Autonomous Edge-Deployed AI System for Real-Time MPD Control and Optimization

Abstract

Abstract Current Managed Pressure Drilling (MPD) systems rely heavily on operator interpretation, rule-based automation, and reactive responses to wellbore events, approaches fundamentally limited by human response times, static control algorithms, and inability to learn from operational experience. This paper introduces a novel autonomous Artificial Intelligence (AI) based control system that extends edge computing beyond conventional rule-based automation to include self-learning hydraulics models, dynamic drilling window updates, and reinforcement learning-based control, capabilities previously requiring offline analysis and manual model tuning. The system integrates AI-driven models operating in real-time at the rig site: a hybrid physics-based/machine learning model that continuously learns and corrects pressure predictions; a geomechanics model that dynamically updates pore and fracture pressure; and a reinforcement learning control that maintains bottomhole pressure within drilling window boundaries during both static and dynamic conditions. The architecture addresses key automation gaps: pattern-based classification distinguishes wellbore events; severity-graded safe modes provide graduated responses matched to anomaly type; and predictive equipment degradation monitoring enables proactive maintenance scheduling. This paper presents the system architecture, AI model integration methodology, and operational implications of transitioning from human-supervised to AI-supervised MPD control while maintaining operator's oversight and override capabilities.