3D Tacoma Narrows Bridge: Torsional Flutter from Wind to Collapse

Raise the wind on a Blender-built suspension bridge: below the onset speed the deck twist dies out, above it negative aerodynamic damping makes every cycle bigger until the span fails.

Switch to 2D

Presets

Playback

Parameters

Camera

Layers

Quality

Model check

The science

What happened in 1940. The Tacoma Narrows suspension bridge (main span 853 m, deck 11.9 m wide and 2.4 m deep, plate girders) twisted itself apart on 7 November 1940 in a wind of roughly 19 m/s (about 42 mph), a speed far below a design storm. The cause was not resonance with vortex shedding (the wind did not oscillate at the deck's twist frequency) but aeroelastic flutter: motion of the deck changes the wind moment on it, and the change feeds energy into the motion. The last stage of the failure was the first torsional mode, antisymmetric about mid-span with antinodes at the quarter spans, at about 0.2 Hz (12 cycles per minute).

Model. One modal degree of freedom, the twist theta of the deck, treated as a sectional model with uniform properties per unit length:

I theta'' + 2 zeta I w0 theta' + I w02 theta = ½ rho U2 B2 CM(alpha), with the quasi-steady incidence alpha = theta + (B/2) theta'/U and CM(alpha) = c1 alpha / (1 + (alpha/a0)2), a linear moment slope that saturates (a stall-like roll-off, a0 = 0.8 rad). Expanding for small twist, the wind adds a term rho U B3 c1 theta' / 4 that has the same form as damping but the opposite sign when c1 > 0: it is negative aerodynamic damping. The structure's own damping 2 zeta I w0 wins at low wind and loses above Uc = 8 zeta I w0 / (rho B3 c1). Above Uc the amplitude grows as exp(sigma t) with sigma = (rho U B3 c1/4 - 2 zeta I w0)/(2 I) until the saturating moment lets the energy input match the dissipation (a limit cycle; the graph shows the harmonic-balance estimate). In the 3D view the twist is multiplied by the mode shape sin(2 pi (x + L/2)/L) along the main span. When the twist at an antinode exceeds 30 degrees the model declares the hangers overloaded and the centre span falls (a ballistic animation, not a structural simulation).

Honest simplifications and calibration. The constants are of the right order for a long, shallow plate-girder deck but are not measured values: I = 5e5 kg m2/m is an assumed modal inertia including the cables' share, structural damping 0.5 % is typical for a welded or riveted steel suspension bridge, and the moment slope c1 = 0.65 per radian was chosen so that the unaided deck flutters near the observed 19 m/s. That makes the onset wind a calibration, not a prediction. Real flutter analysis (Scanlan) uses frequency-dependent flutter derivatives A2*(K) measured in a wind tunnel, couples twist with vertical bending, includes turbulence and the cables' geometric stiffness, and the real 1940 motion also involved a change of mode (from a vertical mode to the torsional one). Quasi-steady theory is only a rough guide for such a bluff section. The wind here is steady across the deck with small gust modulation. Vertical bending, cable sag changes and the real structure's non-linear stiffness are not modelled. The collapse is an illustration, not a prediction of how pieces fall.

Why it does not happen to modern bridges. Deeper trusses or streamlined box girders (low c1, the "Open, streamlined deck" preset), added damping, higher torsional frequency (which raises Uc in proportion to w0) and wind-tunnel testing of section models. In the model the onset speed scales with zeta and w0 and inversely with c1, so you can test each remedy.

Checks. (1) The numerical growth rate at three wind speeds against the closed-form sigma; (2) the zero crossing of the measured growth rate against the closed-form Uc; (3) the limit-cycle amplitude of the full nonlinear integration against the harmonic-balance estimate (energy in = energy out over one cycle, solved by quadrature); (4) the energy balance: change of kinetic plus elastic energy equals the work of the wind moment minus the work dissipated in damping.

Loading model...
Loading Blender model...
Deck twist at the quarter-span antinode
Growth rate versus wind speed
Limit-cycle twist versus wind speed
Made with 🟠 Blender · Three.js