Predictive maintenance usually arrives in a hatchery as a quotation for sensors. That is the expensive door, and rarely the first one worth opening. On a vaccination machine, the parameter that warns you earliest is generally one your team already writes down and nobody reads back.
Key takeaways
- Monitoring is only worth anything if the warning arrives early enough to act on. That gap is what the P-F interval describes.
- Elapsed months are the wrong denominator for a machine that works in short bursts around transfer. Count cycles and eggs.
- Delivered dose, sanitising circuit behaviour and consumable counters are signals you can already read without buying instrumentation.
- Choosing to keep monitoring and do nothing is a legitimate outcome of the method, not a failure of it.
What predictive means once the sensor catalogue is closed
Predictive maintenance is a decision rule, not a product. The general guidelines for condition monitoring and diagnostics of machines, published as ISO 17359 in its 2011 edition, set out the logic plainly: you select a parameter that reflects the health of the machine, you set alarm criteria for it, you monitor at an interval, and you decide what to do when the criterion is crossed.
Two points in that standard matter more than the hardware. The first is that the monitoring interval depends on the type of fault and on how fast it progresses. A defect that develops over months tolerates a monthly reading. A blockage that develops over one shift is not a condition monitoring problem at all, it is an operator check.
The second is that the decision after an alarm is not automatically to repair. On a machine of low criticality, the standard’s own guidance allows for taking no immediate action and continuing to monitor at normal intervals. That is the part most vendor presentations skip, and the part that protects your budget.
Underneath sits the P-F curve familiar from reliability work: a potential failure becomes detectable at some point, and a functional failure follows later. The distance between the two is the only thing that makes prediction useful. If the injector gives you eight hours of warning and your spare part sits four days away at a distributor, you have bought a diagnosis, not a solution.
Counting hours is the wrong denominator
A hatchery vaccination line does not run like a factory conveyor. It runs hard for a few hours on transfer day and sits idle the rest of the week. Published throughput figures give a sense of the burst: full-size in ovo systems of the Embrex Inovoject family are quoted at up to around 70,000 eggs per hour depending on configuration, while the compact Inovoject m aimed at smaller hatcheries is described at 12,000 and 20,000 eggs per hour according to the incubation system used.
Translate that into maintenance terms. A machine that clears a week’s transfer in three hours accumulates very few running hours per year, which is exactly why calendar-based servicing drifts out of step with real wear. Wear on the injection tooling follows the number of eggs presented to it, not the number of weeks since the technician last visited.
The same arithmetic explains why algorithmic prediction disappoints on this class of equipment. Models learn from failures, and a single injector in a single hatchery simply does not generate enough of them. Where machine learning does pay in this sector, it pays at fleet level, in the hands of a supplier watching hundreds of installations, which is a service you buy rather than a system you build.
Four signals the machine gives before it stops
The useful starting point is the set of parameters already available to you, at no capital cost.
- Delivered dose per egg or per bird. Drift in delivered volume is the most consequential signal on any injection system, because it degrades vaccination quality long before it degrades availability. Measuring the volume across a fixed sample of heads, at the same point in the cycle, turns a vague suspicion into a trend line.
- The sanitising circuit. In ovo systems of this type pump sanitising fluid over the needle and punch assembly after every injection, precisely to limit carry-over between eggs. A fall in flow or pressure there is a biosecurity event first and a mechanical event second, which is why it deserves a daily check.
- Tooling condition against a counter. The dual tooling used in ovo punches the shell with one needle and delivers a preset volume through a second. Both are consumables. Recording eggs injected since the last change, rather than months since the last change, is the cheapest improvement most sites can make.
- Air and fluid pressure at the head. Slow deviation from the reference setting, logged at start-up, catches leaks and partial obstructions while they are still adjustments.
| Signal | What it warns of first | Sensible reading interval |
|---|---|---|
| Delivered dose volume | Vaccination quality, then pump or seal wear | Every vaccination session, on a fixed sample |
| Sanitising fluid flow and pressure | Cross-contamination risk between eggs | Daily, before the first tray |
| Eggs injected since tooling change | Needle and punch wear, injection accuracy | Continuous counter, reviewed weekly |
| Air and fluid pressure at start-up | Leaks, partial blockage, regulator drift | Every start-up, logged not eyeballed |
The case for leaving some of it alone
Not every component deserves a monitoring strategy. The test is the one the standard applies, criticality, and in a hatchery criticality is measured in deadlines rather than in euros. Eggs reach day 18 or 19 on their own schedule and the hatch does not wait, so anything that can stop the line during transfer is critical by definition. A conveyor motor with a spare on the shelf is not.
That distinction is what turns condition data into money. Monitoring a part you already stock buys you nothing, since you were going to swap it either way. Monitoring a part with a long lead time buys you the lead time itself, which is the whole point. The cadences and the spares list behind that judgement sit in our preventive maintenance plan for vaccination machines, and predictive work should be read as an adjustment to that plan rather than a replacement for it.
Building the failure history you do not have yet
Most hatcheries discover, when they finally ask for one, that they have no usable failure record. Work orders exist, but they say what was replaced and not why. A prediction has nothing to learn from that.
Rebuilding the record costs nothing and starts on the next stoppage. Note the point in the cycle, the eggs injected since the last intervention on that assembly, the symptom in the operator’s own words, the part fitted, and the downtime in minutes against the transfer window. Three or four seasons of that produces something no supplier dashboard can hand you: the failure pattern of your machine, in your hatchery, with your egg supply and your water quality.
It also settles arguments. When a supplier proposes a monitoring package, the question is which parameter it watches, how much warning it has produced on comparable installations, and whether that warning is longer than your own supply lead time. Without a history of your own, you have no way to check the answer.
Before the monitoring question, the machine question
Condition monitoring is easier to justify on equipment matched to your vaccination route and your hatch volume in the first place. The selection criteria come before the sensors.
Sources: ISO 17359:2011, Condition monitoring and diagnostics of machines, general guidelines, for the parameter selection logic, the interval rule and the criticality-based decision; published technical descriptions of the Embrex Inovoject and Inovoject m systems, for throughput ranges, the dual needle and punch design and the inter-egg sanitation of the tooling. Consulted August 2026.
Published previously, fully revised on 12 August 2026. General technical guidance for hatchery and livestock professionals. Vaccine handling and any suspected loss of vaccine efficacy should be referred to the responsible veterinarian and to the product’s own instructions for use.