Prediction of convective heat transfer coefficient during deep-fat frying of pantoa using neurocomputing approaches

• Multiple linear regression (MLR), connectionist and adaptive neurofuzzy inference system (ANFIS) models were developed to predict the convective heat transfer coefficient of pantoa.
• The performances of MLR, connectionist and ANFIS models were compared using root mean square.
• The best prediction of heat transfer coefficient was obtained using ANFIS model with frying time and temperature as input factors.
• ANFIS model could be a simple and accurate alternative for prediction of heat transfer coefficient during processing of pantoa.

Deep-fat frying (DFF) is the major processing step in preparation of pantoa, a popular Indian dairy sweetmeat. In this study, the dough for pantoa was rolled into balls of 15 g, and fried in sunflower oil at 125, 135 and 145 °C for 8 min. Convective heat transfer coefficient, which defines the heat transfer characteristics of the product during DFF, was determined using one-dimensional transient heat conduction equation as 92.71–332.92 W·m− 2·K− 1. Neurocomputing techniques such as connectionist models and adaptive neurofuzzy inference system (ANFIS) were compared vis-à-vis multiple linear regression (MLR) models for prediction of heat transfer coefficient. A back-propagation algorithm with Bayesian regularization optimization technique was employed to develop connectionist models while the ANFIS model was based on Sugeno-type fuzzy inference system. Both connectionist and ANFIS models exhibited superior prediction abilities than the classical MLR model. Amongst the three approaches, the hybrid ANFIS model with triangular membership function and frying time and temperature as input factors gave the best fit of convective heat transfer coefficient with R2 as high as 0.9984 (99.84% accuracy) and %RMS value of 0.1649.Industrial relevanceConvective heat transfer coefficient defines the heat transfer characteristics of a product during frying. Accurate prediction of heat transfer coefficient is important for design of process equipment and saving energy during commercial production. Developing models to predict heat transfer and the coefficients have been a challenge. Neurocomputing is one of the emerging intelligent technologies with analogies to biological neural systems. Therefore, it has the capability to predict complex relationships in food systems. Neurocomputing approaches such as connectionist and ANFIS models are now widely used in the food industry to predict various engineering properties of food, optimization of various transport processes, unit operations and formulating new products and product characteristics. No attempt has been made to predict the heat transfer coefficient during frying of pantoa. In this study, the convective heat transfer coefficient of pantoa was predicted using connectionist models and ANFIS techniques. These neurocomputing techniques are expected to alleviate the difficulties in conventional heat transfer modeling.