Model-based condition monitoring: bringing digital twins to existing infrastructure

Germany's federal motorway network contains around 28,000 individual bridge structures. Around 8,000 require repair or modernisation in the long term, with 4,000 of these taking priority, according to the Federal Ministry for Digital and Transport in a 2023 survey (opens in a new window). According to the German Federal Court of Auditors (opens in a new window), however, only around 40 per cent of the planned number were modernised in the past three years.

When funding and capacity are limited, setting priorities becomes the real engineering challenge. Which bridge must be refurbished first, and which still has reserves? Today, the basis for such decisions is primarily close-up inspection and computational models based on design assumptions. How a structure actually behaves has never been measured for most bridges.

The principle: measurement and modelling belong together

Measurements alone do not answer the question of condition. Strain or deflection says little while it remains unclear how much results from temperature, traffic or a genuine structural change. Conversely, a computational model is only as good as its assumptions about stiffness, prestressing and supports.

Model-based condition monitoring combines both. A physical finite element model (FE model) is compared with measurements. The process is called inverse analysis: it seeks the model parameters that best explain the measurements through calculation. The calibrated model then provides quantities engineers can work with: stress reserves, utilisation ratios and forecasts.

This is also what the then Federal Ministry for Digital and Transport means when, in its Framework Document for Digital Twins of Federal Trunk Roads (opens in a new window), it defines the digital twin as a dynamic representation with (partially) automated, bidirectional data exchange to support decisions: the real structure and its digital model remain connected. Measurement and condition data flow from the structure into the digital model, where they are analysed. The results can then inform decisions about inspection, maintenance or repairs. This two-way flow of information is known as bidirectional data exchange. A static BIM model with a dashboard therefore does not yet meet this definition.

Proof of concept: the Terfener Inn Bridge

The Terfener Inn Bridge on the A12 in Tyrol demonstrates that model-based condition monitoring works in operation. It is a three-span prestressed concrete box-girder bridge operated by Austria's ASFINAG. During construction, around seven kilometres of fibre-optic cable were embedded in the concrete to measure strain and temperature at ten-centimetre intervals. The WeStatiX SHM cloud platform recalculates the FE model daily and automatically recalibrates it.

Validation after more than four years of operation was published at the 2nd Latin American Symposium on Structural Health Monitoring in January 2026 in Santiago de Chile (Weissenbach & Penasa, 2026 (opens in a new window)). The main findings:

• The model is quantitatively reliable. During load testing, the calculated mid-span deflection differed from the measurement by less than 1 per cent. During operation, measured and calculated strains agree within ±5 per cent.
• The real structure behaves differently from the original design assumptions. Compressive stresses of -4.9 and -4.2 MPa had been predicted for the deck and bottom slabs respectively; measurements were approximately -3.0 and -2.3 MPa. Put simply, the bridge responds more stiffly to loading than expected and deforms less. The calibrated model can explain the difference in part through an effective stiffness - the entire structure's resistance to deformation - that is 31 to 36 per cent higher than originally assumed.
• The model separates temperature effects from structural behaviour. The model can also distinguish changes originating in the structure from those caused solely by temperature. Between 7 and 32 °C, temperature alone changes stresses by around ±2 MPa. Such daily and seasonal fluctuations can obscure slower changes in the structure. The model removes the temperature effect, revealing the long-term trend in structural behaviour.

Sensors are the bottleneck

The effort involved in such a twin no longer lies in ongoing analysis. Data checks, FE calculations, calibration and reporting run automatically every day at the Terfener Inn Bridge. Specialist engineers can set up this type of twin using a largely standardised, almost plug-and-play process: create the model, connect measurement data and start calibration. WeStatiX SHM is one example of such a platform. Engineering work shifts from preparing data to interpreting it.

Hardware remains the challenge. The Terfener Inn Bridge was a new structure, with fibre optics installed during concreting. Existing bridges would need sensors retrofitted: purchasing, installing, wiring and maintaining contact sensors, often requiring an access platform and closures. This effort recurs for each structure. It is feasible for a flagship bridge, but not for a portfolio of thousands of structures.

Contactless measurement

This is where another measurement technology comes in. Portable radar interferometers capture structural movements from a distance: the device transmits radar waves and measures the phase shift of echoes from components such as girder edges, joints or railings. This yields a point's displacement along the line of sight, at submillimetre resolution. Photogrammetry assigns spatial positions to the measurement points.

One example is MyMo from IDS GeoRadar, a Hexagon company. The system records up to 50 points simultaneously at up to 400 measurements per second, without sensors or reflectors on the structure. One person sets it up beneath the bridge, and static deformation and vibration behaviour are captured in a single visit. The device moves from structure to structure instead of remaining at one location.

At the Olona Bridge, a steel-composite bridge on the Autostrada 36 Pedemontana near Cassano Magnago, contactless measurement could be compared with an established method. Permanently installed accelerometers determine the bridge's characteristic vibration frequencies, known as natural frequencies. Together with Politecnico di Milano, MyMo measured the same quantities remotely using radar. The results were very close. The project therefore shows that contactless measurement can reproduce conventional acceleration measurements and natural frequencies with high accuracy. Further evidence comes from measurements of the Torino Esposizioni roof, one of the 20th century's most revolutionary roof structures, in central Turin. On the structure, which had recently undergone extensive refurbishment, the mean deviation of acceleration measurements was 0.8 per cent compared with permanently installed accelerometers.

Weissenbach & Penasa 2026

The workflow: from existing structures to decisions

Together, these two elements - mobile contactless measurement and the model - create a workflow that could work across a portfolio of existing structures. The individual steps have been tested, but the following combination has not yet been implemented end to end on a single structure.

• Digitise existing structures. An FE model is created for each structure from existing records, using parametric templates for recurring structural types.
• A one-day measurement campaign. Contactless load testing with vehicles of known weight and vibration measurements under traffic. Deflections and natural frequencies are recorded.
• Calibrate the model. The measurements feed into the computational model. Model parameters, such as the bridge's effective stiffness or support conditions, are adjusted until the calculated behaviour matches the measured behaviour as closely as possible. Inferring suitable model parameters from measurements is called inverse analysis. A platform such as WeStatiX SHM can automate this process.
• Initial assessment and baseline. The calibrated model provides a realistic assessment and a documented reference condition.
• Decide. If the structure shows reserves, repeat measurements within the inspection cycle are sufficient. If anomalies emerge, permanent monitoring is justified at that location, and the calibrated model is already available.

Zusammenspiel zwischen Messung und Sensorik

Permanent sensors are then allocated on the basis of measurement rather than a structure's prominence.

What a one-day measurement cannot do

A measurement taken on one day is a snapshot of overall structural behaviour: stiffness, supports and vibrations. It does not reveal prestressing steel corrosion, internal stresses or local cracks. Above all, it does not capture time-dependent effects: the prestressing findings at the Terfener Inn Bridge only became apparent over years of continuous data. Natural frequencies also depend on temperature, so repeat measurements are only comparable with documented boundary conditions. This approach does not replace continuous monitoring. It helps target its use.

What is needed

• Model templates instead of individual models, so that modelling does not become the next bottleneck.
• Open data formats, so that measurements can flow into analysis and asset management independently of the manufacturer.
• Services suitable for tendering: measurement campaigns, calibration and assessment as a defined service rather than a research project.

Conclusion

Model-based condition monitoring has been tested and validated in practice. Analysis can be automated. The remaining bottleneck is data collection at individual structures. If this becomes mobile and contactless, the digital twin can expand from individual projects to the existing asset portfolio: first measure, then calibrate, then decide.