A Combined Measurement System for Fast Classification of Water Contamination in Lubricant Oil

In this paper, a measurement system aimed to the fast classification of water contamination in oil samples will be presented. The transduction principle is based on the permittivity change of an interdigital capacitor which changes its capacitance value while immersed in oil samples with different water concentrations. Differently from other works, the presented system proposes a circuit and a measurement approach. It combines the broadband excitation property of MLS-based impulse response (IR) measurements with the support vector machine (SVM) machine-learning (ML) model. This approach allows to speed up the measurements, thus reducing the energy-per-measurement parameter in order to make the system suitable for battery-powered portable devices. The theoretical foundations, the circuit-level description of the analog front-end, and the used ML model will be presented in detail. The classification capability of the system will be proved by evaluating 40 IRs from 6 prepared oil samples at water concentrations of 0 vol%, 0.2 vol%, 0.5 vol%, 1 vol%, 2 vol%, and 3 vol%. The proposed system is able to measure a 1023-point IR in 700 ms, which is better than the state-of-the-art. Finally, an overall classification accuracy of 90% is obtained after the SVM training process with a 10 fold cross-validation.