Why the Spring Model Doesn’t Work The Abelian Group Edward Lee, Seonhee Kim, Adrian So
Model Equation of Motion Our model equation of motion for the beam is based on the spring equation: where y is the displacement, C is the (normalized) damping constant, and K is the (normalized) spring constant. Initial conditions: y(0) = y 0 and y’(0) = 0.
Methodology Data for the vibrating beam were collected (total 9 runs) Subsequent analysis of data was performed in Matlab using statistical methods
Assumptions Physical assumptions Initial conditions were precisely as described Density, size of beam were not important (mass is not well-defined) Mathematical assumptions Errors were identically, independently distributed Gaussian N(0,σ 2 ) Errors independent of measured displacements (homoscedasticity) and of time
Matlab Analysis Parameters (C, K) for model equation allowed to vary in order to fit the observed data Fit was determined using a residual-least squares cost function Set of optimal (C, K) determined for each run
Observed Values of C and K Data for all 9 runs fit according to the procedure before Mean C: s –1, σ = s –1 Mean K: 1539 s –2, σ = 4.749s –2
Residuals Residuals clearly time dependent (structure in residual-time plot) Expected no dependence on time – variance in errors should be same for each measurement Expected scatter with no patterns Homoscedasticity assumption fails Expected no dependence on measurement – variance in errors should be same for each measurement Expected scatter with no patterns about the x-axis (fitted value)
QQ-Plot Data do not lie on the expected straight line There are more extreme values than in a Gaussian Residual distribution decays slower than in a normal distribution
Conclusions and Remarks All assumptions on the errors were violated Model decayed much faster than observed The spring model does not adequately describe the problem of the vibrating beam To less than 95% confidence, C = ± s –1 and K = 1539 ± 9.3 s –2
Improvements Tried using least-fourths fitting Penalizes large deviations from experimental data much more
Other Improvements Use beam equation and model to fit data Drawbacks: more parameters required for fitting Fourier analysis Model frequency spectrum of observed data and obtain parameters Use time- and space-dependent error terms ε = ε(x,t)