In concrete block manufacturing, strength and water absorption tests require curing periods and destructive testing, which incur substantial time and cost and make it difficult to provide real-time quality feedback for all produced units. To address this limitation, this paper integrates process logs collected from the mixer, production-related equipment, the tilting machine, and the curing chamber at an hourly resolution, and constructs an integrated dataset by linking them with measured test results.
Using a multi-output regression framework, we compared candidate models including GBDT, RF, ET, and XGB. The results show that, for Facility A, ET achieved strong performance with an average R² of 0.8846 (average MAE of 0.1661), while for Facility B, GBDT achieved high performance with an average R² of 0.9558 (average MAE of 0.0729). Moreover, SHAP analysis confirmed that operating-condition variables such as vibration time, mortar feeding time, and temperature fluctuations were relatively more influential for Facility A, whereas input-quantity and production-state variables such as production count, hopper weighing values, and mixing count were relatively more influential for Facility B. By integrating PLC process data with quality test data and combining predictive modeling with explainable analysis, this paper provides practical evidence to support proactive quality prediction and the prioritization of process-control factors.