Debasish Jana

Earthquakes damage transportation infrastructure over wide areas, and post-event assessment is often limited by restricted ground access. Optical satellite imaging with optical flow applied to pre-and post-event satellite imagery offers a way to estimate surface displacement remotely. However, the deep learning-based optical flow model has not been systematically studied for optical satellite images. Therefore, this paper evaluates how different pretrained models of Recurrent All-Pairs Field Transforms (RAFT) affect displacement accuracy after domain-specific fine-tuning for earthquake-induced ground deformation mapping. This paper develops a variant of RAFT called RAFT-Quake for earthquake-induced ground displacement estimation on optical satellite imagery. Five pretrained RAFT variants were independently fine-tuned using an identical training configuration on the FaultDeform dataset, a synthetic satellite dataset derived from Landsat-8 imagery containing simulated strike-slip earthquake displacement fields. Each model was trained for 30 epochs and evaluated on 100 synthetic validation samples using end-point error (EPE) and the L2 norm to quantify the displacement error field. Fine-tuned models were also applied to a real WorldView (WV)-2 image pair from the 2019 Ridgecrest earthquake and compared against COSI-Corr reference displacement using mean absolute error (MAE) and root mean squared error (RMSE). The RAFT-Quake achieved a reduced EPE by 32.7-72.9% and reduced L2 norm by 32.5-64.5% across all five variants on the synthetic validation set compared with pretrained RAFT models. The fine-tuned RAFT models applied directly to the realistic WV-2 imagery achieved an MAE reduction of 22.6-83.3% and RMSE reduction of 25.2-83.3% relative to COSI-Corr reference displacement fields compared with pretrained RAFT models. Specifically, RAFT-Sintel and RAFT-KITTI achieved the closest agreement with COSI-Corr. This study develops variants of RAFT optical flow called RAFT-Quake for earthquakeinduced ground displacement estimation and systematically evaluates the influence of pretrained initialization under a synthetic-to-real fine-tuning strategy. These results support automated, wide-area displacement mapping for post-earthquake assessment of transportation infrastructure when field access is limited, and identify RAFT-Sintel and RAFT-KITTI models as reliable choices for real satellite applications.

Transportation networks play a crucial role in society by enabling the smooth movement of people and goods during regular times and acting as arteries for evacuations during catastrophes and natural disasters. Identifying the critical road segments in a large and complex network is essential for planners and emergency managers to enhance the network’s efficiency, robustness, and resilience to such stressors. We propose a novel approach to rapidly identify critical and vital network components (road segments in a transportation network) for resilience improvement or post-disaster recovery. We pose the transportation network as a graph with roads as edges and intersections as nodes and deploy a Graph Neural Network (GNN) trained on a broad range of network parameter changes and disruption events to rank the importance of road segments. The trained GNN model can rapidly estimate the criticality rank of individual road segments in the modified network resulting from an interruption. We address two main limitations in the existing literature that can arise in capital planning or during emergencies: ranking a complete network after changes to components and addressing situations in post-disaster recovery sequencing where some critical segments cannot be recovered. Importantly, our approach overcomes the computational overhead associated with the repeated calculation of network performance metrics, which can limit its use in large networks. To highlight scenarios where our method can prove beneficial, we present examples of synthetic graphs and two real-world transportation networks. Through these examples, we show how our method can support planners and emergency managers in undertaking rapid decisions for planning infrastructure hardening measures in large networks or during emergencies, which otherwise would require repeated ranking calculations for the entire network.

In structural vibration response sensing, mobile sensors offer outstanding benefits as they are not dedicated to a certain structure; they also possess the ability to acquire dense spatial information. Currently, most of the existing literature concerning mobile sensing involves human drivers manually driving through the bridges multiple times. While self-driving automated vehicles could serve for such studies, they might entail substantial costs when applied to structural health monitoring tasks. Therefore, in order to tackle this challenge, we introduce a formation control framework that facilitates automatic multi-agent mobile sensing. Notably, our findings demonstrate that the proposed formation control algorithm can effectively control the behavior of the multi-agent systems for structural response sensing purposes based on user choice. We leverage vibration data collected by these mobile sensors to estimate the full-field vibration response of the structure, utilizing a compressive sensing algorithm in the spatial domain. The task of estimating the full-field response can be represented as a spatiotemporal response matrix completion task, wherein the suite of multi-agent mobile sensors sparsely populates some of the matrix’s elements. Subsequently, we deploy the compressive sensing technique to obtain the dense full-field vibration complete response of the structure and estimate the reconstruction accuracy. Results obtained from two different formations on a simply supported bridge are presented in this paper, and the high level of accuracy in reconstruction underscores the efficacy of our proposed framework. This multi-agent mobile sensing approach showcases the significant potential for automated structural response measurement, directly applicable to health monitoring and resilience assessment objectives.

Many networks, such as transportation, power, and water distribution, can be represented as graphs. Crucial challenge in graph representations is identifying the importance of graph edges and their influence on overall network efficiency and information flow performance. For example, important edges in a transportation network are those roads that, when affected, will significantly alter the network's overall efficiency. Commonly used approach to finding such important edges is ``edge betweenness centrality'' (EBC), an edge ranking measure to determine the influential edges of the graph based on connectivity and information spread. Computing the EBC utilizing the common Brandes algorithm involves calculating the shortest paths for every node pair, which can be computationally expensive and restrictive, especially for large graphs. Changes in the graph parameters, e.g., in the edge weight or the addition and deletion of nodes or edges, require the recalculation of the EBC. As the main contribution, we propose an approximate method to estimate the EBC using a Graph Neural Network (GNN), a deep learning-based approach. We show that it is computationally efficient compared to the conventional method, especially for large graphs. The proposed method of GNN-based edge ranking is evaluated on several synthetic graphs and a real-world transportation data set. We show that this framework can estimate the approximate edge ranking much faster compared to the conventional method. This approach is inductive, i.e., training and testing are performed on different sets of graphs with varying numbers of nodes and edges. The proposed method is especially suitable for applications on large-scale networks when edge information is desired, for example, in urban infrastructure improvement projects, power, and water network resilience analyses, and optimizing resource allocations in engineering networks.

In civil, mechanical, and aerospace structures, full-field measurement has become necessary to estimate the precise location of precise damage and controlling purposes. Conventional full-field sensing requires dense installation of contact-based sensors, which is uneconomical and mostly impractical in a real-life scenario. Recent developments in computer vision-based measurement instruments have the ability to measure full-field responses, but implementation for long-term sensing could be impractical and sometimes uneconomical. To circumvent this issue, in this paper, we propose a technique to accurately estimate the full-field responses of the structural system from a few contact/non-contact sensors randomly placed on the system. We adopt the Compressive Sensing technique in the spatial domain to estimate the full-field spatial vibration profile from the few actual sensors placed on the structure for a particular time instant, and executing this procedure repeatedly for all the temporal instances will result in real-time estimation of full-field response. The basis function in the Compressive Sensing framework is obtained from the closed-form solution of the generalized partial differential equation of the system; hence, partial knowledge of the system/model dynamics is needed, which makes this framework physics-guided. The accuracy of reconstruction in the proposed full-field sensing method demonstrates significant potential in the domain of health monitoring and control of civil, mechanical, and aerospace engineering systems.