Critical infrastructure

The situation calls for the special art of setting the right priorities. When an overpass needs maintenance or timely rehabilitation – before it collapses like the Carola Bridge in Dresden in the worst case – solid real-world data is required. One important criterion in considering how much strain is placed on a structure and whether repairs are needed as a result is the axle load, the force transferred from the vehicle axle to the road and thus to the structure. About half of the more than 28,000 bridge and overpass structures that are part of Germany’s federal highway network were built before 1985, when there were many fewer trucks on the road and trucks were much lighter as well. These days, it is not unusual to see gross vehicle weights of 40 metric tons and individual axle loads of ten metric tons. “While it almost doesn’t matter how many cars cross an overpass on a given day in terms of wear, excessive truck axle loads can lead to cracks and material fatigue in the structure,” says Dirk Koster, Chief Scientist at the Fraunhofer Institute for Nondestructive Testing IZFP.

A joint research project called ImaBEdge aims to lay a more solid foundation for evaluating these structures. Alongside Fraunhofer IZFP, the Fraunhofer Institute for Photonic Microsystems IPMS and Autobahn GmbH, which is part of the German federal government, seven other partners are involved. ImaB-Edge has received about 5.6 million euros in funding from the German Federal Ministry of Research, Technology and Space (BMFTR). The project’s centerpiece is a sensor system consisting of a central node, the edge gateway, to which numerous sensor edges can be connected, each one of them currently capable of serving 32 sensors in turn. “The main achievements in development lie in data reduction and evaluation right at the sensor, using high-performance electronics, and in the data model, which supplies important information so the engineers can make sound predictions of a bridge’s condition, including on site,” Koster explains. The information the sensors collect in and on the bridge represents a vast amount of data, so it is not sent into the cloud but analyzed and compressed on site using artificial intelligence. All that winds up on the operator’s server is compiled information on the condition of the overpass.

The researchers have already put the system to the test under real-world conditions on the Fraunhofer IZFP parking lot. The goal is to continuously monitor the structures to detect any deterioration in their condition early on. This way, the experts will have reliable data at their disposal at all times and can intervene promptly when needed. To achieve this, vibration and temperature sensors built into the roadbed collect 500 gigabytes of data day after day and then transmit the information via LAN or Bluetooth to the sensor edge, where it is pre-processed together with information from the weather station and a connected camera. The vibration data generated when a vehicle passes over the overpass and the conditions of the asphalt and surrounding environment can be used to estimate the axle loads of the vehicles passing by. The camera combines this data with real-world information on the vehicles, such as speed, the number of axles and the distance between them. In the edge gateway, this data and data on aspects such as the results of structural inspections is analyzed by the AI and linked to the digital model of the bridge. The team also plans to incorporate the ImaB-Edge system developed in the project into a highway overpass by late October 2025. Farther down the road, the researchers plan to use their system to detect critical conditions in railway infrastructure, tunnels and dams early on as well.

ImaB-Edge is a high-performance system that gathers data from dozens or even hundreds of sensors with precision accuracy. But with some structures, complex measurements take a backseat to speed and low power consumption. If there is any sign of a crack emerging in a bridge, monitoring needs to be started right away. Cases like these are where the MAUS monitoring system (short for Multimodal Autonomous Sensor Platform) from Fraunhofer IZFP can really shine. “The MAUS system combines the features of the edge gateway and the sensor network into a single platform, so you arrive at a practical solution within just a few days. Adjustments for successful implementation are possible within a few weeks,” says Christoph Weingard, a research scientist at Fraunhofer IZFP. “ The only requirements are a power supply, like a standard household 230-volt connection, and a communication channel such as LoRaWan or 5G/LTE.”

A solar cell can be used to provide power in areas of the structure that are remote or hard to access. Setup and installation is simple, too: The four electronic component assemblies that MAUS comprises — base module with processor core, sensor module, communication module and power supply – can be plugged into each other in any order. MAUS can accommodate not only simple, standard commercially available sensors for aspects such as expansion, displacement, deflection, inclination, moisture, humidity and temperature but also special vibration, acoustic, ultrasonic or eddy current sensors. Depending on the requirements, ultrasound can be used to detect cracks, for example, while acceleration sensors can log the structure’s natural oscillation and structure-borne sound sensors detect when tension wires have split. “Using this full range of sensors in bridge construction and simply adding versatile monitoring electronics will make a crucial contribution to solving the problems we are seeing with critical infrastructure these days,” Weingard says. The data gathered is encrypted and transmitted directly to the bridge operator. A specific configuration of the MAUS system has been implemented on a bridge in Munich, for example, where it is already helping the operator reduce the time and money invested in monitoring the structure’s condition.

