Physics-Anchored Predictive Maintenance Synthetic Dataset and Computational Results

This dataset contains the synthetic, physics-consistent data and computational results generated for the study titled “Physics-Anchored Predictive Maintenance and Process Safety Optimization in Chemical Engineering: A Deep Learning Framework.” The dataset comprises 3,600 observations representing progressive degradation trajectories for three representative chemical-process equipment classes: centrifugal pumps, shell-and-tube heat exchangers, and packed-bed catalytic reactors. The data include process and condition-monitoring variables, physics-derived degradation indicators, health-state labels, and remaining useful life targets. The repository also contains the model-performance metrics and confusion matrices obtained from the Extreme Gradient Boosting (XGBoost) classification experiments and the Long Short-Term Memory (LSTM) remaining useful life prediction experiment. The dataset was generated computationally using physics-consistent degradation relationships and controlled stochastic variability. It is provided to support transparency, reproducibility, and independent evaluation of the computational methodology reported in the manuscript. The data are synthetic and should not be interpreted as measurements from an operating industrial plant.