A Machine Learning-Based Decision Support System for Predicting and Repairing Cracks in Undisturbed Loess Using Microbial Mineralization and the Internet of Things
Recent years have seen a significant increase in interest across several sectors in the application of learning techniques to extract ground object information, such as soil cracks, from remote sensing high-resolution images. Out of the many technologies, the microbial-induced carbonate deposition (MICP) technology is used to inject bacteria and cementation liquid containing specific bacteria into the cracks of soil to be repaired. Calcium carbonate types of cement are produced by bacterial metabolism so that cracks in the soil could be repaired for disaster management. However, detection of cracks and taking appropriate decisions for repairing are the most fundamental issues that researchers’ attention. Machine learning algorithms can be trained to detect and predict cracks in undisturbed loess using various data sources, such as images captured using the internet of things (IoT), devices, drones, and/or ground-based sensors. These algorithms can be designed to identify different types of cracks based on their shapes, sizes, and orientations, and can be trained on large datasets of labelled crack images to improve their accuracy over time. In this paper, we propose a decision support system (DSS) that detects and predicts cracks and recommends a suitable crack repair methodology. Our results show that our system is highly accurate. Our system provides real-time recommendations to engineers working on crack repair projects in undisturbed loess, guiding them on where and how to apply microbial mineralization treatments based on the predicted crack locations and treatment effectiveness. We noted that the accuracy of the crack detection and prediction can be increased significantly (up to 9.57%) when the proposed DSS approach is considered. Moreover, if PSO is implemented as the optimization model, then we can see that the accuracy can be significantly improved by as much as 21.67% to no DSS approach and 11.32% to the DSS approach.