When a Ship Design Changes, Which AI Checks Have to Run Again?
A ship design team completes a check on a three-dimensional model, then a designer moves a component or changes an opening before the next review. The earlier check may still be visible as complete, even though its result applies to a version the team no longer plans to build. A blanket rerun would consume review time, while relying on the old result could leave a changed requirement unexamined. The team needs to distinguish checks affected by the revision from those whose inputs stayed the same. That requires a traceable link between the finding, the model revision and the checklist requirement it was tested against.
Noahlogy has signed an agreement to join the AVEVA partner ecosystem and is testing whether AI can help engineers check 3D ship design data. The company has demonstrated data reading, visualization and checklist checks, Yusuke Fukuju, its president and CEO, told The Supply Chainer. The work remains at the technical verification stage. The demonstration gives the team a way to test whether findings remain traceable as the model and checklist change.
“An AI check should be tied to an identifiable revision of the design data and the checklist used,” Fukuju said. A result for one version of a model may remain useful as a record of that version. It cannot establish that a revised design still satisfies the requirement. Fukuju recommends reviewing the affected checks and rerunning them against the appropriate revision, with a defined workflow for review.
A result an engineer can inspect
A useful result needs to tell the engineer which source data and checklist requirement were used and show the evidence for the finding. A review screen might, for example, link a flagged component to the relevant drawing revision and point to the feature that triggered the flag. Fukuju says teams also need to measure missed issues, false alarms and cases where the available information is insufficient. A check that works reliably for one type of component or requirement needs its own evidence before being used for another.
Sergio Viera Velasco, territory manager for North America at CONTACT Software, described a similar record-linking problem in earlier coverage of recurring manufacturing defects. In his example, production and quality data “can be directly linked to the corresponding product structures, components, and design revisions in the PLM system.” His comments concerned the connection from the factory floor back to engineering. For ship design, the comparable test is whether a reviewer can trace a finding to the exact component and revision, then decide if it still applies.
Keep the release decision with the engineer
Fukuju sees AI as a way to draw attention to items that warrant review. Interpreting design intent, resolving ambiguity and making decisions about safety, compliance and design release remain with the responsible engineers and approval authorities. Noahlogy has not reported validation results for individual checks or established a production release process. For now, the test is whether the demonstration can produce findings that engineers can reproduce, inspect and act on as the model changes.
