Gasification Theory - Biomass Gasification, Pyrolysis and Torrefaction - page 239

The input layer has two values associated with it: inputs and weights.
Weights are used to transfer data from layer to layer. In the first step, the
information is processed at the nodes and then added up (summation); the
result is passed through an activation function. The outcome is the node's
“activation value,” which is multiplied by the specific weight and transferred
to the next node.

7.5.2.8 Network Training
Training modifies the connection weights in some orderly fashion using
learning methods (Guo et al., 2001). It begins with a set of data (with inputs
and outputs targeted); the weights are adjusted until the difference between
the neural network output and the corresponding target is minimum
(Kalogirou et al., 1999). When the training process satisfies the required tol-
erance, the network holds the weights constant and uses the network to make
output predictions. After training, the weights contain meaningful informa-
tion. A back-propagation algorithm is used to train the network. Multilayer
feed-forward neural networks are used to approximate the function.
A neural network may return poor results for data that differ from the orig-
inal data it was trained with. This happens sometimes when limited data are
available to calibrate and evaluate the constants of the model (Hajek and Judd,
1995). After structuring the neural network, information starts to flow from
the input layer to the output layer according to the concepts described here.

7.5.2.9 CFD Models
CFD can have an important role in the modeling of a fluidized-bed gasifier. A
CFD-based code involves a solution of conservation of mass, momentum, spe-
cies, and energy over a defined domain or region. The equations can be writ-
ten for an element, where the flux of the just-mentioned quantities moving in
and out of the element is considered with suitable boundary conditions.
A CFD code for gasification typically includes a set of submodels for the
sequence of operations such as the vaporization of a biomass particle, its
pyrolysis (devolatilization), the secondary reaction in pyrolysis, and char oxi-
dation (Babu and Chaurasia, 2004a,b; Di Blasi, 2008). Further sophistica-
tions such as a subroutine for fragmentation of fuels during gasification and
combustion are also developed (Syred et al., 2007). These subroutines can be
coupled with the transport phenomenon, especially in the case of a fluidized-
bed gasifier.
The hydrodynamic or transport phenomenon for a laminar flow situation
is completely defined by the Navier

Stokes equation, but in the case of
turbulent flow, a solution becomes difficult. A complete time-dependent
solution of the instantaneous Navier Stokes equation is beyond today's

computation capabilities (Wang and Yan, 2008), so it is necessary to assume
some models for the turbulence. The Reynolds-averaged Navier Stokes