Gains:
- Ability to plan block production, welding and assembly processes with AI and accelerate quality control
- Ability to use digital twin and production data within an AI-supported project tracking framework
- Ability to verify AI's production and quality comments with NDT/inspection result and class approval
A ship is born by cutting and bending steel plates, welding blocks, pulling pipes and cables, and assembling giant blocks into a pool. A shipyard is a complex production environment where thousands of parts, hundreds of subcontractors and a tight schedule meet. A small delay or a welding error cascades and erodes the delivery date, cost and class approval. Artificial intelligence (AI) helps manage this complexity: in production planning, welding quality control, material logistics and project risk tracking. But production is ultimately a physical and safety-critical business; The output of AI is not a substitute for qualified inspection and class survey.
Limit: AI's outputs such as "this weld seam is acceptable", "this block is within tolerance", "this plan is realistic" are not accepted without confirmation by non-destructive testing (NDT), physical measurement and approval of the competent welding/quality engineer and class surveyor. A structural resource determines the integrity of the ship; False "acceptance" is the seed of a rift in the sea.
Concepts: Block construction: The ship is produced separately in large sections (blocks) and then assembled. Welding seam (weld): The bond where two metal parts are joined by heat; the basis of ship integrity. NDT: Non-Destructive Testing; searching for flaws inside without disturbing the source (x-ray, ultrasound). Tolerance: The allowable deviation range of a measurement. WPS: Welding Procedure Specification. Critical path: The chain of work that delays delivery if the project is delayed.
Production Planning and Project Risk
The shipyard project consists of hundreds of interconnected jobs: steel cutting, block assembly, painting, piping/cable installation, machinery placement, testing and delivery. AI models these jobs and their dependencies and helps determine the critical path, see bottlenecks early, and quickly calculate scenarios such as "how will delivery be affected if a subcontractor is 2 weeks late?" In material logistics, it plans which part needs to be on site and when and flags the risk of delays.
But the planning model is only as good as the data it inputs: unrealistic time estimates, missing dependencies, or weather/workforce uncertainty will fool the model. AI can produce an optimistic “everything in its time” scenario; The experienced project manager corrects this with the actual site condition and buffer logic. The value of AI is not to give a single definitive date, but to make scenarios and risks visible.
Tip: When having AI project plan, ask for three “worst / most likely / best” scenarios and the impact of each delay item on delivery. A single optimistic date is the most dangerous outcome of planning; The uncertainty range is necessary for the real decision.
Welding Quality Control and NDT
Welding is the most safety-critical job in shipbuilding; A ship's hull consists of thousands of meters of welded seams, and each one must be strong. In modern shipyards, AI-powered image processing scans the weld surface or x-ray/ultrasound images, marking possible defect candidates (pores, cracks, inadequate penetration) and speeding up human inspection. This is a powerful screening tool, but it carries two-fold risks: marking an actual defect as "acceptable" (missed defect) is very dangerous; Marking a good seam as "defective" is unnecessary repair and cost.
Critical principle: The “acceptable” output of AI is not a substitute for welding procedure (WPS) and class acceptance criteria. Welding quality; qualified NDT operator, appropriate inspection method and approved by class surveyor. AI draws attention to areas that humans might miss; But the final acceptance/rejection decision belongs to the certified person and the principle of "inspect if suspicious" prevails in terms of safety.
area
AI contribution
Human/class role
Production plan
Critical path, scenario, bottleneck
Realistic time, buffer, decision
material logistics
Delay risk flagging
Supply decision, priority
Source browsing
Marking a defect candidate
NDT approval, WPS/class acceptance
Dimension/tolerance check
Deviation detection
Physical measurement, correction decision
Project report
draft, summary
Accuracy and decision responsibility
Mini Cases
Case 1 — Missed weld defect. An AI-powered system marks a structural block junction source as “acceptable”. The quality engineer still orders ultrasound inspection because it is a critical link; inspection finds an internal penetration defect that is not visible from the surface. The weld is redone. If the AI output were to be relied upon alone, the fault would go out to sea. Lesson: on critical welds, NDT and class approval are retained even if AI says "accepted".
Case 2 — The optimistic plan trap. A shipyard commits to delivery according to the AI-generated “everything on time” schedule. The plan did not account for weather, subcontractor delays and rework allowance. In reality, the project is delayed by 5 weeks and a penalty is incurred. The commitment would have been more realistic if the experienced PM had asked for a three-scenario and buffered plan from the beginning. Lesson: not the only optimistic date, but the uncertainty gap and buffer are essential.
