Solenoid life testing: what to measure, and what most rigs miss

Why this happens

A life test is usually built to answer a yes or no question: does the unit still work after a million operations. That is the least useful version of the test, because it reports the end of life and nothing about the path to it.

Counting operations measures the failure, not the degradation. The quantity that matters is not when the unit stopped but how much margin remained at a thousand, ten thousand and a hundred thousand operations. A design that loses 40 % of its force over the first ten thousand operations and then flatlines is a different design from one that loses nothing until it fails suddenly, and both look identical on a rig that only counts.

Degradation is continuous and failure is a threshold. A coil whose force falls from 0.5 N to 0.25 N over three million operations reports “still operating” for the whole of that period, and then reports failure. The useful question is how long it stayed above the load. If the load is 0.20 N and the force is falling steadily, the unit is on a schedule, and the schedule can be read from the trend rather than waited for.

The current waveform is the cheapest sensor available, and it is already wired. Force goes with the square of current, and current is set by the supply voltage and the winding’s resistance and inductance. The inductance depends on the plunger position, because the air gap is part of the magnetic circuit. That single fact turns a voltage source and a current probe into a position sensor, a temperature sensor and a force indicator, on the actual production part, with no mechanism modifications.

To make that concrete: take a 12 V winding rated 5 W, which puts its resistance near 28.8 ohms. If its inductance runs from 60 mH with the plunger out to 200 mH with it seated, the electrical time constant ranges from 2.1 ms to 6.9 ms, a factor of three. The time for the current to reach 90 % of its steady value is about 2.3 time constants, so it moves from roughly 5 ms to roughly 16 ms as the plunger travels. A shift of that size is easy to measure and easy to trend.

The waveform carries more than timing. The steady current tells you the winding resistance, and therefore its temperature, if the supply is regulated. The point where the current trace bends is the point where the plunger starts to move, because the resulting back-EMF changes the slope. If that bend moves later in the cycle, the plunger is starting later, which means friction has risen or force has fallen. If the trace never reaches its previous steady value, the winding has changed. And if the bend disappears entirely, the plunger is not moving at all, which is the failure the counting rig would eventually have reported as a number.

Acceleration is where most programmes quietly change the product. A million operations at one per second takes 11.6 days of continuous running, which is why nobody runs it that way. Raising the rate to five per second brings it down to 2.3 days. But if each operation is energised for 100 ms, then at one per second the duty is 10 %, and at five per second with the same on-time it is 50 %. The winding is now being driven five times as hard thermally, and the test is measuring a product that does not exist. Shortening the on-time keeps the duty where it was, but only down to the point where the plunger no longer has time to complete its stroke, at which point the mechanism being tested has also changed. Choosing which of the three to compromise is an engineering decision, and it should be written down, because a reader of the report cannot reconstruct it.

Check these in order

1. Define the failure criterion before the first cycle. Not “stopped working” but a threshold: force below a stated value with the stated load, response time beyond a stated limit, release time beyond a limit, or leakage past a limit. Write it as a number: 0.25 N minimum against a 0.20 N load, 25 ms maximum to full stroke, 30 ms maximum to release. A criterion defined after the test will be defined by whatever happened.

2. Decide what is recorded continuously and what is sampled. Continuous current logging for a million cycles is a data volume problem. A workable compromise is a full waveform capture at the start, at a logarithmic series of points along the test, and at any anomaly, with a simple cycle counter running throughout. Points at 0, 10, 100, 1,000, 10,000, 100,000 and 1,000,000 operations give seven snapshots, which is enough to see the shape of a trend.

3. Define the electrical conditions, and use the worst case rather than the nominal. The lowest voltage in the tolerance band, the highest ambient the enclosure will see, and the duty cycle from the real control software rather than the datasheet. A test at nominal voltage on a bench supply measures the most favourable state the product will ever be in. On a 24 V system with a 10 % tolerance, the coil sees 21.6 V at the low end, and force falls by about 19 % there before the winding has warmed at all.

4. Attach the real load. Spring, detent, linkage or whatever the application presents. A test with the plunger running free measures the coil and nothing else, and most field failures are the mechanism rather than the coil.

5. Mount it the way it is mounted. Conduction into a metal frame, or isolation in a plastic housing, changes the winding temperature by 20 K or more and therefore force by tens of percent. A unit tested in free air is not the unit that ships.