“Most monitoring systems are designed for concrete bridges, but there is a lack of technologies for steel bridges,” says Christoph Heinze, a research scientist at the Fraunhofer Institute for Large Structures in Production Engineering IGP. That is “The MAUS system combines the features of the edge gateway and the sensor network into a single platform, so you arrive at a practical solution within just a few days.” Christoph Weingard, Fraunhofer IZFP probably because 90 percent of all bridges and overpasses in Germany consist of concrete. Even so, the lack of technologies in this area is a problem. After all, steel bridges experience corrosion and fatigue cracking, different kinds of damage than those found in concrete bridges, where major problems include spalling or degradation from moisture. Heinze and his team aim to close this gap. In the BIMLeB (BIM-Based Life Cycle Assessment for Bridges) project, they are working with TU Dortmund University on an allin- one solution for monitoring damage in steel and steel composite bridges. “We’re looking at the entire cycle, from surveying the structure’s current condition to continuously monitoring changes, which we feed into a BIM model, to predicting how the damage will evolve, along with recommendations for actions to take,” Heinze explains. The BIM (building information modeling) model describes computer-assisted methods to represent a structure digitally. 

There are two key questions. First, how does the bridge change over time? And second, what are the conditions and loads affecting it? “To answer those questions, we are working with the Chair of Steel Construction and the Chair of Computer Graphics at TU Dortmund University to build digital models for measures like load-bearing structure calculations,” Heinze explains. The first step involves the researchers taking photos by drone. They then use photogrammetric methods to create a model from the photos. If there are especially strict requirements for the accuracy of the measurements or the structure is difficult to access, they also use other methods, such as laser scanners or mobile mapping systems. With any system, the result is a 3D point cloud, which the researchers combine with information on the individual structures to create a BIM model. The team has already created digital models of four bridges in this way, including a 20-meter railway bridge. The researchers at TU Dortmund University use artificial intelligence to predict how the damage will evolve based on the models and current sensor data. Will the damage remain relatively confined over time? Or will there be a rapid deterioration? 

A team of researchers at the Fraunhofer Institute for Experimental Software Engineering IESE and the University of the Bundeswehr in Munich also view digitalization as the solution to the bridge dilemma. Specifically, they see digital twins as the answer. A digital twin is intended to combine all of the data that exists about a specific bridge in a single location, where it can be accessed at any time. This includes everything from details on planning and construction to operational data and information on dismantling. Which materials were installed where and in what way? “A digital twin opens a window on a bridge’s current condition while also allowing for predictive analysis and efficient lifecycle management,” says Tagline Treichel, a computer scientist at Fraunhofer IESE. “Certainty increases, the structure’s lifespan is extended and predictive maintenance becomes possible.”

 Digital twins are familiar from manufacturing, where they are used to depict current conditions and predict how a system will respond to change. For their bridge monitoring activities, the researchers at Fraunhofer IESE use open-source software that they originally developed for digital twins in industry and adjust it accordingly. The biggest challenge is integrating the vast amounts of data involved. Since there are different data formats, many of them incompatible with each other, the researchers translate the data into a single standardized language. This gets even more difficult if the information only exists on paper; the team is also working on another project aimed at ways to transfer this data automatically.

The “translation” has worked so far. The researchers have already digitalized five bridges. The Schwindegg bridge in the Mühldorf am Inn district is one of them. It also contains nearly 140 sensors that gather data on acceleration, expansion, weather and more, which also goes into the digital twin. So is it realistic to think all of Germany’s bridges could go digital in the next few years? “It’s definitely feasible from a technological perspective. The bases are there, and initial pilot projects show how it could work. The real challenge is coordinating all of the stakeholders involved, since responsibility for bridge infrastructure cuts across different structures at the federal, state and municipal government levels,” Treichel says.

How can Germany’s aging bridges be monitored on a large scale? By listening closely, say researchers at Fraunhofer IDMT.

