A Machine Learning Approach for Drive-By Condition Monitoring of Bridges
Year: 2023
Type: dissertation
Abstract: The deterioration of transport infrastructure as it ages poses a difficult problem for infrastructure managers who are faced with the maintenance of large transport networks with limited funding. Bridges are critical elements, particularly for road and rail transport, and failure of a bridge can have significant impacts on the functioning of transport networks while also posing a risk to safety. In recent years there have been a number of major bridge collapses which have highlighted the importance of ongoing inspection and monitoring of bridges. Traditionally, bridge inspection regimes have relied on visual approaches, or where deemed necessary, sensors can be installed on bridges to gather more detailed information on the structural performance. These methods are time consuming and labour intensive and are not scalable to facilitate regular monitoring of all bridge structures on the network. In addition to the costs, these methods also pose various logistical challenges and safety issues, meaning that there is a clear need in the industry for quicker and cheaper methods for monitoring bridge health. This thesis presents an alternative approach, which utilises in-vehicle sensors to monitor the condition of bridges. This ‘drive-by’ bridge health monitoring approach uses vehicles which already travel on the road network to indirectly measure the condition of the bridges. This cost-effective solution eliminates the logistical and safety-related challenges of existing methods, and this thesis presents solutions to some of the main technical challenges which must be overcome before the concept can be fully commercialised. Firstly, novel techniques are developed for converting the measured vehicle vibrations into features which relate to the damage-sensitive vibration response of the bridge. Secondly, machine learning techniques are leveraged to remove the influence of operational and environmental factors which can also influence the way that the bridge vibrates (e.g. vehicle speed or ambient air temperature) and also to locate and classify the damage. Results from numerical studies and laboratory experiments demonstrate that the proposed techniques can be used to detect bridge damage and represent a promising development towards the commercialisation of drive-by bridge monitoring techniques.
Author: Corbally, Robert
Language: English
Subfield: Civil and Structural Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7159903509
