What question did this study set out to answer?

The aim is to develop a system for adaptive lubrication pressure prediction using machine learning models and sensor data.

April 5, 2026Open Access

Smart Autolube: Optimized machine learning-based pressure prediction for AIoT lubrication systems

Key Points

The aim is to develop a system for adaptive lubrication pressure prediction using machine learning models and sensor data.
Developed Smart Autolube incorporating AIoT and sensor-based monitoring.
Tested under mining simulation conditions with pressure, temperature, and stress sensors.
Utilized feature engineering techniques like lag and rolling statistics during preprocessing.
Employed ensemble algorithms, Random Forest and Gradient Boosting Regressor, for prediction modeling.
Conducted explainability analysis using SHAP to assess feature contributions.
Base model initially showed low prediction accuracy (R² < 0.1).
Improved performance after feature engineering, achieving R² of 0.9816 for GBR and 0.9711 for RFR.
Identified key contributing features such as trends and lag variables during explainability analysis.

Abstract

Autolube systems have been widely adopted in the mining industry to improve equipment reliability, but most still operate at fixed time intervals without adapting to real conditions in the field, and monitoring systems use LED lights, making it difficult to diagnose failures due to the minimal system interface. To overcome these issues, this study developed Smart Autolube based on Artificial Intelligence of Things (AIoT), which integrates sensor-based monitoring with machine learning models for adaptive lubrication pressure prediction. With industry support from PT. Multindo Technology Utama, the system was tested under mining simulation conditions using pressure, temperature, and stress sensor data. After preprocessing, which includes winsorization, feature engineering (lag, rolling statistics, and trends), two ensemble algorithms, Random Forest Regressor (RFR) and Gradient Boosting Regressor (GBR), are used to build a prediction model. The base model showed low accuracy (R² < 0. 1), but after feature engineering and extreme hyperparameter tuning, the performance improved significantly with an R² of 0. 9816 for GBR and 0. 9711 for RFR. Explanability analysis using SHAP (SHapley Additive exPlanations) shows that engineering features such as trends, lag₂, and rollingₘean₃ contribute the most to the predictions compared to native features such as temperature and voltage. This study proves that Smart Autolube can provide accurate and explainable lubrication pressure predictions. Further research is suggested to expand the scope of the data, add other mechanical parameters, and test the generalization of the model in different industrial environments.

AI से पूछें

Bookmark

View Full Paper