Monitoring the status of machinery is a crucial aspect of production and service units to uphold operational efficiency. Timely changes in engine lubricant significantly contribute to enhanced performance and extended engine lifespan. However, determining the precise replacement time remains a challenge. Oil spectral analysis, while effective, is both expensive and time-intensive. This study aims to introduce an alternative method to engine lubricant spectral analysis. The investigation involves analyzing the results of spectral analysis and dielectric coefficients of 17 engine lubricant samples through statistical methods. The primary objective is to develop models for predicting oil contaminants based on dielectric properties, offering a substitute for spectral analysis. To achieve this, several intermediate goals are pursued. Multilayer perceptron artificial neural networks (MLP-ANN) and support vector machine (SVM) methods are employed for modeling. The performance of the two models is assessed using indicators such as Root Mean Square Error (RMSE), model efficiency, and R-squared (R2). The results indicate that the SVM model consistently demonstrates an efficiency exceeding 0.95 for all predicted indices (Fe, Pb, Cu, Al, Mo, Na, Si, and Vis@100). Consequently, dielectric spectroscopy of lubricant emerges as a viable alternative to traditional oil spectral analysis.

Keywords