Raw parts or pallets: how should the robot load the machine?

In Modern Machine Shop's 2018 Top Shops survey of 471 shops, 59% of the top-performing group ran unattended machining, but only 23% used machine-tending robots. Unattended machining is possible without a robot, the magazine noted, with equipment such as bar feeders, parts catchers, fixtures that hold several workpieces and pallet pools. The robot is the easy part. What decides whether a robot cell runs through the night is how the part gets from storage into the machine: as a raw blank into a vise that stays on the machine, or clamped on a pallet that was set up somewhere else.

What goes wrong in real cells

In the published accounts of robot cells below, the robot is not what failed. Alexandria Industries in Minnesota built its first robot cell in 2001 around two machines and a family of telecom parts. The telecom work disappeared, and whenever one machine was down for a new-job setup or maintenance, the robot and the second machine had to stop too. After a couple of years the shop removed one machine and dedicated the cell to a single part family, and spindle uptime and output on the remaining machine rose by more than 30%. The two cells it installed in 2005, each built around a family of like parts, ran 24/7 within six months on as many as 30 different jobs (Modern Machine Shop, 2019). Rekluse in Idaho engineered a dedicated robot process for one part family, and by the time the engineering was done, design changes to those parts had made it obsolete (Modern Machine Shop, 2015). Maskinfabrikken HMA in Denmark found the opposite bottleneck: its robots suddenly produced more than the staff could prepare and finish, so the shop hired new people. "The robot arms are 100 percent dependent on us," its CEO told DI in 2023.

The published accounts hold smaller lessons too. Alexandria left 4th-axis indexers mounted on its machines even for parts that did not need them, because reinstalling one and adjusting the robot program could take two hours; the shop judged the extra indexer fixtures worth it because they sped up changeovers. Toolcraft in Seattle found its robot busy for about 6 minutes of a 56-minute cycle, so it gave the robot rinsing and drying work as well (Modern Machine Shop, 2019). In the Chalmers University study described further down, one company's press-brake robot ran about 1,000 cycles before commissioning problems set in, and with deliveries taking priority, the company had not found the time or resources to solve them.

Process comes before the robot, too. In a Gardner Intelligence analysis of about 1,100 shops, presented in Modern Machine Shop in 2019, shops without a formal improvement method rarely ran unattended machining, and in the few that did, it made virtually no difference to profit margin. In shops that had one, profit margin was about 8% with unattended machining, against about 5% without it. The survey shows a correlation and cannot prove the cause, but it points the same way as the shop stories: the cells that worked were built on a process that already ran well.

The pattern behind these stories is variability. A raw-part cell fails when the blanks, the chips or the part mix vary more than the cell was built for. A pallet cell absorbs it at the setup station, where a person sees each part before it goes in. The rest of this article is about where you want that variability to land.

How each route works

Raw parts into a machine vise

Blanks sit in trays or drawers beside the machine. The robot picks one, reaches in through the automation door and places it in the vise, and the machine clamps it. After the cycle the robot takes the finished part out and puts the next blank in. The vise never leaves the machine, so the gripper carries only the part. A part that needs its gripped face machined gets a second operation: a second clamping position, or a turnover station where the robot regrips it. Gimbel Automation, which sells an in-machine alternative, writes (Today's Machining World, 2026) that such a two-operation cell needs the robot calibrated to two vise positions, and that the robot pedestal and flip station take 1.2 to 1.8 m (4 to 6 ft) of floor beside a typical vertical machining center.

Pallets prepared off the machine

An operator clamps parts onto zero-point pallets at a setup station (in a vise, in a fixture, or several parts per pallet) and puts the pallets into the cell's storage. The robot carries a whole pallet to the machine and sets it on the zero-point interface, which locates and clamps it. The setup for tomorrow's job happens while the spindle cuts today's.

Setup time decides more than the robot does

Al Youngwerth, president of the automation maker VersaBuilt, proposed a rule of thumb in Production Machining in 2018, written about turning: if setup time is more than 10% of the batch run time, the job is low-volume work. Run the numbers on a 90-minute setup, an example figure you should replace with your own. The rule puts the threshold at 15 hours of run time, which at 20 minutes per part is a batch of 45. A batch of 10 of the same part runs 200 minutes, so the 90-minute setup is 45% of the run time.

