Gains:
- Ability to accelerate trade-off studies from mission requirements to conceptual aircraft/spacecraft sizing with AI support
- Ability to carry out requirements writing, traceability and consistency control with configured prompts
- Ability to eliminate unreasonable outputs by cross-validating AI-suggested design points with physics-based hand calculations
An airplane or spacecraft is born from a list of requirements, not on a drawing board. The sentence "Carry this much load to this range, at this speed, at this cost" is the seed of the design that will last for years to come. Conceptual design (the earliest design stage in which the rough size, weight and configuration of the vehicle are first determined) is the process of extracting an initial shape from this seed. The decisions made here are cheap, but their effects are the most expensive: one wrong conceptual choice can throw away millions of dollars of detailed design.
In this unit, we will use artificial intelligence (AI) to alleviate three burdens of conceptual design: translating vague task requests into measurable requirements, maintaining traceability and consistency between requirements, and accelerating trade-off analysis between competing design options. But let's lay out the critical principle from the beginning: Every size, every weight estimate that the AI suggests is necessarily cross-validated by a physics-based hand calculation. AI is a brainstorming and accounting partner; does not own the sizing decision.
Concepts: Requirement: A single verifiable statement that the tool must satisfy. Traceability: The chain of record of where each requirement came from and with what design/test it was met. Trade space: The set of all options created by changeable design variables. MTOW: Maximum take-off weight (Maximum Take-Off Weight), the highest mass that the vehicle can take off.
From Task Need to Requirement: Step by Step
A good requirement has three qualities: singular (says one thing), measurable (contains a number and a unit), and verifiable (it's obvious how to check it). "The aircraft must be fast" is not a requirement; "The aircraft must achieve 250 kt caliber airspeed (KCAS) at sea level" is a requirement.
The first step is to gather stakeholder requests and distill them into draft requirements with AI. The second step is to make each requirement measurable: replace each vague adjective (“lightweight,” “durable,” “efficient”) with a metric and threshold. The third step is to look for contradictions between requirements: range and payload often compete with each other, AI quickly brings these contradictions to the surface. The fourth step is to establish a traceability matrix: tabulating which top-level task objective leads to which sub-requirement.
On the conceptual dimensioning side, AI can recall known engineering relationships (weight fractions, thrust-to-weight ratio, wing loading) and establish an initial chain of calculations. For example, for a passenger aircraft, the fuel weight fraction can be estimated with the Breguet range equation and the empty weight fraction can be estimated with statistical relations. But each link of this chain is confirmed by hand calculation.
Tip: When writing requirements, explicitly give the AI the task to “replace each ambiguous adjective with a metric, unit, and verification method.” The model works much more disciplined when you say "make measurable" unverifiable statements like "user friendly".
Configuring Trade-off Analysis
The heart of trade-off analysis is juxtaposing several candidate configurations with common criteria. The classic tool is the weighted decision matrix: criteria are given importance weight, each candidate is scored on each criterion, the weighted total is taken. AI quickly builds this matrix, but beware of two pitfalls: weights are a subjective engineering judgment (not left to the AI) and scores must be based on physical justification (not made up).
Criterion (weight)
Config. A: High wing
Config. B: Low wing
Config. C: Canard
Range, km (0.30)
1200
1350
1250
Curb weight, kg (0.25)
640
610
690
Production cost (0.20)
medium
low
high
Stability (0.15)
high
medium
medium
Ease of maintenance (0.10)
high
medium
low
You can have AI build this table; But range and weight numbers must come from an independent preliminary calculation, and qualitative scores such as "stability: high" must also be justified. The power of AI is that it tabulates ten different configurations in minutes and performs sentiment analysis (how the ranking changes if a weight changes).
Weak prompt / Strong prompt
Weak prompt:
I am designing a training aircraft. Tell me the best configuration.
Powerful prompt:
Role: You are a conceptual aircraft design consultant. Context: Preliminary trade-off study for a two-seat trainer aircraft (values are representative): target range ~1000 km, cruise speed ~55 m/s, payload ~220 kg, low production cost priority. Task:1) Suggest 3 candidate configurations (wing position, engine type).2) Set up a table comparing each candidate on: range, estimated empty weight, production cost, stability.3) Use for your numerical estimates write the relation and assumption.Constraint: State which statistical relation you used in your weight estimates; write unit; Don't make a final decision, submit a candidate. I will weigh in and make the final decision myself.
