Gains:
- Ability to speed up initial sizing and design discovery by giving hull form, resistance and propulsion parameters to AI in a structured format
- Ability to query the setup of CFD (computational fluid dynamics) studies, network and boundary condition logic with AI support
- Ability to capture fitting coefficients and results by cross-checking AI's hydrodynamic predictions with model experimentation, empirical method, and physical plausibility
The entire economy of a ship is hidden in its submerged form. When the hull form (hull form; the ship's underwater and surface hull geometry) changes by even a millimeter, the fuel it will burn, the cargo it will carry and the emission bill will change throughout the life of that ship. Therefore, form design and hydrodynamics (the science of the flow of water around the boat and the forces it causes) are the stages where the most expensive decisions of naval engineering are made. Artificial intelligence (AI) accelerates three tasks at this stage: initial sizing and design exploration, installation of empirical resistance methods, and planning of computational fluid dynamics (CFD) studies.
But let's nail it from the start: no resistance curve, no form coefficient given by AI translates into a propulsion decision without being confirmed by model experimentation or a validated method. Water does not read formulas; The actual behavior of the boat is known only by measurement and physical plausibility. AI is an accelerator and idea generator here; hydrodynamics is not a witness to reality.
Concepts: Resistance: The opposing force exerted by the water on the boat as it moves forward; It consists of friction, wave and vortex components. Block coefficient (Cb): The ratio of the underwater volume of the hull to the rectangular prism surrounding it; is a measure of fullness. Froude number (Fn): Dimensionless number relating speed to boat length; determines the wave resistance. Propulsion efficiency: How much of the power produced by the engine is converted into effective thrust. EHP/DHP: Effective power (Effective Horse Power, the power required to overcome the resistance of the boat) and delivered power (Delivered Horse Power, the power reaching the propeller).
Initial Sizing and Design Exploration
Form design is a search problem: there are endless combinations between main dimensions (length Lpp, width B, draft T), form coefficients (Cb, prismatic coefficient Cp, mid-section coefficient Cm) and speed. In the classical method, the engineer starts from the data of similar ships (main length series, regression relations). AI accelerates this discovery: in minutes it tabulates dozens of candidate major size combinations, the approximate displacement of each, and whether the L/B, B/T ratios fall within typical ranges.
Here is the step by step process. The first step is to fix the mission requirement: cargo to be carried, range, service speed, channel/port constraints (e.g. Panama, Suez width-draft limits). The second step is to roughly determine the first major lengths with similar ship data. The third step is to select the form coefficients within a reasonable range according to the speed-aspect ratio (Froude number). The fourth step is to make a preliminary estimate of resistance and power by an empirical method (e.g. Holtrop-Mennen regression). At each of these steps, AI builds the formula and generates numbers; but it is imperative that you ask the source of each coefficient and compare it with the typical range.
Tip: When you have the AI suggest a form coefficient, ask "is this Cb value typical for this Froude number; in what range should it fall?" For example, in a high-speed form, a full Cb (0.80+) explodes wave resistance; AI can remind you of this, but the final choice is yours with similar ship data.
Empirical Method and AI in CFD Setup
In the preliminary stage, resistance is estimated by empirical regression methods rather than model experimentation. Methods such as Holtrop-Mennen break down resistance into its components: friction resistance, form effect, wave resistance, bulge, and air resistance, with formulas derived from a large database of ships. AI recalls the steps of this method, calculates intermediate coefficients and builds a spreadsheet skeleton. But the method has its validity limit: Holtrop is derived for certain Cb, Fn and L/B ranges; If you go outside this range, the result will be unreliable. AI should remind you of this limit; If he doesn't remind you, you should ask.
On the CFD side, AI doesn't do the solution itself, but it helps you query the setup: where the mesh resolution should be tightened, what the initial cell height (y+) should be for the boundary layer, why the turbulence model (like k-ω SST) is chosen, how to control convergence. This is a teacher-student dialogue; What confirms the numerical result is the mesh independence study (mesh independence; the result does not change when you tighten the network) and, if possible, model experiment calibration.
Method
what gives
strong point
Limit/verification
Similar ship/regression
First major size, brute force
Very fast, data is cheap
Only valid for similar geometry
Holtrop-Mennen
Component resistance curve
Standard for front design
Cb/Fn/L·B range limit; Need model confirmation
CFDs
Detailed flow, pressure field
Sees local effects
Sensitive to network and turbulence model; calibration is necessary
Model experiment (pool)
Measured actual resistance
Most reliable
Expensive, time consuming; scale effect correction
Mini Cases
Case 1 — Fitted correction coefficient. A team calculates the total drag with Holtrop for a 190 m tanker as 620 kN. When they tell the AI to “improve the result,” the model adds a “0.80 form correction multiplier” of unknown origin, reducing the resistance to 496 kN. This 20% reduction reduces propulsion by approximately one host class. The team asks about the basis of the multiplier; AI cannot cite sources. When they compare it with the model test data, they see that the real resistance is around 610 kN. Lesson: no correction factor without a source enters the result.
