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

Predicting
the solid–liquid mass transfer coefficient
(kSL) in stirred tanks is of great importance
in the chemical, metallurgical, and allied process industries. While
there are several correlations available in literature to predict
this parameter, they are only applicable to a narrow range of variables.
In this work, 1117 data points are collected from 13 research papers.
First, three machine learning models are developed for the prediction
of the Sherwood Number (Sh) using two approaches,
viz., (a) incorporating engineered features (EFs) as inputs and (b)
using independent variables as inputs. The CatBoost regressor (CatB)
is found to be more accurate than the random forest regressor and
artificial neural network regressor, achieving an impressive R2 > 0.95 for test data. Next, the CatB model
is interpreted by two approaches: (a) sensitivity analysis of the
EFs and (b) determination of the functional form of the correlation
based on CatB model’s predictions that developed using independent
features. Furthermore, this novel, human-readable correlation is validated
by fitting it to 816 experimental data points for the disc turbine
impeller, yielding an impressive R2 =
0.92, which is superior to the current state-of-the-art.