Data-Driven Prediction Of Minimum Fluidization Velocity In Gas-Fluidized Beds Using Data Extracted By Text Mining

Jibin Zhou,Duiping Liu,Mao Ye,Zhongmin Liu

DOI: https://doi.org/10.1021/acs.iecr.1c02307

2021-01-01

Industrial & Engineering Chemistry Research

Abstract:Minimum fluidization velocity (U-mf) is of fundamental importance in gas fluidization. Lots of empirical correlations have so far been reported in the literature to calculate U-mf. However, U-mf is affected by numerous factors, including, among others, the operation conditions and physical properties of both solids and gases. The applicability of empirical correlations relies essentially on the experiments upon which they were developed, and in practice, the choice of U-mf- is a matter of the knowledge and experience of chemical engineers. In this work, we proposed to establish a database by extracting experimental data of U-mf from open literature using the text mining technique. We first presented a pipeline of natural language processing to identify and extract the functional parameters related to U-mf with 83% accuracy from similar to 40 000 papers. A database of U-mf containing eight impacting factors, i.e., particle diameter, particle density, particle sphericity, bed voidage at minimum fluidization, gas density, gas viscosity, operating temperature, and pressure, was created. We then used a data-driven machine learning method with the extracting data to predict U-mf which is shown superior over the empirical correlations by achieving higher accuracy for a much wider range of gas-solid systems. We expect this work illustrates a potential and promising approach to make use of the huge amount of experimental data in the literature and replace the empirical correlations in practical chemical engineering design and operations.