An Exploratory Investigation into Image-Data-Driven Deep Learning for Stability Analysis of Geosystems

Abstract

This study investigates image-data-driven deep learning in the stability analysis of geosystems. For the purpose, the recent breakthrough in computer vision represented by the Convolutional Neural Network (CNN), which was later used as a core technique in developing Google’s AlphaGo, was studied for its capacity in assessing the stability of retaining walls. The concept used in the famous Dogs vs. Cats Kaggle challenge, in which machine learning algorithms are used to classify whether an image contains a dog or a cat, was employed. A CNN was used to analyze images for retaining walls to tell whether a wall is “cat” (safe) or “dog” (failed). For quantitative analysis, 2D images for retaining walls, organized as datasets of sizes from 500 to 200,000, were generated using a stochastic method and labeled using a traditional mechanistic method. An accuracy of 97.94% was achieved for predicting whether the retaining wall is safe via binary classifications with the CNN. Testing via the analysis of 20,000 additional images, which were independent and identically distributed, confirmed the results. Further investigations into the dataset sizes and computational power yielded quantitative insights into the influence of data and computing resources on the application of deep learning in the stability analysis of geosystems. The study, for the first time, proves the feasibility of stability analysis of geosystems with image data and provides a potential big data solution for geotechnical engineering as well as other civil engineering areas.