Generative Pattern Learning for Missing Sensor Data Imputation in Industrial IoT

In industrial IoT (IIoT) systems, sensor data integrity is crucial for real-time decision-making, predictive maintenance, and operational efficiency. However, missing sensor readings-caused by hardware malfunctions, environmental interference, or communication failures-can significantly degrade anomaly detection, impair system diagnostics and reduce decision accuracy. To address this challenge, we propose a generative imputation framework that combines symbolic pattern learning with latent-space modeling to accurately reconstruct missing values in temperature time-series data. We validate our method using real-world datasets from industrial stack, curing, and drying ovens, collected via wired PT100 sensors. Experimental results show that our approach reduces the Root Mean Squared Error (RMSE) by up to 50% compared to traditional statistical techniques. These findings underscore the potential of generative models to enhance sensor data integrity, improve system resilience, and ensure robust, data-driven decision-making in industrial environments.