A Predictive Maintenance Concept for Lubricant Oil Usage

The condition of lubricant oil is crucial for industrial machinery, as it minimizes wear and extends the equipment’s lifetime. A predictive maintenance approach that assesses oil condition in real time and forecasts the remaining useful lifetime can prevent costly component failures and machine downtime. The predictive maintenance concept presented here includes anomaly detection and real time oil condition estimation. Data is collected using an equivalent set of sensors at both a laboratory evaluation testbed and a production environment facility. A gold standard laboratory test provides information on the current oil condition. For anomaly detection, it is demonstrated that a Gaussian mixture model effectively identifies distribution shifts in temperature dependent sensor readings. For real time oil condition estimation, the phosphorus concentration is selected as the target value. A random forest regressor and a Gaussian process regressor are evaluated using a grouped cross-validation dataset. As an example, the predicted phosphorus concentration from the field data aligns closely with laboratory analysis results. To further validate these findings, additional field data from various use cases is needed. In the future, this data could help refine the model and incorporate an estimate of the remaining useful life.