Prediction of feed flowrates, based on vibration patterns, generated by vibration sensors on an industrial circular vibrating screen using a set of ML models

• Flow rate predicted from screen vibration of six accelerometers. • MLP reaches ∼ 90 % test R2; error near stereo system’s ± 10 % limit. • Inlet-left sensor is most informative; model prediction robust under other single-sensor failures. • Adding mean ambient temperature improves test R 2 by ∼ 0.7 %. Machine monitoring systems in the mineral and waste processing industries are typically separate from process monitoring systems, including those for vibrating screens. This study presents a novel approach integrating flowrate measurement capabilities into a circular vibrating screen’s vibration monitoring system embedded in the I-Step Operator system developed in collaboration with IFE Aufbereitungstechnik GmbH. A set of machine learning models was trained to predict feed flowrates using vibration patterns captured by smart acceleration sensors on the screen, including tri-axial accelerations and the angle of the major vibration axis relative to the horizontal plane. A volume flowrate monitoring system was temporary installed above the screen’s feeding conveyor belt to collect data for model training and testing. Among the models evaluated, the multilayer perceptron regressor achieved the best performance, with an R2 score of 90.10 ± 1.00 %, matching the flowrate monitoring system’s measurement error margin of 10 %, as provided by the manufacturer. The study also identified the inlet left sensor as the most critical input, demonstrating the model’s low sensitivity to other single-sensor failures. Additionally, incorporating ambient temperature improved the model’s performance, suggesting that factors beyond vibration patterns can enhance prediction accuracy. This indicates that such factors can be successfully integrated into the model, further extending its predictive capabilities.