Predictive Neural Network Modeling for Diesel Engine Part Wear Assessment via Analysis of Wear Element Concentration in Used Oil

Abstract

Abstract
This paper introduces a nondestructive method for estimating wear in
diesel engine parts. Ulaanbaatar Railway, Mongolia's major railway
company, has utilized Russian-manufactured 2TE116Um series diesel
locomotives since 2010, following maintenance schedules outlined by the
manufacturer. However, observations during the initial maintenance
period from 2010 to 2016 necessitated adjustments to align maintenance
schedules with Mongolia's unique operating conditions. Assessing diesel
engine wear and predicting part lifespans based on wear element
concentrations in engine oil has global applicability. Nonetheless, the
existing Russian-approved methodology, different chemical compositions
in diesel engine parts compared to other locomotive manufacturers, poses
challenges in implementing recent approaches like neural networks (NN)
for accurate predictive maintenance scheduling. Addressing this
challenge, our study conducted spectral analysis of engine oil under
Mongolia's operational conditions, analyzing wear element
concentrations and their fluctuations. Furthermore, during maintenance
periods, engine parts were disassembled and measured. Subsequently,
data were utilized to train a neural network model to predict remaining
useful life of the parts. Our two-stage neural network model
demonstrated a remarkable improvement in predictive accuracy
compared to traditional mathematical models, with an R=0.99, R=0.82
MSE. This enhanced model accurately assesses component wear,
optimizing locomotive repair schedules, thereby potentially reducing
maintenance expenses, and enhancing locomotive performance
significantly.