Data-Driven Fault Detection for HVAC Control Systems in Pharmaceutical Manufacturing Workshops

Large-scale heating, ventilation, and air conditioning (HVAC) control systems in pharmaceutical manufacturing are characterized by numerous operational parame-ters, delayed fault detection, and challenges in fault diagnosis. This study proposes a data-driven early warning method for equipment parameters, integrating Principal Component Analysis (PCA) and Nonlinear State Estimation Technology (NSET). Ini-tially, operational data collected from air conditioning units are preprocessed and an-alyzed to extract parameters relevant to fault detection. A nonlinear state estimation predictive model is constructed by minimizing the residuals, and its performance is further optimized using PCA. Actual operational data from the SCADA system are then utilized to predict and analyze deviations in the mixed-room temperature during production. Fault detection is achieved by evaluating whether the prediction residuals exceed a predefined critical threshold, allowing for timely identification of abnormal operating conditions. Comparative analysis of system data before and after faults is conducted to further validate the approach. Experimental results demonstrate that the proposed PCA-NSET model is feasible and effective for fault detection in HVAC con-trol systems within pharmaceutical workshops.