Öppna denna publikation i ny flik eller fönster >> Visa övriga...
2026 (Engelska) Ingår i: Computers & structures, ISSN 0045-7949, E-ISSN 1879-2243, Vol. 333, artikel-id 108499Artikel i tidskrift (Refereegranskat) Published
Abstract [en] Rapid vibration assessment is needed to evaluate timber assemblies across material and design variants. However, conventional computational models remain costly for repeated reanalysis because wood’s natural variability and component-geometry changes require new component-level dynamic calculations. This paper extends data-driven dynamic substructuring (DDDS) by learning a reusable machine-learning (ML) substructure for cross-laminated timber (CLT) panels. Finite-element models of sixteen tested CLT panels are calibrated against modal measurements, and the calibrated properties are expanded to 1000 material realisations using log-normal fitting and Latin hypercube sampling. A deep neural network (DNN), trained for four panel geometries, predicts the complex-valued condensed dynamic stiffness matrix of a CLT panel over 1–300 Hz, which is coupled with finite-element column substructures through standard interface equations. Predicted responses for single-floor, two-floor, and multi-floor assemblies closely match FE–FE references. In five-floor benchmarks, the FE–ML procedure is faster than both the FE–FE analysis and Craig-Bampton reduction in the cases examined, with computational savings that depend on the problem size and increase with panel size, supporting efficient screening of timber assembly designs.
Ort, förlag, år, upplaga, sidor
Elsevier Ltd, 2026
Nyckelord Timber structures, Data-driven dynamic substructuring, Condensed dynamic stiffness, Cross-laminated timber (CLT), Reduced-order modelling, Machine learning
Nationell ämneskategori
Husbyggnad
Forskningsämne
Byggkonstruktion
Identifikatorer urn:nbn:se:ltu:diva-120541 (URN) 10.1016/j.compstruc.2026.108499 (DOI) 001923153400001 () 2-s2.0-105055154457 (Scopus ID)
Forskningsfinansiär Forskningsrådet FormasStiftelsen för internationalisering av högre utbildning och forskning (STINT)
Anmärkning Funder: National Research Foundationof Korea (RS-2024-00350389);
2026-11-132026-11-132026-11-13 Bibliografiskt granskad