พูนพัฒน์ พูนน้อย. Using Artificial Neural Network Approach for Modeling of Microwave-vacuum Drying of Plant Materials. Doctoral Degree(Food Engineering). King Mongkut's University of Technology Thonburi. : King Mongkut's University of Technology Thonburi, 2006.
Using Artificial Neural Network Approach for Modeling of Microwave-vacuum Drying of Plant Materials
Abstract:
The transport of electromagnetic wave energy, heat, and mass within a plant material
undergoing microwave-vacuum (MV) drying (MVD) is considerably complex due to
the dynamic change of products physical properties. The process control parameters
such as microwave power and vacuum pressure should be adjusted appropriately
corresponding to those time-dependent physical properties to prevent thermal damage
and to maintain drying efficiency. Information regarding the effects of microwave
power and vacuum pressure on the temperature and moisture content are essential for
process design, optimization, and control. Simulation of Maxwells equations together
with heat and mass transfer equations may give details in temperature and moisture
distributions within the product during MVD. However, the prediction accuracy of such
approach significantly depends on a good estimation of the physical properties.
However, the prediction of such physical properties, particularly the dielectric constant
and dielectric loss factor, during the MVD process is extremely difficult since these
properties change with temperature, moisture content, chemical component, and
density. Artificial neural network (ANN) modeling can handle such complex
relationships; therefore, such technique can be utilized to model MVD. The main
objective of this research was to develop ANN models for temperature and moisture
content/moisture ratio predictions of some selected plant materials undergoing MVD.
In a comparative study on errors of moisture ratio prediction in MV-dried mushroom, it
was observed that the one-hidden-layer feed-forward ANN model provided lower
reduced chi-square (?2), root mean square error (RMSE), and residual sum of squares
(RSS) (2.197?10-5, 3.955?10-3, and 3.097?10-3, respectively) than those of the diffusion
approximation model. The errors for the diffusion approximation model were 2.793
?10-4, 1.646?10-2, and 5.362?10-2 for ?2, RMSE, and RSS, respectively. These results
indicated that the ANN model represented the drying characteristics of mushrooms
better than did the simple mathematical model. Therefore, the ANN model could be
considered as a better tool for estimation of the moisture ratio of mushrooms.
In cases of temperature and moisture content predictions, numerous static ANN models
were trained and validated with the experimental data obtained from MVD of tomato
slices at different drying conditions. Inputs for single static ANN models were time
(ti+1), initial temperature (T0), moisture content (MC0), microwave power, and vacuum
pressure. The outputs were temperature (Ti+1) and moisture content (MCi+1) at a given
ti+1. The mean relative error (MRE) and mean absolute error (MAE) of this model for
iii
Ti+1 were 1.53% and 0.77 ?C, respectively. In the case of MCi+1, the MRE and MAE
were 11.48% and 0.04 d.b., respectively. The computation errors were found to be
significantly reduced when adding temperature and moisture content values at ti in the
input layer. The MRE and MAE for Ti+1 were 0.35% and 0.18 ํC, respectively. In
contrast, these error values for MCi+1 were 1.78% (MRE) and 0.01 d.b. (MAE). It was
noted that, however, such a model either provided lowest error in temperature prediction
or in moisture content prediction but not the lowest error in both the prediction
parameters simultaneously. The parallel static ANN model consisted of two doublehidden-
layer feed-forward ANN models were then separately trained, simultaneously
for moisture content as well as for temperature. Inputs for the ANN models were
magnetron on-off status, vacuum pressure, temperature, and moisture content at time
ti. The previous temperature and moisture content data at time ti-1, i-2, , i-n where n =
0, 10, 20, and 30 were also added to the input layer. Outputs from the ANN models
were temperature and moisture content at time ti+1. The results indicated that the static
ANN model working in parallel with the previous temperature and moisture content
data provided more accurate results and required less training time than those of single
static ANN models. The prediction errors of the parallel static ANN model for Ti+1 were
0.03% (MRE) and 0.02 ํC (MAE). In case of MCi+1, the MRE and MAE were 0.77%
and 0.01 d.b., respectively.
Simulation of the ANN model may supply essential information regarding the
temperature and moisture content of plant materials corresponding to microwave power
and vacuum pressure levels to a control system. Therefore, improved drying efficiencies
and thermal damage prevention may be achieved.