Some bridges are so broken down that they practically shout it out. “When welded crossbeams supporting the structure break, they make an ear-splitting crash that can be heard for quite some distance around,” explains research scientist Olivia Treuheit. “We want to keep things from getting to that point,” she says. Together with her team from the Fraunhofer Institute for Digital Media Technology IDMT in Ilmenau, she listens closely when cars and trucks thunder across bridges, causing them to vibrate. The researchers’ goal is to identify even minor damage such as microscopic cracks or loose screws, along with soiling, early on by sound alone.

The team uses AI to analyze the audio data they collect. “This lets us filter out other kinds of noise, like traffic, rain, birdsong and music from car stereos. All we are left with is the sound of the bridge itself, the result of the vibrations created by the crossing,” Treuheit explains. Differences in sound indicate potential damage even before it becomes visible.
 

The goal is ongoing acoustic monitoring of at-risk bridges more than 25 years old. To accomplish that, the researchers plan to install highly sensitive, lowcost MEMS microphones on bridges in the long term. The innovative microelectronic sound converters are especially suitable for applications in which the available space and energy consumption are major factors, which is why they are used in smartphones, for example. As the first step, measurements will be taken on two test bridges in the town of Pirna, one with screw connections and the other welded. The team of researchers is receiving support in these efforts from MKP GmbH, an engineering firm that specializes in technologies to monitor bridges through measurements. Treuheit comments: “We want to log as much acoustic variation as possible, not just the ideal condition.” To achieve that, the researchers are deliberately adding dirt to road surfaces or loosening individual screws. This is the only way the AI can learn to not only detect anomalies but associate the sounds with specific causes as well. “We know from acoustic monitoring in industrial manufacturing, which is one of our areas of focus at Fraunhofer IDMT, how the sound of welds changes if they have any cracks, for example. Now, we want to put our expertise to work for bridges.”

The researchers are hoping to receive follow-up funding for their AIrBSound project, which has received an initial round of financing through June 2026 from the German Federal Highway and Transport Research Institute. This is a promising time for research in this field, as the collapse of the Carola Bridge in Dresden in September 2024 garnered widespread public attention. “It’s extremely motivating to do research on a topic with so much social and economic importance. In particular, working with other Fraunhofer institutes and connecting with experts in the areas of construction physics, prototyping and audio hardware offers tremendous potential for identifying good solutions,” Treuheit says.

Technologies such as sensor networks and digital twins can also be applied to other structures. However, laser technologies are in the lead when it comes to monitoring tunnels and railways as infrastructure elements. With tunnels, the key challenges are voids and water intrusion. How many tunnels in Germany are affected by damage like this, requiring repairs? Solid figures are lacking. The existing practice is to manually inspect tunnels, with construction crews using a specially defined hammer to tap the tunnel’s surface and listening to the reverberations. Does it sound hollow? This might sound like a makeshift solution, but it is actually a standardized, state-of-theart process. However, localizing damage is quite imprecise, as inspectors can only roughly map it on tunnel diagrams.

“We’re replacing mechanical hammers with laser pulses,” says Alexander Reiterer, a department head at the Fraunhofer Institute for Physical Measurement Techniques IPM, describing the LaserBeat project, on which he is working in tandem with Fraunhofer IGP. “To do this, we focus a laser beam so tightly that a plasma flash ignites before it hits the tunnel surface, and that in turn induces a shock wave on the surface itself.” It is almost like the researchers are tapping the surface without actually touching it. A laser microphone records the sound this makes. Just like with manual inspections, the question here is whether there is a hollow sound. The tunnel surface can be analyzed over a distance of two meters in any direction from a given site – with precision localization and, unlike in human hearing tests, objectively.

While the researchers at Fraunhofer IPM focus mainly on hardware and parts of the software, their colleagues at Fraunhofer IGP use AI to analyze the data generated. The team has already surveyed the eight-kilometer-long Albvorland Tunnel between the cities of Stuttgart and Ulm, and now two other test tunnels are to follow. Their method is not limited to tunnels, either. It is suitable for all kinds of concrete structures. “The interest we are seeing is a sign that the market has been waiting for this technology,” Reiterer says. The “laser hammer” for tunnel operators and others in the field could be ready for use two years from now.

