison of the different model specifications are conducted with the Akaike Information Criterion (AIC). The results show that (i) the multivariate model is the most suitable and (ii) the overall AIC value of a multivariate model is lower than its sum for the univariate linear mixed models.