Computer-Vision based Structural Health Monitoring: Application to Buildings & Bridges

Katyal, Krishn and Somala, S N (2022) Computer-Vision based Structural Health Monitoring: Application to Buildings & Bridges. Masters thesis, Indian Institute of Technology Hyderabad..

Text
CE20MTECH14005_ Krishn Katyal_T.pdf - Submitted Version
Restricted to Repository staff only until June 2027.
Download (2MB) | Request a copy

Abstract

Artificial intelligence is revolutionizing many industries, including construction industry. Many subfields of AI can be successfully applied in construction industry to achieve better or increased profitability, efficiency, safety and security. One of the areas where AI has shown significant success with-in the construction industry is in the field of civil infrastructure condition assessment using visual recognition methods. Computer-vision based SHM (Structural Health Monitoring) technique can be used in the identification & localization of critical structural components, as well as in the detection & quantification of structural damage. To train AI algorithms, we typically require large amounts of training data that consists of images and corresponding ground truth annotations. Data can be generated in two ways, one is by acquiring real data using remote cameras or Unmanned Aerial vehicles and the other way is to generate training data can by leveraging synthetic environments. Lot of challenges persists in collecting large amount of real data, and also the images are annotated manually in most existing approaches. Hence lot of research is carried out in the recent past to leverage synthetic environments to develop a unified system for automating vision-based SHM. This thesis discusses on various different ways by reviewing some past literature, on how synthetic environment can be built and how synthetic training data can be generated from the same along with it¶s automated ground truth data. To add to this discussion further, a synthetic environment is built using Blender software and synthetic image data of a building is rendered along with its corresponding automated labelled images. The thesis just not stop at explaining the way in which automated synthetic data for SHM applications can be generated, it also explains on how deep learning algorithms are used in order to bring about automation in data processing stage also. To illustrate this automated data processing stage three datasets, one each of bridge and building available in the public forum and also the one built in Blender software were used. Fully convolutional network-based semantic segmentation algorithms were trained and tested on these three datasets and results so obtained are reported. In the end, the thesis is concluded by a briefly discussing on the challenges still persisting with computer-vision based SHM, further followed by a brief discussion on potential future research direction

[error in script]

IITH Creators: