skip to content

As-is BIM model generation system for bridge inspection

There are more than 600,000 bridges along public roads in the US and 9.1% of them were classified as structurally deficient in 2016. Assuming a similar number of bridges in the EU, and a two-year inspection cycle, there is a need for more than 500,000 bridge inspections per annum across the EU and the US. Specifically, $12.8 billion was spent in the US in 2013 for bridge maintenance while £4 billion was spent in England in 2012-2013. The current practice of bridge inspection in Bridge Management Systems (BMS) is mainly a manual process: it is either based on inspectors’ on-site visual observation with traditional tools, or based on remote survey techniques combining long-hours manual modelling work in the office.

The need for innovative solutions for timely and intelligent inspection and assessment methods has led to numerous research efforts towards Building Information Modelling (BIM) bridge models. However, the BIM models are still not being used in current BMS despite the implementation of high-density surveying technologies (e.g. laser scanning and photogrammetry) for data collection over the past few years. Because (1) existing methods are not efficient to produce geometric bridge models from point cloud data (PCD) and (2) the models generated do not carry any information regarding to the bridge inspection history.

Motivated by the abovementioned challenges, PhD candidates Philipp Hüthwohl and Ruodan Lu with Laing O'Rouke reader Dr Ioannis Brilakis, propose an automated BIM model generation system that integrates the following two major novelties to produce semantically enriched BIM models: 

  • A bridge object detection and modelling software tool for automated compilation of solid 3D geometry from PCD. This includes two tasks: 1) detecting bridge objects in the form of labelled point clusters; and 2) fitting of geometric primitives and detecting relationships needed to constitute a BIM in an established data format (e.g. IFC).
  • A damage measurement tool for damage detection, classification and spatial/visual properties measurement and integration of this information into a BIM model.

They use laser scanning as well as photogrammetry to generate detailed spatial raw data with registered imagery for inspection of slab and beam-slab bridges around Cambridge. These bridges are the most representative bridge types in the UK (73% existing highway bridges and 86% of planned future bridges in the UK). They pursue what can be called a top-down strategy, to reconstruct the raw 3D geometry of the bridges with an object detection and fitting software module. The result is an enriched bridge BIM model in IFC format. In the meanwhile, they train a deep neural network from surface texture to detect the bridge defects such as cracks, spalling, or efflorescence, whose properties are directly mapped to the enriched textured BrIM model so that a semantically enriched IFC bridge is generated.

This as-is BIM model generation system aims to achieve the next step in as-is bridge modelling by taking isolated geometric 3D objects derived from the point cloud, and using them to create a complete, semantically-rich as-is model. In precise terms, the system supplements 3D as-is model with semantic information based on objects’ geometry and damage. This cannot be done with existing systems.