Anomaly Detection on Volve Field: Constructed Labels, Baseline, and Dual-Head Model

A new arXiv paper (2608.05685) addresses the challenge of anomaly detection in real-world industrial settings, using the open Volve field dataset released by Equinor. Unlike public benchmarks from test rigs where faults are induced and known, real production fields provide sensor histories without fault logs, requiring methods to invent their own labels. The authors take two key steps: they construct anomaly labels that are validated against the field's engineering documents to ensure physical plausibility, and they release the reasoning behind each label. They then test whether these labels are learnable using an unsupervised baseline and a small dual-head model that identifies when an event occurs and its type, an idea carried over from earlier work. The paper is announced as new on arXiv and is available at the provided URL.