When designing a long-term structural health monitoring system, it is incredibly easy to focus all our attention on the health of the physical assets themselves, such as bridges, dams, pipelines and wind turbines. However, there is a critical and often overlooked prerequisite: you cannot accurately monitor asset health if you are not simultaneously monitoring data health.
Safeguarding data health requires a proactive, multi-layered engineering approach. We must continuously assess sensor performance using signal quality indicators, built-in diagnostics, and automated data validation checks. To address the inevitable challenge of sensor drift, scheduled calibration verification procedures must be implemented to maintain accuracy over time.
Furthermore, physical redundancy is one of the most effective tools available. For example, if multiple strain gauges are installed in close proximity on a major bridge structure, neighbouring sensors can be compared in real time. If one sensor begins to drift or fail, it is possible to quickly identify the outlier, use interpolation techniques to compensate for data gaps, and maintain the overall integrity of the monitoring system.
However, it is equally important to recognise that critical assets operate in highly dynamic environments. Variations in load, temperature and operating conditions directly influence sensor readings. In many cases, what appears to be sensor drift is a normal physical response to changing environmental conditions.
This is one of the reasons why purely data-driven AI models often struggle in structural health monitoring. Without physical context, it can be difficult to distinguish between a failing sensor and genuine structural degradation. By using physics-informed models, we gain the engineering context to differentiate between the two.
Ultimately, the impact of sensor drift depends heavily on the operational context and the criticality of the asset being monitored. In safety-critical applications, such as aircraft systems, even a minuscule amount of sensor drift may be unacceptable and could require immediate, corrective action. In contrast, for some infrastructure monitoring applications where the objective is to identify long-term trends and trigger routine maintenance activities, a small amount of drift may be perfectly acceptable.
By continuously validating sensor performance and evaluating both measurement and decision uncertainty side by side, organisations can ensure that their monitoring systems provide data that supports safe, reliable and cost-effective operational decisions.
This will bring together HBM, Brüel & Kjær, nCode, ReliaSoft, MicroStrain and Discom brands, helping you innovate faster for a cleaner, healthier, and more productive world.
