1. The Digital Crystal Ball: Predicting Bridge Failures Before They Happen

Recent tragic collapses like the Polcevera viaduct in Genoa have shown us that traditional visual inspections aren't enough. Your systems claim they can detect changes so subtle that engineers joke about "seeing fish hitting the structure." Can you walk us through a real-world scenario where your monitoring technology caught a critical issue that would have been invisible to conventional inspection methods?

Bridges are getting old; they now experience much higher loads due to traffic; climate change is impacting them. Traditional inspections are expensive, time-consuming and very difficult in remote regions. For example, prestressed concrete relies on hidden tendons and concrete strength; elements you cannot see.

That is why SHM is gaining momentum. Detecting dynamic properties of a structure (natural frequencies, damping ratios and mode shapes) is becoming very popular. It is the most cost-effective method to get an overview of the performance of any structure. It is done by performing OMA (Operational Modal Analysis) which is an output-only method.

This means that the structure is not artificially excited but the ambient forces are used as excitation. Since structures are very stiff and excitation forces very low, it is fundamental to be able to measure very low acceleration levels at very low frequencies to be able to perform the analysis - that is where the fish joke comes from.

In practice, synchronized vibration devices capture dynamic responses of the structure periodically to calculate natural frequencies and mode shapes. The information is uploaded to the cloud and compared with a normal operation/baseline status. If current readings are outside acceptance bands, then an alert is triggered for further investigation. In the most sophisticated projects the data are also used to update mathematical models (FEM) and simulations are run to localize and estimate the damage.

What's the most surprising early warning sign your systems have detected?

Although we have sold thousands of our sensors to system integrators for installation on bridges and structures throughout the world, use of dynamic properties for health monitoring is still quite new. Dewesoft is the provider of tools to measure these dynamic properties – we do not interpret or own the data; this is for the consultants. Thankfully, bridge failures are quite rare – but they do grab attention when they happen.

Here is an example of a dynamic system that was installed in Norway a few years ago with the backing of the Roads Authority:

While not our devices, and the supplier is no longer involved in these activities, the principle is the same: synchronised vibration devices capture dynamic responses of the structure periodically to calculate natural frequencies and mode shapes. In this case, they were able to close the bridge and divert traffic before a serious incident occurred. This is not possible with statutory annual or 5-yearly inspections.

Remember bridges do not fail every day and do not accumulate the same number of cycles as, say, a gas turbine in a very short time… our customers are doing same as SAP in the above example – continuously monitoring over longer periods and observing small changes which can herald an impending incident. By continuously monitoring, the asset owners are able to plan appropriate interventions with the least disruption and in the most cost-effective manner. Repair is certainly going to be less expensive than replacement.

2. From Ancient Masonry to Modern Marvels: The Scottish Challenge

Scotland's bridge infrastructure spans centuries—from historic stone arches to cutting-edge designs like the Queensferry Crossing. Each presents unique monitoring challenges. How does your technology adapt to monitor a 200-year-old masonry bridge versus a modern cable-stayed structure?

This is a great question. As you noted above, some of our acquisition units have a massive 160 dB dynamic range, meaning that tiny responses can be detected. To implement a system with that range would involve our “more traditional” convertors in the SIRIUS range and individual cables to sensors distributed on the structure in question.

In the past, massive structures – like masonry arch bridges – had to cope mainly with the slow-acting force of gravity. They are incredibly stiff and heavy. For these structures, static measurements such as tilt or displacement are generally sufficient. Regular spot measurements are made and diurnal/temperature effects backed out in the analysis to give an indication of how much a crack is changing, or the change in angle of a pier. Crack gauges and telltales could be used and checked on a weekly basis without concern. The measurement frequency might increase during, say, local construction activities, but spot periodic spot measurements are generally enough with such structures.

Modern bridges are optimised using Finite Element Models to provide an aesthetically pleasing structure that performs the same function as well as being able to handle modern loading and coping with the elements. A consequence is that these structures are lighter and flexible with a more “lively” response (think about the famous Millennium Bridge!). Consequently, there is not so much of a need for the massive signal range and our standard, low noise, module with a 25 μg/√Hz noise density and 96 dB dynamic range is perfectly suitable.  In the case of structural health monitoring on a stiff structure like a masonry arch bridge (or a pedestrian bridge with low excitation levels) our seismic triaxial module can be used. This has a dynamic range of 137 dB and a noise density of only 0.7 μg/√Hz which is comparable to analogue force-balance transducers.

What our customers love about the IOLITEi-3xMEMS-ACC range of devices is the way that they daisy chain together using standard CAT6 Ethernet cable. This means that there is only one connection to the data collection PC and this cable provides power, transfers data and ensures that the data points are perfectly synchronised to microsecond precision. By having the digitiser local to the sensor, our devices are immune to electromagnetic interference and have been installed on electrified railway infrastructure; they neither pick up any of the interference associated with the high voltages, nor do they impart any noise to the railway systems.

And given Scotland's notoriously harsh weather conditions, how do your systems perform when facing everything from North Sea gales to Highland freeze-thaw cycles?

Our IOLITEiw range of devices are placed in IP67 Aluminium housings. To maintain the easy installation process, we now have a new, compact version of the housing – one with glands which accommodate the connectors as well as the cables and one with rugged Harting Push-Pull connectors. The advantage of the larger gland and split seal is that premade bespoke cables with standard RJ45 connectors can be used without needing to cut and crimp on site – a bonus on a cold, wet and windy site!

