Abstract: Traditional CNC machining quality prediction (CMAP) methods primarily rely on statistical learning models to establish mapping relationships between parameters and quality. However, these methods fail to effectively express causal evolutionary mechanisms under high-dimensional parameter coupling and nonlinear working conditions, resulting in predictions that lack physical interpretability and exhibit poor generalization performance when working conditions change. While knowledge graphs can incorporate expert experience and domain knowledge, existing methods focus solely on static knowledge representation and struggle to capture dynamic causal relationships between status singles and machining quality. To address these challenges, a causally-iterative machining knowledge graph (CIMKG) is proposed, that hierarchically and iteratively integrates temporal evolution patterns and physical mechanism constraints into the knowledge graph structure, achieving formal representation of dynamic causal relationships in machining processes. Experimental results demonstrate that the CIMKG effectively captures parameter-quality causal associations in basic and typical machining processes, showing superior prediction accuracy compared to traditional CMAP methods. The proposed approach not only enables forward prediction from process parameters to quality characteristics but also facilitates reverse causality analysis for quality anomalies.

Key words: CNC machining, quality optimization, knowledge graph, causal relationship iteration