At Andy Marine, we believe that practical manufacturing knowledge is one of the most valuable assets built over time.
Inside our workshop, there are engineers, production specialists, and experienced technicians who have spent years working with stainless steel marine hardware — from material selection and casting processes to finishing and quality control.
Many of these insights are learned through daily production experience: understanding why a component fails, how different materials perform in marine environments, and what details truly matter when developing reliable hardware.
However, this practical knowledge is often shared only within the workshop and rarely reaches customers who are making sourcing and engineering decisions.
ENGINEER TALKS is our effort to organize these conversations and share the manufacturing experience behind marine hardware.
Through this series, we discuss materials, production processes, design considerations, and real-world lessons from the perspective of a marine hardware manufacturer.
Today's ENGINEER TALKS Ep.27 is presented by Engineer Zhang, written by me the editor.
Why a Sample Fits and the Batch Still Needs Rework
A custom bracket, hinge or latch arrives as a sample. On the bench it goes together cleanly: the pin slides through, the leaf swings, the fastener pulls the joint closed without a fight. It is signed off. Then the first production batch lands on the same bench, and a share of it needs a touch with a file, a tap with a mallet, or a fastener torqued down to pull two faces flat.
Nothing about the drawing changed, and nothing about the supplier changed. What changed is the number of parts being judged. The sample was one piece, and it was very often assembled by a person who had the mating part in hand and adjusted it until it fitted. A batch is not one piece; it is a spread of pieces, and the person who fitted the sample is not standing at the production line.
That gap — between "this one fits" and "any one fits" — is not a temperament problem on the supplier's side. It is a drawing, process and acceptance problem, and it is the subject of this episode. This article is written for the buyer or project engineer who has to decide what "custom hardware that fits" actually means before the purchase order goes out, because that is the cheapest place to fix it.
One Sample Proves One Part, Not a Process
Every manufacturing process produces variation. A sample is a single observation drawn from that spread.
| A sample | A batch | |
|---|---|---|
| What it is | one unit measured | every unit the process makes |
| What it shows | one point, between LSL and USL | the spread the units are drawn from, against the same limits |
| What it can support | a statement about one part | a statement about the whole population |
The stronger claim — the one a batch production line and a field spare-parts chain both depend on — is interchangeability: a part is interchangeable when it is made to a specification so nearly identical to others that it will fit any assembly of the same type and can freely replace another part without custom fitting such as filing. That is a statement about the whole population, and one measured sample cannot support it.
Bench fitting narrows the gap in a specific way, and it is worth naming the technique honestly. When two halves are measured and sorted so that a loose half is paired with a tight half, the deviation is absorbed at the assembly point rather than removed from the process; that is selective assembly. When a part is filed, tapped or pulled into place by hand, that is manual fitting. They are different techniques, and what they share is the important part: neither one tightens the manufacturing process, and neither one proves that an arbitrary pair of parts will fit. They guarantee the pair in front of you and nothing beyond it.
A first-article inspection has real, separate value. It validates that the setup is correct and the drawing was understood — the right program, the right offset, the right material. A setup error is systematic, so it repeats on every unit, and one thorough check at the start protects the whole run from a mistake that would otherwise be cloned thousands of times. What a first article cannot do is prove that the process will stay centred. Tool wear, gradual thermal growth and a fixture working loose all arrive after the first piece, and the first piece was genuinely correct when it was measured. Expecting the run to hold because the first article passed is one of the most common assumptions in custom hardware sourcing, and it is the assumption that produces a reworking second batch.
Start From the Fit, Not From the Dimension
The phrase "critical dimension" means nothing until it is tied to a function. On a mating component the function is a fit: a clearance, an interference, a seal compression, an alignment, a stop position, a latch engagement. The dimension that matters is the one on the shortest physical path that closes that gap — the dimension chain — and the real work is to trace that chain rather than to circle every number on the drawing.
Two parts can each sit comfortably inside their own tolerances and still interfere or rattle together, because the fit is decided by the sum of the chain, not by any single dimension. This is why "we hold the same tolerance on everything" is not an answer to a fit problem: the composite of several in-tolerance dimensions can still exceed the gap they are meant to close.
So the first DFM question is not "how tight can you hold it?" It is "which gap are we protecting, and which dimensions close it?" A buyer who can answer that in the request for quotation gets a very different, and usually more useful, answer from one who sends a drawing and asks for a good part.
