Your weighers are already capable of more.
Wingman is an autonomous control layer for multi-head combination weighers. It tunes every head, every cycle, every shift. It does the work nobody has time to do by hand.
Schematic. Hopper states are real, recorded during the audited session; the combination shown is solved live in your browser. Measured giveaway figures are on the results page.
Mean fill per head over the armed window, in grams: head 1, 9.4; head 2, 11.7; head 3, 18.3; head 4, 22.6; head 5, 19.5; head 6, 8; head 7, 7.2; head 8, 6; head 9, 7; head 10, 21.2; head 11, 34.2; head 12, 24.1; head 13, 16.7; head 14, 10.2.
A combination weigher isn't one machine.
It's fourteen feeders, each behaving slightly differently, drifting all shift. Somebody has to notice that and correct it. In practice, nobody has the time, so the line runs on settings that were right this morning.
The weighers are excellent. The problem is that a human has to sit there and tune them by feel, and that human cannot be everywhere at once, forever. The enemy is the tuning burden, not the equipment.
Five things, in this order.
It watches first
Installed in monitor mode, it reads every head, every cycle, and records what the machine does under your operators. It writes nothing.
It takes over gently
When you arm it, it starts from your existing settings and walks them in small, bounded steps. Nothing lurches.
It solves the whole machine at once
Every cycle it answers one question: which feeder settings bring every head closer to its ideal fill, without violating the machine's limits?
It learns your heads
In the background it builds a model of how each individual head responds, pairing every change with the weight that change genuinely caused.
It only acts on learning it can prove
The learned model adjusts the proven solution only where it is confident of a strict improvement. Otherwise the proven solution is used, untouched.
Then it locks in success, and teaches the next machine what it learned.
It cannot make things worse.
The learned model can only ever improve on a proven control move. It can never replace it with a guess. The worst case is exactly as good as the proven controller.
Four independent permissions must all agree before a single value reaches your machine. Any one of them says no and nothing is written.
It found a fault nobody was looking for.
Under identical commanded settings on every head, four adjacent heads were filling at roughly one third the rate of the rest of the machine. A product-distribution fault upstream of the weigher, silently capping the line's efficiency, invisible in any number anyone was looking at.
It compensated automatically and named the heads, so a human could go and fix the actual cause.
Watching by Wednesday. Running your line by Friday.
Install and connect
One on-premise computer, connected over your weighers' standard interface. No machine modifications. No line downtime.
Monitor mode
It runs alongside your line, watching every head every cycle and learning your products, while writing nothing at all.
Fully autonomous
You arm the first machine. It starts from your existing settings and improves from there. Return it to monitor mode at any moment.
Book a line assessment.
We'll tell you what your line is actually doing before you buy anything.