Cognitive Analytics for Oil and Gas

Oil and gas companies are having problems learning from the data to understand the different operational states and failure modes of assets, and uses this learning to provide adequate warning before failures occur so operators can plan for corrective actions thus optimizing their Operations and Maintenance budgets.

SparkPredict has been deployed on Upstream assets such as Drillstrings and Electrical Submersible Pumps as well as Downstream assets such as pumps in refineries. For assets with no labeled failures, SparkPredict analyzes events and identifies anomalies (unknown operating states, failure conditions, etc.) automatically. SparkPredict uses the identified anomalies to recognize patterns of deviation and raise alarms if significant deviation from normal is observed. SparkPredict leverages cutting edge, cognitive, machine learning techniques to additionally predict asset failures. The cognitive, or reasoning based, nature of our algorithms mean SparkPredict can be deployed to any asset in any location and the insights will adapt to the unique characteristics of that particular asset. In addition, SparkPredict integrates with already installed Asset Monitoring systems or works with data historians, like OSI PI, to leverage pre-existing and/or live streaming data for improved failure predictions.

• [Process Optimization - Remote Diagnostics]
  SparkPredict detects both Stuck Pipe and Wash Out conditions with high accuracy.
• [Efficiency Improvement - Operation]
  SparkPredict provides an in-context advisory system technicians can use to quickly find documents and other digital resources to address issues, automatically provide meaningful remediation steps, and seamlessly communicate and share data with team member
• [Efficiency Improvement - Maintenance]
  For Electrical Submersible Pumps (ESPs), SparkPredict predicts failures days in advance.

• The average lead times for detection of these failures is: Stuck Pipe: 1.2 hours, Wash Out: 2.3 hours.

• SparkCognition targeted two failure modes, accounting for 85% of all ESP failures, and were able to provide the following median forewarning: Electrical Short: 5.5 days, Mechanical Breakdown: 6 days.