Inside the Build: Automating Precision at Scale

Featuring Troy Criego, Senior Automation & Robotics Engineer; Adam Kress, Automation & Robotics Engineer; David Franklin, Senior Product Development Engineer.

“That’s good math. That’s the math you want to have.”

Welcome to Inside the Build, a podcast series that dives deep into how Forj Medical solves complex manufacturing challenges, told by the people whose knowledge, expertise, and ingenuity make Forj Medical the partner medtech companies trust when the technology is complex and the stakes are high.

The manual process, and why it was breaking down

The original process for this molding operation looked straightforward on paper. A 150-ton vertical press with a four-station rotary table. Each station held a mold base with eight cavities. An operator at each of two stations would place four needles, one into each cavity, before the table rotated and the press cycled. The cycle time to beat was 22 to 25 seconds. Four needles, eight cavities total, every cycle, every shift.

The needles themselves are the complication. Each one is roughly an inch long, U-shaped, with a small flag on the end containing a hole that has to line up with an alignment pin in the mold. The clearance is about ten thousandths of an inch. The stamped flag is three thousandths thick, easily bent, and the operators are wearing gloves. Even with a nest in the tray to hold orientation, a bumped tray meant needle points scattering across the bench, a safety risk, and time lost putting everything back.

The failure modes compounded. A bent flag would change the way it reflected light at the vision inspection station, flagging a part that then had to be pulled. A miss on the pilot hole could mean the needle resting next to the pin instead of on it. Mold a part like that and you don’t just scrap the part. You can damage the tool. A tool repair runs a half shift to a full shift, and when you’re producing eight parts every 22 seconds, that’s a lot of output left on the table.

And operators get tired. Over an eight-hour shift, the cycle time drifts. The motion changes from a precise vertical placement to a sideways laying motion that’s easier on the wrist but harder on the needle. Throughput became unpredictable, which made planning and forecasting harder than it needed to be.

Then the customer asked about going from eight cavities to sixteen.

Why automation, and why a SuperTrak

Sixteen cavities in 22 to 25 seconds, by hand, was not realistic. Two operators loading sixteen needles at that speed would have been a feat to watch and impossible to sustain. Qualification builds at sixteen cavities by hand confirmed what everyone already knew. The cycle time blew past target.

The team chose a SuperTrak magnetic conveyor as the backbone of the automated cell. The decision came down to flexibility. A traditional rotary table moves in one direction at one speed and locks every station to the same cycle. A SuperTrak runs independent pallets on a magnetic track. Each pallet moves on its own. Forward, backward, fast through empty sections, slow around the corners where a needle could fly off a pin. New segments can be added if the line ever needs to expand. Branches can be added if a problem station needs to be pulled out of the queue and reinserted later.

The cell layout uses four robots arranged to minimize floor space. Two robots handle tray management from racks. Two more pick needles from the trays using vision. The needles arrive from the supplier upside-down relative to how they need to load into the mold, so the cell flips them onto a rotary station in groups of four, builds inventory of flipped sets on staged pallets, and a final robot picks four needles at a time and places them into the sixteen-cavity mold in four quadrants. Vision verifies placement before the press fires.

The cycle time on the new cell is 18 seconds for sixteen needles. Twice the parts in less time than the manual cycle for eight.

Details that mattered

Two design choices stand out. The first is using the top of the robot quill, not the end effector, to handle trays. The robots sit buried in the cell, working from the top down, and either pick robot can pull a tray from either rack. While one rack is being changed over by the operator, the other keeps feeding. The robots can cross between racks if one runs short. Less operator interaction, no second-shift dependency on tray management.

The second is the vision system. Each tray gets two camera passes. The first uses an angled side light that catches the two top posts of the U-channel and reads them as two lines. If the needle is flipped, those lines don’t appear, and the cell knows to skip it. The second pass is a backlit image that finds X-Y position and rotation for the pick. Building the backlight was its own problem. Off-the-shelf options didn’t fit the tray geometry, so the team built a custom backlight using strip LEDs with a Delrin diffuser. Too good a vacuum seal on the thermoformed tray distorted the part. Too weak and the tray bounced. The end-of-arm tool went through eight or nine revisions to balance hold, distortion, and dispersion.

Cycle time also came from robot path optimization. Simulation software let the team position the trays and flip stations in spots where the robot joints moved through their fastest envelopes. Joint speeds aren’t uniform, and the difference between an arbitrary layout and a tuned one shows up immediately in cycles per minute.

What’s still being refined

The flip station nests are the current focus. Each needle is held by vacuum during the flip and released by a small puff of air on the handoff. The bottoms of the needles aren’t perfectly flat, so the vacuum is really pulling air past them rather than sealing, and they flutter on the nest under certain conditions. Lot variation in needle geometry plays into it. Additional vacuum holes and material changes for wear are in progress, and after roughly ten million cycles on each machine, the team has data on where the wear is showing up that no one predicted in the initial design.

There’s also a gap in inspection between the pick station and the press. Once a needle is on a pallet, the next time it’s seen is when it lands in the mold. A camera over the staged pallets would let the cell flag a misaligned or damaged needle and either reject it or replace it in queue, since the SuperTrak makes that kind of reroute mechanically simple.

What we’d do again

The SuperTrak earned its place. So did the Epson SCARA robots, for repeatability and low maintenance, and the decision to use the top of the robot for tray handling. The vision system architecture, two passes per tray with the custom backlight, would carry forward.

The one thing the team would change, if they could rewind fifteen years, is the needle orientation in the supplier’s tray. The flip station exists because the needles arrive 180 degrees off from how they load into the mold. If that conversation had happened at the start of the program, the cell would be simpler, faster, and cheaper. It’s the clearest case study in this build for why design-for-automation conversations need to happen earlier than they usually do.

This has been Inside the Build, a podcast series by Forj Medical, where we’re shaping the future of life-saving devices, one build at a time. Thank you for listening. For more information, visit us at forjmedical.com.