Detecting pilferage and leaks for Oil & Gas Industry
About the Customer: Healthcare Provider ($5B+ market cap) Consistently ranked under top 15 providers in the US In North America with over 45 hospitals and >2+ million patients
Compelling Event
Identify leakages in Gas stations.
The Pain point
- Handling massive sensor data is problematic for many organizations and our client was no exception.
- To perform analytics on sensor data is a nightmare because of instrumentation errors, you can observe values changing haphazardly with in milliseconds.
- Loss due to False Alarms:
Type 1 error: Cost of investigation incurred because of false alarm.
Type 2 error: Loss of fuel or accident cost, if any, for not carrying the investigation, assuming that alarm is false.
Tech stack
- Tableau
- Amazon Redshift
- SQL Workbench
- Predera- KAI
Predera Benefit
Optimization of Anomaly Detection technique
- Solid Feature Engineering - Several heuristics pertinent to seasonality of data were added. Dimensionality reduction techniques were used to come up with optimum set of features for ML model
- Clear comprehensive process limits were set up to differentiate leakages from false alarms
- Root Cause investigation is made easier by providing visualizations across various dimensions
- Errors in existing summarization techniques have been identified and rectified, thereby improving the quality of sensor analytics
Business Impact
Avoiding the false alarms by optimizing anomaly detection technique using AI model has helped our client to identify losses at the earliest, thereby saving huge costs. Out of pocket cost of investigation on false alarms was also avoided.