FILL-WEIGHT CONTROL
Real-time weight and level control per format and per head, pushing the mean down toward target without breaching the minimum.
STAGE 03 // FILL & PACK
Fillix
stops the giveaway.
Fillix controls filling and fill weight or level to minimise giveaway while holding compliance, and drives robotic assembly, capping, decoration and case-packing across jars, tubes, bottles and compacts. It fuses checkweigher data with fill-level vision to trim filler setpoints in real time.
PER FORMATFILL CONTROL
ISAACROBOT CELLS
REAL-TIMESETPOINT TRIM
JETSON + IGXGPU PROFILE
$ fillix giveaway --shift current TARGET FILL 50.0 g min 49.0 g MEAN FILL 50.42 g (was 51.30 g) STD DEV 0.21 g Cpk 1.71 GIVEAWAY 0.84% baseline 2.60% UNDERFILL 0 units compliance HELD HEAD BALANCE h1 +0.03 h2 -0.06 h3 +0.12 -> trim head 3 by -0.09 g (auto, in envelope) ROBOT CELL cap + decorate NOMINAL RELEASE GATE stabion CLEARED
FILL-WEIGHT CONTROLSENSOR FUSIONROBOTIC CELLSCHANGEOVER ASSISTRELEASE GATEIMMUTABLE AUDIT
THE PROBLEM
Giveaway is invisible per jar and enormous per year
After compounding, product is filled into jars, tubes, bottles and compacts to a precise weight and level, then sealed, decorated and packed. Overfill given to guarantee minimum fill weight is pure product handed away at scale; underfill risks compliance failure.
Short runs and constant new-SKU launches explode changeovers, and the line technicians who tune fillers, cappers and decoration are scarce.
Most plants run blind on real-time fill weight, giveaway and pack quality across formats.
GIVEAWAY MATHS / ONE LINE
- TARGET FILL
- 50.0 g
- ACTUAL MEAN
- 51.3 g
- GIVEAWAY
- 2.6%
- UNITS / YEAR
- 18M
- PRODUCT GIVEN AWAY
- 23.4 tonnes
- ILLUSTRATIVE VALUE
- $0.4M–$1.6M
ILLUSTRATIVE SITE FIGURES [ASPIRATIONAL UNTIL MEASURED].
CAPABILITIES
What Fillix does
SENSOR FUSION
Checkweigher plus fill-level vision fused into one estimate, so rejects and setpoint nudges act on the same truth.
ROBOTIC CELLS
Assembly, capping, decoration and case-packing driven by simulation-trained manipulation policies on Isaac and Jetson.
CHANGEOVER ASSIST
Format changeover assistance and recipe recall, with predicted changeover time per sequence.
RELEASE GATE
Fillix honours Stabion batch status: product is not packed out ahead of its release decision.
IMMUTABLE AUDIT
Every setpoint trim, reject and robot program version is logged with its approver and evidence.
THE LOOP
Four beats, every batch
-
01
MEASURE
Checkweigher, fill-level vision and head telemetry stream into one live distribution per head and per format.
-
02
OPTIMISE
The giveaway model computes the lowest safe mean fill given observed variance and the compliance floor.
-
03
TRIM
Head-level setpoint trims are applied inside the envelope; anything larger escalates to the line technician.
-
04
PACK
Robot cells cap, decorate and case-pack, releasing only product cleared by Shadera and Stabion.
AI CAPABILITIES
What is real today, and what is not
Serumon labels its own maturity. Anything not yet validated in production is marked aspirational, on this site and in the room.
REAL
FUSION + TRIM CONTROL
Fill-level vision and checkweigher fusion with closed-loop setpoint trimming per head.
REAL
GIVEAWAY OPTIMISATION
Constrained optimisation of mean fill under compliance limits and observed process variance.
SIM REAL / LIVE ASPIRATIONAL
ROBOT POLICIES
Manipulation and handling policies trained in simulation; on-line deployment follows per-cell validation.
REAL
CHANGEOVER PREDICTION
Predicted changeover duration per format sequence, feeding Twynex campaign scheduling.
NVIDIA ALIGNMENT
GPU is must-have, not nice-to-have
JETSON + IGX ORIN
Line-side control and robot-cell inference with deterministic latency and safety segregation.
ISAAC
Simulation-trained manipulation, capping and packout policies transferred to physical cells.
TENSORRT
Quantised fill-level vision models to keep pace with full line rate.
TRITON
On-site serving of fill, defect and changeover models across multiple lines from one node.
MEASUREMENT
What we get judged on
GIVEAWAY %PRIMARY METRIC
CHANGEOVER HRSSECONDARY METRIC
PACK DEFECTSTHIRD METRIC
$9k/moLINE TIER PRICE
PILOTS ARE CONTRACTED AGAINST A NAMED ROI METRIC AND A BASELINE WE RECORD BEFORE ANYTHING CHANGES.
THE MOAT
Checkweighers reject. Fillix re-shapes the distribution.
A fixed threshold on a checkweigher protects compliance by pushing the mean up. That is a rational, expensive choice made in the absence of a live model of process variance.
Fillix models the distribution per head, per format and per product viscosity, then walks the mean down as variance tightens — and it learns the relationship between how the batch was made and how it fills.
- Per-head, per-format fill distributions over time.
- Viscosity-to-fill behaviour linked back to Emulson traces.
- Robot cell telemetry and failure modes.
- Changeover durations by real sequence, not by standard.
DEPLOYMENT SHAPE
- TIER
- LINE / PLANT / ENT
- LINE PRICE
- $9,000 / mo
- PLANT PRICE
- $60,000 / mo
- ENTERPRISE
- $400k–$5M ACV
- EDGE NODE
- PER VESSEL CLUSTER
- AUTONOMY
- GRADUATED
ECOSYSTEM
How FILLIX plugs into the loop
Every agent hands the next one a better decision. Buy one; the others make it sharper.
EMULSON
undefined
SHADERA
undefined
STABION
undefined
TWYNEX
undefined
FILLIX FAQ
Straight answers
The compliance floor is a hard constraint, not an objective. Fillix optimises the mean subject to keeping underfill probability below the limit your quality team sets, and it widens the safety margin automatically when variance rises.
No. Fillix writes setpoints into existing fillers through their control interface. Older fillers with coarse setpoint resolution see smaller gains, and we will tell you that during the survey.
The robotic cell work targets standard industrial and collaborative arms driven through their controllers, with policies trained in Isaac. Cell-by-cell feasibility is confirmed in the site survey.
The measurement is immediate; the control benefit lands once variance is characterised, typically within a few production weeks on a stable SKU. Highly variable products take longer and improve less.
WEDGE PILOT
Run Fillix on one line
Instrument, baseline, shadow, then assist — against a metric you name before we start. If the numbers do not move, you have lost a quarter of telemetry and gained a very detailed picture of your own process.