In monitoring any structure, from bridges to tunnels and train tracks, one goal is constant: to disrupt traffic as little as possible. With tunnels, this could mean that instead of having to close the tunnel at least in part for inspections, the measurement units would be carried by trains that are passing through anyway, or in the case of road tunnels, by inspection vehicles. Making this a reality is what Reiterer is working on via new laser scanning methods. “We can now map several million 3D points per second using ultrafast modulated laser light and special rotating mirrors. One key point is that not only the geometry but also other features such as moisture can be determined,” Reiterer explains. And it can all be done essentially while simply driving by.

These kinds of methods from Fraunhofer IPM are also used for railways, especially to determine their movements. If the ballast under the tracks sinks, the position of the tracks themselves may change as well. As one example, defective concrete ties seem to have been responsible for the Burgrain train derailment in the summer of 2022, in which five people were killed and 78 people were injured, some of them severely. Installed on a train, the laser scanner can detect these kinds of displacement with outstanding accuracy. The system is functional and is already in use by Deutsche Bahn and by other operators and companies outside Germany. So far, the scans are still being performed from measurement vehicles, but the researchers are working to move the system to regularly scheduled trains. “The first step was to shrink the installation space needed. We’ve done that, and next comes the evaluable phase,” Reiterer explains. Initial tests of the miniaturized system in live railway operation are scheduled to take place over the next six to nine months. After that, the goal is to install it on regular trains serving routes around the country.

If these inspection systems are to be installed on regular trains in the future, data analysis is also needed to go with them. With trains being constantly in use, the data needs to be analyzed in real time and transmitted via radio. The MUM-Mini system from Fraunhofer IPM can achieve this. Only about the size of a shoe box, it incorporates multiple laser scanners, a camera system and a processing unit for data analysis. The system was originally developed for roads, where it detects objects such as street lights, manhole covers, curbs and different road surfaces. The researchers adjusted many different factors to adapt this initial version to railway settings and miniaturize it for use on regular trains. “Among other things, we integrated a measurement method that uses ultra-tiny mirrors to capture the reflected laser light,” Reiterer says. But the biggest challenge lies in using AI to analyze the data right in the measurement unit itself in real time. One to two years of development work will still be needed before the system can be used in regular trains.

Beyond that, another elementary aspect of track maintenance is wear affecting the upper part of the rails themselves. Known as the “heads” of the rails, these parts undergo abrasion over time, especially on the inward-facing side. Travel gets bumpier, which contributes to further wear. If microscopic cracks form, the track can crack open for hundreds of meters, a scenario that must be avoided at all costs. The current practice is to use special measurement trains for this as well. Sensors on the underside of the train project a laser line onto the rail head, and a camera takes pictures of the line. Deformation in the laser line reveals just how the rail heads are doing, with accuracy to within a fraction of a millimeter. “Less-expensive measurement technology has allowed us to significantly reduce the system’s costs, plus we have a new algorithm that analyzes the data right in the system,” Reiterer says.

Thus, the technology is paving the way for the test systems to be used on regular trains for inspecting rail heads as well. If the costs are still too high, the next place to look is substitutes for the optical systems. One option is the low-cost acceleration sensors found in any cell phone these days. Over the next few months, Reiterer and his team plan to study how they can be used. “If the head is in a poor condition, the car rocks back and forth. The characteristics of this motion can be used to calculate the condition of the heads,” Reiterer explains. So should we just stick three or four acceleration sensors onto every high-speed train to know how the tracks are doing? It’s not quite that simple. Data analysis is crucial. “The data from the sensor has a different structure with intact tracks than if spots are missing, but these patterns are individual. This means it takes huge amounts of data to train the AI,” Reiterer says. The researchers plan to obtain this data from a secondary rail operator that will allow them to attach sensors to passenger trains.

Roads and bridges are increasingly jampacked, but that isn’t all. According to current forecasts, passenger and freight traffic on the railways is also expected to double between now and 2040. Österreichische Bundesbahnen (ÖBB), the main Austrian rail operator, is looking to digitalization as the answer. How can the entire track network be scanned, evaluated and predicted using digital tools? To find out, ÖBB teamed up with about 20 partners from industry and the research sector for the Rail4Future project. The company brought Fraunhofer IZFP and the Chair of Road, Railway and Airfield Construction at the Technical University of Munich on board for the Smart Rail subproject. “Climate change is causing tracks to heat up to more than 60 degrees Celsius in some cases, a few degrees higher than was assumed when they were originally designed,” says Michael Becker, who was in charge of the subproject. “This means the tracks are under greater longitudinal tension than expected.” And that can cause warping, especially where tracks are permanently connected to the ground below them, like at stations or bridges, which in turn can prompt rail cars to jump the track.

