On the Bench: What We're Writing Next
- Justin Eckhardt

- 15 hours ago
- 4 min read
The Skill Repository publishes what two decades on the shop floor actually taught us. Here is what is on the bench right now - drafts in progress, coming over the next weeks. No fluff, no vendor brochures, just the stuff that costs you a weekend if nobody tells you.
Why vision installations should be done with two people
One person at the HMI, one at the camera - and the real reason has nothing to do with speed. It is about what happens to focus, aim, and honesty when the person adjusting the lens is also the person judging the image.
There are two ways to face your monitor with respect to the door in your office
An engineering-shop meditation on trust, focus, and floor plans. You already know which way yours faces. We will argue both sides are wrong for reasons neither camp expects.
Lighting is Everything
Every failed vision system we have ever been called in to rescue had the same root cause, and it was never the camera. Darkfield, dome, collimated, polarized - and the five-dollar trick that outperformed a five-figure retrofit.
Tightening is a Snap
Torque, preload, and why the click of a properly set wrench is the most underrated sound in machine building. Also: the fasteners that came loose at 2am on a customer floor and what they taught us about assembly discipline.
What do whiteout, nail polish, Loctite and void stickers have in common?
They are all torque-witness and tamper-evidence tools, and each one has a rightful place on an industrial machine. A field guide to marking, sealing, and knowing when somebody touched what they should not have.
Has anyone ever successfully implemented an internally triggered vision system?
Free-running cameras, software triggers, and the seductive idea that you do not need an encoder or a photoeye. We have opinions, we have scars, and we are genuinely asking.
SolidWorks, Fusion 360, SketchUp, Inventor, AutoCAD... or ????
The CAD package question, answered the only honest way: by what actually ships machines. What we use for concept, what we use for fab, what we stopped paying for, and where AI-driven drawing generation is quietly changing the answer.
Milking a cow?
Automation in a dairy parlor is a controls problem where the product weighs 1,500 pounds and has opinions. Sensors, washdown, and patience.
Stacking furniture
A robot that handles furniture sounds simple until you meet the payload, the reach, and the showroom floor it has to live on.
A whole city's water treatment plant
Municipal water is the quiet giant of controls work: redundancy, alarms, and code that cannot fall over on a holiday weekend.
The geek squad for a major car company
Years of on-call support inside an auto plant. What breaks, what never breaks, and what the maintenance crew knows that the manual doesn't.
I saw it all, from American sedans and Japanese pickups to electric cars - Milpitas or Fremont
Three decades of Bay Area manufacturing through one integrator's eyes, from the joint-venture years to the EV lines that replaced them.
Dog food is breakfast cereal?
Extrusion is extrusion. Why pet food, breakfast cereal, and plastic pellets are the same process with different marketing.
Staying in every casino in Reno
Long commissioning runs mean long hotel stays. Field-service notes from a town where every room comes with a slot machine.
Museum curators are the pickiest customers ever.
Tolerances measured in emotion. Motion control for exhibits where the spec is 'it has to feel right' and the client can see half a millimeter.
Can you provide traceability across lettuce, ASIC chips, and diagnostic trays?
Same question, three industries, three wildly different answers. Serialization from the field to the fab to the lab.
What is the best way to roast vegetables with just electricity?
No gas line, no flame, production rates anyway. Electric roasting as a controls and thermal-uniformity problem.
Potatoes: from fishbones, to steam peeling, to bruise rot
A tour of potato processing - the defects, the peeling physics, and the vision systems that sort a crop that never looks the same twice.
Did we install the first AI vision system for manufacturing in California?
A neural-net classifier running on a production line years before anyone called it AI. We are digging through the archive to make the case.
Milling the state of California on a 4x8 Zenbot - backwards
Sourcing real topography data, turning it into toolpaths, cutting the whole state on a 4-by-8-foot CNC router, and discovering we had milled it mirrored. Plus what it was like collaborating with a celebrity sculptor.
The day the coffee-can wax line caught fire
Blow-molded containers, hot wax impregnation, and the afternoon the two stopped cooperating. What a fire teaches you about process design that no FMEA ever will.
Flying to the powdered-drink factory
A commissioning trip in a small plane with a former employee who now flies for a major airline. Some service calls start on the ramp, not in the parking lot.
When a search giant took over the green-card factory
One Silicon Valley building, four eras: a graphics-workstation pioneer, a drug-delivery company, a microneedle startup, then a search giant - while the plant that printed America's green cards carries on next door. An integrator's view of the Valley reinventing itself around the machines.
When a glucose test strip was the most profitable product in the Bay Area
Before software ate the Valley, a glucose test strip quietly out-earned everyone. What it was like building automation in that era, when a square inch of chemistry was worth more than a rack of servers.
How AI can help find Black Bart's treasure
The gentleman bandit worked the stage roads right through our backyard. Terrain models, old maps, and modern pattern-matching applied to Gold Country's favorite unsolved problem.
Want one of these prioritized? Tell us through the contact form. The full pattern library behind all of them lives in the Saber knowledge base.



Comments