Hysteresis and non-linearity are both quoted as a percentage of full scale, both are small, and both are routinely confused. They are measured in the same run and they mean different things: non-linearity is how far the curve departs from a straight line, hysteresis is how far the up-curve departs from the down-curve.
A cell can be perfectly linear and still have hysteresis, and the other way round. Only one of them depends on how fast you load it.
The sequence that produces a valid number
| Step | Load | Purpose |
| Preload | Three cycles to rated capacity | Settles the fixture and the cell |
| Increasing run | 0, 20, 40, 60, 80, 100 percent | The up-curve |
| Decreasing run | 100, 80, 60, 40, 20, 0 percent | The down-curve |
| Hysteresis | Largest difference at the same load, as percent FS | The datasheet value |
| Dwell at each point | Long enough for the reading to settle | A short dwell hides hysteresis |
| Temperature | Constant within 2 K | Thermal drift looks like hysteresis |
Dwell time is where most in-house calibrations go wrong. A cell read two seconds after the load is applied is still creeping, so the up-curve is measured while the reading is climbing and the down-curve while it is falling. The apparent hysteresis is then the sum of the creep in both directions, which can be several times the true value.
What causes it
Real hysteresis in a strain gauge cell comes from friction and from the gauge itself. Friction at the load introduction points, in a rocker pin or a bolted joint, produces a difference between loading and unloading that is proportional to the friction force and therefore to the load. The gauge and adhesive contribute a smaller amount through the same viscoelastic mechanism that produces creep.
This is why hysteresis is worse on a cell with a poor mounting surface and why it improves when the cell is loaded through a proper rocker or a ball. A cell bolted flat to a base that is not flat has built-in hysteresis that no calibration can remove.
Reading the three specifications together
- Non-linearity 0.02 percent, hysteresis 0.02 percent: a normal precision cell, and the combined error is not the sum in practice.
- Non-linearity 0.02 percent, hysteresis 0.10 percent: the mechanical installation is suspect, not the electronics.
- Both large and repeatable: the cell is being loaded off-axis or the base is flexing.
Checking a cell in service
A full calibration needs dead weights or a reference cell. A quick check does not: load to 80 percent, hold for a minute, record, unload, and repeat three times. The spread between the three loaded readings is repeatability; the difference between the loaded and unloaded readings at the same nominal point is hysteresis plus creep. If the spread grows over months, the installation is moving.
Record the loading rate as well as the values. A cell rated at 0.02 percent hysteresis will show 0.1 percent if it is loaded in half a second, because the spring element rings and the gauge sees a dynamic overshoot. A load cell with a documented calibration sequence is easier to defend than one with a single certificate line, and the site index lists the mounting accessories that keep the load path axial.
Standards referenced: OIML R60.
