Remaining useful life prediction of railway wheelsets using a composite wear index under sparse monitoring data

Abstract

Accurate prediction of the Remaining Useful Life (RUL) of railway wheelsets is important for operational safety and efficient maintenance planning. This study proposes a physics-informed composite wear index integrating wheel diameter and flange wear and validates it against field operational data. The composite index, W=f(D)∘f(L), combines diameter wear and flange wear through a reprofiling coefficient k, representing diameter reduction per unit flange restoration. Using machining records from a 60-wagon freight train operated by PT Kereta Api Indonesia (1,791 monthly records over 3 years), the field-based median k was estimated at 2.75 mm/mm for the train set and 3.00 mm/mm fleet-wide. The selected modelling value of k = 3.2 mm/mm lies near the upper range of field observations and provides a conservative approximation. Applied to the operational dataset, the index successfully tracked coupled wear progression, showing that 62% of wagons had exceeded the midlife threshold (W ≥ 0.50). A deterministic benchmark using 201 observations across five reprofiling cycles was used to compare five machine-learning models under 30–60% monitoring densities. Linear regression achieved the lowest error (R² ≈ 1.000), while gradient boosting showed the most reliable non-linear performance. The results support the proposed composite wear index as a practical basis for wheelset RUL estimation under limited monitoring data.