Simulation-Based Wheel-Level Fault Diagnosis Dataset for a Six-Wheel Independently Driven Off-Road Chassis
Description
This dataset supports the fault diagnosis experiments presented in the paper "Simulation-Based Wheel-Level Fault Diagnosis of a Six-Wheel Independently Driven Off-Road Chassis Using Torque Observations and Cross-Channel Attention" (Control Engineering Practice). The dataset contains two NumPy NPZ files: 1. fault_dataset.npz (main experiment): - X: shape (25398, 21, 1000), dtype float16. 25398 sliding windows, 21 input channels, 1000 samples per window. - y: shape (25398,), integer labels in 0-38 (39 healthy/fault classes). - split: shape (25398,), train/validation/test assignment (0=train, 1=val, 2=test; 8466 windows each). - This file contains the actual (simulated) motor torque channels used in the main fault diagnosis experiments. 2. fault_dataset_no_torque.npz (ablation): - Same 21-channel structure and dimensions as the main file. - The six actual motor torque channels are replaced by the upstream reference/command torque signals, used for the torque-observability / signal ablation study. Despite the file name, it still contains 21 channels. All data are generated from a vehicle simulation of a six-wheel independent drive/steer off-road chassis. The 39 class definitions, channel descriptions, and split convention are documented in README.md and labels.csv.