Preprint / Version 1

Operating Speed Effects on Vibration-Data-Driven Discrimination of Field-Used Lubricant Samples in an Internal Combustion Engine

##article.authors##

  • Chamuditha Keerthisena University of Peradeniya
  • Asanga Ratnaweera

DOI:

https://doi.org/10.31224/8855

Keywords:

Experimental Machine Learning, Vibration, Condition Monitoring, Engine Health

Abstract

The concept of inferring lubricant condition from engine vibration is already established in the literature, particularly at fixed or narrow operating conditions. This paper presents a feasibility study for lubricant property prediction across varying operating speeds, investigating whether lubricant-related vibration signatures can be separated from operating-speed variability well enough for vibration to serve as a practical surrogate for direct lubricant condition sensing. Vibration was recorded from a petrol engine at six speeds under no load with five field-used lubricant samples, and discrimination between the samples, a necessary first step, was tested using engineered statistical descriptors and supervised machine-learning classification. Three feature representations, differing in whether lubricant information is assumed to sit in a few interpretable descriptors or to be spread across many, were evaluated under two strategies in which the test speed was either represented in or withheld from the training data. The samples were discriminable to a modest extent (accuracy up to 0.63 against a chance level of 0.20), indicating that exploitable lubricant-related information exists in a concentrated minority of the global feature set, although weak and diffuse across the full feature space. Discrimination varied with operating speed and thermal history of the engine, although per-speed estimates are descriptive. An auxiliary analysis identified the most contributing descriptors, which may be transferable unlike the trained models, and showed a descriptive tendency for confusion between samples of similar viscosity.

Downloads

Download data is not yet available.

Downloads

Posted

2026-11-20