Experimental data-centric prediction of penetration depth and holding capacity of dynamically installed anchors using machine learning

The visualization of experimental research phenomena and the reliability of experimental data enable their utilization in validating theoretical and numerical methods. However, the existing literature on dynamically installed anchors (DIAs) lacks cohesion, resulting in a reliance on onshore driven pile design theory, as well as empirical or semi-empirical calculation methods for DIA design. Therefore, this study aims to establish an experimental database and propose an data-centric design method for DIAs. The established experimental database comprises 503 sets of experimental data from various sources, including 254 sets of field tests, 210 sets of centrifuge tests and 39 sets of 1g-model tests of seven representative DIAs (i.e., torpedo anchor, DPA, OMNIMax anchor, DEPLA, L-GIPLA, DPAIII, and fish anchor). The geometric characteristics as well as in-soil installation and loading performance of these DIAs are systematically summarized and explored. Based on this comprehensive experimental database, a novel machine learning (ML) algorithm-based approach is proposed for predicting the penetration depth and holding capacity of DIAs. This research provides a new perspective centered around existing experimental data for the design methodology of DIAs.