Intelligent production control method for cigaret manufacturing based on industrial data and machine learning

Abstract

By combining multi-stage industrial data collection with machine learning-driven prediction and feedback regulation, an integrated quality control architecture for cigaret production has been constructed. Thru standardized coding, dynamic batch-based data partitioning, and real-time sensor networks, high-fidelity process monitoring and fine-grained data traceability can be achieved. In order to predict quality outcomes and detect anomalies across production cycles, a prediction engine based on random forests and recurrent neural networks has been added. The engine also enhances discrimination capability in the presence of multi-source noise and nonlinear process drift. The closed-loop control strategy uses a data-model hybrid feedback mechanism to automatically adjust production parameters. This method combines model-based predictions with on-site quality measurements. Robust data storage solutions, including elastic archiving and low-latency operations, ensure quick access to models as well as compliance and auditability. By implementing the system on a full-scale industrial production line, product variations are significantly reduced, anomaly traceability speed is accelerated, and output is increased. Quantitative evaluation, using traceability efficiency, defect-free yield, and variance index, demonstrates the continuous improvement of this technology over traditional open-loop control. Comparative analysis further demonstrates the system's ability to maintain product consistency during environmental or operational disturbances, supporting rapid process adaptation and expert supervision. The results indicate that the proposed method achieves robust, adaptive, and transparent quality control in dynamic, data-intensive manufacturing processes. This method also establishes a replicable digital foundation for process optimization at an industrial scale.