Nonlin benefit with different weighting

[avikde]

find previous issue

• found before that helps more at lower Qdt? check

[avikde]

Nonlin benefit debugging - describe when it is beneficial

lower power solutions nonlinBenefit(ret1, range(0, 3, length=6), range(180, 360, length=6); Φ=120, Qdt=0, Rpow=1e1, τ2eq=true) nonlin_eq_Qdt0_Rpow1e1.zip

nonlinBenefit(ret1, range(0, 3, length=6), range(180, 360, length=6); Φ=120, Qdt=1e2, τ2eq=true) nonlin_eq_Qdt1e2.zip

nonlinBenefit(ret1, range(0, 3, length=6), range(180, 360, length=6); Φ=120, Qdt=1e3, τ2eq=true) nonlin_eq_Qdt1e3.zip

nonlinBenefit(ret1, range(0, 3, length=6), range(180, 360, length=6); Φ=120, Qdt=1e4, τ2eq=true) nonlin_eq_Qdt1e4.zip

nonlinBenefit(ret1, range(0, 3, length=6), range(180, 360, length=6); Φ=120, Qdt=1e6, τ2eq=true, tol=5e-2) nonlin_eq_Qdt1e6.zip

[avikde]

From https://github.com/avikde/robobee3d/pull/140#issuecomment-583086308

Nonlinbenefit:

• chooses nonlin more for larger wings slower flapping -> NEED TO EXPLAIN THIS
• nonlin transmission helps more for efficiency

[avikde]

Nonlin benefit where

Helpful picture for nonlin benefit. Seems like the band of helpfulness is in general when T is higher.

Generally correlated with higher stroke

Qdt = 1e3

Qdt = 34e3

Qdt = 67e3

Qdt = 100e3

nonlin_phi120.pdf nonlin_phi120.zip nonlin_phi120_Qdt_0_Rpow_1e1.pdf nonlin_phi120_Qdt_0_Rpow_1e1.zip

[avikde]

Things that set some "scale" that may affect the effectiveness of the nonlin transmission

• output stiffness
• act strain limit

[avikde]

Qdt=1e4, with ko=50 nonlin_eq_Qdt1e4_ko50.zip

Same with phi 90 nonlin_eq_Qdt1e4_ko50_phi90.zip

^ bad due to a lot of -13 (infeasible?)

with phi90, ko=80 - idea to try and get a similar ratio of output to act reflected stiffness

phi90, ko=50

[avikde]

Phi=90

ko=20 with phi=90 - NO NONLIN BENEFIT opt would pick lin in this instance

i.,e. if act stiffness dominates, this is bad

ko=60, Qdt=1e4

ko=60, Qdt=5e4

Qdt moves the "preferred FL" band up and down

ko=55, Qdt=5e4

moved preferred band to the right

ko=55, Qdt=1e4

[avikde]

• Can try ko 30:60 or something
• How to pick the Qdt,FL that provides a good benefit?

[avikde]

Phi=90 redo (deltay was wrong)

Qdt=1e4, ko=30

Qdt=5e3, ko=60

Qdt=1e4, ko=60

3e4, ko=60

1e4, ko=55

1e4, ko=50

5e3, ko=55

5e3, ko=57.5

[avikde]

Observations:

• ko=30 is too low for 90deg
• ko=60 is OK, but should check intermediate
• changing Qdt moves the preferred-FL up/down
• changing ko moves the preferred region to the right?
• the preferred region in Tratio is also where it lowers the freq and increases area

Band of preferred ko seems quite small

[avikde]

nonlinbig.zip

with nonlinbig2.zip

nonlinbig4.zip

nonlinbig5.zip

[avikde]

by experimenting with openloopplot, can get an idea: nonlin transmission helps more when ko is low and mw is high

[avikde]

Try:

• create axis called ko/(ko+ka/T1^2)
• normalize w.r.t at Tratio=0 only ko changing (this for all)
• lump together different wing mass trials and create a ko/m axis

need to combine with the openloopplot at - call out a few parts of the param space nlbenefit-nonlinbig3.pdf

EDIT: some old slices here https://github.com/avikde/robobee3d/issues/141#issuecomment-585442168 but hidden now

To improve and consolidate:

• should aggregate some of the data and make a scatter3d to show that the performance is highly correlated with the koratio
• next to it show a contour plot with the aggregated data
• point out 2 regions of this contour plot; draw a cartoon of the shapes next to it

[avikde]

Openloop plot normalize based on output displacement. Automatic implemented but doesn't work that well due to dynamic effects. Manually tuned T1 scale

olplot.pdf