Error accumulation and traceability difficulties during the assembly of complex products are critical challenges constraining the improvement of manufacturing precision. Traditional methods, which rely on periodic sampling and manual diagnosis, face engineering limitations such as significant feedback lag and ambiguous identification of root causes. To address these issues, this paper proposes a digital twin-driven assembly error modeling and intelligent traceability method. First, an intelligent analysis system for assembly quality traceability is established by integrating virtual twin technology. A three-dimensional digital twin architecture, comprising a data layer, a model layer, and an analysis layer, is developed to provide a foundational framework for error propagation modeling. Second, a Time-Sensitive Networking (TSN) framework is adopted as the underlying architecture, and dynamic optimization modeling is implemented through an attributed graph network. Finally, a traceability algorithm based on an error propagation network is introduced. By calculating the cumulative weights of error propagation paths and backtracking each path, the error propagation intensity of each node is quantified. The node with the highest intensity is identified as the primary error source, requiring subsequent optimization. Experimental validation demonstrates that, compared to traditional offline detection methods, the proposed approach significantly enhances the timeliness of quality control, with notable improvements in response speed and resolution accuracy for assembly process anomalies. This study provides a novel methodology for dynamic error control and intelligent decision-making in assembly processes, offering substantial engineering value for improving the assembly quality of complex products.