Digital twin reconstruction for structural analysis for infrastructure monitoring
In collaboration with
Scenario
Movyon is a leader in the development and integration of Intelligent Transport Systems, Tolling, and Infrastructure Management solutions, and a center of excellence for research and innovation within the Autostrade per l'Italia Group. The R&D department of Movyon approached Deix with the goal of experimenting with an innovative way to periodically monitor and analyze roads and highways.
With the advent of new technologies, data acquisition has become more accessible and faster, but data processing still requires precision and significant effort to ensure that collected data can be analyzed and made available to decision-makers.
The challenge
The need was to label individual elements from data collected from aerial surveys and LIDAR data aquisitions
The main challenge of the project was to digitally reconstruct a 3D model of the individual elements that make up the infrastructure, precisely identifying columns, beams, abutments, joints, and other typical components of road structures, following a well-defined classification.
The need was to create a digital twin with the highest level of conformity to reality, training algorithms to accurately "label" individual elements from data collected from aerial surveys conducted with drones through recordings and LIDAR scanning technologies.
The large amount of data collected by the drones translates into a cloud of hundreds of millions of points, whose segmentation process into its subcomponents represented the main challenge of the project.
The solution
The project has seen the use of different approaches to reach the solution. In particular, two methods have been considered
The use of Machine Learning, through the use of neural networks for automatic learning
An Unsupervised Approach, using CAD models (BIM - Building Information Modeling) to create associations between the models and drone measurements
Both approaches showed some difficulties that were overcome thanks to constant collaboration with the client.
In the case of machine learning and neural networks, the need to work with a reduced availability of labeled data led us to work together with the client to make the best use of the available information to generate an automatic tagging system capable of providing artificial intelligence with sufficient information to initiate automatic learning.
Regarding the unsupervised approach, the challenge was to create alignments and associations between the images acquired by drones and the CAD models used as a reference, which were not always exactly faithful to the real infrastructure. This was made possible through the use of ad hoc developed multistart optimization algorithms.
The R&D work performed in total synergy, which combined data labeling, cleaning and integration activities with algorithmic development, was crucial for the successful completion of the project.
Results achieved
The project was successfully completed and all the accuracy objectives for the digital reconstructions requested by the client were achieved. In particular, it was possible to create digital twins of the structures, identifying individual elements with a precision of over 90%. This paved the way for a new phase of the project aimed at leveraging what has been created to improve structural monitoring and analysis activities in a path of continuous innovation and growth.
Future developments
The project represents only the first step towards an increasingly effective digital monitoring and classification activity. The ultimate goal is to implement effective predictive maintenance tools, using this technology to perform automated scans over time. The aim is to have faster and more efficient support for structural analysis and maintenance forecasting activities, making processes more effective and quicker in a context where currently many activities are time-consuming and require a significant amount of energy.
If you are interested in learning more about how the development of advanced algorithms can support predictive maintenance and other activities, please do not hesitate to contact us for consultation. We are ready to help you develop solutions tailored to your needs