Novel Correlation for Critical Speed for Solid Suspension in Stirred Tanks Developed Using Machine Learning Models Trained on Literature Data

Abstract

Critical speed for solid suspension (N js ) in stirred tanks is an important design parameter in several chemical processes. There is a need to develop a generalized correlation that applies to a broad range of literature data. Also, the literature lacks a comparative study of different machine learning (ML) models for the prediction of N js . In this paper, 3240 data points have been extracted from 35 papers on solid suspension and N js has been modeled as the dependent variable, initially using ML models. Three ML models (random forest regression (RF), CatBoost regression (CatB), and artificial neural network regression (ANN)) are compared. The CatB model was the most effective, resulting in R 2 = 0.99 for the testing dataset, better than any empirical correlation published. Further, we have successfully used the CatB model to obtain the functional form of the behavior of various parameters. We then developed novel correlations (in closed form and with parametric uncertainty) for HE-3, PTD, and DT impellers with R 2 = 0.89, 0.84, and 0.86, respectively, where the correlation constants were tuned using experimental data published in the literature. Our closed-form correlations significantly outperform earlier correlations published in the literature. Such a methodology has not been reported in the literature we have surveyed, and we believe that it is novel and original.

Novel Correlation for the Solid–Liquid Mass Transfer Coefficient in Stirred Tanks Developed by Interpreting Machine Learning Models Trained on Literature Data

“…The detailed modeling framework of all algorithms was explained in Joshi et al (2023) . The following subsections briefly explain the algorithms those were used in this work.…”

“…A brief overview on machine learning applications to design the multiphase reactors was given in previous study . It has been found that a study on machine learning correlations to correlate Sh () or k SL .25em ( normalm normals ) for agitated vessels is absent in the literature.…”

“…The comparison with state-of-the-art correlations is discussed in Section . The machine learning correlations have been interpreted in the following ways: (i) examining the EF model’s sensitivity to its individual as well as groups of parameters; (ii) using the method described in previous publication . Furthermore, a novel correlation has been successfully developed to predict the Sh () for the disc turbine (DT) impeller.…”

“…Real-world systems can exhibit intricate relationships that may not be adequately captured by the simplified correlation models. Machine learning techniques, which can handle nonlinear and complex relationships, may offer better alternatives in such cases …”

“…Moreover, in previous publication, an interpretable machine learning approach to correlate N js (rpm) with the corresponding variables is successfully examined . In this work, a similar methodology was used to develop the correlation to estimate the Sh ().…”