A vibration-based machine learning approach for roller gap detection in biomass pellet production

This research focuses on optimising biomass pellet manufacturing processes by detecting roller gap variations in rotary ring die pelleting (RRDP) technology. Integrating experimental testing, response surface modelling (RSM), and vibration-based machine learning, this study aims to ensure optimal conditions for biomass pellet mill operation. Vibration-based machine learning techniques offer an approach for detecting roller gap variations, while RSM provides mathematical models to understand process dynamics for identifying optimisation criteria. Experimental testing explores the impact of process variables on pellet quality metrics. Results demonstrate machine learning model performance in detecting roller gap variations with F1-scores ranging from 88.1% to 100.0% across a pilot- and industrial-scaled setup. ANOVA results underscore significant relationships between roller gap, feedstock layer mass, and pelleting process metrics, while the created RSM models all have determination coefficients R 2 of ≥ 0.90 . Overall, this comprehensive approach contributes valuable insights into optimising efficiency and product quality in the biomass industry through an integrated framework. • Biomass pellet production critical for sustainable energy supply chain. • Sensor-based monitoring increase efficiency connecting algorithms to operations. • ML models exceed results K-nearest neighbours, random forest & linear discriminant. • Response surface models reflect well roller gap and feedstock mass. • Demonstrated scale-up suggest algorithmic stability of rotary ring die pelletisers.