Artificial intelligence‐based correlation: Process side heat transfer coefficient for helical coils in stirred tank reactors

Abstract

AbstractStirred tank reactor systems often employ jackets/limpets, single/multiple internal coils for heat transfer. Internal coils are used for heat removal in the case of exothermic reaction systems. The desired heat transfer area can be accurately estimated based on the overall heat transfer coefficient (HTC) considering the process side, utility side, and wall conduction resistance. For estimation of process side HTC, a large number of empirical/semi‐empirical correlations are available in the published literature. These correlations account for impeller Reynolds number, fluid Prandtl number, viscosity ratio, dimensionless geometric factors, and so forth. However, all the factors over the entire range of industrial operation do not get covered in a single study, hence the predictions may involve significant errors for industrial cases. The advanced artificial intelligence‐based techniques can provide unified correlation based on the experimental data set extracted from various investigations and the predictions can be accurate and reliable. In the present work, three techniques random forests (RF), artificial neural networks with Bayesian optimization (ANN‐BO), and support vector regression with grid search optimization (SVR‐GS) have been used with a data set of 1297 points with R2 values of 0.995, 0.972, 0.990 on the training data set and 0.962, 0.945, 0.905 respectively on the testing data set. Furthermore, the statistical measures show that RF provides a better fit as compared with other AI models, so RF is used for parametric sensitivity analysis. The permutation feature importance shows that the agitation speed, impeller dimensions, and positioning has a significant impact on the HTC.