Case 3 — Unnecessary cost of rejection. An image processing system flags many sources as "defective" due to over-fine tuning; The team tries to repair them all. When confirmed by NDT, most of these are actually within acceptance limits. Unnecessary repairs waste both time and money. Lesson: Not every defect candidate that AI flags goes for repair until it is confirmed by acceptance criteria; False alarms are also a cost.
Copiable Prompt Templates
Template 1 — Project plan scenario analysis:
Role: You are the shipyard project management consultant. Context (representation): [project scope], main work items and estimated times are attached. Task: 1) Model dependencies and determine the critical path. 2) Generate three worst / most likely / best scenarios. 3) Show the impact of each major delay item on delivery. Constraint: Don't give a single exact date; Add buffer and uncertainty range. Write that time estimates must be verified with the field team.
Template 2 — Material/logistics risk screening:
Mark (representative) delay risks in the following supply plan:[part/subcontractor/delivery dates].1) Which parts directly affect the critical path?2) Mark long-lead items.3) If there is a delay, suggest an alternative/priority.Constraint: Suggestions are not decisions; The procurement decision lies with the project team.
Template 3 — Weld defect assessment support:
Evaluate (representative) an NDT/visual inspection finding:[defect type, size, location, weld type].1) What acceptance criteria is this defect type evaluated against?2) Which inspection method (RT/UT/MT/PT) is appropriate?3) Mark "doubtful" cases.Constraint: You don't make the accept/reject decision; The decision is made by a qualified NDT operator and surveyor according to WPS and class acceptance criteria.
Template 4 — Quality control false alarm balance:
I will evaluate the output of the AI-supported source scanning. [number of defects flagged, number confirmed by NDT].1) What is the false alarm rate, what is the unnecessary repair cost?2) What is the risk of missed defects (false negative)?3) How should this tool be positioned in critical resources?
Weak prompt / Strong prompt
Weak prompt:
Look at this source photo, is it acceptable?
Powerful prompt:
Role: You are a weld quality control consultant. Context (representation): a structural block joint weld; visual finding [definition], location-critical load-bearing connection.Task:1) What type of defect might this finding indicate?2) Which NDT method should it be confirmed with?3) Which acceptance criteria (WPS/class) apply?Constraint: You do not make the acceptance/rejection decision; NDT and classification are required in critical welding. Apply the "if in doubt, examine" principle.
The weak prompt asks for direct "acceptance" from the photo; The powerful prompt introduces the NDT method, acceptance criteria and human/class approval.
Common mistakes
- Substituting AI “acceptance” for NDT. Even if AI is accepted in the critical weld, non-destructive testing and class approval are preserved.
- Committing to a single optimistic date. A promise of delivery without buffers and scripts is a risk of punishment.
- Thinking of a false alarm as costless. Sending unnecessary "defective" markings for repair without confirmation wastes money and time.
- Not verifying duration estimates with the field. If the model's times are unrealistic, the plan is wrong from the beginning.
- Overlooking long supply items. Long-lead parts silently determine the critical path; should be marked early.
In summary
AI accelerates planning, logistics risk and weld quality screening in shipyard production; Makes the critical path and bottlenecks visible. However, production is safety-critical and physical: weld acceptance is confirmed by NDT and class approval, the plan is confirmed by realistic time and buffer, defect candidates are confirmed by acceptance criteria. AI's "acceptable" or "on-time" output is used as a preparation for, rather than a substitute for, qualified human and surveyor judgment.
Application task
For a representative shipyard project (main work items and durations), have the AI build a critical path and a three-scenario plan; Ask for the delivery impact of each delay item. Then, consider a weld defect finding with the "weld defect evaluation support" template and write down which NDT method and acceptance criteria it will be confirmed with. Debate the false alarm/missed defect trade-off of AI-assisted scanning. Indicate who has ultimate responsibility at each decision point.
checklist
- [ ] I retained NDT and class approval even though AI said "accepted" in critical sources.
- [ ] I set up the project plan with a scenario and a buffer, not a single date.
- [ ] I marked the time estimates to be verified with the field team.
- [ ] I identified long lead time items early on the critical path.
- [ ] I evaluated false alarm and missed defect costs together.
- [ ] I have subjected each acceptance/rejection and delivery commitment to the approval of a competent person/surveyor.