Over-rejection — acceptable product being blown into the reject stream — is the single most common complaint we hear from processors, and the most economically damaging. Unlike escaped defects, which are a quality risk, false rejects are direct revenue destruction, occurring continuously for every hour the machine runs.
The encouraging news is that over-rejection is rarely a machine defect. It is usually configuration, presentation or maintenance, and therefore largely fixable without capital expenditure.
This guide works through the genuine causes in order of likelihood, with diagnostics you can perform on your own line.
Start here: quantify the problem before changing anything
Do not adjust anything until you know your baseline. Twenty minutes with a scale:
```
- Run production at normal settings and feed rate
- Collect reject output for a fixed period — say 10 minutes
- Weigh it
- Manually separate genuine defects from acceptable product
- Weigh both fractions
Carry-over ratio = acceptable product in reject ÷ total reject output ```
This works for the same reasons it works as an ongoing KPI — it requires no laboratory, and responses directly to configuration change. Our accuracy explainer covers measurement method in greater depth.
Record the figure. Every fix should be validated against it. Resist the urge to change several settings at once — you will not know which change helped.
Cause 1: sensitivity set too aggressively
The most obvious cause, and often not the real one, but check it first because it costs nothing.
If parameters were set during commissioning by someone optimising for defect removal — or set once by an operator who wanted to be certain nothing escaped — you may simply be running tighter than economics justify.
Diagnostic. Reduce sensitivity incrementally and re-measure carry-over after each change. Watch accept quality carefully as you loosen; find the point where escaped defects begin rising materially. That knee is your optimum.
Important framing. Some carry-over is normal and even desirable — the cost of few escaped defects. The goal is not zero carry-over, it is minimum total cost, balancing lost product against defect penalties:
``` Total cost = (escaped defects × penalty per unit) + (false rejects × selling price per unit) ```
On high-value products — cashew kernels, premium roasted coffee, tree nuts — the second term dominates heavily, so bias toward yield protection. On thin-margin high-volume products the balance shifts. Full economic framing in our ROI guide.
Cause 2: unstable or excessive feed — the most common real cause
This is the cause most often missed, and frequently the whole problem.
When particles arrive faster than they can stabilise into a single layer, three bad things happen simultaneously:
- Units overlap, so the camera reads a merged surface rather than individuals,
- Material tumbles, presenting inconsistent surfaces,
- Small defects get hidden while acceptable units get misjudged as larger errors.
The result appears identical to "sensitivity too high" — lots of false rejects — but increasing sensitivity is exactly the wrong response, since the problem is presentation, not judgement.
Diagnostics and fixes:
| Check | Fix |
|---|---|
| Feed rate as percentage of rating | Reduce toward 60–75% of rated maximum |
| Feeder vibration amplitude | Adjust to stabilise the curtain |
| Material depth on chute / belt | Too deep causes stacking — reduce |
| Upstream surge / starvation | Install or commission a surge hopper |
| Chute surface condition | Build-up disrupts laminar flow — clean |
Feed stability deserves emphasis. Machines starved then flooded cannot present consistently regardless of settings. A surge hopper upstream often resolves persistent instability more effectively than any parameter change.
Cause 3: fouled optics or drifting illumination
Optical degradation and excess rejection are strongly linked, and this is the second most common true cause.
A dust film on a camera window scatters light, lowering effective contrast. Low contrast means legitimate natural variation in your product begins to look like defect signals — so the machine rejects acceptable units it previously accepted.
The progression is insidious: performance declines gradually, carry-over climbs slowly, and operators may compensate by adjusting sensitivity, making things worse.
Diagnostics and fixes:
- Clean all optical surfaces with approved lint-free cloths and agents, then re-measure carry-over. On dusty materials this alone frequently restores baseline.
- Check illumination output. LEDs decline gradually. Compare against commissioning values if recorded.
- Verify reference / background plate cleanliness — reference drift shifts every subsequent decision.
- Run post-cleaning verification. If cleaning restores performance, you have identified the cause and should adjust cleaning frequency upward.
Schedule detail is in the maintenance checklist. Note that roasted coffee, nuts and plastic generate oils and dust that foul optics far faster than clean grains — operations applying rice-line cleaning schedules to these materials routinely underperform.
Cause 4: ejector problems — neighbouring units caught in the blast
Ejector behaviour causes over-rejection in ways that look mysterious until understood.
- Excess air pressure deflects not just the target but neighbours within the airstream.
- Slow valve response fires late, missing its target and disturbing the wrong unit.
- Leaking valves create continuous turbulence, disrupting nearby material.
- Inconsistent air supply makes ejection unpredictable, sometimes firing unnecessarily.
