Content area
In industrial settings, quality monitoring is essential to ensure the reliability, traceability, and efficiency of production processes. This study presents an analysis of principant component (PCA) approach for multivariate statistical process control of signals collected from sensors installed throughout the stages of a brewing plant’s production cycle. The strategic placement of sensors enables continuous and integrated monitoring, capturing meaningful data from raw materials to the finished product. The methodology is structured in two phases: Phase I involves building a statistical model using historical in-control data; Phase II applies this model to monitor new observations. The PCA enables dimensionality reduction and highlights the main directions of process variability. Anomalies are detected through two control indices: Hotelling’s T2, measuring variation within the principal component subspace, and the Squared Prediction Error (SPE), capturing residual variance. Control limits are derived from theoretical distributions: β and F for T², and a moment-based approach (up to third order) for SPE. A significance level α aligned with Six Sigma practices ensures a balanced trade-off between sensitivity and false alarm rate. The model proved effective in automatically identifying out-of-control observations and in iteratively improving the quality of the monitoring system. Although validated on a specific case study, the proposed framework is generalizable to any multi-stage, sensor-monitored production system. It offers a flexible and robust tool for predictive maintenance, early anomaly detection, and quality control optimization in complex industrial environments.