A 25 person optics shop runs its Haas overnight with a vision guided cobot
High precision optical work for scientific camera assemblies; precision aluminum parts · Pick and place · Vendor published
This account comes from a robot maker, integrator, or their marketing partner. Treat every performance and payback figure as advertising until an engineer verifies it against your own job.
The job
Walt Machine Inc. of Lumberton, Mississippi receives one order each spring for about 6,000 camera housings and runs the whole thing on a single Haas VM-3, with 30 to 45 minutes of machining per side. Working eight hour days that took several weeks and left the machine idle all night, and hiring a temporary operator meant the order would be nearly finished by the time the person was trained. The shop now sets 15 to 20 unmachined blanks on a table, and a UR10 with a wrist camera photographs the table, moves in for a detailed shot of one part, loads it into the machine vise and signals the control to cycle. The president reports teaching the camera a new part takes about 10 minutes.
The shop
25 employees, published-by-vendor. Shift pattern: not published, though the published account describes an eight hour attended day plus four to six unattended hours at night.
The equipment
Universal Robots UR10 with a Robotiq Plug and Play wrist camera mounted directly above a Robotiq 2 Finger 140 gripper, loading a Haas VM-3 vertical machining center. In a later vendor follow up the shop added more robots and mounted a UR10 on a ceiling rail because of the small floor footprint, with that arm tending three different machines.
The published numbers
Output: twice as many parts machined per day with the first robot, published-by-vendor. Unattended running: an extra four to six hours per night, published-by-vendor. Annual order: about 6,000 camera housings inside a two month window, published-by-vendor. Machining time: 30 to 45 minutes per side, published-by-vendor. Teaching time: about 10 minutes per part, published-by-vendor. Cycle time improvement: explicitly none, the published account states the cobot does not improve cycle time and improves job complete time instead. Later expansion: a 2,500 percent production increase, published-by-vendor. Payback: not published. Scrap: not published. Cell cost: not published.
Where a source did not publish a figure, this entry says so rather than guessing.
What we would check before believing it applies to you
This is the cleanest example in the set of automation buying machine hours rather than cutting cycle time, and the source says so outright, which is unusually honest for a case study. It only pencils out because the part runs 30 to 45 minutes a side, so one blank load buys a long unattended stretch and the vision task is trivial, finding a flat blank on an uncluttered table under shop lighting. A shop with two minute cycles, or with parts that arrive nested, oily or overlapping, should expect a very different outcome, because load frequency and part presentation are what actually break these cells. Grade this as vendor material even though you will find it under a Modern Machine Shop byline, because Robotiq published the original account first and the magazine byline reads Edited by rather than By. Treat the 2,500 percent figure with real caution, since it comes from a later vendor follow up and blends added robot capacity with a large jump in customer demand.
Sources
Is your job like this one?
Tell Joe what you are running. He will give you a straight preliminary read, and he will tell you if a robot is the wrong answer for your situation.
Read the pick and place guide