Solenoid inconsistent in production: batch variation or design margin?
The distinction is in the lower tail of the distribution, not the mean. A design problem moves the whole distribution down. A batch problem widens it, usually with the mean unchanged. Measure at least 30 units, compare spread as well as average, and remember that a 10% shift in air gap produces roughly a 17% shift in force — so gap variation amplifies into everything downstream.
Why this happens
Inconsistency in production is a distribution problem, and most factories measure it as an average problem. The result is a decision made on the wrong number.
Separate the two families properly and the diagnosis gets much easier. A design or margin problem moves the whole distribution: the mean itself sits too close to the requirement, so even perfect consistency fails. A batch problem keeps the mean where it was and widens the curve, so the average looks fine while the lower tail crosses the requirement. The first is fixed by changing the design or accepting a lower specification. The second is fixed by finding which variable started moving.
The complication is that physics amplifies. Force in the working region varies roughly with the inverse square of air gap, so a 10% increase in gap costs about 17% of force. A tolerance that looks reasonable on a drawing at ±0.05 mm becomes a far larger percentage on the force output. That is why the culprit is so often an air gap or travel dimension that nobody flagged, while everyone argues about core material, which is a smaller lever.
Check these in order
1. Validate the measurement before you analyse the data. Take one unit and measure it ten times with the same operator and rig, then swap operators. If repeatability is poor, the spread you are about to investigate may not exist in the product at all. I have run this test after a week of process investigation and found the variation was in the fixture.
2. Measure at least 30 consecutive units and plot them. Not 10 selected good ones. You need the tails. Look at three numbers together: mean, standard deviation, and the lowest units. The lowest units are what will generate the complaint, because a customer notices the one that is weak, not the thirty that are fine.
3. Classify before you investigate. Mean low, spread normal: a design or margin problem — go back to the requirement and the temperature derating. Mean normal, spread wide: a process variation problem — find the variable that started moving. Both mean and spread wrong: usually a design margin problem that a process shift exposed.
4. Rank variables by force sensitivity, not by observed variation. This is the step that saves the most time. Work out how much force changes per unit of gap, per unit of concentricity, per unit of spring preload, per turn of winding. Gap usually wins by a wide margin because of the squared relationship. Then measure the actual variation in each of those, and multiply. Highest sensitivity times highest observed variation is where you should spend money.
5. Check the process variables that move in production but not on a bench. Insulation thickness on the bobbin moulding, plunger plating thickness, washer grade, seating depth after press, spring batch, winding turns if the machine counts rather than measures. Several of these are invisible in the finished unit.
6. Re-measure after each change, with the same sample size. If the spread does not move, the change did not address the cause. Reducing the spread by tightening a tolerance that physics does not care about is expensive and does nothing.
What actually to change
| Finding | What to change | Why not the other thing |
|---|---|---|
| Mean too close to requirement | Increase design margin, or re-derate for temperature | Tightening process tolerances will not move a mean |
| Spread wide, gap dominant | Tighten the gap-controlling dimension: spacer, stop, seating depth | Core material has a small sensitivity, so it buys little |
| Spread wide, unit-to-unit rub | Improve concentricity control in the bore and on the plunger | Material change does not address alignment |
| Spread wide, spring batch | Specify spring force and rate, not just free length | Length is easy to measure and weakly correlated with force |
| Measurement repeatability poor | Fix the fixture, control the plunger axis, log temperature | You cannot reduce variation you cannot measure |
| Two variables each near limit | Budget them together, not individually | Individually acceptable is how a batch ends up below requirement |
When it IS the harder problem
Tolerance stacking with no single out-of-spec item. One unit in ten is weak, and the weakness follows the unit rather than the position on the bench. This is the pattern I have misjudged twice: the mean sat comfortably above requirement and every inspection record was within tolerance, but several dimensions were each at their own limit, in the same direction, on the same unit. Nothing was defective. The product was over-toleranced. The remedy is to budget the variables as a stack with a worst-case sum and check that the worst case still passes, which is a design task rather than a production task.
A spread that appears only after assembly. Force measured on loose components can look tight while force measured after press and staking is wide, because the assembly step deforms one of the gap-setting features. The variation is introduced by the process, not by the parts. If your parts are consistent and your assembled units are not, the assembly operation is your variable, and the acceptance criteria for the parts will never catch it.
A change that improved the mean and widened the spread. This is the worst outcome, and it happens when a process improvement is adopted without checking the tails. A new plating, a new lubricant, a different supplier for the plunger: the average force improves, so the change is declared a success, and then complaints continue because the low end of the distribution did not improve or got worse. Always compare spread, not only mean, when you accept a change.
A note on what this page is
This is a personal notebook, not a product page. I write down the checks that actually decide the outcome and the order to run them in, including the ones I got wrong first.
solenoid inconsistent forcesolenoid batch variationsolenoid production tolerancesolenoid force distributionsolenoid air gap tolerance
Frequently asked
- My average force is comfortably above requirement. Why are customers complaining?
- Because averages do not ship, individual units do. If your mean sits at 5.6 N against a 4.5 N requirement but your spread has widened, the lower tail is below requirement even though every summary report looks green. Design against the batch minimum, then subtract the temperature derating on top of that.
- How many units do I need to measure to tell the difference?
- At least 30 per group, and they should be consecutive or from several cavities rather than ten hand-picked good ones. With 30 you can see whether the spread changed even if the mean did not. With 5 you are looking at noise, and you will make a decision on it.
- Which tolerance should I tighten first?
- Whichever one the force is most sensitive to per unit of variation, not whichever one looks worst on the drawing. Air gap sits in an approximately squared relationship, so a 10% gap shift is a 17% force shift. That amplification means gap variation usually dominates the budget even when it is a smaller percentage than everything else.
- The spread widened after we changed suppliers. Is that the supplier's fault?
- It is the supplier's variation, but not necessarily the supplier's fault, because your drawing may have given them room they were entitled to use. A tolerance is only as good as its sensitivity: if the drawing allows a gap variation that costs 17% force, a supplier meeting the drawing exactly can still cost you the product.