Analysis of the particle characteristics of solid-liquid turbulent flow by triaxial vibration statistic and machine learning methods

Abstract

The real-time monitoring of particle information in turbulent flow is critical for safe and efficient production. A new method based on the fusion of particle-wall free collision and triaxial vibration method are proposed to measure particle information in turbulent flow. The characteristic frequency band of the triaxial vibration response excited by the particle-wall collision is successfully characterized by time-domain statistical, Zoom-FFT and STFT analysis, etc. Further, SVM and CNN were used to evaluate the performance of the particle characterization and analysis methods, high recognition accuracy were obtained for whether particles were present (98.48%), the flow velocities (85%), and the particle sizes (96.67%), which also proved the validity of the triaxial vibration response analysis method. Finally, a particle-wall collision energy model expressed as Energycollision=a1∙m+a2∙p2+b1∙m+b2∙p+c1∙m+c2 was established for solid-liquid flow, and the average error rates for each vibration direction were 11.40% (x-direction), 3.17% (y-direction), 13.92% (z-direction) and 7.90% (multidirectional joint).