Diagnostics and fixes:
| Check | Action |
|---|---|
| Air pressure at manifold | Reduce to minimum that reliably deflects target particles |
| Air quality — moisture and oil | Drain receivers, check dryers, test for contamination |
| Audible leaking valves | Listen at manifold with line running; replace leaking units |
| Valve response testing | Test individual solenoids monthly; replace slow units |
| Receiver capacity | Ensure volume adequate for peak simultaneous firing |
Reducing air pressure is frequently the single highest-value adjustment available. Many lines run far higher pressure than necessary. Tune downward gradually, re-measuring carry-over and escaped defects after each step — you will typically find substantial excess.
Air supply also accounts for the largest ongoing running cost, so this fix improves your operating economics twice. See the energy consumption analysis.
Cause 5: legitimate product variation not modelled
Sometimes the machine is working correctly and the model of acceptable variation is wrong.
Every natural product varies. New-crop material differs from stored material. A different origin or variety shifts hue and size. Unexpected surface sheen — common on roasted coffee through oil migration, and on some nuts — reads as variance the machine has not learned.
If your material's appearance changed and defaults were never re-established, misclassification follows.
Fixes:
- Re-establish the acceptable-variation baseline for current material, rather than reusing stale configuration.
- Store recipes per material and per season, and revalidate with each new crop.
- Consider trained recognition models. Systems learning from many real examples tolerate natural variation considerably better than single-threshold approaches, which is precisely why they matter most on cashew and premium coffee where over-rejection costs most. Technical background in AI deep learning for sorting.
Cause 6: mechanical breakage masquerading as poor sorting
Particularly relevant on fragile products. If materials are damaged during handling — through excessive drop height, hard impact surfaces, or turbulent transfer — damaged pieces are genuinely defective and correctly rejected. Operators perceive this as over-rejection when it is really upstream-created damage.
Diagnostic. Run a known mass through with ejection disabled. Quantify breakage entering versus leaving the machine. The difference is machine-induced damage.
Fixes. Reduce drop heights, add impact-absorbing food-grade surfaces at transitions, eliminate sharp conveying transitions, reduce ejector pressure, and consider whether a belt-type platform suits the material better than free-fall — see the chute versus crawler comparison.
Systematic diagnostic order
Work in this sequence. Each step is cheap and most problems resolve by step three.
```
- Measure baseline carry-over → establish the number
- Clean all optics, re-measure → often resolves entirely
- Check and reduce feed rate toward 60–75% rating → most common true cause
- Verify air supply quality and reduce pressure → second most common
- Inspect for leaking or slow ejector valves → replace identified units
- Review sensitivity against escaped defects → find economic optimum
- Re-establish baseline variation for current crop → if material changed
- Measure mechanical breakage with ejection off → if fragile product
```
Change one variable at a time, re-measuring carry-over after each. Simultaneous changes make results uninterpretable, which is how acceptable settings drift permanently into poor ones over months of small adjustments.
Prevention: keeping it fixed
Once tuned, protect the improvement:
- Measure carry-over monthly and log it. Gradual rise signals developing problems before output complaints appear.
- Clean optics at material-appropriate intervals — daily minimum, more often for oily or dusty products.
- Store and back up working parameters so they can be restored after any intervention. Covered in the maintenance checklist.
- Revalidate seasonally with each new crop. This single habit prevents most recurrence.
- Consider remote diagnostics where available, which surfaces developing faults earlier — see remote debugging and IoT.
FAQ
How much carry-over is normal?
Product-dependent, so establish your own baseline rather than chasing a published figure. Well-run lines typically achieve single-digit carry-over, with some installations below one percent on favourable materials. If yours is materially higher, the diagnostic sequence above will usually find the reason.
Will reducing sensitivity cause quality problems?
It can, which is why you adjust incrementally and measure both directions — carry-over falling and escaped defects rising. Stop at the point where escapes begin climbing materially. That is your economic optimum, and it differs per product value.
My carry-over is fine in the morning and worse by afternoon — why?
Classic optical fouling or illumination drift developing over the shift. This pattern almost always indicates cleaning frequency is insufficient for your material, or LED temperature stability issues. Add a mid-shift clean and measure whether the afternoon decline disappears.
Can worn parts cause over-rejection?
Yes, particularly ejector valves. A valve that responds slowly mistimes its pulse, and one that leaks disturbs nearby material. Monthly valve testing identifies decline before it affects output materially.
Is this covered under warranty?
Configuration and presentation are not warranty matters — they are operational. Genuine component failure generally is. Worth noting, though, that most over-rejection is not component failure, so a service call may find nothing mechanically wrong. Diagnose first; it is usually quicker and cheaper.
We changed varieties and rejects jumped — normal?
Completely expected. Different varieties have different appearance envelopes, so the machine is judging new material against stale criteria. Re-establish baseline variation for current material and store it as its own recipe. This is why recipe management matters where multiple varieties are processed.
Getting help with your specific case
Over-rejection is diagnosable. Work the sequence above, change one thing at a time, and measure. Most cases resolve through cleaning, feed rate, or air pressure — none requiring expenditure.
Review configuration guidance by material across the product range, understand the financial trade-off in the ROI guide, and if you want a second opinion specific to your line, contact our technical team with your material details and measured carry-over figure.
