Bridge monitoring: what to measure, how to read data, and when to act

In short

• A bridge "breathes" daily and seasonally. Deflection, tilt, and strain that change with temperature are normal behaviour, not damage. The problem is the trend that remains after temperature is removed.
• Seven quantities are typically measured on a bridge: strain (vibrating wire sensors), deflection, tilt (inclinometers), bearing and expansion joint movement, structural temperature, acceleration and natural frequencies (MEMS accelerometers), and, on cable-stayed structures, cable force.
• A threshold on a raw value lies: on a hot afternoon it will trigger an alarm, on a freezing night it will miss a real change. Set the threshold on a compensated value or on a deviation from the temperature-dependent expected behaviour.
• A change in natural frequency is a symptom, not a diagnosis. First exclude temperature, mass, and bearings, then think about stiffness.
• A human closes the lane, not a sensor. WARNING, ALARM, and NO_DATA must each have an assigned person, action, and deadline, plus a path to exclude temperature and the channel before anyone rolls out cones.

A bridge lives daily and seasonally, and that is normal

If you manage a bridge, you know the feeling: the basic inspection was in spring, the next one is in a year, and between them you only see the structure from the road during a drive-by. "What if I miss something?" and "I do not want to close a lane without reason" are two fears that only seem to cancel each other out. Continuous measurement addresses both at once, provided you know how to read what it shows.

The first thing you see after connecting sensors to any bridge is a wave. In the morning the girder is shorter and cooler, in the afternoon longer and warmer; the span rises and falls by fractions of a millimetre or by millimetres, the support tilts to one side and returns. Add a year on top of that: in winter the structure contracts, in summer it expands, sliding bearings travel their full range back and forth. For someone seeing such a chart for the first time, the wave looks worrying. For a bridge engineer it proves that the bearings and expansion joints are working.

A documented case of a prestressed bridge about 260 m long, equipped with point sensors and fibre optic sensors, shows this clearly: after the first year of measurement, structural behaviour proved to be dominated by temperature changes, with no concerning trends and no local cracking (industry publication (2025) on hybrid systems using bridge case studies). That is the most common and best monitoring outcome: confirmation that the structure is doing exactly what it should.

The value of continuous measurement comes from two things. First, from reference: to recognise change, you need a recorded normal state, at least one full annual cycle. Second, from time: processes that lead to loss of load-bearing capacity, such as cable corrosion, a seized bearing, support settlement, or crack growth, develop over months. A quarterly periodic measurement sees two points and guesses what happened in between. A measurement every 15 minutes by default, or more often, shows the shape of the process.

Monitoring does not replace the periodic inspections required under Article 62(1) of the Polish Building Law, which must be carried out at least once a year and at least once every 5 years by persons holding construction licences in the relevant discipline. On national roads, this obligation is fulfilled through inspections under the GDDKiA instructions: routine inspections during drive-bys, basic inspections at least once a year, extended inspections at least once every 5 years, and, for specified structures, detailed inspections and expert assessments. Managers of other roads have the same inspection obligation under Article 62 and the Act on Public Roads; bridge structures are kept in a separate register under road regulations, not in the building logbook (Article 64(2)).

An inspection is the eye and the hammer: loss of concrete, rust, insulation condition, drainage. Monitoring is the behaviour of the structure under load and temperature between inspections. Each tells you something different; together they give the full picture. We describe where monitoring fits into the overall maintenance system in the main pillar of the structural monitoring series.

What this means for you: between inspections, the bridge does not have to be blind, and the first year of measurement usually gives you what you need most: documented "everything is within limits", which you can take to management and the inspector.

What to measure on a bridge: element → quantity → sensor

The sensor set on a bridge does not come from a catalogue, but from two questions: which element is likely to fail first, and what symptom can be measured earliest. A prestressed girder does not "crack suddenly" - first the strain changes in the critical section. A support does not "tip over" - first its tilt changes by tenths of a milliradian. A bearing does not "seize overnight" - first its movement stops keeping up with temperature. MEMS accelerometers are used for dynamics, and vibrating wire sensors, inclinometers, and displacement sensors for static behaviour.

A few practical notes on the table. A vibrating wire sensor is a transducer in which the measured quantity is read from the vibration frequency of a tensioned wire, which gives stability over decades and resistance to long cable runs. That is why it remains the standard for strain in elements that must be observed over the full life of the structure. We explain why this is so and how temperature compensation works for such a sensor separately in the post on vibrating wire sensors. Vibrating wire sensors are not suitable for dynamics, where MEMS are used instead.

