Summary and Conclusion
This first test case is really promising. We have seen that it is possible to compress a long thermal transient using SVD, achieving a reduction factor of about 20, with completely negligible differences in structural results obtain with either the regular, “full” approach, and the reduced one. Actually, the structural pass is so economical that is can be done “on the fly”, hence there is no need for storage (only pre-calculated results need to be linearly combined, hence the limiting factor for solution time becomes I/O and not CPU. For example, in my case the gain factor on elapsed time was 7 and not 22). Needless to say, the method is only applicable as long as the structure remains elastic,
Importantly, here we applied this method to the structural pass only of the thermal elastic simulation. However, once a first thermal transient has been solved, and POD vectors are obtained, there is no reason not to use that vector basis to speed up the thermal transient analysis (i.e. search for thermal solutions in this vector space, that is, solve for less than 10 unknowns instead of 105. This will be covered in a next post.
That’s all for today!