Building a Custom Machine for a Production Line

Somewhere in most manufacturing plants there is a station where a person does the same motion 900 times a shift. It is slow, inconsistent, hard to staff, and it is where the line backs up. Automating it sounds like buying a machine, but no catalog machine does that exact job, so what gets bought is a custom machine — a one-off piece of capital equipment designed, built, debugged, and commissioned for one process in one plant. There is no volume to amortize development over: it is deliverable and prototype at once, and it must be maintainable by the plant's own technicians.

Requirements Come From the Line, Not From a Meeting

The biggest cause of failed automation projects is a requirements document written in a conference room by people describing what they believe happens on the floor. A proper capture starts by standing at the station and watching it run across shifts, including a bad day. Three categories come out.

The part

Not the nominal part — the whole population. What is the real dimensional variation, including the tail? How many variants, now and next year? How does the part arrive: bulk in a tote, in a tray, on a conveyor? Ask for a box of the worst parts the plant has, because feeding and gripping are sized by the worst case.

The process

Cycle time, uptime target, changeover time, quality criteria and how they are measured, and what counts as a fault. Also: what does the operator do that nobody wrote down? The judgment call, the little tap, the visual check — each gets engineered in or explicitly handed back to a person.

The plant

Floor space and installation access, power, compressed air, IT policy. Many manufacturers mandate a specific PLC brand, safety vendor, and HMI so their maintenance team can support the machine — find that out before design. Then write it up as a formal specification with measurable acceptance criteria, the same discipline as capturing customer requirements before design starts.

Cycle Time Is a Budget, Not a Target

"Six seconds per part" is not a specification until it is broken into a budget. Every motion, dwell, and settling time gets a line, and the sum has to close with margin. A pick-place-and-inspect station, for example:

Two levers appear immediately. Parallelism: overlapping operations — the return move during vision processing, a second nest loading while the first is worked — is cheaper than making individual motions faster. Margin: design to 70 to 80 percent of the required cycle time, because real machines lose time to jams and retries.

Then convert cycle time into throughput honestly. A well-built custom machine achieves 85 to 95 percent availability once debugged, less at first. If the business case only closes at 100 percent uptime, the case is wrong — the same arithmetic decides whether to automate at all, as in manual assembly or an automated line.

Controls Architecture: PLC or Embedded

This decision is made early and is expensive to reverse.

For a one-off machine in a customer's plant, a PLC is almost always right — usually the brand the plant already runs. Any industrial electrician can troubleshoot it, the certified safety modules are mature, and at 2 a.m. the plant's technician needs to open the program, find the rung that is not making, and fix it. Embedded or PC-based control earns its place when the machine will be replicated or the algorithm is beyond ladder logic, at the price of depending on its developer. Many good machines are hybrids: a PLC for sequencing and safety, an industrial PC for vision. Either way the electrical build follows NFPA 79, with a UL 508A listed panel where required.

Safety Is Designed In, Not Added

A custom machine is machinery under OSHA, and the employer carries the legal obligation for guarding. The builder's job is to hand over a machine that lets them meet it, documented.

US practice follows ANSI B11.0 for risk assessment, ANSI B11.19 for safeguarding performance, and ANSI/RIA R15.06 where a robot is involved. Run it before the mechanical design freezes:

1. Identify every task a person performs on or near the machine — loading, clearing jams, changeover, cleaning, maintenance, not just operating. Most injuries happen during those.
2. Identify each task's hazards: pinch, shear, crush, entanglement, stored energy, electrical, thermal.
3. Score severity, exposure, and avoidance to get a risk level.
4. Reduce risk in order of precedence: eliminate by design, guard, engineering controls, warn, then procedure and PPE.
5. Re-score, document residual risk, carry it into the manual.

Safety-rated functions — emergency stop, guard interlocks, light curtains, safe torque off — use certified components rated to a performance level set by the assessment, per ISO 13849-1. Do not build them from ordinary relays and PLC logic; use safety relays or a safety controller, separate from process logic. Lockout/tagout under OSHA 1910.147 needs physical isolation of every energy source, including stored pneumatic and gravitational energy: an axis that falls when air is dumped is a hazard the assessment must catch.

FAT and SAT: How the Machine Gets Accepted

Two acceptance gates, and the difference matters commercially. Structure payment around them — deposit, design freeze, FAT, final at SAT — so both parties share a written definition of "done," as in industrial machine development.

Factory Acceptance Test (FAT) happens at the builder's facility before shipment. The machine runs the customer's real parts at rate, in front of the customer, and every acceptance criterion is demonstrated and recorded: a continuous run of several hours measuring cycle time, first-pass yield, and fault frequency, plus tests of every safety function, fault-recovery path, and changeover. Watching it make three good parts is not a FAT.

Site Acceptance Test (SAT) happens after installation, on the plant's utilities, with the plant's operators, feeding real upstream and downstream equipment. It repeats the demonstration in the real environment and adds integration: connectivity, data logging, interfaces to adjacent equipment.

Spares, Service, and the Next Ten Years

The machine runs long after the build team moves on, and the decisions that set its cost of ownership are early ones.

• Use off-the-shelf components wherever possible — standard bearings, cylinders, sensors, and drives the plant can source in a day. Custom machining is for geometry that has to be custom, not a bracket that could be a catalog item. Motion tradeoffs are in linear motion guides and pneumatic vs electric actuators.
• Design wear items to be replaced fast. Grippers, tooling, belts, and cups wear; make them accessible without disassembly and repeatable without realignment.
• Ship a spares kit: sensors, valves, fuses, belts. A machine down two days for a $40 proximity sensor is what plants complain about most.
• Document properly: electrical and pneumatic schematics, assembly with part numbers, commented PLC program, operator manual, maintenance schedule — source files, not PDFs.

Cost and Schedule

A single-station custom machine commonly lands between $80,000 and $250,000 and takes four to eight months to SAT; a multi-station cell with robots and vision runs several hundred thousand and up. Controls engineering, debug, and commissioning account for 40 to 50 percent of that effort, so schedule a real debug phase. Broader economics: setting up an assembly line.

Projects House designs and delivers custom automation equipment — requirements capture on the floor, mechanical and controls design, risk assessment, build through a global manufacturing network, and FAT and SAT commissioning. If a station on your line needs a machine that does not exist yet, describe it in our contact form.