Multilevel Stability and Load-Transfer Behaviour of a Large-Span Chinese Solar Greenhouse Under Non-Uniform Snow Loading and As-Built Joint Conditions

The stability of large-span Chinese solar greenhouses depends not only on snow-load distribution but also on support continuity and the progressive engagement of as-built joints. To distinguish these effects, this study presents a multilevel investigation combining geometrically and materially nonlinear frame analysis, a full-scale sequential hanging-load test with shell-model comparison, and refined solid modelling of design-reference and as-built joints. These three levels are complementary but mechanically uncoupled. Four global cases were analysed: supported-uniform (SU), supported-non-uniform (SN), unsupported-uniform (UU), and unsupported-non-uniform (UN). Their ultimate total vertical loads were 12.336, 11.594, 2.592, and 2.069 kN, respectively. Non-uniform snow reduced the ultimate load by 6.0% for the supported frame and by 20.2% for the unsupported frame, while removal of the intermediate columns reduced the capacity by 79.0% under uniform snow and by 82.2% under non-uniform snow. For the SN case, changing the initial-imperfection amplitude from L/2000 to L/1000 reduced the ultimate load by only 0.285%, indicating that the principal capacity hierarchy was insensitive to the investigated amplitude range. The shell model reproduced the measured strain signs and principal loading trends. At the final 40 kg state, the absolute strain differences ranged from 0.091 to 0.931 µε, whereas the sequence-level root mean square errors ranged from 1.289 to 6.361 µε, indicating larger path-dependent discrepancies at some intermediate loading stages. The refined as-built joint model indicated progressive closure of the column-top–longitudinal-member gap, with an interpolated contact-onset load of approximately 129.7 N and non-zero normal contact pressure at the next prescribed load of 149.4 N. Under the baseline truncation boundary, the simulated column-reaction share increased to 87.8% at 1000 N; alternative spring boundaries produced shares of 93.9–97.4%, demonstrating that the exact value was boundary-dependent while the progressive engagement mechanism remained consistent. Overall, support continuity dominated the stability capacity of the investigated frame, while as-built gaps and contact conditions governed the delayed mobilisation of the local support path. The local joint results are used for mechanism interpretation and are not propagated into the global ultimate-capacity calculation.