6. Include a hold at operating temperature. Creep and stress relaxation are driven by time at temperature under load, not by cycle count, so a test that counts operations without long energised holds will not see the mechanisms that produce a slow force loss, as set out in why force drops after thermal cycling. A 100 hour hold at the working temperature, repeated at intervals through the test, costs four days and reaches mechanisms a cycle count never touches.

7. Keep new reference units aside and measure them alongside. Every measurement on a cycled unit should be compared with the same measurement on an uncycled unit of the same batch, stored under the same conditions. Three reference units kept at room temperature and never energised are enough for most programmes. Without that reference, a drift in the rig or the ambient is indistinguishable from degradation in the part.

8. Trend the waveform features, not just the pass or fail. Current rise time, the position of the bend, steady current and release time. Four numbers per sample point turn the test into a curve, and the curve tells you how much margin remains rather than only whether any remains. Seven sample points and four features is 28 measurements, which is a spreadsheet rather than a data acquisition problem. That curve is also what supplies the alert limit in production.

9. Record the ambient continuously. Every thermal mechanism in the list depends on it, and an unrecorded ambient makes a temperature-driven result look like a random one. A 5 K drift in the room between the first sample point and the last moves the steady current by about 2 %, which is the same size as the drift you are trying to detect.

What actually to change

What the trend shows	What to change	Why not the obvious thing
Rise time lengthening steadily	Check the winding and the gap, in that order	A drive change hides a mechanical drift
Steady current drifting up	Check winding temperature and supply regulation	Rewinding treats a thermal problem as a magnetic one
Bend moving later in the cycle	Investigate friction at the guided surfaces	Force loss and friction rise look identical electrically
Force falls, timing unchanged	Check the gap and the return spring	Timing is the last thing the geometry affects
One unit fails early, others fine	That is a distribution, not a design fault	A single early failure is a scatter result, not a batch result
All units drift together	A design margin problem, not a manufacturing one	Screening will not fix a shared cause
Failure only in the accelerated group	Review the acceleration factor	The acceleration activated a different mechanism
Test passes, field fails	Reproduce the real supply, ambient and load	The product was never tested, only the coil

The distinction in the fifth and sixth rows is worth making explicitly, because it decides where the next month of work goes. When one unit out of ten fails early and the rest carry on, that is the shape of a distribution and the response is to understand the spread, which is the same measurement that supports incoming inspection, as set out in inconsistent force across a batch. When all ten drift in the same direction at the same rate, that is a design margin and no amount of sorting will change it. The two look similar in a report and require opposite responses.

When it IS the harder problem

The target is a million operations and the schedule will not allow eleven days. Then the acceleration has to be chosen deliberately and validated. Pick two stress levels, run both, and check that the failure mechanism is the same in each before applying any factor to a third. A factor fitted at 60 C and 90 C is not automatically valid at 120 C, where the lowest-rated component in the assembly may become the limiting mechanism instead. If the accelerated units fail with a different mechanism, such as a winding failure in place of a mechanical one, the factor is not transferable and the test has measured a different product. This is the case where the cheapest correct answer is usually to reduce the sample count rather than to raise the stress.

The requirement is a mean time between failures rather than a life. Then the deliverable is a distribution, and a distribution needs enough failures to fit. Running ten units to a million operations with no failures gives a lower confidence bound and nothing more; it cannot produce a mean. Ten units with three failures gives a fit and a slope, which is what a reliability figure is actually made of. If the customer needs a reliability figure, agree on the statistical basis before the test, because the sample count and the run length are set by it, and both are expensive to change afterwards.

The field failure is real and the bench test passes every time. Then the difference between the two is the interesting object, and it is worth systematically listing what the rig does differently: supply impedance, voltage tolerance, ambient, mounting, load, duty cycle, the presence of moisture, and whether the unit is energised for long periods. In the experience of most programmes, one line of that list accounts for nearly all of the discrepancy, and the rest of the programme has been spent optimising the wrong variable. The insulation side of the same question, where bench testing also misses the mechanism, is covered in testing coil insulation properly.

The design has a known weak point and the test is being used to prove it is acceptable. Then the useful test is not a life test at all but a margin test: on a part rated for 100,000 operations, run 20,000, measure what is left, and compare it with the sensitivity of that residue to the dominant variable. A life test answers how long. A margin test answers how much remains, and on a new design the second question is the one with an actionable answer. Where the margin depends on the winding’s temperature, the response time side of the same waveform is treated in why response is set by inductance.