An oilfield service provider sought to automate manual monitoring of rig signals, such as temperature and pressure, to alert drilling staff of potential issues. However, the complexity of the drilling environment meant a rules-based approach was insufficient, and historical label data was unavailable for supervised machine learning.

‍

We developed unsupervised anomaly detection models that used techniques like Bayesian filtering and isolation forests to detect critical events in time-series data. A cloud-based architecture was deployed to ingest real-time drilling data and generate alerts. By automating 75-90% of field technician tasks, the solution improved operational efficiency, and the client can now commercialize this end-to-end data product.