Knowledge / Shipyards and production

Dimensional control in block assembly: datums and tolerance chains

Distinguish datums, thermal effects, tolerance chains and measurement uncertainty in ship-block assembly.

On this page

Two blocks can each pass a workshop inspection and still fail to fit together. The difficulty may lie in their individual shapes, but it may also come from incompatible reference systems, different support conditions or an incomplete definition of the interface. Dimensional control is the discipline of making those distinctions before steel is cut or forced into place. A dense point cloud is useful evidence only when the measurement question, coordinate system and uncertainty are understood.

Define the feature and the decision

Begin with the engineering decision: whether a mating edge can be assembled, whether a foundation lies within its required envelope, or whether a shaft-related feature has the specified alignment. Then identify the measurand, meaning the quantity actually being measured. “Block accuracy” is too broad. A distance between named datums, an angular deviation about a specified axis or a surface profile relative to the design model can be checked and discussed.

The drawing revision and applicable acceptance criteria must be fixed before interpretation. Design coordinates are not automatically inspection datums, and a convenient temporary target is not necessarily a stable reference. Record which physical points realise the coordinate system and how they can be recovered after transport, coating or partial removal. Otherwise, a later survey may be precise in its own frame but incomparable with the earlier one.

Keep coordinate transformations visible

A block survey often uses a local frame, while erection uses vessel coordinates. A transformation aligns the two through translation and rotation; scale should not be freely adjusted unless the measurement model explicitly justifies it. A best-fit algorithm can reduce average residuals while concealing an important mismatch at the actual assembly interface. The fitted result therefore needs an engineering interpretation, not just a favourable root-mean-square number.

Consider the difference between fitting an entire block and fitting its mating plane. Either may answer a legitimate question, but they do not answer the same question. Report the points used, rejected observations, fitting method and remaining residuals. If an outlier is removed, preserve the reason. Automatically discarding inconvenient data can erase evidence of a distorted panel, moved target or incorrectly identified feature.

Measurement traceability belongs to results

NIST's traceability policy explains that metrological traceability concerns a result linked to a reference through a documented calibration chain, with uncertainty contributions. An instrument carrying a calibration label does not make every field result fit for every purpose. The survey still depends on setup, environmental effects, target definition, operator actions and data processing.

The uncertainty question is practical: is the measurement capable of supporting the intended decision? A display resolving tenths of a millimetre may still have a much larger uncertainty over a long outdoor baseline. Repeated measurements under unchanged conditions reveal repeatability, but may not reveal a shared bias. Independent checks should address different failure modes rather than merely repeat the same setup and calculation.

Temperature and support conditions can change the answer

Dimensional reference conditions matter. NIST's metrology FAQ identifies 20 °C as the reference temperature for industrial dimensional measurement. That is not an instruction to maintain an outdoor hull block at exactly 20 °C. It highlights the need to define reference conditions and account for relevant thermal effects when comparing measurements with specifications.

For an invented uniform steel member, assume length 20 m, expansion coefficient 12 × 10⁻⁶/K and a uniform temperature increase of 15 K. The linear model gives ΔL = αLΔT = 0.0036 m, or 3.6 mm. The coefficient is an explicit teaching assumption, not a certified property for a particular heat of steel. Sunlit and shaded regions may have different temperatures, making a single-temperature correction inadequate.

Support conditions introduce another distinction. A long block on workshop stools, lifted at selected points and resting in its erection supports can have different elastic shapes. Record the condition in which the dimension is intended to apply. A survey taken during lifting is not automatically a substitute for a survey in the specified assembly state. Structural assessment is needed when support-related deformation could materially affect the result.

An original tolerance-chain example

Assume a fictional interface gap is g = D − a − b. Here D is the separation of two reference planes, and a and b are the distances from those planes to two mating features. Let D = 10,000 mm, a = 4,000 mm and b = 5,996 mm. The nominal gap is 4 mm. These dimensions define a one-dimensional teaching model; real block interfaces also involve angles and surface form.

Suppose bounded dimensional allowances are ±2 mm for D and ±1 mm for each of a and b. Worst-case stacking produces a gap range of 0 to 8 mm, because the possible absolute contributions sum to 4 mm. This is a geometric bound under the stated assumptions, not a probability distribution. It must not be labelled a 95% confidence interval.

If, in a separate statistical model, independent standard uncertainties were 2, 1 and 1 mm, the combined standard uncertainty would be √6 = 2.45 mm. That model has different input meanings from the bounded-allowance example. Correlation, shared datums and systematic effects can invalidate a simple root-sum-square calculation. Keeping tolerance, uncertainty and statistical variation distinct prevents an attractive calculation from answering the wrong question.

Design the measurement sequence around decisions

Measure features when correction remains manageable and when their final meaning is clear. A pre-weld survey can reveal fit-up problems; a post-weld survey can reveal process effects; a post-transport check can establish whether references survived. The sequence should follow the assembly strategy rather than collect every possible point at every stage. Excess data can obscure the few features that determine whether the next operation is ready.

The public NSRP accuracy-control project summary describes connecting accuracy data across work centres. That is a useful organisational direction, not proof that a particular data platform eliminates rework. A practical record joins feature identity, design revision, measured value, uncertainty statement, support and temperature conditions, acceptance decision and authorised response. It should remain understandable without reopening a proprietary measurement file.

Acceptance needs a stated decision rule

A measured value near a limit requires more thought than a red or green cell. The decision should use the applicable contractual or technical rule for accounting for measurement uncertainty. Do not invent a universal guard band. If no decision rule has been agreed, identify that gap before presenting a borderline result as a definitive acceptance or rejection.

NIST's discussion of uncertainty in dimensional calibration provides metrological background, rather than shipbuilding acceptance limits. Likewise, a general quality reference does not replace the vessel's applicable class and project criteria. Preserve the measurement report even when an engineering concession is granted: acceptance by assessment and conformance to the original geometry are different outcomes.

Common mistakes and a useful handover

Common errors include mixing millimetres and metres, reversing an axis, changing the best-fit point set without recording it, treating nominal model surfaces as measured steel, and interpreting scanner resolution as field accuracy. Another is comparing warm, restrained workshop measurements with cool, released erection measurements as if the assembly had not changed state.

A useful handover contains the recoverable coordinate frame, named interface features, conditions, uncertainty basis, deviations and open decisions. This guide does not supply fabrication tolerances or authorise cutting, forcing or heat correction. Its purpose is to make the measurement defensible: another competent team should be able to understand what was measured, reproduce the interpretation and see why the next assembly decision follows.

Sources