Novel machine learning agent adds predictive capability to automated generator controls
What happened
Patterson-UTI presented a reinforcement-learning agent trained on historical rig data that advises generator scheduling and integrates recommendations into the rig HMI for driller acceptance. The system was trained on data from over 100 land rigs and is designed to balance fuel efficiency, generator stability and reserve sufficiency while returning recommendations at the edge. Watch for pilot acceptance rates, HMI usability, and how vendors propose licensing, data rights and rollback controls
Buyer takeaway
Treat this as an operationally real integration that will change sourcing from purely hardware to software-plus-edge procurement because it creates ongoing data and performance obligations
Cost / money
Directionally reduces fuel and maintenance costs if accepted operationally, but requires upfront integration, edge compute and licensing terms that add near-term procurement spend
Supplier / commercial
Vendors will seek software licensing, data-access and maintenance contracts; procurement should negotiate trial/evaluation periods, data ownership, update cadence and rollback rights
Safety / operations
Changes operator workflows and creates a dependency on connectivity and model performance; acceptance tests, HMI ergonomics and manual-fallback procedures are essential
What to watch
Watch pilot acceptance rates, HMI usability, vendor SLAs for edge compute, and how data-access or update clauses are priced
Key facts
- Trained on historical data from 104 land rigs
- Advisory integrates into rig HMI as accept/dismiss recommendations
- Designed to trade off fuel efficiency, generator stability and reserve sufficiency
Source excerpts
In this setup, real-time power data from each engine is transmitted to the rig control system through a PLC, which then transfers that data to an edge device
These mistimed shutdowns disrupted drilling operations and reduced operator confidence in the automation. As a result, many drillers chose to disable or override automatic shutdowns, preferring the predictability of manual control
We’ll see things like rig power demand, the operational state, generator status and so forth
