Medical Device Assembly Machines Built in Singapore

Motionwell Automation builds medical device assembly machines in Singapore, meaning multi-station machines that assemble a device and produce the evidence that it was assembled correctly. The delivered reference is a 12-station rotary syringe assembly machine on a 15-second cycle, servo-indexed, with press stations that monitor insertion force and log every press against the part serial number, so an out-of-band assembly is rejected at the station that caused it rather than at final inspection. Machine-level indexing repeatability of plus or minus 0.05 mm was verified with a laser tracker over 1,000 consecutive index cycles at commissioning, jig contact surfaces are machined to plus or minus 0.02 mm, construction is cleanroom-compatible to ISO Class 7 and Class 8, and the machine shipped with an IQ/OQ/PQ package prepared under GAMP 5. Machines are designed, assembled and tested at our Woodlands Link facility, and the company has delivered more than 150 special purpose machines since 2014 under ISO 9001:2015 and bizSAFE Level 3.

Where we stand, said plainly before you read further. We do not manufacture devices or device components, we do not write your design history file or your technical documentation, we are not a notified body and we do not issue CE certificates. We are also not your quality unit: you own the validation and your QA approves it, and what we deliver into it is the machine, its instrumentation, its records and the protocols we execute alongside your team.

This page takes the decisions in the order they get made: what this class of machine has to do that a general one does not, why the precision core usually ends up on a dial, force-monitored insertion and what the curve tells you, particles, materials and cleaning, unit identity, the evidence each qualification stage wants, adjustability after qualification, when this is the wrong purchase, and what has to be settled before design freeze. The industry context sits on our medical device automation page, and the machine described above is written up in full in the 12-station rotary assembly case study. If you have a device drawing and a target cycle time, skip ahead and talk to an engineer.

What Makes a Medical Assembly Machine Different From a General One?

On a general assembly machine the deliverable is the assembled part. Here the deliverable is the assembled part and the record of how it was made, and the record is part of the product in a practical sense: a unit you cannot evidence is a unit you cannot release. Everything else on this page follows from that one sentence.

The design consequence is that the unit of engineering stops being the operation and becomes the operation plus the evidence it produces. Every station that can affect the device needs three things a general station does not carry: an instrument measuring what the station changes, a window the measurement is judged against inside the machine, and a log tying the result to the unit. A transfer or an empty index position needs none of that, and instrumenting it buys data nobody will read.

Which stations fall on which side of that line is a concept-stage decision rather than a documentation exercise at the end, because it sets the control architecture. Three things follow that reshape the machine rather than the paperwork. Verification moves upstream, from a final inspection that tells you a unit is bad to a station measurement that tells you which operation made it bad. The reject path becomes a record rather than a bin. And every parameter that can affect the device acquires an owner and a permission. Each is taken up below.

Medical automation equipment is bought against that list rather than against a cycle time, which is why two machines with the same throughput are not the same machine. The design-area comparison with a general industrial build is set out in our note on medical device assembly automation.

Why Does the Precision Core Usually Go on a Dial?

Two reasons that are specific to regulated device assembly, and one counterweight that overrides both.

The first is registration. On a rotary table the part is loaded into a nest and stays in it until the last station releases it, so the orientation established at load is the orientation every later station works from. That is what makes the tolerances above meaningful together: an index repeatability figure is only worth having if nothing between the first station and the last is allowed to disturb what it positioned.

The second is that classified air is expensive floor. Twelve operations in a circle need a smaller clean envelope than the same twelve in a corridor, and a machine can carry that envelope itself rather than requiring a room around it: our cleanroom automated test equipment holds ISO Class 7 and 8 over its fixtures with 4 to 6 fan filter units sealed into the enclosure ceiling, standing in a normal factory bay. Station density therefore buys classified volume as well as rent, and a compact guarded perimeter with one operator position rather than an access point every few metres.

The counterweight is that a dial’s station count is geometry, fixed the day the table diameter and nest pitch are drawn. Device processes acquire stations: a corrective action adds an inspection, a customer audit adds a check, a market adds a unique device identifier and the mark and its verification arrive as two more positions. If your sequence is not finished, the architecture argument goes the other way, and the full comparison, hybrid included, is on our rotary indexing versus inline assembly page.