Deflection is the most intuitive quantity for an asset owner and the hardest to measure continuously, because it requires a reference point below the span. Where none exists, such as over a river or valley, deflection is determined indirectly, from the tilt profile along the span or from strain, and you must remember that it is a calculated quantity, not a directly measured one.

Temperature is not "just another channel". It is the background without which the rest of the table cannot be interpreted. It is measured at several points in the cross-section, top of the deck, web, and bottom, because the temperature gradient through the depth of the section bends the span independently of the mean temperature.

Where to mount sensors

Critical sections from the structural analysis, midspan, regions over supports in continuous systems, top and bottom of the same section so you can distinguish bending from a change in mean temperature, every support with a reason to settle, every bearing with clear movement. Fewer points in the right places are worth more than many points in random places. If you have the design documentation, the list of control sections is already there, because the designer knows where the smallest safety margin is.

Temperature as background: the multi-axis chart and why thresholds lie without it

An illustrative example to feel the scale. Take a span 100 m long and an annual temperature range of the structure of 40 °C. Eurocode 2 takes the linear expansion coefficient of concrete as 10×10⁻⁶ per kelvin unless more accurate data are available (PN-EN 1992-1-1, clause 3.1.3). Change in span length:

ΔL = L × α × ΔT = 100 000 mm × 10×10⁻⁶ × 40 = 40 mm

Forty millimetres of movement at the bearings and expansion joints over a year, from temperature alone, with no technical change whatsoever. The pier tilt also changes as its sunny side warms, and the strain in a sensor embedded in concrete changes between morning and afternoon. Each of these quantities has a thermal component, often larger than the change you are looking for.

This leads to three consequences for monitoring.

First: a threshold on a raw value is useless. If you set an alarm at bearing movement "greater than 30 mm", you will get it on every July day and learn to ignore it. Then in winter, when the bearing should return by 30 mm and only returns by 10, the threshold will not trigger, because the absolute value is small. A threshold only makes sense on a compensated value or on the difference between measured behaviour and what is expected for a given temperature.

Second: temperature compensation is two different things, and they must not be confused. Sensor compensation is the correction for the transducer itself reacting to heat, using the manufacturer's coefficients. Structural compensation is the removal from the measurement of the portion the bridge produces "by nature" under the influence of temperature. The first is calculated from a formula. The second is derived statistically from the reference period: the dependence of tilt or strain on temperature is plotted as a scatter of points, a straight line or hysteresis loop is fitted, and what stands out beyond the cloud is the signal.

Third: a multi-axis chart, with tilt or strain on one axis, temperature on the other, both on the same time axis, is a core engineering tool, not a dashboard decoration. If the two curves run in parallel, the bridge is breathing. If the temperature has returned to the spring value but the tilt has not, you have a trend. If the tilt curve has started to flatten relative to temperature, meaning a smaller daily amplitude at the same thermal amplitude, something has stopped moving. These three shapes are visible by eye on a chart and almost never in a table of numbers.

There is also the gradient: a temperature difference between the top and bottom of the section bends the span independently of the mean. A sunny afternoon in March, when the deck is warm and the web is cold, can produce a larger deflection than a uniformly hot August day. That is why one temperature sensor is not enough. PN-EN 1991-1-5 describes thermal actions on bridges exactly this way, as a uniform temperature component and temperature difference components (gradient) through the depth and width of the section, and both must be in the measurement if you want to separate them.

What this means for you: before you accept any threshold, ask what it is based on, a raw reading or a temperature-referenced value. That one question eliminates most false alarms before they appear.

Natural frequencies: what a change may mean, and what it may not

A natural frequency is the frequency at which a structure vibrates freely after being disturbed from equilibrium. It depends on stiffness, mass, and support conditions. On a bridge, it is determined from MEMS accelerometer records. After a heavy truck passes, the span "rings" for a moment, and the spectrum of that ringing (FFT, fast Fourier transform) shows a peak at the first bending frequency, often also at the second and the torsional one.

