Write down the observation you need
“Monitor the bridge” is too broad to guide an installation. An owner might want to understand movement at an expansion joint, changes in a vibration pattern, or the response of a particular member to passing vehicles. These questions involve different locations, sampling rates and supporting records. The measurement plan should explain what evidence would support an interpretation and what evidence would remain unavailable.
The structure matters as much as the instrument. A long-span suspension bridge such as the Humber Bridge responds differently from a short reinforced-concrete deck. The Forth Road Bridge, the Millau Viaduct and the Øresund Bridge are useful reminders of structural diversity, not evidence that one monitoring configuration fits every asset. Drawings, previous inspections, repair history and the load path should inform sensor selection.
Match the sensor to its limits
Accelerometers record acceleration at their attachment points. Engineers may use those signals to examine dynamic response and estimate modal properties. A strain gauge records local deformation over a defined gauge length; converting that measurement into a stress estimate requires material, installation and temperature assumptions. Displacement sensors describe relative movement between a target and a reference, so the stability of that reference is essential.
Temperature measurements are not background decoration. Air temperature, surface temperature and temperature within a member can differ substantially. A sunlit deck can develop gradients that a single shaded thermometer will miss. Wind and traffic information may also help explain response. None of these channels measures “bridge safety” directly, and an AI model cannot recover information that the sensing arrangement never captured.
| Measurement | Useful question | Important limitation |
|---|---|---|
| Acceleration | Has dynamic response changed? | Response also depends on excitation, mounting and noise. |
| Strain | How does a local member deform? | Local readings do not describe every possible defect. |
| Displacement | How much relative movement occurs? | The reference point may itself move. |
| Temperature | Could thermal conditions explain response? | One location may miss important gradients. |
Make the measurement trustworthy
Installation quality affects everything downstream. A loose bracket can alter an acceleration signal. Moisture can affect connections. Cable damage, battery deterioration or wireless interruptions can introduce gaps. These are not minor housekeeping details: they can produce exactly the kind of abrupt pattern change an anomaly detector is designed to notice. A sensor register should record identity, position, orientation, units, range, calibration information and maintenance events.
Timing also needs explicit attention. Comparing channels with unsynchronised clocks can create misleading relationships. Sampling must suit the phenomenon being observed, and suitable filtering is needed to avoid aliasing. There is no universal sampling frequency for bridge monitoring. The required rate depends on the response of interest, the instruments and the analysis method. Keep original data where practicable, with a documented record of processing.
ISO 18649 addresses dynamic testing and investigations of bridges. ISO 16063 concerns calibration methods for vibration and shock transducers. ISO/IEC 17025 sets requirements for testing and calibration laboratory competence. These references have different scopes; citing them does not demonstrate that a particular installation is compliant. The relevant edition, application and evidence of calibration still need checking by competent people.
Prepare data for interpretation, not just storage
A useful data-quality process flags missing values, saturation, flat-lined channels, implausible jumps and clock problems before modelling begins. It preserves the distinction between a real zero and an absent observation. If a gap is filled computationally, the replacement should remain identifiable. Otherwise an apparently smooth history may conceal an instrument outage or an assumption made by the analyst.
Coverage should be described honestly. A small sensor network samples selected behaviour; it does not inspect every bearing, tendon, weld or foundation. Visual inspection, non-destructive examination and assessment answer questions that continuous sensing may not reach. The monitoring plan should say which failure mechanisms are outside its intended scope, especially when a dashboard might otherwise suggest comprehensive visibility.
Before moving on, ask whether another person could reconstruct the meaning of a reading from its metadata. If its units, reference, timing or processing history are unclear, model sophistication will not repair that ambiguity. Once the measurement chain is dependable, the next task is to establish how this particular bridge behaves across ordinary environmental and operating conditions.
Reference starting points
Consult current editions and local requirements rather than treating a reference name as a design instruction.