What Does a Force-Displacement Curve Tell You That a Pass or Fail Cannot?

Of everything a medical assembly machine records, this is the one we would keep if we could keep only one, and it is worth being precise about why.

The insertion stations on the delivered rotary machine are pneumatic press-fit actuators, each carrying a strain-gauge load cell that measures insertion force in real time through the whole press stroke rather than at the end of it. The PLC records the complete force-displacement curve for every assembly cycle. The control is closed-loop against a window: if the measured force leaves the programmed tolerance band at any point during the stroke, the press halts immediately and the part is flagged for rejection, and the data is logged per serial number.

A press fit is an interference joint, and the joint is invisible once it is made: a correctly seated component and one that stopped a fraction short both present as an assembled device. The only other way to separate them is a test that pulls the joint apart, which destroys the unit and therefore samples. Recording force against displacement converts a destructively sampled attribute into a record held for every unit, and that is the whole argument.

The last two rows are why the record is force against position rather than a force reading. Force alone can be right for the wrong reason; force plotted against displacement is a corridor in two dimensions, and a joint that arrives at the correct force in the wrong place falls outside it.

Three further returns come from the same data at different times. Immediately, it localises the fault, because the station that made the joint is the station that rejects it. Over a shift, it trends: a population sliding towards the edge of the window shows as movement in the curve well before it shows as a reject, which is how a moulding tool wearing at a component supplier reaches you as a chart rather than a complaint. At audit, it is attributable, because each curve carries a serial number rather than a batch.

State the limit as clearly as the benefit. The curve proves the press, not the device. It says a component entered with the expected resistance and stopped in the expected place, and nothing about whether the finished device functions, seals or holds pressure; those are separate stations with their own instruments, on our automated test equipment and machine vision inspection pages. One consequence to settle early: a full curve per press, per cycle, per serial number is a meaningful data volume over a production year, so where it lands and how long it is kept is a design decision rather than an IT decision found later, as covered on the electronic device history record page.

Where Do the Particles Actually Come From on an Assembly Machine?

Assembly machines have a harder version of the cleanroom problem than conveying or packing equipment, because their particle sources sit where the product is rather than upstream of it. Four are specific enough to this class of machine to name.

Pneumatic exhaust at the press. A standard cylinder vents through a muffler on every stroke, which means it releases air into the room at the precise moment the tooling is over the device. Cleanroom-rated cylinders from suppliers we already buy, such as SMC and Festo, with the exhaust piped back to a manifold outside the clean zone, is the answer, and it is far cheaper to specify than to retrofit.

Bowl feeders. A vibratory feeder is a mechanism whose entire function is components sliding against a steel track for a whole shift. On the delivered rotary machine the feeders orient and present components from outside the ring, so the generating mechanism sits outside the classified envelope even though the loading position is a station inside it. That separation is decided at concept, and there is no later fix for it.

The nests themselves. Every load and unload is a rubbing contact against the part, repeated for the life of the tooling, which is why nest material is selected for wear as much as location and why nests are planned as consumables.

Everything that moves under the product plane. Guides, screws, belts and cable carriers behave here as on any cleanroom machine, and the treatments are the same ones worked through in our cleanroom automation guide.

The design move underneath all four is a boundary: the generating mechanism on one side, the product plane on the other, filtered downflow between them, and every crossing sealed rather than slotted.

How Do Material Choice and Cleaning Chemistry Decide Each Other?

An easy gap to leave in a specification is to name a steel grade for the frame and say nothing about the surfaces that touch the device. On an assembly machine most product-contact area is not steel at all: it is nests, gripper pads, press tooling faces and the chute the finished unit slides down, and those are the surfaces a cleaning validation has to cover.

The choice between SUS304 and SUS316L is usually driven by the cleaning chemistry, chlorides in particular, rather than by the device. Engineering plastics are chosen for different reasons: POM holds dimension at a friction coefficient of 0.2 to 0.35 against steel, and neither POM nor nylon sheds conductive debris into a Class 7 or 8 room the way a metal nest can, or brings a static discharge path to sub-components that do not tolerate one.

