Railway safety through predictive vertical displacement analysis using the PINN-EKF synergy

Scientific Machine Learning (SciML) finds extensive application in daily life, industry, and scientific research. Specifically, in railway data analysis, it utilizes tools such as time series analysis, classification, and data visualization. Among these, monitoring vertical displacement, or the movement of railway components relative to a fixed point, is vital. This indicates changes in the elevation of railway infrastructure, highlighting track settlement and structural shifts. However, relying solely on single sensor measurements for these displacement calculations can introduce inaccuracies. To overcome this, sensor fusion techniques are employed. These techniques employ advanced algorithms to combine data from multiple sensors, thereby enhancing accuracy. They consider the individual characteristics of each sensor, effectively mitigating the limitations of any single sensor. This study introduces a hybrid approach that combines the Extended Kalman Filter (EKF) with the Physics-Informed Neural Network (PINN) to refine predictive data analytics in dynamic railway environments. The experimental results underscore the efficacy of this innovative methodology.