Hi,
Bear with me, a fairly long question and based on the global sensitivity plot below
Can anyone suggest how to find C and D?
Is there a function that can be used as a computed measure in a global sensitivity study that stores the maximum or minimum value of another variable?
It is clear by visual inspection that the global max and min of the function is at C and D.
The optimiser will find A and B.
The reasons for this:
- The x variable moves from 0 to 360.
- The optimiser looks at the local gradient.
- To maximise the function it starts at x=0 (A) and moves toward B because the gradient is +ve in that direction
- Once it gets to B, all gradients are then -ve so it stops and reports the figure at point B
- To minimise the function it starts at x=0 (A) and realises that moving to the right is a +ve gradient and therefore does not go toward B
- The optimiser cannot go to the left and toward E in the direction of the downward gradient because the x variable given to the optimiser STARTS at 0 and progresses to 360. (even though it is part of the same curve segment when the graph is wrapped around because in my case, the x variable is angle in degrees, the optimiser cannot go below 0)
Immediate thoughts are:
- You have the global graph so what's the problem. It's C and D. Move on.
- I have a lots and lots of measures, lots of geometries, lots of load cases. Graphing them all will result in repetitive strain injury.
- The method needs to be as automated as possible.
- Why don't you change the start angle
- The model angle relative to world is arbitrary
- Geometry is variable
- Loading is variable
- I would need to know what to change the start angle to
In the instance where the function height (shown in green) that A and B lie on is within the optimiser tolerance, the optimiser will put A and B are at the same location; at A. This may in of itself not be an issue because it is telling us that the function is flat and this is useful information. The optimiser realises that in order to change the value of the measure on this flat function it has to move a very long way from zero. It doesn't do this, the optimiser is short sighted and can't see the quantum leap to C because it's too far away.
Reducing the optimiser tolerance will find point B with more accuracy.
QUESTIONS
Is there a function that can be used as a computed measure to store the maximum or minimum value of another variable?
So as the global sensitivity study moves from 0 to 360, the measure is continuously updated but only in one direction.
If there isn't, can we have one please?
Thanks