Binary classification method for efficient and accurate system reliability analyses of layered soil slopes

To enhance the computational efficiency in slope reliability analyses, a binary classification method (BCM) that takes advantage of a judgement-based strength reduction method (SRM) and an active-learning support vector machine (SVM) was developed to conduct system reliability analyses of layered soil slopes. The SVM was naturally employed to establish a binary classifier via the judgment technique to approximate the true limit state function because the stability state (e.g. stable or unstable) of a slope can be determined without calculating its exact factor of safety according to the SRM. An active-learning technique was developed to iteratively search training samples in the vicinity of the border between safe and failure domains, to update the SVM classifier with the aid of a modified initial sampling rule. Then, Latin hypercube sampling was employed, together with the obtained SVM classifier, to compute the slope system probability of failure. Three representative examples taken from the literature were employed to evaluate the performance of the proposed method. The proposed BCM shows great computational efficiency compared with existing methods. It reduced the computational cost to several minutes for simple slopes and to approximately 30 minutes for a complex real case, while maintaining a good computational accuracy.

Automated summary

A binary classification method (BCM) was proposed to enhance the computational efficiency in slope reliability analyses. By combining judgement-based strength reduction method and active-learning support vector machine, the BCM approximated the true limit state function and calculated the probability of failure in slope systems.