Manufacturing systems today are heavily instrumented and connected, producing large amounts of diverse operational data and encouraging the use of AI for prediction and decision support. However, models based mainly on correlations often fall short in robustness, generalization across settings, and clear mechanistic understanding, which are essential in high-value and safety-critical production environments. This review examines the integration of causal reasoning and knowledge graph infrastructures as a structural foundation for explainable and decision-oriented manufacturing intelligence. Pearl’s three-level causal hierarchy (association, intervention, and counterfactual) is adopted to classify recent research articles according to the depth of causal capability they operationalize. The analysis shows that most existing work remains concentrated at structured associational reasoning and interventional effect estimation, while fully operational counterfactual systems are comparatively rare. Knowledge graphs are widely employed to provide structured semantic representation and traceability, supporting human-centered and machine-interpretable reasoning. However, formal interventional and counterfactual semantics remain partially integrated. Recurring consensus and open challenges are identified, including hidden confounding, partial observability, scalability of expert-driven modeling, underdeveloped dynamic causality, and ambiguity in the use of “counterfactual” terminology. The review clarifies the structural role of causal knowledge graphs in advancing manufacturing intelligence from traceable association toward quantified intervention and counterfactual foresight and outlines future research directions.