A Dimension Without a Datum Is a Different Measurement to Everyone
A critical feature should carry a clear datum reference, a defined measurement state — free or constrained — and a stated verification method. ASME Y14.5 is the reference most drawings point to for communicating design intent so that parts have the intended form, fit, function and interchangeability; ISO 5459 covers datums and datum systems. The procurement consequence is unglamorous but decisive: with no agreed datum, the supplier's dimensional report and your own inspection can both be competently executed and still disagree, because they measured from different references, or in different states, or with a different setup.
Measurement also carries conditions. The standard reference temperature is 20 °C, and measuring at another temperature adds uncertainty, which matters most on the tightest features. These are not bureaucratic details to be settled at the end; they are the agreement that makes "in tolerance" a shared quantity instead of two private opinions.
Casting, Machining and Finishing Do Not Move the Same Surfaces
By the time a custom part is finished, its surfaces sit under two different tolerance systems, and confusing them is a common source of "it was in tolerance and it still did not fit."
Cast surfaces are covered by general casting tolerances. ISO 8062-3 provides the framework for casting dimensional and geometrical tolerances and for machining allowances. Machined surfaces are covered by whatever tolerances the drawing indicates individually. These are two systems, not one, and a dimension that runs from an unfinished cast surface to a finished machined surface does not simply inherit the tighter machining tolerance on the other end. If a fit interface lands on a raw cast surface, its achievable accuracy is not the machining tolerance the drawing may suggest elsewhere. General tolerances for dimensions that carry no individual indication — the ISO 2768-1 framework — fill in the rest of the silence, and they are looser than a designer may assume.
The second correction is about what machining and finishing actually do to a part. It is a mistake to picture material removal as shrinking every dimension uniformly. Removing material can make an external dimension smaller, but it can make a hole or a gap larger, and it changes edges, flat faces and contact geometry as well. Finishing — mechanical polishing or electropolishing — removes real material from the surface, in amounts that depend on the process and its purpose. The consequence is simple and easy to get wrong: if a mating dimension is dimensioned to its finished size while the part is still an unpolished blank, the finishing step will move it after the measurement that was supposed to prove it.
The answer is not to machine every feature oversize and then polish it back. Allowance has to be allocated feature by feature and by final state. Decide first which state the drawing governs — the finished state — and let the machinist carry the allowance only on the features that the finishing step will actually move. Note, too, that the cast-versus-machined split is about which tolerance system applies, not about which surface matters. A cast surface can be functionally important, and a machined surface is not automatically the only functional one. The design says which features are critical, which process makes them, and how they are accepted — and where the drawing is silent, the buyer is the one who will eventually discover the gap.
Matched Delivery Is Not Interchangeability
When a supplier delivers hardware as a matched pair or a set, the two halves were measured, sorted and paired to give a tighter fit than their individual tolerances would allow. The same is true of parts that were hand-fitted at the bench: the fit lives at that assembly point and travels no further.
Interchangeability is the stronger and less forgiving requirement: any part must fit any assembly, with no filing and no matching, and that has to hold both on the production line and when a single half is replaced from spares years later. A matched pair is not a worse product in the abstract — for a one-off, or for a complex mechanism where the two halves are genuinely a unit, it can be the right answer — but the two delivery modes need completely different acceptance methods, spare-parts strategies and field-repair expectations.
The trap is that the requirement is often never written down. If the drawing and the order never say "arbitrary interchangeability," a supplier may reasonably deliver pairs that fit each other and nothing else, because that is what the delivered sample demonstrated. Whether you are buying matched sets or truly interchangeable parts is a confirmation item for the buyer and the supplier to settle before the order, not something to read into a product page after the fact.
What Each Verification Step Can and Cannot Prove
Custom hardware is usually checked with some combination of the steps below. They answer different questions, and none of them substitutes for another; the recurring failure is reading one as if it were a different one.
