Multi-Source Information Fusion for Quality Assessment in Prefabricated Components Assembly Considering Uncertainty

Abstract

Prefabricated steel box-girder assembly faces prominent fuzzy and random uncertainty in manufacturing inspection data, and effective multi-source information fusion assessment methods are still insufficient. Targeting this problem, this paper establishes a four-category and 15-indicator hierarchical pre-assembly risk evaluation index system for steel box-girder segments on the basis of specifications, literature review and expert consultation. An improved D–S evidence fusion workflow embedded with evidence similarity screening mechanism is proposed to solve high-conflict multi-source factory inspection data. Combined with cloud-model qualitative–quantitative mapping, Monte Carlo simulation and global sensitivity analysis, the framework realizes quantitative risk grading and dominant manufacturing risk factor identification. A case study yields four main findings: (1) after fusion, uncertainty for all beam segments remained below 0.05, satisfying engineering confidence requirements; (2) the improved D–S evidence theory achieved a high-risk confidence of 0.9944, outperforming classical evidence theory (0.9675) and the weighted average method (0.6229); (3) Monte Carlo simulation confirmed the method’s reliability in distinguishing risk levels among segments; and (4) global sensitivity analysis accurately identified key risk factors for each segment. This framework can realize pre-assembly risk early warning relying on conventional factory inspection records and provides technical support for reducing prefabricated steel box-girder on-site assembly failure risk.