From Ship CAD to the Real Seam: Registration, Seam Finding and Tracking

A digital seam must be related to the real joint

A CAD model can identify the intended geometry of a ship small assembly, but it does not by itself locate the actual joint in the cell. The workpiece has to be identified, placed, related to machine coordinates and assessed locally before a welding path can be executed under the agreed procedure. Separating those operations makes a model-led proposal easier to evaluate and its failures easier to diagnose.

For this article, registration means establishing the relationship between the model or part frame and the machine frame. Seam finding means locating a joint or selected features before welding. Seam tracking means updating the tool's relationship to the joint during welding using the selected feedback method. These definitions are an engineering review vocabulary; the supplier's actual functions and interfaces must be documented for the proposed configuration.

TitanWeld's marine concept remains highly non-standard and project-specific: two arms per basic unit, with a current coordination ceiling of four units and eight arms. A model-led workflow does not remove setup, simulation, welding-procedure definition or engineering verification. Neither the robot count nor the existence of a CAD import establishes that every seam or local variation can be recognized automatically.

Check the information contained in the model

Confirm whether the supplied file contains only solid geometry or also includes usable weld definitions, identifiers and assembly metadata. A visible intersection between parts is not necessarily a required weld. Some edges may be intentionally unwelded, some may need intermittent welds and others may have a termination detail specified only on a separate drawing. Reconcile the digital model with the controlled fabrication documents.

Preserve weld identity when geometry is imported, converted or simplified. A change in file format can alter entity names or divide a curve into different segments. The project should specify how an intended weld remains associated with its drawing requirement and inspection record. Otherwise a regenerated path may look correct while its completion history no longer refers to the same scope.

ABB describes geometry-based offline arc-welding programming that creates robot positions from CAD-defined weld locations. [1] This is an external example of the technology category, not evidence that ABB software is installed in the TitanWeld marine project. Use it to frame a precise demonstration request: show how selected geometry becomes an identified welding task, including approach and departure movements.

Establish coordinate relationships explicitly

Identify the coordinate frames used by the project: machine or cell, robot base, external axis where applicable, tool, sensor, fixture and workpiece. The review does not need to expose every internal implementation detail to an operator, but engineering records must make the important relationships traceable. A local correction applied in the wrong frame can produce a plausible movement in the wrong direction.

Define the features used for registration and the conditions under which they are observed. They should distinguish the intended orientation of the assembly, including cases where repeated stiffeners create similar-looking geometry. The project must explain what happens if the wrong assembly revision is loaded, the part is reversed or a reference feature is obstructed. A low numerical fitting error is not enough if the wrong features were matched.

Verify registration at locations relevant to the weld scope, not only at the features used to calculate it. This is a proposed engineering check rather than a prescribed algorithm. It helps reveal whether a simple placement correction adequately represents the part or whether local variation requires separate assessment. Keep registration results and subsequent local seam corrections identifiable in the record.

Select seam finding by the feature it must observe

Ask what the proposed sensing method measures directly and what assumptions connect that measurement to the desired seam. A detected plate edge, a contact event or an optical profile can provide useful information, but the interpretation depends on the joint and surface condition. The enquiry should include real tacks, local obstructions and the surfaces expected at production handover.

Fronius's assistance-system documentation distinguishes pre-weld seam positioning functions from tracking functions and describes wire-based sensing among its options. [2] These examples demonstrate why the generic phrase automatic sensing is insufficient. They do not imply that those proprietary options are included in the TitanWeld system or that one method is appropriate for every shipbuilding joint.

Define failure outcomes before optimizing speed. A missing edge, inconsistent result or measurement outside the accepted input range should lead to a documented disposition. The system may request another observation or an operator review, depending on the approved design. Avoid an undocumented fallback that simply welds the nominal CAD line whenever a sensor result is inconvenient.

Distinguish tracking from permission to change the process

Tracking addresses the relationship between the tool and the observed joint during travel. It does not automatically authorize unlimited changes to current, voltage, travel speed, deposition or weld size. Any adaptation must remain within the selected process instructions and validated system behavior. The project should show which variables can change, why they change and what limit causes a hold.

