A tiltmeter accuracy problem is almost always a result of a set of tiltmeters being installed incorrectly, much less about one single tiltmeter registering incorrect measurements. When eight units are installed along a retaining wall and the chain is anchored incorrectly, the resulting drift can be much bigger than the movement that is supposed to be measured. To improve accuracy, we need to understand how individual errors combine, correlate, and cancel across the array of sensors.

Resolution, repeatability, accuracy: which number to chase for a set

For a set of sensors, repeatability is much more important than resolution or the absolute accuracy of the sensor. When a pier rotates by fractions of a milliradian over a season, the most important aspect is to have a stable zero to be able to compare the readings between weeks or months. That is why a sensor with modest absolute accuracy but excellent repeatability is much better for infrastructural monitoring than a very accurate sensor with a drifting zero point.

For quasi-static rotation monitoring, repeatability is the most desired trait and absolute accuracy should be treated as secondary, because the structure is compared against its own baseline. That is why the Tiltmeter is specified with a repeatability of ±0.0008° alongside an angle accuracy that changes with range, from ±0.002° near level to ±0.25° out at ±90°.

Errors that can be controlled

Thermal drift

Since solar radiation hits one face of a bridge or a wall, the materials bend in a non-uniform way, so the whole structure rotates over the day. On an exposed steel element, the diurnal thermal swing reaches ten to twenty times the actual settlement signal. The sensor also has its own sensitivity drift in changing temperature. The impact of these changes is called thermal drift.

The fix to this problem is measuring both the tilt and the temperature. Published work on low-cost MEMS tilt sensors shows that a simple polynomial relating error to temperature and inclination can cut RMS drift by around 96%, which is why every acquisition from our sensor carries ambient temperature in the same packet. You then correlate tilt against temperature, model the diurnal component, and subtract it, leaving the slow structural trend you were after. For readers who want the mechanism in detail, we wrote a separate piece on why thermal drift happens in MEMS tiltmeters.

Mounting

A tiltmeter measures the rotation of whatever it is bolted to. If that surface flexes, corrodes at the fixing, or was never level to begin with, the sensor will report noise. Rigid mounting on a non-flexible surface, proper leveling, and a fixing that will not loosen through thermal cycling are necessary for correct measurements.

A unit bolted to a secondary bracket or a cladding panel will not give a faithful reading of the structure's tilt. Put each sensor on primary structural material, on the axis of the rotation that needs to be measured.

Vibration

Broadband vibration can bias a static tilt reading through vibration rectification error, where the DC average of an AC input shifts the apparent zero. On a bridge deck or next to plant machinery this is an offset. Our tiltmeter reports RMS and peak vibration between 0.1 and 31.25 Hz with each reading. If a node sits in a high-vibration zone, you see it in the context data and can weight or filter that channel.

The chain accuracy problem

Reconstructing a deformation profile from a chain of tiltmeters is an integration, which accumulates error. Each node contributes its own small angular uncertainty, and as you sum segment rotations along the wall or the deck to build displacement, those uncertainties add up.

For a correct reading, the chain needs to be anchored. One end has to be tied to something you trust not to move, or independently surveyed, so the reconstructed profile has a fixed datum rather than floating on the sum of its own errors.

A useful property of a set installed in one environment is common-mode rejection, since sensors that share the same daily temperature cycle and the same broad site conditions drift together, and differential measurements between adjacent nodes cancel much of that shared error. Relative movement between neighbours is far more trustworthy than the absolute measurement from one sensor, and often it is the relative movement, a joint opening, a differential settlement, that signals damage. Our Tiltmeter Chain Tool computes both absolute and differential displacement across the chain for exactly this reason.

Frequently Asked Questions

Is a more expensive, higher-accuracy tiltmeter always worth it for structural monitoring?Past a point, paying for tighter absolute accuracy buys nothing if repeatability and thermal handling are unchanged, since the structure is judged against its own baseline. Spend the budget on installation quality, temperature correlation, and a properly anchored chain.

Can software fix a badly installed tiltmeter set after the fact?No. Compensation can remove thermal drift and averaging can suppress random noise, but a sensor bolted to a flexing bracket or an unanchored chain has corrupted the measurement at the source. There is no algorithm for a bad datum.

How dense should a tiltmeter chain be?Denser is not automatically better. More nodes mean more accumulated integration error and more data to manage, and over-instrumented first deployments can lose the asset owner's confidence within a year or two. Place sensors where the rotation gradient is expected to change and where monitoring will inform decisions.

‍