PSI - Issue 80

Tafara E. Makuni et al. / Procedia Structural Integrity 80 (2026) 105–116 Tafara Makuni / Structural Integrity Procedia 00 (2019) 000 – 000

112 8

Figure 4: JavaFoil NACA3418 aerofoil results for Re c = 47400 - 264880 with the characteristic length as the aerofoil chord with α ∞ = -20 o – +20 o .

3.2. Flow Visualisation For the results given in Table 3, the flow field across and around the aerofoil was visualised using flow tufts. Example results are shown in Figure 5 (a) and (b). Here it can be seen that the flow tufts across the aerofoil appear fairly straight, suggesting that the flow here is attached. Towards the base of the aerofoil section, the flow tufts on the wind tunnel floor towards the trailing edge appear to follow that of a recirculation region suggesting that the flow here is separated. This region appears to be limited to the base of the aerofoil section. Figure 5 (c) shows a sketch of the flow field with the region of separated flow highlighted.

(a)

(b)

(c)

Figure 5: Flow visualisation using tufts for U ∞ < 25 ms

-1 and α ∞ = 0 o on (a) the aerofoil, (b) the tunnel floor and (c) the sketched flow field.

3.3. BRBP ML model for Aerodynamic Load Reconstruction The aerodynamic database generated using XFOIL showed the pressure distribution around the aerofoil to be well behaved on both the upper and lower surfaces which is to be expected for attached flow as discussed in Section 3.2. For a combination of M ∞ , α ∞ and h ; there is a corresponding unique pressure distribution. To predict this distribution based on these inputs alone, a ML model was developed using the entire XFOIL generated database. Using this database, 70% of the data was selected for training and the remaining 30% was used for testing. An ANN model was trained using the Levenberg-Marquardt backpropagation, LMBP. This ANN model had two hidden layers. The results were found to be accurate in reproducing the pressure distribution as shown in Figure 6. The same dataset was trained using the Bayesian regularization backpropagation, BRBP and Scaled conjugate gradient backpropagation, SCGBP