Case 2 — Exceeding the Froude limit. A student has Holtrop calculate the resistance of a 50 m ferry aiming for a service speed of 30 knots. For this length-speed the Froude number increases to ~0.70; Well above Holtrop's reliable range. AI gives results without specifying it first. When the student asks about the validity range, it turns out that the method is not suitable in this regime and a different method (Savitsky, etc.) is required for planer/semi-displacement forms. Lesson: question the validity envelope of every empirical method.
Case 3 — Error from network. In a CFD study AI gives the resistance 540 kN with a coarse mesh. When the engineer tightens the network and performs independence studies, the result is 605 kN; the initial value was 11% lower because the aft wave field was not sufficiently resolved. Lesson: CFD result cannot be considered reliable without a network independence study.
Copiable Prompt Templates
Template 1 — Pre-sizing exploration:
Role: You are a ship preliminary design consultant.Context (representative): Payload for a bulk carrier ~40,000 t, service speed ~14.5 knots, within the Suez draft limit.Task:1) Recommend 4 candidate main length combinations (Lpp, B, T).2) Approximate displacement for each candidate, L/B, B/T, Cb ver.3) Mark which ratios fall outside the typical range.Constraint: Write the unit of each number; specify the regression/database you are using; Don't make a final decision, submit a candidate.
Template 2 — Configuring the Holtrop resistor calculation:
Set up a preliminary resistance calculation framework with the Holtrop-Mennen method. Inputs (representative): Lpp, B, T, Cb, Cp, service speed, wet area. I want:1) List the resistance components separately (friction, form, wave, projection, air).2) Write the formula and coefficient you use for each component.3) Specify the validity range of the method (Cb, Fn, L/B) and check whether my inputs are in this range. et.4) Add how to calibrate the result with the model experiment.
Template 3 — CFD setup query:
I'm setting up a boat resistance CFD study. Explain the following decisions with justification and give a typical range of values: - Turbulence model selection and reason - Near-wall solution (y+ target) and initial cell height - Free surface (VOF) settings - How to check convergence and mesh independence Add a "how do I verify this" line for each suggestion.
Template 4 — Result reasonableness check:
Critique the following resistance/power result:[Paste EHP, DHP, speed, resistance values]1) Is the order of magnitude typical for this size ship?2) Is your propulsion efficiency assumption reasonable (typically 0.55-0.70)?3) Calculate the resistance back from the speed I entered, is it consistent?4) Mark any questionable or unfounded values.
Weak prompt / Strong prompt
Weak prompt:
Give the best form and resistance curve for this boat.
Powerful prompt:
Role: You are a hydrodynamic preliminary design consultant. Context (representative, not actual project): Lpp ~180 m, B ~32 m,T ~12.5 m, Cb ~0.84, service speed ~14.5 knots. Task: Estimate component preliminary resistance by Holtrop method; write down the source of each coefficient and the validity range of the method. Constraint: ADD correction factor without source. List how to verify the result with model testing and what measurements I need to take. State units (kN, kW, knots) clearly.
The weak prompt form leaves "best" undefined and does not require validation; The strong prompt requests the context, method, validity limit, and verification path.
Common mistakes
- Ignoring the validity envelope. Using Holtrop or any other regression outside the range from which it was derived gives reasonable-looking but incorrect resistance.
- Adopting an unsourced correction factor. Say "improve" and the flimsy coefficients the AI adds silently distort the result.
- Using CFD without network independence. The coarse mesh cannot resolve the stern wave and vortex field; resistance becomes low.
- Forgetting the scale effect. The model experiment result requires friction correction (ITTC method) when moving to full scale; AI can bypass this.
- Taking dispatch efficiency optimistically. Even if the resistance is correct, an overly optimistic drive efficiency underestimates the required machine power.
In summary
AI accelerates hull form discovery, empirical resistance calculation setup, and CFD run planning; makes dozens of candidate forms comparable in minutes. But every empirical method has a validity envelope, every CFD result has a network dependency, and every full-scale estimate has a scale correction. The resistance and power values produced by AI can only be translated into a referral decision after passing through the filter of model experimentation, empirical confirmation and physical plausibility.
Application task
For a representative ship (e.g. Lpp ~150 m, B ~25 m, T ~9 m, Cb ~0.75, speed ~15 knots), have AI calculate preliminary drag and EHP using the Holtrop method. Then query the order of magnitude and dispatch efficiency assumption by applying the “conclusion plausibility check” template. Ask the AI for the validity range of the method and check if your inputs are within this range. Report at least one inconsistency you found and how you verified it in half a page.
checklist
- [ ] I compared the validity range of the empirical method I used with my inputs.
- [ ] I asked and confirmed the source of each correction factor that the AI added.
- [ ] I compared the resistance and power results to an order of magnitude with a typical ship.
- [ ] If I used CFD, I planned/performed the network independence study.
- [ ] I have verified that the dispatch efficiency assumption is within a reasonable range.
- [ ] I based the final propulsion decision on the model experiment/validated method.