The appeal of this measurement is that one quantity describes the whole span, not one point. That creates the temptation to treat a drop in frequency as a damage alarm: lower stiffness, lower frequency. The physics is correct. The problem is that the same effect is also caused by:

• temperature: in winter asphalt pavement and concrete are stiffer, in summer more compliant, so the natural frequency varies seasonally;
• mass: snow, water in the box section, accumulated material, traffic standing in a queue on the span;
• support conditions: a bearing that has started to seize stiffens the system and raises the frequency, which on a chart looks like "improvement";
• excitation method: a different truck, a different path, a different amplitude, and with the same stiffness it will give a slightly different peak, especially when the structure behaves nonlinearly.

This is not a theoretical caveat. A one-year monitoring campaign on the Z24 bridge (Peeters and De Roeck, 2001) showed natural frequency changes of several percent over the annual cycle, with a clear freezing effect of the pavement, changes of the same order as the damage effects being studied on that structure. A later literature review (Xia et al., 2012) confirms that the influence of temperature on dynamic properties of structures is the rule, not the exception. That is why natural frequency is an excellent trend indicator, provided that: (1) you have records from at least one full year and know its temperature dependence, (2) you compare it with other channels (strain, tilt, bearing movement), and (3) you react to a compensated and persistent change, not to a single reading.

What a change may mean after temperature and mass have been excluded: a drop indicates loss of stiffness (cracking, corrosion of reinforcement or cables, joint damage); an increase indicates a change in support conditions (seized bearing, jammed expansion joint), meaning not better, just different. In stays and hangers, the vibration method makes it possible to estimate cable tension from the natural frequency of the cable, one of the few cases where a static quantity follows directly from a frequency.

The same accelerometer record is used to assess events: passage of an over-limit vehicle, impact on a support, impact works nearby. Here the envelope matters, whether the waveform exceeded the previously recorded maximum, and the distribution of energy in third-octave bands. We describe the method in the post on vibration monitoring during construction under PN-B-02170; on a bridge you assess the structural response, not the effect on a building, but the tools are the same.

What this means for you: a single drop in natural frequency is not a reason to close the structure or panic before management. It is a reason to look at temperature, mass, and bearings, and only then order an inspection.

Monitoring during works: proof load tests, launching, cable erection, repairs

Operational monitoring looks for slow trends against years. Construction monitoring looks for exceedances against hours. The sensors may be the same, but the thresholds, reading frequency, and notification recipients are entirely different.

Acceptance tests under load (proof load test). The technical and construction regulations for public roads (Regulation of the Minister of Infrastructure of 24 June 2022, Journal of Laws 2022 item 1518, section 109) require that a new bridge or viaduct with a theoretical span of at least 30 m undergo acceptance testing under static load, producing internal forces from 50 to 60% of the standard characteristic load, and under dynamic load. Exceptions include pedestrian and bicycle bridges, animal overpasses, soil-encased structures, masonry spans, and non-prestressed reinforced concrete spans. At the request of the road authority, any structure may be tested, especially one of unusual design. In practice, this is called a proof load test, and the methodology is described in WR-M-23 guidelines.

Measured quantities include deflection under successive vehicle positions, strain in control sections, settlement and tilt of supports, and in the dynamic test, the response to a passage at a defined speed and the natural frequencies. The result is compared with the calculation model. If the same sensors remain on the structure, the test becomes the best possible reference measurement for later operational monitoring: you have a measured response of the structure to a known load and a known "new" bridge state.

Longitudinal launching. A box girder moved on sliding bearings changes the static scheme every metre: a section that was over a support is soon in the middle of a cantilever span. Strain in control sections and support tilt, or horizontal friction forces, read in real time, make it possible to stop the operation before an exceedance becomes damage. Here 15 minutes is too slow: during launching you need a faster or continuous reading, agreed with the construction method.

Cable erection and prestressing. The sequence for tensioning stays and cables is calculated to within tens of kilonewtons. Cable force, span deflection, and pylon tilt after each stage are compared with the design values for that stage; a deviation that accumulates is a signal for correction.

Repairs and reconstruction. Lifting a span to replace bearings, milling pavement, demolishing part of the deck, or traffic narrowed to one girder with loading concentrated there, each of these situations changes loading in a way the original design did not foresee. A reference measurement before work starts, tight thresholds on the deviation from the reference, and notification to the works manager, not only to the asset owner, are the minimum.

