Machine tending automation: buffers, part presentation and the machine interface decide the cell

The buffer decides how long the cell runs alone

A tending cell runs until it has nothing left to load or nowhere left to put what it made. Everything else in the installation is subordinate to that. Sizing input and output storage against the shift pattern is the first calculation of the project, and it settles floor space, whether trays or stacks are used, and how often an operator must visit. Buffers fail in specific ways: trays that nest badly, stacks that topple, finished parts needing to cool or drain before they can be stacked. Design the empty-tray and full-tray handling explicitly, since that is where operators end up spending their attention.

Presenting parts in a position something can pick from

Something has to know where each part is. In rising order of cost and falling order of reliability: parts held in fixtured trays, parts in a semi-ordered layer with vision locating them, and parts loose in a bin. Bin picking has improved a great deal and is still the slowest and least predictable route, so price the trays before assuming the bin. Blank condition counts as well. Burrs, coolant film, tangled parts and inconsistent castings defeat grippers that worked perfectly on samples. Where an upstream process supplies blanks, changing that process to deliver them in order is often cheaper than buying sensing to cope with disorder.

Talking to equipment that was never designed to be commanded

Machine tools, presses and moulding machines can nearly all be automated, but the interface ranges from a documented port to nothing at all. New equipment is normally specified with an automation package covering door opening, clamp control, cycle start and status signals, and that package is far cheaper ordered with the machine than added afterwards. Older equipment may need external door actuation, relays wired into the control, and added sensors to confirm what the machine will not report. Two questions follow: whether the modification affects the builder's support position, and whether it disturbs the safety functions already fitted.

Balancing one arm across several machines

One arm serving several machines looks attractive because robot utilisation rises, and it is risky because the machines then queue. It works where cycle times are long relative to load and unload, where the machines run comparable parts, and where a stoppage on one does not idle the others. It works badly with short cycles, since the arm becomes the constraint and every machine waits its turn. Model the interference before committing, using the real spread of cycle times rather than the nominal one, and be honest about what happens when a single machine faults while the arm keeps serving the rest.

What the person on the cell now does

The role shifts from loading to keeping the cell fed and correct. That covers replenishing and clearing buffers, running the opening part check and periodic gauging, swapping tooling or grippers at product change, and diagnosing why the cell stopped. It also covers judgements nobody wrote down: noticing that parts are coming out warmer, that a gripper has begun marking a finished face, that a pick is grazing the tray. Give the cell a named owner. Cells without one drift into a state where somebody loads them by hand because that is quicker than fixing whatever stopped working.

Frequently asked questions

Can we automate a machine that has no automation interface?

Usually, though the cost sits in the workarounds. Door actuation may need adding, clamping may need external control, and cycle status may have to be inferred from added sensors rather than a proper handshake. Each of those is a modification to somebody else's equipment, so check the support and warranty position first and expect the safety functions to need reassessment. Where the machine is near replacement, ordering the next one with an automation package included is the better spend.

Is bin picking worth it, or should we present parts in trays?

Price both before choosing. Trays add handling, storage and either a person or an upstream process to fill them, but they deliver a fast, predictable pick. Bin picking removes that handling and gives back cycle time, occasional failed picks and a sensing system somebody must maintain and re-teach. Where an upstream process could deliver ordered parts cheaply, take that route. Where parts genuinely arrive loose from a supplier or a tumbling operation, bin picking earns its place.

One robot per machine, or one robot serving several?

Compare handling time against machine cycle time. Where the cycle is long, one arm serves several without becoming the constraint and utilisation improves markedly. Where cycles are short, the arm saturates and machines wait, which is worse than giving each one its own. Weigh failure behaviour too: a shared arm is a single point whose loss stops every machine it serves, whereas dedicated units fail one at a time and leave the rest producing.