AI Case Study
Origin achieved 80% accuracy in identifying low production wells and $50M in savings using machine learning applications from C3 IoT
Origin has used the C3 IoT's C3 Predictive Maintenance solution to optimise well placement and output and derive operational value from their real-time data. Furthermore data from a variety of enterprise silos were integrated into a comprehensive data scheme.
Oil And Gas
The goal of the initial 12-week project was to develop two machine learning applications:
* Well Output Forecasting (on the C3 IoT Platform): Predict output of individual wells before drilling commences, optimise well placement by accurately detecting low-producing wells, and identify parameters that maximise well output.
* Well Equipment Health (C3 Predictive Maintenance): Predict and optimise the run-life of installed Progressive Cavity Pumps based on machine learning models that detect failure parameters and optimise equipment design choices for each well.
This was accomplished as a "unified, federated, cloud based data image was built by integrating data from sources including hourly and daily sensor measurements from each well, drilling logs, geology estimates, permeability assessments, well work logs, equipment asset records, among others... Data complexity was overcome through the creation of a single, consistent data schema. Previously, each source system codified individual wells differently or used different terms to refer to similar entities".
All 2,000+ gas wells in Origin Energy’s Integrated Gas business portfolio are equipped with 50+ sensors each, providing a rich data set of real-time data resulting in 200M+ reads per day. However, bringing together and analysing data from different isolated enterprise systems to derive some operational benefit is a difficult task.
The vendor claims the following results after 12 weeks:
* "~80% accuracy in identifying low production wells before breaking ground
* 300+ days simulated increase in run-life of wells
* $50M in identified savings."
From the vendor: "C3 Predictive Maintenance goes beyond traditional rule-based systems by using AI / machine learning to identify failures proactively. Its algorithms analyze all relevant information including sensor data, SCADA data, asset management systems, plus structured and unstructured data such as technician notes, and external data sources such as weather. C3 Predictive Maintenance is capable of predicting failure significantly in advance of legacy rule-based systems, with higher precision and recall. The application's AI / machine learning algorithms are dynamic and learn over time, adapting to emerging failure modes. It also can use unsupervised AI / machine learning algorithms to detect anomalies across all datasets – allowing operators to intercept unusual operating modes".
Well sensor data in real-time (more than 200M reads/day), along with additional structured and unstructured data (e.g. field work, drilling logs, geographical estimates) processed.