A team at Cambridge University is applying advanced machine learning to build BIM models, but help from national infrastructure agencies is needed to realise the technique's vast potential, says Viorica Patraucean
The digital era has fundamentally changed the way infrastructure management operates.
Maintaining high-quality, resilient and sustainable infrastructure is key to economic growth. Digital BIM models* enhance the efficiency and effectiveness of asset management, providing a single source for asset data and information that everyone involved can access during design and construction, and throughout the lifecycle of a built asset.
However, while creating BIM models for future assets is relatively straightforward, documenting existing assets is far more complex and there are no efficient tools to perform this specific task. As a result, the costs outweigh the benefits of using BIM for existing infrastructure. Can it be made to work effectively for existing assets?
At the Centre for Smart Infrastructure & Construction, at the University of Cambridge, we are taking a new approach to creating BIM models for existing assets. Our initial focus is on bridges, because they generally have fewer occluded parts, making the process easier.
In a bid to bypass the costly laser scanning and man hours required to create the BIM model of an existing structure, we are designing and applying advanced machine learning techniques to create the model from videos. The potential of this method is vast, but the transition from prototype to deployment is strictly limited by the availability of training data – videos of bridges and laser-scanning point clouds to facilitate the data labelling.
To further develop this method to the wide benefit of infrastructure and asset managers, we need help to collect sufficient training data. This requires collaborating with national agencies involved in the infrastructure sector and professional surveying companies.
Working together will benefit all parties. We are witnessing significant progress in machine learning and computer vision – self-driving cars are surely only a matter of time. Deep machine learning could also provide the key to making BIM work for existing infrastructure and, in turn, enhance the future of asset management.
*We are using the term here to refer to a digital model that includes information about the (visible) geometry, part labels and materials.
Viorica Patraucean is a research associate in 3D model generation at the Centre for Smart Infrastructure & Construction, University of Cambridge
To find out more about developing BIM for existing infrastructure contact Viorica Patraucean at firstname.lastname@example.org