Discussion Forum

Posted: 1 decade ago 25 sept. 2013, 07:20 UTC−4

Hi, The scaling of any dependent variable has essentially two purposes: 1. Avoid that numbers in the system of equations to be solved have widely varying orders of magnitude. This improves the numerical conditioning of the problem. For this purpose, the scale of a variable needs only to be correct within several orders of magnitude, so this is seldom sensitive. 2. It is also used when checking convergence. Here it is sensitive. If, for example, the scale for a contact pressure is 1e8 [Pa] (the default), but the actual contact pressure is only 1e6[Pa], then the check for convergence will be sloppy, so the accuracy is low. So you should set it to a value which is representative of the actual contact pressure. Most variables in COMSOL as a default uses automatic scaling, which works well for many problems. The contact pressure is one of the exceptions. Note that also the displacement in a contact analysis has a suggested manual scaling which is 1% of the model size. Regards, Henrik

Posted: 1 decade ago 25 sept. 2013, 10:28 UTC−4

Hi Henrik, Thank you for your kind explanation. What should be the amount of order of magnitude that scaling has, for ex when we have a contact pressure of 5e7 Pa? What does mean "a value which is representative of the actual contact pressure"? As far as I understand it means that scaling and contact pressure should be the same. Is it right or is there any relation like the scaling should have a certain amount of order greater then contact pressure? And for the displacement in contact analysis, for the definiton that "scaling which is 1% of the model size"; Does the model size correspond to length of contact region or something else? Best wishes

Posted: 1 decade ago 25 sept. 2013, 11:09 UTC−4

[QUOTE] What should be the amount of order of magnitude that scaling has, for ex when we have a contact pressure of 5e7 Pa? What does mean "a value which is representative of the actual contact pressure"? As far as I understand it means that scaling and contact pressure should be the same. Is it right or is there any relation like the scaling should have a certain amount of order greater then contact pressure? [/QUOTE] I just meant that the contact pressure will vary. In your case, you should probably use 5e7, unless that is a more or less singular peak value in which case you could use something smaller. On the other hand, 5e7 does not differ that much from the default 1e8, so it may not make much of a difference if you change it at all. [QUOTE] And for the displacement in contact analysis, for the definiton that "scaling which is 1% of the model size"; Does the model size correspond to length of contact region or something else? [/QUOTE] The model size is the diagonal of the bounding box around your entire geometry. So if you had a 1x1x1 sized cube, you would see 1e-2*1.732 suggested in the scaling text field. Regards, Henrik

Posted: 1 decade ago 25 sept. 2013, 12:38 UTC−4

Thank you very much. Should we take into account unit? For ex, if the bounding box is 1x1x1 meter and choice for length unit in geometry section is in cm, then should we write 1e-2*1.732e2 ? Or should we always use it in meter for this scaling section? (same thing for contact pressure scaling) Best wishes

Posted: 1 decade ago 27 sept. 2013, 03:40 UTC−4

Hi, The scaling should be in the units of the model, as defined by the Unit System at the root node of your model tree. The default scalings adapt to it. Regards, Henrik

Posted: 1 decade ago 28 sept. 2013, 15:18 UTC−4

Hi, Thank you for your explanation. Best regards