Machine learning assisted measurement of solid mass flow rate in horizontal pneumatic conveying by acoustic emission detection

Abstract This work proposed a series of generalized strategies to establish a machine learning prediction model for solid mass flow rate in horizontal pneumatic conveying using acoustic emission detection. The strategies included adding flow regime parameters to the inputs for different flow regimes and standardizing the inputs for different conveying setups and different conveying materials. To the most important, the standardization of inputs could be evaluated primarily based on theoretical analyses of the generation and propagation mechanisms of acoustic signals. Finally, the strategies were tested by experiments. Adding flow regime parameters reduced the prediction error under different flow regimes by 8.9%. After the standardization of acoustic signals, the prediction errors under different conveying setups (conveying pipes with different diameters) and different conveying materials (polypropylene and polyethylene particles) were significantly reduced by 159.8% and 198.0%. This work may supply a bridge between the lab-scale experiments and real application in commercial plants.