Abstract: Quality data in automotive manufacturing processes is often fragmented, redundant, and structurally rigid, posing challenges to knowledge accumulation and intelligent quality management based on historical data. Considering the implicit assembly relationships and nested causal structures in product quality issue descriptions, a knowledge graph construction method is proposed, comprising two types of entities and three types of relations. “Fault phenomena” and “fault parts” are used to represent the causes, results, and locations described in quality issue texts. A localization relation is introduced to connect “fault phenomena” with “fault parts”, which avoids overly specific entity naming while preserving the spatial location of parts, thereby enhancing spatial semantic representation. Subsequently, a matching relation is further introduced to represent assembly constraints among parts, and a causal-slot decomposition mechanism is established to enhance the structural stability and reasoning capability of the graph. In addition, an information extraction framework based on a large language model is developed to automatically identify key entities and relations and generate structured representations. Experimental results show that the proposed graph structure outperforms existing quality graph construction methods in terms of information completeness, structural simplicity, and causal reasoning capability, effectively supporting intelligent diagnosis and multi-dimensional root cause analysis. Its application value in intelligent quality management is further validated using real-world data from an automotive manufacturing enterprise.