Abstract Determining the frost layer thickness on plates is very important in heat and mass transfer processes in cryogenic equipment. In this study, the intelligent approaches of multi-layer perceptron trained by Bayesian Regulation (MLP-BR) and adaptive neuro fuzzy inference system (ANFIS) are utilized for predicting the frost layer growth on a vertical plate under natural convection and horizontal and parallel plates under forced convection. Dimensionless groups of the relevant parameters are also formed and used in this analysis. In particular, plate temperature, air temperature, air velocity, relative humidity, and time are taken as the models’ inputs. The self-organizing map (SOM) is applied to examine the influences of inputs on the performance of the selected models. It is shown that the MLP-BR-SOM model provides the best results for the test data samples with AARE of 2.57%, 6.55%, and 8.34% for test data, respectively, for the cases of vertical, horizontal, and parallel plates, respectively. In addition, three new semi-empirical equations comprising of dimensionless parameters are developed for the cases of vertical, horizontal, and parallel plates, with AARE of 12.36%, 27.18%, and 22.076%, respectively. Ultimately, the results are compared with those predicted by the existing empirical equations.