Other improvements mean that status LEDs are visible externally and there is no need to open the housing on site – thus minimising potential for moisture ingress. To combat particularly harsh environments, we also offer the housing in 316 grade stainless steel. Glands are offered in a plastic material or stainless steel. The risk of sacrificial corrosion in a salty/damp atmosphere is minimised by an isolating rubber washer between the metal gland and metal housing.

To see these devices in action, here is a video showing a system, making use of Dewesoft’s acquisition hardware, in extreme conditions:

 

3. The AI Revolution: Teaching Bridges to Talk

Your partnership with companies like Cestel is creating "smart bridges" that correlate structural response with traffic loads in real-time. This sounds like we're giving bridges their own nervous system. Where do you see this technology heading in the next decade? Could we reach a point where bridges automatically adjust traffic flow or even self-report their maintenance needs to authorities?

Any continuous monitoring system will generate a lot of data for sure and machine-learning will definitely be needed to sift through the data to assist the engineer to take the final decision. I’m not sure that we are quite there yet in terms of such a smart bridge actively controlling the traffic as once the damage has been identified, dynamic altering of the loading won’t undo the damage and any restriction will be long term while a solution is devised – so I’m not sure if there’d be much advantage in such an ability. I do think that continuously monitored structures will effectively be able to “self-report” need for intervention; this can happen now. Use of OMA is already helping engineers reduce the number of accelerometers from hundreds to a handful. I do know of a company that is using machine learning to provide proactive asset management at scale. As with all things of this nature, the model needs to be trained on “defect free”/normal behaviour before any decisions can be made about interventions. If I recall correctly, they hope to build a very large database of known good condition structures in different “classes” to reduce the amount of time needed in the learning phase. It is all fascinating and moving quickly!

4. The Economics of Prevention: Making the Business Case

With Transport Scotland managing thousands of structures and local authorities facing tight budgets, the elephant in the room is cost. You've mentioned that monitoring costs are becoming "negligible in the total cost of ownership." Can you break down the real numbers for our audience? What's the ROI when you compare the cost of continuous monitoring versus emergency repairs or—worst case scenario—complete reconstruction?

Yet another excellent question! There is an example of a critical railway bridge in Norway – between Oslo and Trondheim. It is made of three identical spans – two of which were badly damaged when a pier failed due to scour washing it away. Several hundred million Euros were lost due to the bridge being out of action for many months (perhaps almost a full year?). It even made it to The Guardian here in the UK:

The Norwegian government required the bridge to be repaired and reinstated as a completely new one would take too long and cost too much money – given what had already been lost. A condition of that reinstatement was to include a monitoring system. The consultants (Ramboll) and the Norwegian University of Science and Technology (NTNU) worked together to design the system and continue to work together to gain as much information about the behaviour of the structure before any future catastrophe can occur.

Each span is fitted with the same sets of Dewesoft accelerometer/inclinometer devices in matching locations. The system comprises 30 IOLITEiw devices, a weather station and a cabinet housing a PC running Dewesoft software which also broadcasts the data to the end users for processing/analysis. The dynamic data (frequencies/mode shapes) are compared between spans and any significant difference warrants further investigation. Similarly, the average inclination signals are monitored over time and assessed in larger data sets of six and twelve months.

The total cost of the instrumentation is around €100,000 for a system that will allow the operators to understand the environmental and train movement influences on this key structure. This is less than 0.5% of the cost of the bridge being out of action for several months. I think this gives a good indication of the relative cost of monitoring.

5. The Human Factor: Bridging Technology and People

Your systems generate massive amounts of data, but ultimately, humans still need to interpret and act on this information. How do you ensure that a bridge engineer in the Scottish Highlands can make sense of complex modal analysis data at 3 AM during a storm? And looking ahead, how do you balance increasing automation with the need for human expertise and decision-making?

There are so many changes happening in the data processing side of things. For now, in order to do things “sensibly”, a balance has to be struck with the frequency of updates and the expected response time. Dynamic properties tend to change fairly slowly (unless there has been a catastrophic event). In this case, we would expect an appreciable change to be occurring over time – giving the operator/engineer a chance to plan for inspection/closure. The engineer/analysis process will have a good understanding of what is “normal”; the system would be configured to calculate dynamic properties (mode shapes, natural frequencies etc.) a number of times per day – not every few minutes – and these tracked properties allow the analysis process to identify changes in plenty of time before a significant event. Thresholds for action are determined based on what is normal and what is considered outside normal.

In my early career, I was responsible for a structural health monitoring system on an offshore oil platform. There were four pairs of accelerometers and a wave height sensor. At that time we had 9600 baud modems and used a telex machine (!) to inform the operator of the monthly status – or more frequently in the case of storms. I would retrieve the last 24 hour summary file of natural frequencies/wave heights and plot them by hand. If there was a change of something like three standard deviations, we would investigate in more detail. If the significant wave height was more than a certain value (6 m?), we would access the system and download the corresponding high speed data and examine for any changes. We’d monitor the natural frequency more closely for a couple of days afterwards to check for any long term decrease (indicating a loss of integrity). Like an oil platform, there is generally a lot of redundancy in a bridge and it is unlikely that the engineer will be required to make any sudden decisions at 3 a.m. The principles are very similar to what I was doing many years ago – just there’s a lot more data and transfer rates are much faster now.

As mentioned above, machine learning will/does play an ever increasing part in sifting through data. It will be used to determine what is “normal” and what is outside of “normal”, but I suspect it will still be down to a person to decide what the next step is -for a while at least!

 

Catch Dewesoft at Bridges Scotland on Stand E27.

• Road Expo & Bridges Scotland 2025, 26-27 November, SEC Glasgow
• Register to visit, exhibit or sponsor
• Visit the Bridges Scotland website