Skip to main navigation Skip to search Skip to main content

Machine learning-based residual strength prediction for quasi-statically damaged aluminum honeycomb sandwich structures

Research output: Contribution to journalArticlepeer-review

Abstract

Aluminum honeycomb sandwich panels serve as primary load-bearing members in aerospace and satellite structures, but their susceptibility to localised damage from tool drops, ground handling, and runway debris remains a recurring concern for damage-tolerance assessment. Most published compression-after-impact studies address single-impact damage; the influence of multiple, spatially distributed dents on the residual edgewise compressive strength has received limited attention. In the present work, AA6061-T6/AA5056 honeycomb panels with 1–4 hemispherical indents (viz. 5, 10, and 15 mm diameter at 30 mm centre-to-centre spacing) were tested under quasi-static indentation followed by edgewise compression after indentation (CAI). A cohesive-zone finite-element model reproduced the experimental load–displacement response to within 4.7% mean peak-load error, and a six-feature physical descriptor set was used to train Ridge and Gaussian-process surrogates which interpolate within the tested envelope. The undamaged CAI strength was 329 MPa; specimens damaged with 5 mm indenters retained over 91% of this baseline regardless of the number of dents, while four 15 mm indents reduced the strength by 29.5%. A change in degradation rate was identified at approximately 17 J of total absorbed energy, separating a regime which retained over 85% of baseline from one governed by debond-driven face-sheet buckling. Under grouped leave-one-out cross-validation, in which both the held-out experimental sample and its finite-element twin are excluded from training, Ridge regression predicted CAI strength with (Formula presented) and RMSE = 7.2 MPa; a simple lookup baseline collapsed from (Formula presented) to (Formula presented) between the two protocols, owing to the FEA-twin leak which inflates the originally reported scores. The validated surrogate enables rapid CAI-strength estimation within the tested envelope of 5–15 mm indenter diameter, 1–4 indents per panel, 30 mm centre-to-centre spacing, and quasi-static loading at room temperature.

Original languageEnglish
Article number123139
JournalEngineering Structures
Volume364
DOIs
Publication statusPublished - Oct 1 2026

Keywords

  • Aluminum honeycomb sandwich
  • Compression after indentation
  • Debond length
  • Finite element analysis
  • Gaussian process regression
  • Machine learning
  • Quasi-static indentation
  • Residual strength
  • Ridge regression
  • Surrogate model

ASJC Scopus subject areas

  • Civil and Structural Engineering

Fingerprint

Dive into the research topics of 'Machine learning-based residual strength prediction for quasi-statically damaged aluminum honeycomb sandwich structures'. Together they form a unique fingerprint.

Cite this