That is where the two routes split. With raw parts, the setup for a new part happens at the machine while the spindle stands: vise jaws, gripper fingers, the tray layout if the blank size changes, and the first-part prove-out. If a part family shares one stock size and one jaw set, the changeover between its members shrinks to a program call, and the cell can mix them in one night. With pallets, the next job's fixture is mounted and checked at the setup station while the machine runs, and the changeover at the machine is a pallet exchange.

The setup work does not disappear on the pallet route. It moves to the daytime, and it repeats for every part: someone clamps each part on its pallet before the night starts.

The money: what each part number costs you

Raw-part loading needs one jaw set and one set of gripper fingers per stock size, plus blank storage. Its hardware scales with the number of stock sizes. Pallet loading needs a pallet and its fixture for every load in storage, and its hardware scales with the number of part numbers you keep ready and the number of parts in a night.

Size the pallet count from the hours you want covered: pallets = hours × 60 ÷ (cycle time in minutes × parts per pallet). A 12-hour night (18:00 to 06:00) at 20 minutes per part needs 12 × 60 ÷ (20 × 1) = 36 pallets with one part each. A fixture that holds three parts brings that to 12 × 60 ÷ (20 × 3) = 12 pallets. The raw-part cell covers the same night with 36 blanks in storage and one vise.

The €2,000 per pallet with its fixture is an assumed figure for the arithmetic, not a quote; ask your workholding supplier for the real one. The shape of the result holds at any price: on short cycles, pallets with one part each tie up the most money, and multi-part fixtures or raw-part loading cut it.

The raw-part route pays its bill upstream, in blanks. The robot needs blanks of a consistent size, so sawing and stock preparation become part of the process. In a 2009 example in Modern Machine Shop, John Belzer of TCI Precision Metals, a supplier of machine-ready blanks, showed a plate part taking 81 minutes on the machine from a rough oversize blank and 54.83 minutes from a machine-ready blank. The 26 minutes saved (11 minutes of prep machining and 15 minutes of operator work such as changing clamps) were worth USD 40 per part against a USD 30 premium for the prepared blank. Squaring blanks on a 5-axis machine spends its most expensive hours on the least demanding work.

Storage: how long the night lasts

The robot does not set the unattended hours; the storage does. Unattended hours equal the blanks in storage times the cycle time. Vention, an automation vendor, says it often sees users load 16 to 32 parts on a tray base plate with nylon datum pins, and that stacked drawer systems of 3 to 5 drawers often hold 160 to 250 parts. Its comparison assumes a 2-minute cycle, where a tray lasts 30 minutes to an hour and a drawer tower up to 8 hours. 5-axis parts usually run longer. At 10 minutes per part, one tray of 32 blanks covers 32 × 10 = 320 minutes, about 5.3 hours, and three trays (96 blanks, 960 minutes, 16 hours) cover a 12-hour night. Pallet storage follows the same arithmetic with fewer, heavier units, as the pallet count above shows.

Storage can be simple. Grenå Spåntagning in Denmark lays round aluminum blanks of about 1 kg on a table in front of the machine, held in rows of teardrop-shaped holes that take blanks of different diameters, and for shafts the length of its chute sets how long the robot runs alone. The shop reported just over 300 robot hours in one month, 75% of them unattended, in runs of 3 to 4 hours (Metal Supply, 2018). Random bins with a vision system are the hard case: the robot has to find a part in a pile where, in the words of the robot industry association A3, the parts "keep shifting positions and orientations every time a part is removed."

Accuracy: the robot is the loosest link, and it matters least

Each maker in the chain publishes a repeatability figure. FANUC lists ±0.01 mm for the LR Mate 200iD/14L. Spreitzer states 0.003 mm repetition accuracy for the HZS 120-60 A centric vise, and ±0.005 mm centering accuracy measured to the alignment holes in its base. AMF states repeatability below 0.005 mm for its zero-point clamping modules, including the hydraulically released KH10.2 size. The robot's figure is the largest, and the numbers are not measured the same way: each maker uses its own method, and the robot's figure applies at its wrist, before the gripper and the blank's position in the fingers add their own scatter. It matters least anyway, because something else sets the final position in most directions.

