Robot cells: fixturing, part presentation and getting out of a fault

Presentation is the problem you are really solving

A robot picks confidently from a known position. Getting parts into a known position is the engineering. Options run from trays and pallets loaded upstream, through tracks and escapements, vibratory or flexible feeders, to vision locating parts lying loose. Each buys tolerance for disorder at a price in cost, footprint, changeover time and reliability. The choice interacts with the supplier: parts delivered in moulded trays cost more per unit and remove a whole subsystem from your cell, which is frequently the better deal. Decide this before the layout, because presentation determines the cell's size, shape and most of its failure modes.

Fixtures decide the quality, and they wear

Whatever holds the part while the process happens sets the achievable precision, and it does so with no feedback at all unless you add some. Datum surfaces need to be the same ones the drawing uses, clamping must not distort a thin section, and swarf, adhesive or plastic flash must have somewhere to go rather than accumulating under a locator. Every fixture drifts as locating pins wear and clamps lose adjustment, which appears as a slow quality trend nobody attributes to the fixture. Build in a verification method — a reference part run at intervals, or sensors confirming the part is fully seated before the cycle starts.

Sequencing for the day something goes wrong

Most cell logic is written for the happy path and gets its exception handling improvised during commissioning. That is where unattended running is lost. Work through the failure cases deliberately: a part gripped but not detected, a part dropped mid-move, a downstream machine that will not accept, an aborted cycle with tooling still in the fixture. Each needs a defined outcome — retry, quarantine, stop safely — and a state the machine can be recovered from without an engineer. Cells that require someone to jog axes manually after every minor fault will only ever run as well as the availability of the person who can do it.

Guarding shapes throughput more than people expect

Access design is not just a safety obligation, it is a production decision. A cell where clearing a jam means stopping everything, isolating and opening a main gate loses far more output than one designed with a safe loading position, a light-guarded interface or a segmented safe area that lets part of the cell keep working. Those choices must be made with the risk assessment, not bolted on afterwards, because retrofitting a divided safe zone into a finished cell is expensive. National safety regulators publish guidance on machinery access and interlocking, and involving whoever will maintain the cell in that design conversation pays back quickly.

Cycle time is decided by the slowest thing, which is rarely the arm

Teams optimise robot motion, shaving fractions off a path, while the cell waits on a process step, a feeder refill, a fixture clamping or a downstream transfer. Map the whole cycle including the waiting and you usually find the arm idle for a good part of it. That has design implications: a second fixture that lets the arm load while the process runs, a buffer that decouples the cell from an unreliable neighbour, or a slower and cheaper arm than the one specified. Time the real sequence with real parts before concluding that a faster robot is what the cell needs.

Frequently asked questions

Should we buy parts in trays or build a feeder?

Compare the true cost of each. A feeder is capital you own, floor space you lose, a changeover you perform, and a recurring source of jams. Trays or pallets shift the ordering work to the supplier and appear as a piece-part premium plus a returnable packaging loop to manage. For fragile, awkward or high-mix components, presented packaging is usually worth it. For cheap, robust, high-volume parts of one geometry, bulk feeding tends to win. The decision belongs in the sourcing discussion, not only in the engineering one.

How do we stop a cell needing an engineer for every minor stoppage?

Design recovery as a feature. Give the cell a defined home state it can always be driven to, write fault messages that name the physical location and the action needed, provide a guarded way to remove a stuck part without full isolation where the risk assessment permits, and train operators on the specific faults that actually occur. Then track which stoppages still escalate to engineering and fix those individually. Most cells have a handful of recurring nuisances that account for most escalations.

How much buffer should sit between a cell and the operations around it?

Enough to absorb the typical short stoppage of the neighbouring operation without starving or blocking the cell. Directly coupling equipment means every small fault anywhere propagates immediately, and combined availability falls well below that of any individual machine. The counter-argument is that buffers hide problems and tie up work in progress, so size them from observed stoppage patterns rather than by habit, and keep visibility of how full they run.