First variant of the air propeller metamodel creation method architecture

Key Idea

The model learns to correct BEMT errors using a small number of CFD calculations performed at strategically selected points.

Task

Develop a model that calculates propeller thrust and torque at BEMT speed with accuracy close to CFD.

Data Generation

Low-fidelity data generator (my BEMT, 0 points)

Input: geometry, RPM, V

Output: T, P, efficiency vs J (RPM)

High-fidelity data generator (CADFlo, test, 0 points)

Input: geometry, RPM, V

Output: T, P, efficiency vs J (RPM)

Multi-fidelity Architecture
GPLF = GaussianProcess() →
LF(x) over entire domain
GPδ = GaussianProcess() →
Error HF(x) − ρ·LF(x)
« Smart Error Correction »
HF(x) = ρ · LF(x) + δ(x)
Tprediction = 0.0 N
Tstd = 0.0 N
Active Learning (Adaptive infilling)

Optimization

→ Xbest

New HF point (Expected Improvement)

→ New HF point

EI(x) = (η(x) − ηbest) · Φ(Z) + σ(x) · φ(Z)
Z = (η(x) − ηbest) / σ(x)  if  σ(x) > 0
Made on
Tilda