| Step | What it does establish | What it does not establish |
|---|---|---|
| Sample / first-article inspection | That the setup and the reading of the drawing are correct, so a systematic setup error is caught before it is repeated across the run. Formalised first-article documentation (as in the AS9102 method) captures this as a record. | Process stability or capability, or that a later unit will fit. In-run drift — tool wear, gradual heating, a fixture working loose — arrives after the first piece. |
| Functional gauge / go/no-go | Whether the interface assembles, as an attribute result (go / no-go), built from the drawing's functional intent and datum structure. Limit gauging is used to check interchangeability of parts. | A size reading; that every function of the part is good; long-term gauge adequacy. A gauge proves only what its own design and datum structure control, and limit gauging is practised on clean parts at a controlled temperature. |
| Batch sampling (acceptance sampling, as in ISO 2859-1, indexed by AQL) | Whether to accept or reject a lot under an agreed plan. AQL indexes a lot-acceptance plan; it is not a fixed proportion of defects that a good lot is permitted to contain. | That every part in the lot is good, or that the process is capable. An accepted lot can still contain nonconforming parts. |
| Process capability (as in ISO 22514) | How a stable process spreads relative to the specification and how well it is centred, given suitable data, stable process conditions and a reliable measurement system. | Anything about an unstable process, or anything beyond the characteristic actually measured. It does not replace correct drawing limits and a correct datum scheme. Capability and performance indices are not interchangeable terms. |
| Measurement system analysis (as in the AIAG MSA method) | Whether the measuring system itself — repeatability and reproducibility, gauge R&R — is good enough for the readings to be trusted. | It does not improve the part. A sound gauge on a drifting process still reports a drifting process. |
The last two rows are the ones buyers most often over-claim. Capability numbers describe one characteristic on one stable process; they are not a general warranty on a part. And none of these steps costs nothing to set up, so the honest DFM conversation is which of them the feature actually needs — usually a first article plus one functional check on the critical interface, with sampling for the rest — rather than asking for all of them on every dimension.
Drawing Versions and the Changes That Trigger Re-Verification
A drawing is a living document, and a batch is only as controlled as the revision it was made to. Methods borrowed from regulated industries are useful here as method rather than as mandatory marine requirements: a documented first-article requirement such as AS9102, and a part-approval process such as the AIAG PPAP framework, both exist to keep a change from quietly moving a dimension.
The part of those methods worth keeping is the trigger list and the record. A change to the design or material of the part, a change to the manufacturing process, a change of supplier or sub-supplier, a change of manufacturing location, tooling that is moved, replaced or refurbished, and production restarted after a long idle period are all capable of moving a dimension, so any of them should re-open the first-article agreement, fully or partially. An engineering change notice documents a change and drives its review, approval and release before it is implemented rather than after.
The practical control is to write this into the order: which changes the supplier must notify, and what re-verification follows each one. A revised drawing sent after the batch has shipped controls nothing at all.
The Buyer's Checklist: RFQ and Acceptance
Most custom-hardware fit disputes are settled before production, in what the request for quotation and the purchase order did and did not say. The items below are the ones that make the difference between a sample that fits and a batch that keeps fitting.
- Mating part and assembly state — what this part fits against, and whether the fit is judged free or assembled (bolted, clamped, seated).
- The function to be held — the clearance, interference, seal compression, alignment or stop position the joint must deliver.
- Functional dimensions with datums — the dimensions on the shortest chain that close that function, each with a controlling datum, a measurement state, and a stated verification method.
- Final surface state — which surfaces are functional machined faces and which are as-cast or finish surfaces, and the finish specified for each.
- Removal allowance — where finishing removes material, the drawing should govern the finished state and locate the allowance on the features the finishing step will move.
- Delivery mode — matched pairs or sets, or arbitrary interchangeability, stated explicitly rather than assumed.
- Acceptance plan — what the first article contains, whether a functional gauge is used, which sampling plan applies, and whether any process-capability or measurement-system evidence is requested; plus the record that travels with each batch.
- Change notification and re-verification — which changes require notice, and what re-verification (full or partial) each one triggers.
- Records and deviations — inspection records, the drawing revision each batch was built to, and how a deviation or concession is approved.
Two cautions belong with this list. First, do not simply add tolerance to every dimension on the drawing: tightening dimensions that do not sit on a functional chain raises cost and cycle time and improves nothing. The tolerance belongs where the function is. Second, none of this is a promise of zero rework. It is a way to make the process and the interfaces controlled and verifiable, so that rework, when it happens, is a visible, traceable exception rather than an unexplained surprise in a crate.
How We Approach It at Andy Marine
Our customization page describes how custom work is meant to run in our workshop: a design review with in-house tooling; a functional prototype that is checked for fit, movement and the agreed application requirements; then transfer of the approved requirements to production, with later batches checked against the same agreed criteria. The same discipline runs through our OEM/ODM marine hardware work. The logic of this episode is the one written the other way around — the sample is judged against agreed requirements, and it is the requirements, not the sample itself, that carry into production.
If you are sourcing custom marine hardware and want to work through the functional dimensions, the mating part, the delivery mode and the acceptance plan before a single part is made, send us the drawing or a clear photograph, the mating component, and the function the joint has to hold. We can discuss the DFM points and help you set acceptance criteria that hold up at handover.
