Put a weight on a load cell, leave it there, and the reading drifts. Take the weight off and the zero does not come back where it started. Those two behaviours have separate names, separate limits and separate causes, and a datasheet that quotes one without the other has told you half a story.
Creep is not damage. It is the viscoelastic response of the adhesive and the backing under the strain gauge, and it happens on every cell to some degree.
What the test does
| Step | Condition | What is recorded |
| Preload | Rated capacity, a few times, to settle the cell | Zero after each return |
| Load application | Rated capacity applied in a few seconds | Output immediately after loading |
| Hold | 30 minutes at constant load and constant temperature | Output at 30 minutes |
| Creep | Difference between the two, as a fraction of full scale | The number on the datasheet |
| Unload | Return to zero | Zero offset after unloading |
| Recovery | Zero reading after a further 30 minutes | Creep recovery |
| Temperature | Held within 2 K throughout | A drift in temperature looks exactly like creep |
Thirty minutes is a convention that catches the fast component of creep. A cell that is still drifting at 30 minutes has a slow component that the test does not capture, and applications that hold a load for hours need that number, which is why some specifications ask for a 2 hour or a 24 hour curve.
Where creep comes from
Three sources, and only one is the gauge itself:
- The adhesive layer. The glue between the gauge and the metal deforms slowly under shear. This is the dominant source in a well-made cell and it is temperature dependent.
- The backing film. Polyimide backing creeps less than epoxy backing, which is one reason foil gauges with a thin polyimide carrier are preferred for precision cells.
- The spring element. A poorly heat-treated beam or a bolted joint that is settling will creep mechanically, and that creep is permanent rather than recoverable.
Telling creep from something else
Load a cell, hold it, and plot the output. Creep produces a curve that flattens: fast at first, then asymptotic. A curve that keeps rising linearly is not creep, it is thermal drift from the electronics warming up. A curve that steps is a mechanical slip, usually a fixture.
Repeat the test at two temperatures. Creep roughly doubles for every 10 K above room temperature on many adhesives, so a cell that is fine at 20 °C and marginal at 50 °C will drift in a real installation and the datasheet number taken at 20 °C will not explain it.
Designing around it
If the application holds a load for minutes rather than seconds, specify creep rather than accuracy. A 0.02 percent creep over 30 minutes is 2 kg on a 10 tonne cell, and a batching system that assumes the reading is stable will overfill by that amount on every cycle. The fixes are a shorter weighing cycle, a hold-time compensation in the indicator, or a cell with a better creep specification, in that order of cost.
For creep recovery, the practical limit is the zero return: a cell that comes back within 0.01 percent of full scale after a full load has a healthy spring element. One that returns to 0.05 percent has taken a set, and no amount of recalibration will restore the original behaviour. A load cell supplied with a creep curve rather than a single number is worth the extra documentation, and the site index lists cells by capacity and by ingress protection.
Standards referenced: OIML R60.
