Safely responding to critical events in industrial processes

Drilling wells is a complex and potentially dangerous process in which equipment failures, unexpected environmental effects due to the poorly understood subsurface, and human errors can lead to potentially catastrophic outcomes. In an effort to reduce the likelihood of these as well as the cost of drilling, Schlumberger is developing systems that can operate autonomously or semi-autonomously as part of human machine teams in complex, dangerous and uncertain environments. Such systems must be able to respond appropriately to unexpected events and one significant challenge in deploying them is in verifying that the system is safe, and that it will detect and respond to critical events appropriately. This is particularly challenging in drilling as the systems tend to be sensor poor and even leaving the system in a safe state in response to an event often involves a non-trivial sequence of activities.

In this project we aim to provide automated support that helps operators to make decisions when unexpected events occur during operations. We will explore techniques to support decision making in critical situations. In particular, we will focus on verifying that proposed sequences of actions that would lead to a safe state once unexpected events have been detected are in fact safe. Our approach will exploit recent developments in model-based test generation, especially the use of multi-agent systems combined with machine learning for test generation and for online testing of complex human-interactive systems.