In a centric vise the jaws center the blank in the clamping direction whatever the robot did, as long as the blank lands between the open jaws. The jaw stroke (2 × 5 mm on the HZS 120-60 A, per Spreitzer) has to cover both that clearance and the variation in blank width for one jaw setting. A fixed-jaw vise behaves differently: a blank 0.4 mm oversize moves its far face 0.4 mm and its center 0.2 mm. Along the jaws and in height the robot and the support face still decide, so let the gripper open before the jaws close. If it holds on while the vise clamps, gripper and vise fight, and the robot's error gets clamped into the part.

On the pallet route the robot only has to set the pallet down within the interface's lead-in, and the zero-point clamping module pulls it into place. Repeatable is not the same as known, though. A fixture mounted on a pallet at the setup station sits at its own offset until that offset is measured once and stored with the pallet. A cell that runs ten pallets needs ten stored offsets, and the program has to know which pallet it is cutting, through a fixed storage position or a pallet ID.

Chips: what stops the night

Chips come up in several of these trade press accounts, and the two routes expose different surfaces to them. On the raw-part route the jaw faces and the workpiece support open to the chips in every cycle. On the pallet route the zero-point face opens only at a pallet change, and the pallet is cleaned at the setup station. KAD Models, a prototyping shop, has its robots move pallets that people load. Its automation supplier Trinity told Modern Machine Shop in 2024 that human loading removes the risk of chips interfering with the workholding, and KAD's founder described chips in a vise: "it turns into a trash compactor." The same magazine reported in 2026, drawing on the workholding maker Jergens, that chips working into a vise add to jaw lift, and that unchecked jaw lift can add 0.20 to 0.25 mm of runout to a part. Its 2015 report on the robot systems at Rekluse and Pro Moto Billet describes part washing built into the automated system to keep stray chips off the parts and the vise jaws. As early as 2001, Sequoia Automatic in Illinois gave its robot an air nozzle on the gripper base that blew chips off a 12-pocket fixture before it unloaded the finished parts, so stray chips would not fall into the pockets and stop the next blanks from seating (Modern Machine Shop).

The counter-measures exist on the exact products in this chain. AMF lists pneumatic blow-out and support control for its KH10.2 installation clamping module, on its own optional air connection. Spreitzer lists air sensing and cleaning of the workpiece support among the additional functions for the HZS 120-60 A. Air sensing tells the control whether a pallet or part is seated; it reports seated or not seated, not a position, and a chip thinner than its threshold goes unnoticed. A touch-probe check of height and position after each load catches what the air check misses, at the cost of a probing cycle on every load.

A broken tool at night is the other classic stop, and here the pallet route has an edge. KAD Models has programmed an M99 fail-safe that, after a broken tool, loops the CNC back to the automation program: the cell carries on machining its other pallets without damaging the parts that need a new tool, so the broken tool does not cost the rest of the night (Modern Machine Shop, 2024). On the raw-part route a broken tool stops the part number the cell is running, unless a sister tool can take over. On the Unicorn 5-200 Modular, with 78 tools in the magazine and one in the spindle, sister tools stand in when a tool wears, and the optional Renishaw NC4+ Blue F145 tool setter catches breakage between cycles.

Payload: what the robot lifts

The standard cell uses a long-arm FANUC LR Mate 200iD/14L with a 14 kg payload, gripper included. FANUC also limits the allowable wrist moment: 31.0 Nm at J4 and J5 and 13.4 Nm at J6. A load whose center of gravity sits far from the wrist reaches the moment rating before it reaches the mass rating, so long fingers and off-center pallets need checking against FANUC's moment and inertia ratings as well as on the scale.

The arithmetic: 14.0 − 1.5 = 12.5 kg for raw parts, and 14.0 − 1.5 − 2.0 − 4.8 = 5.7 kg for a pallet carrying the vise. For scale, a 60 mm steel cube weighs about 1.7 kg (0.216 dm³ × 7.85 kg/dm³), which both routes carry. A pallet with a light fixture instead of the vise leaves more for the part. The machine sets the second limit: the Unicorn 5-200 Modular takes 45 kg for the part and its fixture together, so in the example pallet, vise and part (2.0 + 4.8 + 5.7 = 12.5 kg) stay well inside it, and the robot's payload is the limit a robot-loaded cell meets first. Height works the same way: the 5-axis work envelope, for the part and its fixture together, is Ø225 mm across and 233 mm high above the machine table, and the 86 mm Spreitzer lists for the HZS 120-60 A, plus its pallet, comes out of that height. A pallet with a low custom fixture leaves more of it for the part.

