Uncertainty Quantification in Measuring Ore Mass Flow Rate with Data-Driven Soft Sensors

In the context of Industry 4.0, there is a growing need for reliable information. Understanding uncertainties in models and data can improve decision-making, safety, and efficiency in the mining sector, especially where machine learning-based soft sensors are utilized. This is particularly relevant in ore transport, where machine learning-based soft sensors estimate the mass flow rate of transported ore. However, these applications often utilize traditional machine-learning techniques to make single-output predictions, limiting their utility. Therefore, this study seeks to advance the performance of such sensors by quantifying the prediction’s uncertainties. As a result, the electrical current from a conveyor belt was used to develop a model utilizing feature selection, engineering, and the Light Gradient Boosting Machine (LightGBM) algorithm, producing interval predictions. Four methods of uncertainty quantification were evaluated: T-Student, Bootstrap, Split Conformal Prediction (SCP), and Full Conformal Prediction (FCP). The results show that LightGBM, together with feature engineering and selection, decreased the soft sensor error compared to the previous study. The Bootstrap and FCP methods provided effective prediction intervals, achieving the intended coverage with minimal interval widths, suggesting their superior ability to balance accuracy and adaptability.