Application of Symbolic Regression to Electrochemical Impedance Spectroscopy Data for Lubricating Oil Health Evaluation

The authors have applied an advanced set of auto-regressive tools for identifying potentially complex, linear and nonlinear relationships in data, wherein the underlying physical relationships are not well described. In this paper these tools and techniques are described in detail, and the results of the application of these tools to evaluation of diesel engine lubricating oil health (based on electrochemical impedance spectroscopy data) is detailed. It is demonstrated that highly accurate models can be constructed which take as input features derived from diesel engine lubricating oil electrochemical impedance spectroscopy data and output estimates of traditional laboratory based oil analysis parameters. The electrochemical impedance spectroscopy and laboratory analytical data used are from a field deployment of oil condition sensors on several long-haul class 8 diesel trucks. The dataset was divided into training and test datasets and goodness of fit metrics were calculated to evaluate model performance. Models were successfully generated for nitration, soot content, total base number, total acid number, and viscosity.