| Περίληψη: | The current thesis deals with the vibration-based fault detection in suspension systems of in-service railway vehicles that operate under varying speed and moving on rails of similar quality (roughness). Vibration signals are measured from three different points on the vehicle which are at the right axle box of the front wheelset, at the right corner at the front side of the car body and at the front right corner of the leading bogie. The modeling of the healthy vehicle partial dynamics under different operational speeds is obtained, in a baseline, training, phase through two Functional Pooled Transmittance Function AutoRegressive with eXogenus input (FP-TF-ARX) models. This modeling corresponds to the formation of a proper subspace through the FP-TF-ARX model parameters, where the vehicle healthy dynamics is represented. Thus, fault detection is achieved using fresh signals obtained from a current unknown vehicle state by testing if they reside within this healthy subspace based on a proper optimization procedure. The vibration signals which are used in the baseline and inspection phase where the fault detection of unknown test cases is attempted, are produced from a detailed and realistic forty-two degrees of freedom railway vehicle model that is constructed in Simpack. Numerous Monte Carlo simulations are performed for healthy and faulty components in the primary and secondary suspensions for the assessment of the employed method. The results indicate that the appropriate extraction of information from vibration signals through advanced stochastic methods that employ TFs corresponding to properly selected measurements from the railway vehicle, may lead from very good up to excellent fault detection of even relatively small faults in the vehicle suspension systems.
|
|---|
