Most of the line's losses sit in the capper
Cap feeding, cap grabbing and cap torque on the filler-capper monoblock caused the most stops and the most rejects, and long formula changeovers added planned downtime on top.
| Loss | Frequency | Value lost |
|---|---|---|
| Cap sorter, chute and transfer jams | 2 stops a shift | ~$5,000 a month |
| Caps missing or misapplied (cap grab) | ~26,000 ppm | ~$6,000–8,000 a month |
| Wrong cap torque | Not identified todayManual check: 4 of 5,400 bottles (~0.07%) | to be quantified |
| Formula changeovers | 45 min per changeover | ~$10,000 a month |
- $50,000–$75,000 a year in potential savings on just the capper with Nova
- Root cause on the first answer for cap sorter jams, using past lessons and the manual's own specs
- One brain for the capper: 1,500 pages of machine manuals, 418 figures and every operator lesson in one place
Nova learns from your operators
Many capper problems can't be solved from data alone. The fix sits in an operator's head: which part to check first, what a bad head looks like. Nova captures that through AI interviews and turns it into an experience layer that feeds the model alongside the line data.
- One week for all capping line operators, versus months for a traditional consultancy to build a loss and value map and an experience layer for everything operators troubleshoot today.
- Reruns every 15 days. Every stop since the last round is captured, and stops with no resolution are marked, so operators can fill in what's missing.
- Solves the adoption problem. Capturing operator experience is a hurdle across manufacturing. Here it's a short conversation instead of a documentation task.
From bulk liquid to packed pallet
Finished liquid product arrives from the making area's finished-product tanks. Empty bottles are oriented by the unscrambler and labelled, then filled and capped on a filler-capper monoblock: a filling carousel and a capping carousel linked by transfer stars, fed with caps by a sorter that orients them and blows the wrong-way-up ones back into the bowl.
Capped bottles are checkweighed, coded, packed, sealed and palletized. Every machine downstream waits on the monoblock, so a cap jam or a run of loose caps stops the whole line, and bad-capped bottles end up on the reject conveyor.
The fixes lived in a few people's heads
Four problems kept costing the capper time and product.
Cap feeding was the most frequent. The cap sorter and chute jammed about twice a shift, from clogged reject nozzles to tamper-proof caps dropping into the starwheel, costing around $5,000 a month. Capping heads that failed to grab caps pushed missing and misapplied caps to around 26,000 ppm, worth $6,000–8,000 a month.
Wrong cap torque was the one nobody could see. The only check is manual: four bottles every 30 minutes, about 0.07% of the 5,400 bottles in that time. A head with a slipping clutch could run loose caps for a long time before anyone knew.
Formula changeovers took 45 minutes each, around $10,000 a month in lost time.
The answers existed, but not in one place. Specs like the 1–2 mm sorter-to-chute gap were buried in 1,500 pages of machine manuals, partly in French. The knack of where to look first sat with a few senior operators. When a contractor or a new operator was on shift, the line waited.
Where the losses start on the monoblock
For each of the three machine losses, we defined the combination of signals that tells it apart from the others, and paired it with what operators told us actually fixes it. Select a point on the drawing to see the signature, the root cause and the troubleshooting steps.
One line, every system, one case timeline
Every source was connected read-only. Each has a dedicated tool the agent calls when it needs that piece of evidence, so every answer Nova gives can be traced back to where it came from, down to the manual page.
| Source | What it tells Nova on this line | Read by |
|---|---|---|
| PLC / SCADA | Machine state, line speed, applied torque, per-head torque time, jam status, counters | get_live_state get_timeseries |
| Alarms and stops | Capper, checkweigher and filler alarms, micro-stops, faults and changeover downtime | get_alarms get_downtime |
| Quality | Torque checks, checkweigher rejects, vision inspection results | get_quality |
| Maintenance (CMMS) | Open and past work orders, PM schedule, spare parts with stock and bin location | get_work_orders lookup_spare |
| ERP | Running order and SKU, recipe setpoints, cap and bottle lots, suppliers | query_production get_recipe lookup_vendor |
| Manuals and SOPs | 1,500 pages of machine manuals, including HMI manuals in French and English, plus the bottle line SOP, with 418 figures, cited by page | search_docs |
| Operator experience | Lessons from loss interviews and every resolved case, with the knack behind each fix | recall_similar_cases save_case |
| People | Who to call: line manager, materials SPOC, operations leader | escalate |
From loss interviews to a live session on the line
The work moved in five steps, starting with the operators who live with the losses.
Mapped the losses
Ran AI interviews with every capping line operator in one week. Captured 12 lessons: what goes wrong, why, and how they fix it, including the changes made to the line in the last few weeks. The interviews rerun every 15 days.
Indexed the manuals
Turned 1,500 pages of machine manuals and the line SOP into searchable sections and 418 figures, so answers cite the page and show the diagram.
Connected the line
Linked machine signals, alarms, quality, maintenance and ERP data read-only into one view of the line.
Defined signatures
Turned each monoblock loss into a signal pattern Nova can recognise and paired it with the operators' fix.
Demoed on the line
Operators used Nova on a real cap sorter jam during the night shift, asking in English and Hindi. Two new lessons were saved with their knacks. Every process change and fix, from the smallest to the largest, is recorded in the app, so the brain keeps growing in one place.
What Nova could change on the bottling line
Potential with Nova, worked out from the plant's own loss figures. Before figures come from the operator loss interviews.
Time to resolve each cap sorter, chute or transfer jam
2 stops a shift: about 40 minutes saved each shift
Caps missing or misapplied
Formula changeover time
Process knowledge
Specs buried in 1,500 pages of manuals, partly in French, and the knack held by a few senior operators
One brain for the capper: 1,500 pages of machine manuals, 418 figures, 14 operator lessons and live line data
The ppm and changeover figures apply the same 18–30% reduction as the total. Wrong cap torque isn't valued yet, so the real saving could be higher.
Fewer stops
The first answer already points to the likely cause, so the line restarts faster and the same jam doesn't come back.
Fewer rejects
Per-head torque data shows which capping head is slipping before loose caps pile up on the scrap conveyor.
Every shift as good as the best one
Contractors and new operators get the senior operator's knack on their phone, in their language.
What plant teams ask us
Does Nova change anything on the machines?
No. Every connection is read-only. Nova recommends and verifies; the operator acts. Changes like a new cleaning step go to the line manager for approval.
Can operators use it in their own language?
Yes. On site, operators asked Nova in English, Hindi and a mix of both, by text and by voice, and Nova answered from manuals written in English and French.
What happens when Nova can't fix it?
It escalates to the right person, line manager, materials SPOC or operations leader, with a summary of everything already checked, so nobody starts from zero.
What if the root cause is something new?
Nova still gathers the evidence and cites the relevant manual sections. Once the operator resolves it, the fix and the knack are captured, so the next time it's in the playbook.
Which AI models does it use?
Nova is model-agnostic. It runs through a model gateway, so the plant can use Claude, Gemini or another approved model.
Have a loss that keeps coming back?
We start with a loss-mapping session with your operators, rank your losses by cost, and show where Nova pays back first.