Department of Aeronautical Engineering · Erciyes University · Türkiye

Experimental Validation of an Aircraft Digital Twin: From Concept to Validation

Before building a test, check that it could actually prove your model wrong.
Ayşe Nur Dişlitaş · Melih Yıldız
16th EASN International Conference
Toulouse · 27–30 Oct 2026 · Digital Design session

The Criterion: Validation-Resolution Ratio Rv

Rv = u / (δ · y) uncertainty of the corrected measurement ÷ model discrepancy the test must detect. Informative only if Rv < 1 test can falsify the model test cannot tell right from wrong 0.10.31 31030 Rv at δ = 7.5 % (log scale) HAPS wing section, 97 mm, same tunnel L1 component · archival data at matched Re L2 aircraft · certified performance L3 nominal mission · flight manual L4 actual operation · 428 ADS-B flights ATR 72 half model, 1:113, 40 cm tunnel 0.33 0.40 0.67 0.67 1.07 → band set here (±8 %) 16.6 1.96 even with an optimistic flat 10 % correction error

The Problem

Digital twins are usually validated by whatever data can be found. A comparison validates nothing, however, if the correction applied to the measurement is larger than the model error it is meant to reveal. We planned a wind tunnel campaign for an ATR 72‑500 twin (Simcenter Amesim, 5–10 % accuracy claim) and first audited it on paper.

Three Findings

  1. The tunnel campaign was cancelled after a one-afternoon audit. The largest model that fits gives Re = 4.0×10⁴ against 1.4×10⁷ in cruise (×348). The scale correction is 122 % of the drag coefficient itself, giving Rv = 16.6.
  2. Operational data change the question. Across 428 ADS-B profiles the median stage is 371 km, while the design mission is 750 km, a distance only 6 % of flights reach. The unknown mass matters less than expected: ±21.5 % in take-off mass gives only ±3.4 % in block fuel.
  3. Match the facility to the aircraft class. The same tunnel reproduces a 20 km HAPS wing's Reynolds number (1.9×10⁵) exactly with a 97 mm chord: Rv = 0.33.

Three Evidence Sources, One Test

SourceKey numberRv
ATR 72 half model, 1:113, 400×400 mm tunnelRe gap ×348; correction 122 % of CD16.6
ADS-B, OpenSky, 428 complete ATR 72 flightsresolves ≈ 8 %; mass → fuel attenuated ×6.41.07
HAPS wing section, 97 mm chord, same tunnelRe matched exactly, M 0.087 vs 0.100.33

The multi-level acceptance bands (±3 / 5 / 5 / 8 %) are what Rv ≈ 1 gives for the best source at each level. The ratio is not tied to tunnels: flight tests, one-count drag increments and loads models are audited in the same way. When δ is small the criterion is harder to meet.

We are looking for: recorded flight data with known mass (FDM/QAR) to close the operational tier, and partners for low-Reynolds HAPS section testing and certification-by-analysis evidence.

Companion talk at this conference: Machine Learning-Assisted Multidisciplinary Design Optimization in Aircraft Conceptual Design Using Digital Twin Technology (E. Z. Tekin, M. Yıldız), which builds a fast twin of the same aircraft class.