AI-Driven Safety Monitoring for Industrial Boilers: Early Fault Detection with Contrastive Learning

In most Industry 4.0 systems, fault detection typically begins only after components start to malfunction, by then, performance and safety may already be compromised. This is especially critical in industrial boilers, which are essential to manufacturing and process heating. Faults such as excess air, fouling, and scaling can lead to serious safety risks like combustion instability and overheating, while also reducing energy efficiency by hindering heat transfer and increasing fuel consumption. This research proposes a contrastive learning-based anomaly detection approach designed to detect early signs of deviation, before faults become severe enough to impact system functionality. Unlike traditional threshold-based methods, which often raise alarms too late, our method learns patterns of normal behavior and identifies subtle shifts that indicate a developing problem. The model is trained on synthetic data generated using a Matlab/Simulinkbased Simscape model of the Viessmann Vitorond 200 Gas-Fired Boiler VD2 Series 380, simulating both healthy and faulty operational scenarios. By distinguishing between safe and unsafe operating patterns at an early stage, the system enables timely safety interventions and identifies inefficient operating modes that lead to wasted energy. Experimental results show that our approach detects emerging faults more quickly than conventional methods, enabling predictive maintenance and real-time optimization. This contributes to safer, more reliable, and energy-efficient boiler operations.