Work

Influence of manufacturing imperfections on boltless Beam-to-Upright connections: An experimental Data-Driven study

Year: 2026

Type: article

Abstract: The mechanical behaviour of beam-to-upright connections (BUCs) is critical for the global stability of steel storage pallet racks. While design standards rely on expensive experimental testing, the significant variability among nominally identical specimens highlights the under-explored influence of manufacturing-induced deviations, which conventional analytical and finite element approaches struggle to capture without oversimplification. This study proposes an experimental-machine learning framework to predict the initial rotational stiffness ( k 0 ) and ultimate moment capacity ( M u ) of boltless BUCs, explicitly incorporating 11 micro-geometric manufacturing deviations alongside 14 macro-geometric and material variables. Using a database of 220 experimental tests, the largest reported for this typology, six regression models were evaluated under a configuration-level cross-validation scheme that tests genuine generalisation to unseen structural designs, with a nested consensus feature-selection procedure applied to prevent information leakage. Gaussian Process Regression proved the most accurate, stable model, achieving a pooled out-of-fold R 2 of 0.916 for M u and 0.796 for k 0 , and is recommended for its accuracy and native uncertainty quantification. A residual-based importance analysis decoupled the manufacturing deviations from the dominant material and macro-geometric effects, showing that they govern a quantifiable minority of the response variance (15.1% for k 0 , 6.4% for M u ) and consistently identifying slot width for stiffness and tab width for moment capacity among the most influential deviations, while others were consistently negligible. As all specimens share one connector family, extending the framework to further typologies is identified as key future work. The framework offers a computationally efficient complement to high-fidelity FEM and actionable guidance for manufacturing quality control.

Source: Results in Engineering

DOI: 10.1016/j.rineng.2026.112457

Authors: Marcelo Calispa, Jon Larrañaga, Jon Elguezabal, Aurea Iñurritegui, Ibai Ulacia

Institution: Mondragon Unibertsitatea

Language: English

Topic: Structural Load-Bearing Analysis

Subfield: Civil and Structural Engineering

Field: Engineering

Domain: Physical Sciences

SDG: Industry, innovation and infrastructure

Open Access status: gold

Estimated APC paid: $1,940