Two consequences follow that a specification can easily leave out. Polymer nests creep under sustained clamping load, so they are wear parts with a planned replacement interval rather than permanent tooling, and that interval belongs in the maintenance schedule at handover rather than being discovered when a dimension drifts. And the nest has to survive your disinfectant as well as your part: one that swells or crazes in service has lost its tolerance, and tolerance was the reason it existed. Send the cleaning agent and the contact time with the drawing.

How Does a Unit Stay Identified From Component to Finished Device?

The machine is where lot-level component identity is bound to unit-level device identity, and it cannot bind more than it knows.

Upstream, a bowl feeder destroys individual component identity by design. Components are traceable to the load that went into the hopper and the changeover event that ended it, not to the piece. The record should say so, because a traceability claim implying piece-level component identity through a vibratory feeder will not survive an auditor asking how.

Downstream of the load station a dial makes unit identity almost free: the nest position is the part identity for the whole build, so every station’s result is attributable without reading anything off the part. That is what lets the delivered machine store a force-displacement curve at each press station against a serial number, link the vision results, measured values and images to the same one, and sort by inspection result into packaging trays at the output.

So the split to write into the specification is short. Bound at unit level: force and displacement at each press, the vision measurement and its image, the station sequence with timestamps, and the reject reason. Held at lot level: component lots by feeder load, nest sets by change part record, and cleaning, environmental and maintenance events against the machine rather than the unit.

Two things then decide whether that record is usable. Rejects have to be reconciled rather than binned, so what the machine says it made, what it passed and what it removed agree; that discipline is on our code reading and traceability page. And the record has to land somewhere controlled. We deliver that side as its own scope: an electronic device history record system taken as a standalone order in August 2026, rolling out site by site for a medical device manufacturer.

What Evidence Does Each Qualification Stage Want From the Machine?

Three stages, three questions, and on an assembly machine they land on different hardware. What follows is what the machine has to hand over rather than how the protocols are structured; the V-model, GAMP categorisation and traceability matrix are on the computer system validation page.

The operational qualification row is where this class of machine differs most. A filling machine is challenged by dosing at the ends of its declared range; a press station is challenged by presenting it with the defects it exists to catch, which means somebody has to make bad parts on purpose and keep them labelled. Agree who prepares those challenge samples early, because a protocol calling for a misaligned component and a stores cupboard containing none is an easy way to lose a week at site acceptance.

Protocols are prepared under GAMP 5, and on our GMP builds the commissioning and qualification sequence follows ISPE Baseline Guide Vol. 5, which matters commercially rather than academically: engineering commissioning runs first and generates test data, then qualification uses that data where it was collected under controlled, documented conditions instead of repeating it for form. Factory acceptance testing happens at Woodlands Link before shipment and site acceptance testing at your cleanroom, and the package includes protocols, executed test records, deviation reports and a validation summary report. Where force curves and inspection results have to be retained as records rather than as files, 21 CFR Part 11 decides where they are stored, who may alter them and what audit trail follows.

Why Is a Machine That Is Easy to Modify a Liability After Qualification?

Because adjustability and evidence pull in opposite directions, and the second half of that trade is easy to miss when the machine is being bought.

On a general industrial machine, a technician turning a knob, nudging a stop or retuning a setpoint from the HMI is a feature, and a good machine has plenty of them. After performance qualification every one of those is either a controlled parameter, with permissions, an audit trail entry and a change control behind it, or an uncontrolled one, in which case the record no longer describes the machine that made the product. There is no third category, and an audit trail showing a press force changed overnight with no impact assessment attached is worse evidence than no adjustability at all.

The design goal is therefore not a machine without adjustment. It is a machine where everything that can affect the device is a controlled parameter with a bounded re-test, and everything else is fixed in metal.

One design principle keeps the cost of all this down, and it belongs in concept review: keep parameters local. A machine whose settings belong to individual stations, with no tuning constant shared between them, has a re-test scope of one station when something changes. A machine with a global adjustment several stations read has a re-test scope of the whole machine, a difference invisible on a datasheet and expensive for a decade.

So the question to put to your quality unit before detail design is which parameters they intend to treat as qualified. That answer decides the HMI permission model, the recipe structure and the audit trail scope, and retrofitting it into a finished PLC program is an expensive way to learn it.

When Is a Custom Medical Assembly Machine the Wrong Buy?

The section that decides whether the rest of this page is worth trusting.

