The same load, a different number: temperature drift in pressure mapping

A blade sits on the test stand for six hours. The load steps are the same in the morning and in the afternoon. The peak pressure under the saddle is not. Nothing was loosened, nothing was reloaded. The only thing that changed is that the sun came round and the steel got warm.

Two drifts, not one

Piezoresistive film drifts in two separate ways, and they are often discussed as if they were one. The zero moves: with no load applied, the reading is no longer the value it started at. The span moves: at full load, the reading no longer matches the pressure it was calibrated against.

The span effect is mostly in the layer. The polymer matrix that holds the conductive particles expands as it warms, the particles separate slightly, the number of continuous conductive paths drops, and resistance rises. The zero effect is mostly at the boundaries: the sensing layer, the adhesive that holds it, and whatever it is pressed against expand at different rates, and the mismatch builds a small residual stress even with no external load. Different mechanisms, different corrections, and neither one is a constant.

Why blade work is unusually exposed

Three reasons, all of them about the object rather than the sensor. The assembly is large and has thermal mass, so it does not follow the air temperature — a steel saddle in direct sun runs well above ambient while the shaded face of the same blade does not. The film sits at the interface, so it takes the temperature of the two things touching it, not the temperature of the room. And the sequence is slow: a load step held for ten minutes is short compared with the time an interface of that size needs to reach equilibrium, so the film is still drifting while you record.

The result is a number that is repeatable within a run and not repeatable between runs. That is the worst combination available, because it looks like data.

What we claim, and what we do not

Our specification is ±5 % of full scale, and it is quoted at a stated temperature. Outside that band the film is not characterised, and no compensation coefficient ships with it. An uncompensated polymer film of this type can shift by a double-digit percentage of full scale across a wide temperature range; the exact figure depends on the formulation, the load level and the direction of the change, so treat any single number with suspicion — including ours.

Three ways to keep the data comparable

Soak, then zero. Fit the film, apply no load, and log until the reading stops moving at a rate you care about. Record the zero again at the end of the sequence. If the two zeros differ by more than the tolerance your conclusion depends on, the run cannot be compared with yesterday's, whatever the peaks say.

Measure the interface, not the air. A thermocouple laid beside the array costs almost nothing and turns an unknown into a recorded condition. If a logging channel is free, an unloaded reference patch of the same film does the same job and is read by the same software as the map.

Correct only the linear part. A two-temperature calibration gives a zero coefficient and a span coefficient, and removing them is legitimate. Hysteresis, creep and non-linearity are still there afterwards, and none of them are temperature effects. A correction applied as if it were complete is worse than a temperature note in the report, because it hides the uncertainty instead of stating it.

How much drift is too much

The threshold is set by the decision, not by the sensor. If you are comparing two saddle designs that differ by 30 % in peak pressure, a 5 % drift does not threaten the conclusion. If you are tracking a 3 % change across a fatigue sequence, the same 5 % makes the trend meaningless. Work out the smallest difference you need to see, then check whether the temperature swing between your first and last run can move the reading by more than half of it. If it can, change the temperature or change the comparison — not the data.

The version that sidesteps the problem

If the question is "is assembly A better than assembly B", measure them in the same session, alternating, and compare differences. Drift is common-mode: it moves both readings in the same direction by about the same amount, and the difference survives. This is the same logic as comparing clamping pressure instead of torque — the absolute value carries too many conversions between you and the quantity, and the difference carries fewer.

What belongs in the report

Peak pressure, the interface temperature at the moment of the peak, and the zero before and after. Three numbers make a measurement someone can argue with in two years. A peak on its own is a claim.