Optimized design of experiment approaches for the calibration of instrumented wheelset

The presented work is done with the purpose of applying design of experiments concept to tests to be performed for the calibration of an instrumented wheelset and its applicability is investigated. The calibration of instrumented wheelsets is performed in the laboratories of Politecnico di Milano, in order to estimate the contact forces in in-line tests of the train. It is done by building a model, which relates the strains measured on the instrumented wheelset, to the contact forces applied on the test rig. Design of experiments aims to reduce the total time spent on the experiments and the costs related to it, by building an effective and efficient test plan. In literature there are plenty of optimality criterions available in order to design an experiment optimally. Each criterion has different method to develop an optimal design matrix that includes the factors influencing the response. In this work, the factors influencing the response are analyzed and a model is tried to be formed to estimate strains from the contact forces applied on wheels. It is assumed that if a test plan is formed using the model to estimate strains from the contact forces, the same test plan can be used to estimate the contact forces from the measured strains. According to the model developed, design of experiments is applied on the calibration tests. Moreover, different optimality criterions are used to make a comparison for the compatibility with the calibration problem. Furthermore, for different strain types, an alternative model relating the forces, are developed. On the results section, the accuracy of the forces estimated, which are calculated through the calibration experiments obtained with respect to optimality criterions, is presented and its effectiveness are investigated. It is verified that, the accuracy of the estimated forces start to converge at a certain number of elements used in the candidate set, in case optimal design of experiments are used. It can be concluded that, it is possible to reach effective results with less test runs, using optimal experiment designs.