Predictive Analytics

Problem

• Unidentified O&G pump failure cost downtime and loss in revenue

What I did - Enhanced and tested prototype

• Classification: helped domain expert to classify pump failure events and patterns leading to it
• Solution: field data fed to a proprietary algorithm that transformed data into the likelihood of patterns leading to failure. Bayesian Network computed probability of actual failure
• Prototype: planned, developed, and gathered feedback in fast cycles
• Test: trained and tested prototype with operational datasets
• Communicate: presented findings and recommend future steps to leadership

Results

• Found prototype was not ready for commercial development (false positives)
• Prototype paved the way for a range of pump monitoring predictive analytics solutions