Grippers: where small cells stall

A 2019 Chalmers University master's thesis, part of a Swedish research project on movable robots in three metal subcontractors working mainly in sheet metal forming, found the main barrier was "lack of resources in the form of time and competence," which made it hard to adapt the products for automation and to design and make grippers. At one of the companies the products differed so much that each one needed its own dedicated gripper. Raw-part loading meets this problem with every new stock size; pallet loading meets it once, because the gripper holds the pallet's gripping interface and stays the same from job to job.

A gripper that holds a sawn blank by friction carries far less than its gripping force suggests. SCHUNK, for example, rates its PGN-plus-P 100-2-AS-V parallel gripper (1.1 kg, 2,200 N closing force) for a recommended workpiece weight of 8.75 kg, calculated with a friction coefficient of 0.1 and a safety factor of 2, and states that form-fit gripping allows significantly higher weights. A SCHUNK how-to guide puts typical friction coefficients between 0.1 and 0.4 (Assembly magazine, 2021), and coolant on the fingers or the blank pushes it toward the low end. Blanks that vary hurt too: at B.I.C. Precision in Ohio, forged blanks varied in shape, so the original electric gripper on its lathe cell was replaced with a custom pneumatic one (Modern Machine Shop, 2021). Fingers shaped to the blank, or a gripping feature machined or stamped into the stock, turn a friction grip into a form fit. The same idea works in the vise: Lang Technik, a workholding maker, says its stamping unit presses a contour into each blank off the machine with up to 20 tons of force in under 5 seconds, so its 5-axis vise holds the part by form fit with 3 mm of clamping depth. A double gripper that unloads the finished part and loads the next blank in one trip shortens the exchange: Modern Materials Handling reported in 2017 that an OnRobot dual gripper cut the load and unload cycle at the Danish worm-gear maker Osvald Jensen from 27 to 15 seconds. It takes roughly the weight of two grippers out of the payload.

The decision table

Many cells do both

The routes combine. A cell can feed blanks of a repeat part into the vise and keep a few pallets for jobs that need their own fixture, and the vise pallet itself can be exchanged when the night's work changes stock size. On the Unicorn 5-200 Modular both routes use the same machine-side hardware: the CNC-controlled 410 × 308 mm automation door, standard on every machine, the optional hydraulic feed-through machine table with its AMF 6204IHA K10.2 zero-point interface, and the optional Spreitzer HZS 120-60 A hydraulic centric vise on an AMF-compatible pallet with a robot gripping interface. The standard robot cell stands beside the machine and is designed for material loading, pallet loading, or both. It is quoted per configuration, because the gripper, the part storage and the cycle logic follow your parts. See how it is priced.

Shops have found several middle paths between the two routes. Zelos Zerspanung in Germany has its robot move vises instead of parts: operators set up new jobs on vises that wait on trolleys beside the machines, and the robot grips only the standard vise, so one gripper covers every job and a new job needs no robot teach-in. The system takes up to 168 vises on four trolleys, batches run from 10 to 1,000, and the shop reaches up to 20 hours of spindle uptime a day (Production Machining, 2022; the vises and trolleys are Lang Technik's). Rekluse's later robot system uses the part-specific soft jaws as its gripper, so a new part needs new jaws instead of a new gripper (Modern Machine Shop, 2015). Sequoia's robot loaded raw blanks into a fixture on the idle pallet of a two-pallet machine, raw-part loading and pallets in one cell.

To choose, list the parts the cell will run in its first year: the number of part numbers, the stock sizes they share, their cycle times, and the heaviest load the robot has to lift. Alexandria's approach after its restart holds for both routes: one family of like parts per cell, standard equipment and process, and no attempt to take on any and all jobs. The automation page shows how the cell is designed to dock to the machine.