The common denominator: in works, the threshold is set by the technologist or designer from stage calculations, for example a permissible deflection deviation in a given step, not by statistics from the reference period, because there is no reference period yet. What this means for you: if acceptance tests or reconstruction lie ahead, this is the cheapest moment to install sensors, because someone is already mounting them, and you gain a reference measurement for years.

Thresholds and procedures for the road authority: when you close a lane

A sensor does not close a lane. A person closes a lane. That is why an alarm system on a bridge is as much about threshold values as it is about a list of people and actions. We describe the general principles for setting thresholds, from the design, from the standard, from the reference period, on a value and on a trend, in the post on warning and alarm thresholds; here are the bridge-specific points.

On a bridge, two levels are most often used: WARNING and ALARM, plus NO_DATA, which can matter more than it seems. A bridge with no data for two days is a bridge that is not being observed, not a bridge that is safe. The source of the values is, in order of priority:

1. Design documentation: permissible deflections, design strains and stresses in sections, permissible bearing travel from the bearing data sheet, expansion joint range.
2. Acceptance load testing or reference measurement: the actual response of the structure to a known load.
3. Statistics from the reference period, after temperature compensation: the envelope of normal behaviour plus a margin.

An illustrative example, not a recommendation: assume the design permits bearing travel of ±60 mm, and the reference year shows that the bridge uses ±40 mm. A WARNING can be set on the temperature-compensated deviation from the expected value for that temperature, for example more than 10 mm, and an ALARM on approaching the design value or on travel stopping while temperature rises. Base the decision on a deviation that persists for several consecutive readings, not on a single point. Such numbers are selected by the designer or the engineer running monitoring for the specific structure; we are only showing the logic.

The "do I close the lane" procedure in five steps

You are afraid that you will close the structure because of a false alarm, or that you will not close it when needed. The same sequence of actions, written in advance and assigned to named people, solves both fears:

1. Take over the alert. One person confirms the alarm and from that moment is responsible for the assessment. "Everyone got the SMS" means "nobody reacted".
2. Exclude the channel. Adjacent sensors, NO_DATA state, communication log from the device, work on the sensor or logger. A single channel without confirmation from others is usually a measurement path fault, not a structural one.
3. Exclude temperature. A multi-axis chart covering the last 7 days and the same period a year earlier. If the deviation fits the temperature dependence from the reference period, it is breathing.
4. Exclude known loading. Works on or near the structure, passage of an over-limit vehicle, a queue on the span, snow, water in the box section.
5. Is the exceedance still unexplained? If the ALARM persists after steps 2-4 and is confirmed by a second channel, such as support tilt together with strain in the span above it, you order an inspection within hours and restrict traffic according to the authority's procedure: tonnage, lane, or, in the last resort, the structure itself.

WARNING alone does not close anything. An ALARM on one channel, explained by temperature or works, also does not. You close when the exceedance is confirmed, unexplained, and persistent, and you have written that rule down before the phone rings.

Two rules decide in practice whether this works. Every silence has a deadline and a reason - an alarm silenced during works must return on its own when the works end. After the first full year, thresholds are tuned: what looked like a reasonable margin in March may cause false alarms in July and miss a real change in January.

The record of these actions, who confirmed, what they checked, when they silenced it, is documentation for the asset owner in the event of an inspection or incident. The inspection report says what the inspector saw on one day; the alarm history says what the structure and the people did over a year. We describe how such a record later stands up as evidence in the post on monitoring data as evidence in a dispute.

Hybrid systems: DFOS for crack detection, point sensors for trends

Point sensors answer the question "what is happening over time?" in selected sections, every 15 minutes by default, or more often, with an alarm. They do not answer the question "where exactly?" Between sections there is a blind zone. On concrete bridges, especially prestressed ones, the gap is filled by periodic DFOS (distributed fibre optic sensing) measurements: a sensor laid along the element, read by an interrogator by the measurement team, gives a strain distribution along the full length and makes it possible to detect and locate cracking. A prestressed bridge of about 260 m described in the literature had more than 1 000 m of such sensor and 12 point sensors in control sections; fibre optic measurements did not reveal local cracks (industry publication (2025) on hybrid systems).

The division of roles is simple: point sensors for changes over time, trends, and early warning; DFOS for distribution along the length, damage localisation, and verification. Results from periodic DFOS sessions are compared with trends from point sensors, and an alarm from the automatic part can be a reason to order an additional session. A fibre optic line laid during construction can still be read years later, so the installation decision does not force an immediate purchase of measurements. Details, arguments for, and limitations are in the post on hybrid systems.

