Last week I realized that if I switched the wind-tunnel fan to the lower speed setting, it might be able to run at speeds lower than 100r/min. I tried it and it worked! This updated plot has points at Re values less than 10000 and they show significantly less convection than expected. As in an earlier post, there would be many possible explanations for finding too much convection, but not for finding too little.
At these low wind speeds the fan speed wanders through a +/-10% range and the convection measurements are averaged over time. I have rewritten my program to compute the convection over smaller intervals than the whole trial and included both the whole trial and smaller runs in the graph. My program also averages measurements through a trapezoidal window, which reduces the variation between the smaller intervals within an experimental trial.
Here is a link to a pdf of plots splitting the measurements in 2 through 8 pieces. Each trial's dots are distributed vertically, reflecting the noisy temperature readings.
The four lowest trials cluster near the asymptote for turbulent convection from a smooth surface, even though the rough surface asymptote would convect more heat. In the "Measured over time split into 4 intervals" chart it is seen that the plate spends most of its time over the range between the two asymptotes.
If the switching between asymptotes were correlated with fan speed variations, we would expect to see the dots for each trial on a slanted line; but they are roughly vertical. The transition between laminar and turbulent forced convective modes remains an unsolved problem for theory. Modelling the transition between these two turbulent forced convective modes may be as intractable.