As a result, rail operators want to know the exact local neutral temperatures for the entire track network, meaning the temperature at which a section of railway is under no tension at all. “The neutral temperature is between 20 and 30 degrees when the track is installed, but it can change over the operating lifespan,” Becker says. Commonly used measurement methods take about 30 to 40 minutes per site, which would normally make measurement impossible without track closures. But ÖBB would prefer to schedule the measurements for existing gaps in the regularly scheduled train service.

As the first step, the researchers examined existing methods of measurement: Can they be adapted for analysis on the track? Are the inspections sufficiently fast and accurate? Ultrasound and magnetic measurement methods emerged as the top contenders. This was followed by calibration in the lab at the Technical University of Munich. The researchers clamped a two-meter section of track in a test bench and used hydraulics to apply a tensile and compressive load of 70,000 kilograms. In the case of ultrasound, they especially tinkered with ways to tell material differences between track sections apart from tension in the track. For magnetic measurement, the challenge lay in making different unknown track sections comparable to each other. Once these obstacles were overcome, a week of large-scale testing followed at the track bed near the town of Rasdorf. The results are impressive: Measurements take only two minutes and require nothing more than a small measurement unit and a laptop. The methodological development is complete, and the next phase involves validation on the various rail networks.

Obtaining accurate and reliable measurements without impeding rail traffic was also the goal for the SenseTrain project, concluded in January 2025, on which the Fraunhofer Institute for Laser Technology ILT collaborated with DB Systemtechnik and three other partners. The key here was that the sensors measure track wear from inside a component – in this case, a wheel bearing cap on a high-speed train. “We’ve developed a technology we can use to incorporate sensors into 3D-printed components,” explains Tim Lantzsch, a department head at Fraunhofer ILT. “There are two key advantages. First, we get measurements from inside the component instead of just from the surface, and second, the sensors are protected against oil and other hostile environmental conditions.” Because the data not only yields information on the tracks but can also be used for predictive planning of train maintenance cycles, first-generation high-speed trains were chosen as the test fleet. The sensors measure the forces arising in the wheel truck underneath the train. If higher strain is measured only on certain sections of track, it is highly indicative of track wear. But if high forces are always found exclusively on right turns, for example, the train is likely the cause. The system has already successfully completed its first test drive, which found that recording data from a moving train works.

Infrastructure maintenance concerns more than just the track network. It is also about the trains themselves. For example, wheel sets require inspection at regular intervals. This can be done manually, by removing the wheel sets and using a small handheld device to inspect them via ultrasound – at 16,000 to 17,000 euros, a reasonably priced version that takes about 20 to 40 minutes per wheel set. But the data is not stored, which can bring legal difficulties for inspectors. Fully automated AURA wheel set testing devices offer an alternative. They store the data collected but require about a million euros in investment, plus the necessary infrastructure.

Researchers from Fraunhofer IZFP worked together with RailMaint GmbH and Evident GmbH on a project called PASAWIS, in which the partners developed a semi-automated version that comes at a price of only about 250,000 euros. “The test data is stored in DICONDE format, an open standard, so inspectors know they are on the safe side,” says Stefan Caspary, a research scientist at Fraunhofer IZFP. Inspection takes just 15 minutes, making it almost three times faster than the manual version. Another advantage is that while fully automated testing requires two types of tests – ultrasound and eddy current – so employees need training in both, the PASAWIS system developed in the project uses ultrasound alone. “But what really makes it special isn’t so much the testing method as the software,” Caspary points out. “The inspector is guided through the tests from start to finish, and the results are automatically stored in reports digitally signed by the operator. All of the testing data is also recorded, so it can be loaded again anytime, from anywhere.” The researchers have built several PASAWIS units in the meantime, and the VPI- and DB-certified system is now available from industry partner Evident.

The special funding means that money is available. The task now is to put that money to work as soon as possible and where it will have the biggest impact. Fraunhofer technology can help with this – and with minimizing the inconvenience to travelers.