A Rapid Classification of Cross-Contaminations in Aviation Oil Using Impedance-Driven Supervised Machine Learning

Real-time monitoring of lube oil plays a crucial role in ensuring optimal machinery performance, preventing failures, and facilitating timely maintenance strategies. The approach proposed in this work, based on impedance spectroscopy and supervised machine learning (ML), addresses this need by a novel solution to a multiclass classification problem of cross-contaminations in aviation lubricant. Impedance measurements were performed at room temperature by immersing a microfabricated sensor in 16 aged oil samples containing increasing concentrations of water and aviation fuel. Two datasets were constructed: the first based on impedance components spectra and the second based on dissipation factor spectra. A data preprocessing and augmentation method was proposed for generating synthetic examples from the measured data. Both datasets were independently used to train three supervised classifiers, whose performance was evaluated based on three different approaches of dataset split ratio and k-fold cross-validation. The 1-nearest neighbors (NN) classifier proved to be the most effective in reducing false positives (FPs), false negatives (FNs), and computational running time. The best results were obtained by employing a split ratio of 60:40 and threefold cross-validation scheme on the impedance components-based dataset, yielding an accuracy of 99.8%.