I spent today playing (or possibly fighting) with the Feasibility Study in BMX.
I'm doing a flotation analysis, so I've got a model of the hull and a 'waterline' plane at a parametric pitch angle and height, which trims the hull solid. I'm using the Feasibility Study to solve such that the remaining (i.e. submerged) solid has the correct mass / volume; and so that an axis through the (input) craft CofG and the submerged volume CofG is perpendicular to the waterline plane.
So far, so good; I'm pretty happy that the geometry works and tells me what I want to know. (That the initial hull design is horribly nose-heavy is another matter...).
However, the convergence algorithm seems to be useless! It invariably tells me that no solution was found: some of the time what this means is that it failed to achieve the target values by a tiny margin; other times it means it's got nowhere near. I've managed to create some 3D annotations so that I can quickly check whether convergence happened, but how do I make the actual study give me a decent answer? It feels like I could do the iteration manually just as quickly - it's only two variables and two constraints!
There only seem to be a few things to play with: percentage convergence, which I've tried from 0.2% up to 10% with no discernable difference; number of iterations, which I've tried as high as 190 and as low as 5, although 5 still seems to take ages; and the two solver methods, of which I've mostly been using the default one (Gradient something?).
It often solves better if I first manually adjust the values to get it somewhere close to the target, which I really feel shouldn't be necessary.
Also, if I ask it to plot graphs it produces a large number (~8) of them, all of which show just one point at (0,0). Much good they are.
Any pointers gratefully received, especially as I really need to finish this initial study tomorrow! Sorry, no images because a) confidentiality and b) I'm writing this at home.
Just had one thought: I'm assuming that the model is perfectly symmetrical, but if it isn't then there may be a lateral mismatch between the centres. Perhaps I should constrain the angle to be >89.9 and <90.1... but shouldn't the convergence allow it to say "close enough"?
Jonathan Hodgson wrote:
Just had one thought: I'm assuming that the model is perfectly symmetrical, but if it isn't then there may be a lateral mismatch between the centres. Perhaps I should constrain the angle to be >89.9 and <90.1... but shouldn't the convergence allow it to say "close enough"?
Nope, can't do that... apparently that constitutes "conflicting constraints".
This seems to be worse than an Optimization Study in Mechanica!