A comparative study on the efficiency of two modeling approaches for predicting moisture content of apple slice during drying

Abstract

The efficiency of two different modeling approaches for predicting moisture content of apple slices during drying were evaluated and compared. The experiments were performed at four air‐drying temperatures and at three levels of air velocity in the convective hot air dryer. Moisture content of apple during drying was predicted using theoretical model and Artificial Neural Networks (ANNs) models. The theoretical model was developed by solving heat and mass transfer equations simultaneously using numerical technique. A multilayer perceptron (MLP) neural network and radial basis function (RBF) network were used to develop neural network modeling. The agreement between the experimental results and the theoretical model predictions was quite good by considering the shrinkage dependent effective diffusivity in the model. Among different ANN structures, the best results were obtained for the MLP network with two hidden layers based on the statistical criteria (the coefficient of determination [R2] and mean squared error [MSE]). Although both two approaches could reasonably forecast the moisture content of apple slice during drying, the ANN model demonstrated better goodness of fit than theoretical model. Practical application Moisture content and water activity are important parameters in production of dried fruit, which have traditionally been measured by destructive and time‐consuming methods. Hence, quick tools are required to evaluate the moisture concentration and monitor changes during drying. Modeling can be considered as a helpful tool for prediction of moisture content as a function of time and space. The results of this study show that the two types of models can successfully predict changes in moisture content during drying. Applying these models at the commercial level can reduce variability in product quality and improve repeatability of the drying process.

Combination of LF‐NMR and BP‐ANN to monitor the moisture content of rice during hot‐air drying

“…The average error of the BP‐ANN model in predicting the moisture content of rice was 0.2, and the model prediction performance was superior. The results of this paper were similar to those reported in the literature by other researchers, and the artificial neural network fits well with the data obtained from the experimental data (Dalvi‐Isfahan, 2020; Tavakolipour & Mokhtarian, 2012). In addition, similar results were obtained for vacuum drying, spray drying and superheated steam drying, demonstrating the strong adaptability and good simulation performance of artificial neural networks (Chai et al, 2018; Li et al, 2021; Shrivastav & Kumbhar, 2011).…”

“…The empirical models are developed by fitting model parameters to experimental data, based on the relationship between moisture content and drying time. Mathematical modeling methods are subject to limitations in the application, such as the choice of parameters, computing time, the application of assumptions in solving equations and the complexity of solving equations (Dalvi‐Isfahan, 2020). In addition, the moisture content of the materials is significantly nonlinear and time‐varying.…”

“…The artificial neural network model has a complex nonlinear system that can accommodate multiple input and output variables, while the fast calculation speed can meet the needs of online drying control and process optimization (Sun, Zhang, & Mujumdar, 2019; Watanabe et al, 2014). Previous studies have shown that neuronal network models compare favorably with conventional models in predicting the moisture content of materials during drying (Beigi, Torki‐Harchegani, & Mahmoodi‐Eshkaftaki, 2017; Dalvi‐Isfahan, 2020; Tavakolipour & Mokhtarian, 2012). A great deal of work has been done by scholars on the real‐time monitoring of moisture content of agricultural products (Mohebbi et al, 2011; Tohidi et al, 2014).…”

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

“…The number of neurons in the hidden layer, thresholding function and training algorithm are the famous factors that need to be adjusted to construct the model. For determining the number of neuron in the hidden layer, trial and error method was used and based on previous researches the tansig and linear function were used as transfer function in the hidden and output layers, respectively, Levenberg–Marquardt (LM) back‐propagation algorithm with the early stopping method was used for training of the ANN model (Dalvi‐Isfahan, 2020; Khaled et al, 2020). The performance assessment of ANN models was evaluated by the root mean square error (RMSE) and correlation coefficient ( R ) between the predicted values by model and the target data.…”

“…Also, the newest comprehensive discussions about the capability and robustness of ANN for modeling the drying process were done by Khan, Sablani, Joardder, and Karim (2020) and Chen et al (2020). Also, Khaled et al (2020); Dalvi‐Isfahan (2020) and Abbaspour‐Gilandeh, Jahanbakhshi, and Kaveh (2019) results indicated that the ANN had a good strength for dynamic modeling of moisture content during the drying process. Also, Liu et al (2020) applied the ANN benefit for optimizing the drying process in the vacuum dryer.…”