The device design is not frozen. Tooling is cut around a geometry. If the geometry moves after detail engineering starts, the nests, feeders and press tooling move with it, and that is scrapped work rather than a revision. Freeze the product first, or buy the inspection station now and the assembly machine next year.

The volume does not justify hard tooling. Where a variant will not run long enough to pay for nests and feeders across every station, a validated manual process with a good fixture and an operator is the honest answer, and a shorter route to a qualified product.

The bottleneck is inspection, not assembly. Inline verification usually pays back faster, disturbs a validated process less, and produces the defect data needed to specify an assembly machine correctly. If you cannot yet state your escape rate to one decimal place, automate the measurement first.

One operation takes minutes while the others take seconds. A cure, a soak or a functional test running for minutes does not belong inside a machine whose parts all move together. It goes offline or into a parallel bank, whatever architecture surrounds it.

You need one machine across unrelated device families. Twelve stations tuned to a syringe geometry are not re-taskable to a different device by recipe alone: new nests, new feeders and re-qualification is a rebuild presented as a changeover.

A proven standard machine already covers your assembly. Where one does, at a lower price than anything we would build, the useful answer is to say so, and that costs you a conversation rather than a commitment.

One exclusion while we are being direct: we do not build production welding cells. Guarding scope follows from a risk assessment whichever way the decision goes, and on the delivered rotary machine that assessment put the feeder replenishment door at required performance level d, worked through on our machine safety and compliance page.

What Has to Be Settled Before Design Freeze?

These are the decisions that cannot be revisited cheaply once the frame is welded and the nests are cut. Run them in order; the first that has no answer is where your concept review should start.

1. The assembly sequence, operation by operation, with the seconds each one needs. Station count, architecture and cycle all follow from this list, and nothing useful is decided before it exists.
2. Which stations can affect the device. That set decides where instruments go, and it is the difference between a machine you can qualify and one you can only inspect.
3. The window for every measured operation, and who owns it. A press force and displacement band is a product specification rather than a machine setting: it comes from your process development, and we build the station to hold it.
4. Cleanroom class, cleaning agents and contact time. Class drives the enclosure and the airflow; the chemistry drives the gaskets, the labels and the nest material.
5. How components arrive, and whether component identity has to survive. Feeder, tray or magazine is a concept-stage layout decision, and it fixes how far upstream traceability can honestly reach.
6. The reject path and its reconciliation. Where rejected units go, who counts them, and how the count is closed against production.
7. Which parameters your quality unit will treat as qualified, and the variant range the nests must cover. The first sets the permission model, the recipe structure and the audit trail before the first line of PLC code; the second is cut into steel.
8. The qualification split. Who writes, who executes and who signs IQ, OQ and PQ, agreed at kickoff rather than in month five.

Frequently Asked Questions

Why is a force-displacement curve better evidence than a pass or fail flag?

Because a flag records the machine's conclusion and a curve records what happened. A press fit is an interference joint, and once it is made a correct one and a marginal one look identical from outside; the only other way to tell them apart is a pull test, which destroys the unit and therefore samples. Recording force against displacement for every press turns that sampled attribute into a per-unit record. It also localises the fault, because the station that made the joint is the station that rejects it, and it trends: a population drifting towards the edge of the window shows up before it makes a reject.

How long does a validated medical assembly machine take to build?

Lead time runs 16 to 24 weeks from concept approval to factory acceptance testing on a standard build, and 24 to 32 weeks where cleanroom compatibility or a full validation documentation package applies. The qualification scope itself typically adds four to eight weeks, and that time is protocol writing, execution and review rather than fabrication, so it runs partly in parallel with the build if the protocols are started early. The schedule risk that actually moves these dates is not ours: if the device geometry changes after detail engineering starts, the tooling changes with it.

Can force monitoring be added to an assembly machine we already run?

Mechanically it is usually straightforward, since a load cell in the press axis and a displacement sensor across the stroke are bought parts. The cost sits in the control system and the record. The measurement has to be sampled fast enough through the stroke to be a curve rather than a reading, the window has to be judged in the machine and act on the press, and the result has to reach somewhere it can be retained and reviewed. On a qualified machine the change is also a change control with an impact assessment and re-testing attached, which is a legitimate cost and a surprising one after the purchase order.