What this looks like in Inclify

The automatic part described in this post, vibrating wire sensors on girders, inclinometers on supports and pylons, displacement sensors on bearings, thermistors, MEMS accelerometers, is handled by the Inclify platform. Loggers send readings via HTTP/JSON, every 15 minutes by default, or more often; the platform accepts any interval and calculates data completeness against the actual cadence. If loggers are already operating on the bridge, connection usually takes a few days, without replacing sensors. On the chart dashboard you overlay tilt or strain and temperature on two axes, the view that separates breathing from trend.

Each channel has WARNING and ALARM thresholds with hysteresis, and a NO_DATA alarm with a time window. E-mail or SMS notifications, and in-app notifications, go to project users according to their preferences. You confirm the alarm, with who and when recorded, and you always silence it for a defined time, for example until the end of the works. After history is collected, the platform suggests thresholds and lets you apply them. To be clear: the platform does not have an alarm for trend or automatic escalation to another person. You read the trend on the chart and in the 7/30 day risk report, and you write the substitute into the owner's procedure.

Accelerometer events are viewed as a time waveform, FFT spectrum, and 21 third-octave bands in the 1-100 Hz range, with an envelope alarm. Engineering reports include temperature compensation, correlation with temperature before and after compensation, calibration backlog, meaning deviation from reference and 7/30 day drift, and data SLA. The alarm history with confirmations and the configuration change audit log remain in the platform; raw frames from the devices stay in the communication log for the configured period. Export table data to CSV, save the chart as an image for the management note. More on the scope for bridge structures: bridge and viaduct monitoring in Inclify.

Sources and further reading

• Act of 7 July 1994 - Building Law, Articles 61, 62 and 64 (duties of the owner and manager, periodic inspections, building logbook) - isap.sejm.gov.pl.
• Regulation of the Minister of Infrastructure of 24 June 2022 on the technical and construction regulations concerning public roads (Journal of Laws 2022 item 1518), section 109 - acceptance tests under static and dynamic load - isap.sejm.gov.pl.
• WR-M-23 "Guidelines for carrying out road bridge tests under proof load" - guidelines recommended by the Minister of Infrastructure, gov.pl/web/infrastruktura/wr-m.
• Order No. 35 of the General Director for National Roads and Motorways of 28 September 2020 - instructions for carrying out inspections of road engineering structures (routine, basic, extended, detailed inspections).
• PN-EN 1991-1-5 Eurocode 1: Actions on structures - Part 1-5: Thermal actions - sklep.pkn.pl.
• PN-EN 1992-1-1 Eurocode 2: Design of concrete structures - clause 3.1.3, linear expansion coefficient of concrete - sklep.pkn.pl.
• PN-EN 1991-2 Eurocode 1: Actions on structures - Part 2: Traffic loads on bridges - sklep.pkn.pl.
• PN-B-02170:2016-12 Assessment of the harmfulness of vibrations transmitted through the ground to buildings - sklep.pkn.pl.
• Peeters B., De Roeck G., "One-year monitoring of the Z24-Bridge: environmental effects versus damage events", Earthquake Engineering & Structural Dynamics 30(2), 2001, pp. 149-171.
• Xia Y., Chen B., Weng S., Ni Y.-Q., Xu Y.-L., "Temperature effect on vibration properties of civil structures: a literature review and case studies", Journal of Civil Structural Health Monitoring 2(1), 2012, pp. 29-46.
• "Hybrid systems - introduction using bridges as an example" and "Hybrid systems - case study: implementation on a prestressed bridge" - industry publications (2025) on combining point sensors with DFOS measurements, available at shmsystem.pl.
• Manufacturer documentation for vibrating wire sensors and inclinometers: geokon.com, campbellsci.com.

What next

The best time to start is the one followed by a known load: acceptance tests of a new structure, repairs or reconstruction, winter with salt and frost, the heat season. Then the first months of measurement immediately give you a reference point, not just waves. If you manage a bridge or viaduct and want to see what a tilt chart against temperature looks like on a structure similar to yours, how alarm confirmation works, and how to build a procedure for your maintenance team, let's talk about monitoring your structure. We can start with a pilot on one structure, including sensors already installed on it. We reply within 24 hours.