CA Training Data Convolutional Autoencoder (Track A) Curated dataset for training and evaluating the Convolutional Autoencoder anomaly detection model. Approach The autoencoder is trained only on normal (no fault) images . At inference, high reconstruction error indicates an anomaly/fault. Structure Total Counts Split Normal Fault Total Train 11,401 0 11,401 Test 33 517 550 Source Datasets Directory Source Dataset Domain electric motor Electric Motor Thermal image Fault Diagnosis DATASET Electrical (primary) induction motor Thermal Images of Induction Motor Dataset Electrical (primary) pv om inspection Photovoltaic System O&M inspection Solar PV (adjacent) pv thermal inspection Photovoltaic system thermal inspection Solar PV (adjacent) solar modules Infrared Solar Modules (No Anomaly only) Solar PV (adjacent) Notes PV O&M files are prefixed dr (double row) and sr (single row) to avoid filename collisions Solar module images were filtered from module metadata.json (anomaly class == "No Anomaly") Test/normal hold out is ~14 20% of electrical equipment normal images Image formats are mixed (PNG, BMP, TIFF, JPG) preprocessing/normalization is required before training
We use cookies for essential functionality and analytics. You can accept or reject analytics cookies.Cookie policy