A vibrating cantilever beam near one of its modes is modelled as a Duffing oscillator…
PHYSICS · Multi-part · Fermi University Physics · Computational Physics · Research software practice, verification and validation · Research software practice, verification and validation
Scientific Computing for Physicists, Ch Are they the right equations? Solution verification, benchmarks and validation
Problem
A vibrating cantilever beam near one of its modes is modelled as a Duffing oscillator x¨+2ζω0x˙+ω02x+c3x3=mF0cosΩt, with ω0=314rads−1, damping ratio ζ=0.008 and cubic stiffness c3=4.5×107m−2s−2, the last obtained by fitting a small-amplitude frequency sweep. A verified RK4 code time-steps the equation to its steady state and, for a forcing amplitude F0/m=5.05ms−2, reads off the peak displacement at resonance: S=3.20mm. A step-halving study over three step sizes (in the asymptotic range) gives a numerical uncertainty unum=0.02mm. Refitting c3 within the scatter of the calibration data and re-running the simulation gives an input uncertainty uinput=0.15mm. A laser-vibrometer measurement of the same beam under the same forcing, averaged over 20 sweeps, gives D=3.05±0.10mm.
((a)) Compute the comparison error E=S−D.
((b)) Combine unum, uinput and uD (independent sources) into uval.
((c)) Using k=2, give the interval for δmodel and state whether the comparison provides evidence of model-form error at this forcing level.
((d)) The same ω0, ζ and c3, unchanged, are then used to predict the peak amplitude when the forcing is raised to F0/m=202ms−2 (forty times larger), a regime in which the Duffing backbone curve is known to bend enough to produce amplitude jumps and bistability. Is that prediction inside the validation domain established in part (c)? Justify.