Telefono: 02.2399.8456
Skype: anna.derosa91_2
E-mail: anna.derosa(at)

Sustainable Mobility

Relatori: Prof. Stefano Bruni, Prof. Alan Facchinetti

Tutor: Prof.ssa Bianca Maria Colosimo

Università di Provenienza: Politecnico di Milano - Ingegneria Meccanica

Titolo della Tesi: Advanced Methods for Condition Monitoring of Railway Infrastructures using in-service Vehicle Acceleration Measurements

Advanced Methods for Condition Monitoring of Railway Infrastructures using in-service Vehicle Acceleration Measurements


The track geometry irregularity has a relevant influence on the dynamic response of the vehicle. The abnormal vibrating response of the vehicle may affect the ride comfort for the passengers and, in the worst case scenario, their safety. Therefore, analyse and optimize the track condition monitoring procedure is an essential topic. Track irregularity is usually monitored by means of dedicated track recording vehicles, provided with sophisticated laser optical and inertial measurement systems, which periodically acquire the track geometry deviations from the ideal position. Since, only few diagnostic vehicles are available for each railway operating company, due to the high cost required to be built, to run and to be maintained, although optical measurements are very accurate, they cannot be continuously acquired on the entire railway network. So, the time interval between two consecutive acquisitions may be not adequate to optimally schedule the maintenance interventions.


This research is focused on developing new methodologies for track condition monitoring using measurements of the vehicle’s running dynamics that can be acquired daily, onboard the in-service trains with robust and relatively inexpensive transducers, during normal line service. To evaluate the track irregularity status from the vehicle acceleration measurements both model-based and signal-based approaches are developed.


  • Pseudo-inversion of the frequency response function matrix(Frequency Domain Approach: FD)
  • Kalman Filter (KF)


  • Linear Correlation Analysis
  • Machine Learning Classification


The two methodologies are compared by means of virtual measurements produced by a fully non-linear three-dimensional multibody model of the rail vehicle in excess of 100 states. Both these methodologies are able to provide an accurate reconstruction of the lateral irregularity in a range of wavelength from 3 to 180 m as shown in Fig.1 (estimation of lateral irregularity – space domain signals) and in Fig.2 (estimation of lateral irregularity – 1/3 octave band spectra).
In Fig.3 (Regression Line - vertical direction), the correlation between the vertical track irregularity and the vertical bogie acceleration is shown. All the points lay close to the regression line with small dispersion, meaning that the correlation between the signals is very high with a value of 0.91.
In the lateral plane, due to the non-linearities introduced by the wheel-rail contact, the correlation is analysed with a machine learning (ML) based fault classifiers. The linear SVM shows good performances with 92.9 % of accuracy and 70.3 % of precision. The results could be improved identifying further predictors to separate more clearly the data in the two classes.


In this research, different diagnostic and identification techniques are investigated to monitor the track geometry irregularities using vehicle dynamics measurements. The model-based methods provide satisfactory results in the numerical experiments when all the model parameters are known, and a complete set of measurements are available. These methodologies are also tested on field measurements providing satisfactory results in the wavelength range related to running safety.
The linear correlation analysis is suitable to investigate the relationship between the track irregularities and the vehicle acceleration in the vertical plane. In the lateral plane, this relationship is non-linear and requires a more complex approach, such as a machine learning (ML) classification. The results achieved are satisfactory and promising. The ML methodology shows a good potential of data driven classifiers to monitor the track irregularities based on the vehicle dynamics.


[1] S. Alfi, A. De Rosa, S. Bruni, Estimation of lateral track irregularities from on-board measurement: Effect of wheel-rail contact model, in IET Railway Condition Monitoring, 2016
[2] A. De Rosa, S. Alfi, S. Bruni, Estimation of lateral and cross alignment in a railway track based on vehicle dynamics measurements, Mechanical Systems and Signal Processing, vol. 116, pp. 606–623, Feb. 2019.
[3] A. De Rosa et al., Monitoring of lateral and cross level track geometry irregularities through onboard vehicle dynamics measurements using machine learning classification algorithms, Proc. Inst. Mech. Eng. Pt. F J. Rail Rapid Transit, 2020.
[4] F. Cangioli, M. Carnevale, S. Chatterton, A. De Rosa, L. Mazzola, Experimental results on condition monitoring of railway infrastructure and rolling stock, in WCCM 2017 - 1st World Congress on Condition Monitoring 2017, 2017.