Multi-source heterogeneous data fusion and real-time anomaly monitoring method for unmanned production lines of complex products

As manufacturing shifts toward greater intelligence and sustainability, ensuring real-time visibility, quality traceability, and predictive maintenance across the entire product lifecycle has become a central challenge for lifecycle engineering. In high-end manufacturing environments, complex production lines generate vast amounts of multi-source, multi-dimensional, and highly heterogeneous data. Effectively managing this data is essential to support lifecycle-oriented decision-making, from process optimization and fault diagnosis to long-term performance monitoring and quality assurance. However, the sheer volume and heterogeneity of production data pose significant obstacles to real-time analysis and anomaly detection. Conventional methods often fall short in acquiring, integrating, and interpreting such data efficiently, resulting in delayed responses to abnormal events and reduced control over key manufacturing parameters. To address these challenges, this paper proposes a novel method for intelligent data fusion and real-time anomaly detection tailored to complex product production lines. The approach combines automated acquisition and preprocessing of sensor data with complex network theory to evaluate data importance and relevance. It then performs deep fusion and correlation analysis across heterogeneous data sources, enabling timely detection of anomalies and deviations during production. The proposed method was validated on a complex production line for small-pellet pharmaceutical manufacturing. By embedding lifecycle awareness into the production line monitoring framework, the proposed method supports accurate process control, product quality traceability, and predictive maintenance. This contributes to enhancing the safety, efficiency, and resilience of production systems while strengthening their alignment with lifecycle engineering goals.