Good manners and practices for successful multivariate data analysis in oil condition monitoring (OCM) 

Chemometrics can be briefly defined as the discipline that applies machine learning (ML) methods to extract the maximum relevant information from chemical system by multivariate data analysis. According to this definition the results obtained by ML methods should have a chemical meaning that should be considered during calibration model’s validation, helping to the interpretability of the model. This, which seems quite obvious, sometimes is not put in practice when ML methods are applied as a complex black box. The interpretability of the model is a concept emphasized by Explainable Artificial Intelligence (XAI), which in recent years has been gaining importance helping to the successful integration of the ML models in real applications.

In this context, this work aims to give some practical advice for a successful multivariate data analysis in lubrication management since the data pre-processing up to the validation and integration stage. For this purpose, this work shows the results of applying ML to spectroscopic data for oil condition monitoring (OCM) purposes, including real-life examples provided by OCM laboratory.