Case Study

Predictive Analytics to Reduce Impact on Upstream Energy Production

About the client

Our client is one of Europe’s largest Oil & Gas producers.
We partnered with a leading AI & Cognitive computing company to deploy AI solutions in the client’s upstream production hubs.

Overview

The objective was to build predictive capability for the client to forecast impending failures in production at least four days in advance using machine learning models.

Business Challenge

Recurring failures in production subcomponents

Globally, Oil & Gas producing companies are looking to maximize their production capacity, increase efficiency, and improve safety by deploying AI-powered predictive analytics. On the other hand, machine failures lead to millions of dollars in losses for these large industrial companies. A report suggests that large plants lose approximately 323 production hours a year. For our client, recurring failures in production subcomponents resulted in >10% downtime and millions of dollars lost in production.  
The same report estimates that the cost of revenue lost, penalties, idle resource time, and the cost of restarting lines can amount to $172 million per plant in a year.

Our Approach and Solution

The ML Models of the client utilized data from multiple offshore plant components to predict impending failures. Our team trained multiple advanced ML Models to predict output from different subcomponents like Compressors, pumps, Gas dehydration, Oil treaters, degassers, separators, valves, etc. Our team also collected sensor data from the upstream system to train ML models.
 
-Our data scientists started by making sense of thousands of tags, including pressure, vibration, temperature, and others. The team reduced the number of tags to about 130 important ones per subsystem. These tags needed further optimization via dimensionality reduction techniques to remove noise in the data.
-This high-quality dataset can help build unsupervised models for each subcomponent. It can also identify new (previously unknown) operating states of sub-components.  

Tech Stack

15

This is some text inside of a div block.
Team members involved

30+

Days delivery

Business Impact

With our ML models, the client successfully identified 75% of production-impacting events with an average of 8 days in advance.
The solution helped SMEs predict and diagnose impending failures and prioritize work orders accordingly.
With reduced production downtime, the annual production of the plants increased by $30M per plant.
With predictive capability, the client could take action on maintenance and other scheduled activities, reduced deferred production to 2 days with savings worth $10M