Real‐time evaluation of artificial neural network‐developed model of banana slice kinetics in microwave‐hot air dryer

Abstract

The required microwave power input and moisture content prediction in microwave‐hot air dryer are challenging. In this study, two artificial neural networks (ANN) were investigated for modeling of moisture content of banana slices and required microwave power input. The experiments were done in five levels of microwave power density (MPD) (4, 5, 6, 7, and 8 W/g) with fixed mode and two levels (6 and 8 W/g) with variable mode at 40°C hot air temperature. The initial MPD, mode, and time were used as input variables in both ANN models. The ANN with 3–5–5–1 structure had best performance for MC modeling with 0.0148 of RMSE and 0.9998 of R for testing data. Also, the ANN with 3–3–1 structure could predict the microwave power with 14.013 of RMSE and 0.9929 of R for testing data. Real‐time verification of developed models showed that developed ANN models could predict microwave power and MC with fixed MPD with good reliability and confidence for online drying control but did not reasonably satisfy for variable MPD. Also, the appearance quality of the banana slices was monitored by machine vision. The results showed that the percentage of banana charring at the fixed MPDs of 6, 7, and 8 is equal to 12%, 24%, and 29%, respectively, whereas no burns were observed at the other power density levels. Practical applications The main benefit of microwave drying is volumetric and high‐efficiency heating. But the temperature of the product at the end of drying process increased and brunt the product. Using the variable microwave power program is one of the famous methods for temperature control to prevent scrunching and burning the product. But powerful models for predicting the moisture content and microwave power during the drying process are not available. In this study, the drying kinetic of banana slices and input microwave power in the hybrid microwave‐hot air dryer under constant and variable microwave power was investigated and real‐time evaluating of the developed models for dryer control was examined. The results expressed the real‐time evaluation of the developed model had great accuracy and reliability and can be used for microwave dryer control.

Recent application of artificial neural network in microwave drying of foods: a mini‐review

“…The ANN model with optimized model parameters was able to predict the moisture removal level of onion slices after microwave drying. Similar predictive ANN models were also reported to predict the required microwave drying time for shelled corn, 18 black cumin, 19 final moisture content of apple, 20 parboiled paddy, 21 shrinkage and color change of quince, 22 time‐dependent moisture change of banana, 23 pistachios, 24 ginkgo Biloba seeds, 25 pineapple, 26 mango ginger, 27 and mushroom 28 . If including processing time as an input parameter, the trained regression model could be used to study the drying kinetics, including moisture, nutrient degradation, and other time‐related properties.…”