Copiable Prompt Templates
Template 1 — Translating a task need into a measurable requirement:
Transform the following stakeholder requests into singular, measurable and verifiable requirements. For each requirement: a metric, a unit, a threshold value, and a proposed verification method (analysis/test/demonstration).Translate vague adjectives ("light", "fast") into concrete numbers; If there is uncertainty, ask what information you need.Requests: [list here]
Template 2 — Traceability matrix and conflict screening:
Take this list of requirements and: (a) build a traceability table that links each requirement to the high-level task objective; (b) mark any pairs of conflicting or challenging requirements (e.g. range vs. payload); (c) list missing or unverifiable requirements. Let the result be a table. Requirements: [list here]
Template 3 — Weighted decision matrix setup:
Build a weighted decision matrix framework for the following candidate configurations. Make the criteria rows and the candidates columns. Leave the weight column BLANK (I will enter the weights). Explain in a separate note the physical justification for the score you will put in each cell; Do not write a made-up number, mark the place you guessed as "guess". Candidates: [...] Criteria: [...]
Template 4 — Conceptual dimensioning cross-check:
Independently check the following presizing result: MTOW, wing area, thrust. (a) Calculate the wing loading (W/S) and compare it with the typical range for your given aircraft class. (b) Calculate the thrust-to-weight ratio (T/W) and compare it with the typical range. (c) Warn if values are outside typical range and indicate possible error. Write all units explicitly. Input: [MTOW, S, T values]
Mini Cases
Case 1 — Early detection of conflicting need. In a UAV project, stakeholders want both "at least 8 hours of airtime" and "maximum 12 kg total mass". By quantifying the two requirements and roughing out the energy budget, the AI shows that an 8-hour flight with typical battery energy density (about 200 Wh/kg) would require 6-7 kg of battery alone; This does not fit into the 12 kg budget along with the payload and structure. The contradiction is discussed before detailed design begins. Lesson: translating requirements into numbers reveals hidden contradictions early.
Case 2 — Sensitivity analysis changes the decision. One team compares three wing configurations; Config with initial weights. B wins (weighted score 0.82 vs. 0.79). The AI is asked “what happens if I reduce the cost weight from 0.20 to 0.10”; in this scenario range comes to the fore and Config. A takes the lead (0.84 vs. 0.80). The team sees how sensitive the decision is to cost priority and clarifies the weights with the stakeholder. Lesson: it's not a single score that matters, but the stability of the ranking.
Case 3 — Order error in weight estimation. AI gives the empty weight of a light aircraft as 95 kg from a relation. The engineer recalls that the empty weight of two-seat aircraft of a similar class is typically in the range of 350-500 kg; 95 kg is physically impossible. When checked, it is seen that the AI has incorrectly applied a coefficient in the relation and it is corrected. Lesson: each weight estimate is compared by orders of magnitude to the known range of the class.
Common mistakes
- Accepting vague requirement. A requirement that does not contain metrics and units cannot be verified; "It should be light" is a wish, not a requirement.
- Having the AI choose the weights. The criteria weights in the decision matrix are subjective engineering and stakeholder judgment; If left to AI, the decision remains derelict.
- Accepting points without justification. If qualitative scores like "Stability: high" have no physical basis, the matrix becomes a fancy list of guesses.
- Not confirming the weight/size estimate. Statistical correlations are rough estimates; It is not carried into detail design without comparing it with the class range.
- Leaving traceability for later. If it is not recorded where the requirement came from, no one will know what decisions a change will affect in the future.
In summary
Conceptual design is the stage where the cheapest but most effective decisions are made. AI is a powerful accelerator here in translating vague requests into measurable requirements, scanning for traceability and conflict, and comparing multiple configurations across trade-off tables. However, criteria weights are human judgment, numerical predictions are verified by physics-based hand calculation and class intervals, the final configuration decision remains with the engineer. AI produces candidates; does not decide.
Application task
Write three high-level mission objectives for an air or spacecraft of your choice. Using AI, translate these into at least six measurable requirements and set up a traceability table. Then create a weighted decision matrix for the three candidate configurations; Determine the weights yourself. Finally, confirm a weight/size estimate given by the AI with an independent hand calculation and class range. Report at least one pair of requirements where you find a conflict.
checklist
- [ ] I made each requirement singular, measurable, and verifiable.
- [ ] I replaced the ambiguous adjectives with metric, unit and threshold.
- [ ] I set up a traceability table and marked conflicting requirements.
- [ ] I determined the decision matrix weights myself (not AI).
- [ ] I attributed the matrix scores to physical justification.
- [ ] I verified the weight/size estimates with class spacing and hand calculation.