SpectraSim Documentation
Parameter · 07 Run Settings

Δp window

simulation_control.dp_plateau.window_iterations Unit: iterations Step 07 · Run Settings

The least number of iterations, after the start-up and settle iterations, over which the mean of the pressure drop is formed and judged. The program widens it to two oscillation periods when the signal oscillates more slowly.

Last updated

Theory

The convergence rule (Δp mean-stability tolerance) needs a window long enough for its mean to be meaningful. No verdict is given until WW samples exist past the start point, equation (1); before that the status is "insufficient", which is not a failure. The most recent WW samples are searched for a dominant oscillation period PP; the window actually used is NN, equation (2), capped at the samples available. The same window is what the reported pressure drop averages, and the live chart shades exactly the samples the rule used.

A longer window reduces the artificial drift a straight-line fit shows on an oscillating signal (about 1.5 % of the mean at 20 samples, 0.06 % at 500) and the statistical uncertainty of the mean; it also lengthens every run by at least that many iterations.

Equations

istart=Nup+Nsettle,navail=#{i>istart}≥Wi_{start} = N_{up} + N_{settle}, \qquad n_{avail} = \#\{i > i_{start}\} \ge W
(1)
N=min⁡ ⁣(max⁡(W, 2P), navail)N = \min\!\left(\max(W,\ 2P),\ n_{avail}\right)
(2)
wheremeaningunit
WWΔp window (this setting)iterations
NupN_{up}Upwind start-up iterations (later if the solver was restarted for a perforate update)iterations
NsettleN_{settle}Settle iterations after a restart, found from the data (MSER-5, see Upwind iterations before the switch)iterations
navailn_{avail}Samples after the start point-
PPDominant oscillation period found in the last WW samples (0 if none)iterations
NNWindow usediterations

Default and rule

Default
500
Rule

500 iterations.

Source

The window of 500 iterations limits the finite-sample drift of an exactly periodic signal (a square wave: about 1.5 % at N=20N = 20, 0.06 % at N=500N = 500), which makes shorter windows misread stable limit cycles as drifting. Settle iterations: the restart between the two solver phases disturbs the solution (pressure drop 17 % off for about 40 iterations on the two-baffle test muffler), and the disturbance must have died out before samples count; the point where it has is found from the pressure drop itself (White, 1997).

Valid range whole number > 0

Values

Whole numbers above 0. The automatic maximum number of iterations grows with it.

Short windows

A window much shorter than 500 can call an oscillating but stable mean "drifting", or accept a mean that is still wandering.

References 1

  1. White, K. P., Jr. (1997). An effective truncation heuristic for bias reduction in simulation output. Simulation, 69(6), 323-334. DOI 10.1177/003754979706900601