TWI describes geometric sensors as inputs for searching, tracking and adaptive welding, and notes interference from the welding environment. [3] Therefore a pre-weld demonstration under clean lighting is incomplete evidence for in-process performance. Where tracking is proposed, examine its behavior with the actual arc, fumes, spatter conditions and tool arrangement represented in the development trial.

Keep the tracking record understandable. Useful evidence identifies the seam, the measured or inferred correction, the accepted limits and any interruption or loss of valid data. A dense graph without location, units or a stated meaning cannot support a practical acceptance decision. Likewise, a successful path trace does not substitute for examination of the finished weld under the agreed quality plan.

Manage corners, tacks and interrupted geometry

Long straight fillets are only part of the sensing problem. At a tack, an end return or a nearby bracket, the observed feature may change abruptly. The project should identify these regions in the weld map and show how the chosen method distinguishes a real seam transition from an unrelated feature. If a detail requires separate handling, preserve that boundary in the program and operator instructions.

Do not assume that a tracking correction can continue safely beyond the last valid observation. Establish the permitted response to lost or ambiguous data with the integrator and welding engineer. The appropriate action depends on the validated method, tool access and joint requirements. This article supplies no universal tracking window, correction limit or blind-travel allowance.

Use representative difficult features during the demonstration rather than removing them from the test piece. Record which details are automatically completed, which require a separate measurement and which remain excluded from the proposed automatic scope. This makes the coverage result more useful than a single percentage derived from the easiest continuous seam lengths.

Control the model, recipe and calibration together

A production operation should identify its drawing revision, imported model revision, generated path revision and welding-recipe reference. Add the relevant tool and sensor calibration state through the project's configuration method. The purpose is to reconstruct what the system believed it was welding and how that belief was translated into motion and process commands.

Define what changes trigger a review. Replacing a torch neck, moving a sensor bracket, servicing an external axis or changing a fixture datum may affect geometric relationships. A file update may affect seam identity or path extent. The response should be based on the changed relationship, not solely on whether the robot program file happens to retain the same name.

Include a practical recovery test after a representative approved intervention. Confirm that the operator can restore the required setup, perform the specified checks and select the correct remaining work. This is particularly relevant when an interrupted assembly has been moved for inspection or repair. A stored job number cannot guarantee that the previous workpiece relationship is still valid.

Validate the chain with deliberate test cases

Build a test matrix that follows the information chain: correct model and placement, accepted placement variation, local joint variation, obstructed reference, ambiguous seam feature and controlled interruption. Select cases within the project's agreed safety and process development methods. For each case, state the expected detection and disposition before conducting the trial.

Record separate results for identification, registration, seam finding, tracking where used, welding completion and final acceptance. This prevents one successful stage from being reported as a successful complete workflow. It also makes failure actionable: an incorrect assembly selection needs a different remedy from a valid seam location followed by an unacceptable deposited weld.

Measure the time required for these checks as part of the complete cycle. More extensive sensing may improve confidence while increasing cycle time; improved upstream location may reduce the sensing workload. The buyer should see those tradeoffs in representative trials. A fast nominal path generator is useful only when the complete model-to-workpiece process produces accepted assemblies at the required production rhythm.

Ask for evidence that follows one real weld

An effective demonstration follows one identified weld from its drawing requirement through model import, registration, local observation, tool movement and examination record. Repeat that trace for an ordinary seam and a demanding detail. The resulting evidence is understandable to both the automation team and the yard's production and quality functions.

Require the supplier to label proposed functions, demonstrated functions and excluded conditions separately. If a sensing feature belongs to a selected third-party component, verify the actual option, integration and trial evidence rather than transferring a manufacturer-wide capability statement into the cell specification. The physical and software configuration tested must match the claim being made.

Model-led welding becomes useful when it reduces repetitive preparation while preserving control of the real joint. The strongest proposal makes that relationship observable and recoverable. It shows where the model provides intent, where measurement supplies actual position, where the process instructions constrain welding and where inspection determines acceptance, without collapsing all four responsibilities into the word intelligence.

Project review checklist

• Confirm model revision, weld identity and the information available in the export.
• Check the relationship between nominal geometry and the actual assembly.
• Separate registration, initial joint finding, process tracking and inspection.
• Define responses to missing observations, changed geometry and interrupted work.
• Trace a representative weld from drawing requirement to examination record.