Fig. 1. An overview of the thermal system under test, showing the heat transfer process within the system.

Fig. 2. Side view of the long rectangular duct with transverse ribs of equal spacing, showing all the essential parameters for heat transfer study within the Test Section of the thermal system.

Fig. 3. Control configuration of a heat transfer model, with the dimensions of ribs and their spacing (i.e. z/H, s/H and h/H) inside the rectangular duct depending mainly on the magnitude of the overall heat transfer.

Fig. 4. The flow diagram of proposed neural-fuzzy model consisting of a neural network for acquiring the knowledge between the input and output data, and a fuzzy logic reasoning mechanism for generating a more reliable suggestion for modifying the induced output values from the trained neural network.

Fig. 5. Layer structure of the neural network, with the input and output parameters stated.

Fig. 6. The training parameters of neural network, indicating a 100,000-iteration training operation would be carried out, in order to develop a fairly accurate trained neural network.

Fig. 7. The graph plot illustrating RMS error variation in a 100,000-iteration training, indicating a particular steady-state value has been reached, which can be interpreted that the trained neural network is fairly accurate up to a certain point.

Fig. 8. The fuzzification of dimensions Q/Q_o and Q/Q_r, where the fuzzy sets are represented by S (small), RS (relatively small), N (normal), RL (relatively large) and L (large).

Fig. 9. The COA defuzzification of process model parameters.

Table 1. The two key parameters of the heat transfer model (Q/Q_o and Q/Qr) HeatTransferPerformance (%)=(DesignOutput−ExperimentalOutput)×100/ExperimentalOutput

Table 2. The dimensional deviation of the heat transfer model

Table 3. The comparison of test results after each iteration