"C-MAPSS Spatiotemporal Graph Datasets"

"This dataset is built upon the NASA C-MAPSS (Commercial Modular Aero-Propulsion System Simulation) benchmark for aircraft engine remaining useful life (RUL) prediction. The raw data undergoes three preprocessing stages: (1) clustered normalization via K-Means on operational settings followed by per-cluster StandardScaler, (2) sliding window segmentation with a fixed window size of 50 time steps, and (3) hierarchical graph construction encoding 14 sensors into 6 engine components (Fan, LPC, HPC, Combustor, HPT, LPT) and one global engine node, resulting in a 21-node graph with bidirectional edges and self-loops. Each sample is a PyTorch Geometric Data object with node features of shape [21, 50], an edge_index defining the graph topology, and a scalar RUL label clipped to the [0, 125] range. The dataset covers four sub-datasets \u2014 FD001 through FD004 \u2014 with varying operating conditions and fault modes, producing eight processed .pt files (train\/test pairs) suitable for graph neural network based prognostic tasks. "