Transformer Vibration Forecasting for Advancing Rail Safety and Maintenance 4.0

A mechanical problem, a modeling question.

Railway axles generate vibration signals that reflect their operating conditions. Understanding those signals can support research into condition monitoring. Changing speeds and loads make the vibrations difficult to forecast.

Forecasting in a changing signal.

This study investigates transformer-based autoregressive models using experimental vibration recordings from accelerometers on train axles. It introduces ShaftFormer and explores spectral methods and enhanced observation models for representing complex time-series behavior.

Why generate vibration signals?

Collecting experimental recordings for every operating condition and fault scenario can be costly. Modeling and simulating signals offers a way to investigate a broader range of conditions. The challenge is to represent their changing structure rather than memorize a single waveform.

Findings and limits.

The research connects sequence modeling to railway vibration forecasting using experimental signals. The results are research evaluations. They do not establish a commercially deployed safety system, guarantee fault detection or replace the validation required for operational railway use. Generalization to new equipment and conditions remains an important question.

My role.

I am the first author of this collaborative paper, published in the IEEE Open Journal of Intelligent Transportation Systems in 2025.