A bad pallet is cheap to fix on the build station and expensive everywhere after it. TesseRack verifies the load while it is still being built — tier height, pack lean, layout conformance, per-face overhang — and calls the fault in the window where a fault is still correctable, before the stretch-wrap goes on.
Almost anything can photograph a finished pallet and tell you it was bad. That finding arrives after the wrap is on, after the load has left the station, and often after it has left the building — at which point the only options left are re-work and credit notes.
TesseRack is built around the interval between the last pack placed and the wrap going on. Inside that window a leaning tier can be pushed back, a misplaced pack can be re-seated, and a pattern error can be caught before it is repeated on the next four layers. Everything about the system — where the cameras go, how fast a decision has to be made, how the alarms are tiered — follows from holding that window open.
The failure modes that put a load on the floor are few, and they recur across products. TesseRack measures each of them directly and in engineering units — not as a pass/fail score whose meaning nobody can reconstruct three weeks later.
How far a pack projects past the pallet edge, reported per face in millimetres against that face’s own limit — the four sides of a load do not share one tolerance, and treating them as if they do either passes a real defect or rejects a good pallet. Measured from calibrated geometry, so the number means the same thing on every line it is quoted on.
The built height of each tier as it lands, and the cumulative height of the load against its spec. Catches a tier that is riding high because a pack is not seated, and the load that will not clear the racking beam before it is wrapped and weighed.
The lean of individual packs and of the tier as a whole. A leaning stack is the single most useful early warning that a load will fail in transit, and it is visible long before anything actually falls.
Whether the pattern actually built matches the pattern specified for that SKU — pack positions, orientations and count per tier. A single rotated pack changes how the whole load interlocks and carries load, and it is the error most easily missed by eye at line speed.
The layout is the design. Everything else follows from where the cameras can actually go on your line.
Every measurement is reported in millimetres derived from a calibrated view of the station. A pixel figure is unusable across cameras, lenses and stack heights; a millimetre figure can be written into a spec, argued about, and audited.
Two supported layouts. A turntable presents every face of the load to a fixed viewpoint; a build station does not, and needs the viewpoints distributed instead. Which one you have changes the camera count and the decision timing, so it is settled before anything is quoted.
A load that is out of spec, a load that is drifting toward it, and a load the system could not assess with confidence are three different situations for the operator. Collapsing them into one alarm is how a quality system gets ignored by week three.
The system has to return a verdict inside the interval between placement and wrap. That budget is set from your actual line rate at design time and the architecture is sized to meet it — not measured hopefully afterwards.
The first-visit feasibility work was done with no model trained on customer data — worst-case handheld capture, and still readable results. What we learn on your line stays on your line.
The same load-on-pallet failure modes recur across FMCG — beverages, shampoo, oil, juice, cans. The per-product module is deliberately thin; the measurement platform underneath is the part that carries over.
TesseRack never commands a machine stop. It raises a quality flag or a reject and hands it to the operator or the line controller to act on. It is not a protective device and it holds no authority over the palletizer.
If a robot cell needs guarding that is Monolith, and if the hazard is at the tool itself that is Forge — separate products, separately deployed. They can share a site and even a camera position, but never a verdict: a quality reject must never be capable of reading as a safety event, or the reverse.
Tell us about your facility. We'll identify the right sensor configuration, generate a full BOM, and walk you through compliance requirements — at no cost, no obligation.