Adaptive Sampling Strategies for SVM?

I am an Engineer interested in creating a surrogate model of a certain phenomenon in the context of reliability engineering. Essentially my quantity of interest is the Limit state function/stability boundary (The split between stable and unstable regions of a certain system) that can be extracted from the surrogate model.

I have used SVM (Regression) previously to create a surrogate model of a ‘system’ based on a Latin Hypercube Sampling plan to which I was able to extract the stability boundary, although this required many sample points (some even redundant) and thus a high computational cost.

Essentially, my previous work treated my SVM model as black box with no prescribed intelligence.

My aim is to use an adaptive sampling strategy, with the aim of focusing ONLY on the stability boundary as opposed to creating an accurate surrogate of the WHOLE system, to increase the ‘intelligence’ of my SVM model. Are there any suggestions to any strategies in research/literature relevant to SVM?
i.e. Contour location is of interest, contrast to a global surrogate.