When is maintenance of a bridge needed? How long does a dike remain reliable? Artificial Intelligence (AI) can improve the quality of control over large structures without significantly increasing costs. TNO is investigating how this can be done.
Would you like to know more?
Would you like to further discuss Artificial Intelligence and our applications? Please contact Henk Miedema.
Structures in our built environment age and may require thorough maintenance or even replacement. This applies, for example, to our road network and its structures, such as viaducts and bridges, and to hydraulic engineering works. We also want to know, for example, the condition of buildings that are founded on wooden poles and of industrial installations.
Keep in mind that circumstances have changed over time, such as increased traffic density. It is clear that we need to keep a constant eye on such structures, and that monitoring is becoming increasingly important. There are also specific questions, such as the consequences of earthquakes in Groningen.
Labour-intensive and complex
The current approach to this inspection work is labour-intensive and complex, partly because the early signs of a structure’s deterioration are difficult to detect. More and more measurement data is being used, such as laser scans of road pavement, photo images and sensors on structures. This makes it possible to use modern data techniques that extract information that is important for management, maintenance and replacement.
This is where AI comes into play, according to TNO researchers Henk Miedema and Willy Peelen. Miedema is responsible for TNO’s fundamental research programme in the field of structural integrity. Peelen is an infrastructure manager. They describe four types of application areas for AI.
“AI can detect changes in the data and indicate that there is a reason to be alert”
Switches or locks
The first type monitors deviations from basic patterns of moving structures, such as railway switches or locks. If, for example, they use more electrical power, this could be a signal that there is a problem. Peelen: “It concerns structures that move frequently and therefore generate large amounts of data. AI can detect changes and indicate that there is a reason to be alert.” This approach is relatively simple and is already in the pilot phase at several companies as well as at ProRail and Rijkswaterstaat.
Image analysis for crack detection
Detecting damage in images is the second area of application. Miedema: “An AI system analyses photos of a structure and indicates whether a line in the photo is a real crack or something else. If it is a crack, the system analyses its features to make suggestions regarding the cause, such as overloading or movements in the subsurface.” TNO is conducting extensive research into this AI application.
“An AI system analyses photos of a structure and indicates whether a line in the photo is a real crack or something else”
The third application is based on the fact that small changes in a structure can say something about risks. Peelen: “This could be due, for example, to subsidence or corrosion. We apply sensors to the structure that measure different variables and transmit large amounts of data. The challenge for an AI system is to detect anomalies in the data flow and to warn us when a bridge needs inspection or measures need to be taken, because a dike is becoming more unstable.”
Patterns in inspection data
As the fourth application, Peelen refers to the analysis of existing data on the ageing processes of infrastructure: “There is a lot of historical information from inspections and we want to link this to degradations that we will see over time. AI can help us to discover new connections.”
Better explanations of how AI works
Miedema mentions that there are a some specific points of interest for the application of AI in this domain: “The decisions to which AI gives input can have major consequences. We need to be able to explain how the invisible algorithms of the AI systems arrive at their recommendations.”
“We need to be able to explain how the invisible algorithms of the AI systems arrive at their recommendations”
In addition, there is sometimes very little learning data to train and improve the AI systems: “We are happy about this, of course, because an administrator tackles a structure’s degradation in time. But therefore we will have to apply advanced methods that work on the basis of limited learning sets about signals of degradation.”
Hundreds of billions of euros
Both TNO researchers point out that some developments are on the road to success, such as the detection of cracks, but that there are still few concrete applications. They say that it’s too early, that the research is still ongoing. Peelen: “But we are working hard on it. The structures built in our country are worth hundreds of billions of euros. They must remain safe and available, without driving a sharp increase of the cost.”