The Arc Loss Dataset

The Arc Loss Dataset is a benchmark dataset designed for time series classification in industrial applications, particularly for fault detection and diagnosis (FDD). It was collected from a large-scale pyrometallurgical plant and captures the real-world complexities of industrial processes, including high dimensionality, sensor noise, and variable system dynamics. The dataset consists of 3,226 multivariate time series samples, each containing 1,101 time steps (equivalent to 55 minutes) with 96 process variables. The dataset is split into three subsets: I) train.pt (70% of the samples, 2,258 samples), II) val.pt (10%, 323 samples), and III) test.pt (20%, 645 samples). More information can be found in the README.txt