An integrated physics-based and data-driven approach to structural condition identification

Start date

01 April 2018

End date

18 September 2019

Project website

View

Overview

Infrastructure performance is important for a nation's economy and its people's quality of life. Inadequate infrastructure is estimated to cost the UK £2 million a day, in terms of maintenance and management. To manage and protect infrastructure efficiently and effectively, the proposed project aims to develop an integrated algorithm to create a reliable and effective approach for structural health monitoring, which can find different applications.

Metallic structures are widely employed in both transport and energy infrastructure. As load transferring elements, connections in such structures are vulnerable due to stress concentrations, with localised damage being particularly hard to detect even under regular inspections. Therefore, the case study of this research will focus on the monitoring of connection condition in bolted or riveted structures.

The project will commence with an experimental investigation of a steel beam with end bolt connections under different damage scenarios due to loosening/lack-of-fit. Monitoring data from strain gauges and accelerometers will be processed to determine the beam's dynamic features. A finite element model will also be constructed and calibrated using the experimental results. Last but not least, an integrated deep learning algorithm will be developed for structural condition identification. There are two innovations in the suggested approach. Firstly, it integrates physics-based and data-driven methods. Secondly, the exploitation of deep learning enables the identification and optimisation of non-linear features, due to the existence of multiple hidden layers.

Thus, the proposed project aims to make a novel contribution to structural health monitoring with diverse applications in different structural types.

Aims and objectives

  1. Develop an integrated deep learning algorithm for structural condition identification
  2. Perform strain and acceleration monitoring tests on a physical archetype, namely a steel beam with end connections
  3. Calibrate finite element (FE) models of the physical archetype based on monitoring test results through model updating
  4. Simulate structural responses under different scenarios by using the calibrated FE model to create initial training data for the deep learning algorithm
  5. Apply the integrated algorithm to structural condition identification and assess its performance.

Funding amount

£100,803

Team