Recognizing an opportunity to reduce drilling time in the presence of large volumes of historical streaming data is difficult. Integrating machine learning algorithms and big data visualization tools enables the engineers to determine the optimal landing zone, rpm and Weight On Bit (WOB) thus achieving a higher Rate Of Penetration (ROP) while maintaining steerability.
The value of an unconventional asset is determined by the number of drilling locations, spacing, stacking and their individual performances. Machine learning and modern optimization algorithms will maximize ROI by recommending well completions designs and spacing. After identifying production drivers, our intelligent design of experimentation process recommends how to proceed and learn through fewer experiments. Learning faster, smarter, we improve the economics of a larger set of wells.
Drilling and well completions costs vary over time. To accelerate the capture of current cost data a machine learning assisted mobile app can be developed to process field tickets and update current spend closer to real-time. This new cost data is then injected into forecasting tools to guide expected spend through to year-end.
Operators face increasing pressure to maintain capital spend within guidance. Machine learning systems can interrogate historic spend, scheduling and operations data to forecast expected expenditures. By providing accurate visualizations of expected and actual costs, planning teams have the flexibility to scale operations up or down in response to available capital, thereby avoiding shutting down operations too soon or not investing all available capital.
Our engineers perform anomaly detection using in-house trained predictive models to provide early warning alerts and diagnostic guidance to our customers. In a complex industry setting such as power plants, aviation etc., anomaly detection is critical to raise alarms beforehand to prevent significant damages and mishaps. Here, we propose a novel approach of using a multimodal neural network-based autoencoder. These detected anomalies generally tend to be more accurate and robust than anomalies detected by a single data source as latent representations capture inter-dependencies between different parts of the system through those multiple data sources.