adds a quick study to test NF robustness to CST noise.

studies/performance_with_cst_noise.py

@@ -0,0 +1,114 @@
+import aerosandbox as asb
+import aerosandbox.numpy as np
+import neuralfoil as nf
+
+af = asb.KulfanAirfoil("naca0012")
+
+typical_weight_magnitude = np.median(
+    np.concatenate([np.abs(af.lower_weights), np.abs(af.upper_weights)])
+)  # 0.1487
+
+wiggle_basis_vector = (-1) ** np.arange(len(af.lower_weights))  # [1, -1, 1, -1, ...]
+
+
+@np.vectorize
+def get_airfoil_with_wiggly_noise(relative_noise: float) -> asb.KulfanAirfoil:
+    noise = relative_noise * typical_weight_magnitude
+    return asb.KulfanAirfoil(
+        name=f"{af.name} + Wiggly Noise of {relative_noise * .2:%}",
+        lower_weights=af.lower_weights + noise * wiggle_basis_vector,
+        upper_weights=af.upper_weights + noise * wiggle_basis_vector,
+        leading_edge_weight=af.leading_edge_weight,
+        TE_thickness=af.TE_thickness,
+    )
+
+
+relative_noises = np.linspace(0, 1, 21)
+wiggly_airfoils = get_airfoil_with_wiggly_noise(relative_noises)
+nf_aeros = [waf.get_aero_from_neuralfoil(alpha=5, Re=1e6) for waf in wiggly_airfoils]
+xf_aeros = [
+    asb.XFoil(
+        airfoil=waf,
+        Re=1e6,
+        mach=0,
+    ).alpha(5)
+    for waf in wiggly_airfoils
+]
+
+### Below is just plotting code
+
+import matplotlib.pyplot as plt
+import aerosandbox.tools.pretty_plots as p
+
+fig, ax = plt.subplots(2, 1)
+
+ax[0].plot(relative_noises, [aero["CD"] for aero in nf_aeros], label=f"NeuralFoil")
+ax[0].plot(relative_noises, [aero["CD"] for aero in xf_aeros], label=f"XFoil")
+ax[0].set_ylabel("$C_D$")
+ax[0].legend()
+
+ax[1].plot(
+    relative_noises,
+    [aero["analysis_confidence"] for aero in nf_aeros],
+)
+ax[1].set_ylabel("NeuralFoil\nAnalysis\nConfidence")
+
+for a in ax:
+    a.xaxis.set_major_formatter(p.mpl.ticker.PercentFormatter(xmax=1))
+
+plt.tight_layout(rect=[0.05, 0.1, 1, 0.93], h_pad=2)
+
+# Draw the airfoils
+for waf, noise in tuple(zip(wiggly_airfoils, relative_noises))[::5]:
+    # Gets the figure-coordinates of the data point on ax[2]
+    x, y = ax[1].transData.transform([noise, 0])
+    display_center = np.array(
+        [
+            ax[1].transData.transform([noise, 0])[0],
+            ax[1].transAxes.transform([0, 0])[1] - 120,
+        ]
+    )
+    display_size = np.array(
+        [
+            fig.transFigure.transform([0.1, 0])[0]
+            - fig.transFigure.transform([0, 0])[0],
+            fig.transFigure.transform([0, 0.1])[1]
+            - fig.transFigure.transform([0, 0])[1],
+        ]
+    )
+    display_lowerleft = display_center - display_size / 2
+    display_upperright = display_center + display_size / 2
+    fig_lowerleft = fig.transFigure.inverted().transform(display_lowerleft)
+    fig_upperright = fig.transFigure.inverted().transform(display_upperright)
+
+    afax = fig.add_axes(
+        [
+            fig_lowerleft[0],
+            fig_lowerleft[1],
+            fig_upperright[0] - fig_lowerleft[0],
+            fig_upperright[1] - fig_lowerleft[1],
+        ],
+        zorder=10,
+    )
+    afax.fill(
+        waf.x(),
+        waf.y(),
+        facecolor=(0, 0, 0, 0.2),
+        linewidth=1,
+        edgecolor=(0, 0, 0, 0.7),
+    )
+    afax.grid(False)
+    afax.axis("off")
+
+    afax.set_xlim(-0.05, 1.05)
+    afax.set_ylim(-0.28, 0.28)
+    afax.set_aspect("equal", adjustable="box")
+
+p.show_plot(
+    title="Effect of CST Noise on NeuralFoil Performance",
+    xlabel="Relative Noise Magnitude [%]",
+    tight_layout=False,
+    legend=False,
+    show=True,
+    savefig="performance_with_cst_noise.svg",
+)