R
ig crews work in a stressful, loud environment where they endure
a barrage of sensory input. Movingmachinery, connections being
made. Instrumentation and sensors of every kind displaying
values and trace graphs. Alarms sounding.
In this environment detecting dangerous and expensive events such
as kicks, mud losses, and spills can be challenging for crews. So one
issue designers of detection systems need to consider is alarm fatigue.
Traditionally, systems are designed to alert the crewwhenmud levels
and flow rates move outside a defined range, which can be a sign of
lost circulation or impending kicks. However, mud levels and flow rates
fluctuate widely during normal drilling operations, so even a correctly
calibrated and functioning systemwill generatemany alarms that do not
signify lost circulation or kicks.
This fosters an environment where drillers can become desensitised
to alarms. Theymight cope with this alarm fatigue by decreasing the
sensitivity of alarms, which decreases the effectiveness of the detection
system. Or theymay turn off the alarms, either temporarily or altogether.
This is like taking off a seat-belt because it chafes the shoulders.
In developing the Enhanced Pit Volume Totalizer (ePVT) Event
Detection system, Pason has grappled with this challenge of reducing
alarm fatigue. The system’s approach begins with an interface designed
from the ground up in co-operation with drillers in the field. The
interface, presented on the large touch-screen Rig Display, is designed
so rig crews can get the information they need within seconds, and
determine if they need to take action.
Under the hood, the newest version of the system includes a
powerful feature called adaptive alarms, which exploits machine
learning to enable a far more robust approach to the detection of kicks
andmud losses.
Machine learning is a field of computer science related to pattern
recognition and statistics. It enables computers to learn fromdata in
order tomake predictions, and to act like rational agents. Machine
learning techniques are used to identify credit card fraud by recognising
deviation fromnormal spending behaviours. Machine learning is also
behind targeted online advertising, search engines, and email spam
filtering. It was also used to teach computers to automatically identify
NO CAUSE FOR ALARM
SEAN UNRAU, PASON, CANADA, REVEALS THE WAYS IN WHICH ALARM FATIGUE IS BEING REDUCED.
|
67
