Predictive Model reduced Breakdowns by 17% for Precious Metals Refinery
Before
A large precious metals refinery faced significant challenges with their cooling systems. Brief machinery breakdowns led to substantial financial losses due to the high value of the refined metals. The company needed a robust big-data analytics solution to implement a preventive maintenance strategy, aiming to reduce the frequency and impact of these breakdowns.
Solution
Our data scientist developed a comprehensive big-data analytics solution tailored for the refinery’s cooling systems. The approach included:
- Isolation Forest for anomaly detection, identifying unusual patterns in sensor data that could indicate early signs of system failure.
- Time-series forecasting using gradient boosting models on both raw and aggregated data to predict future breakdowns. The models achieved T+1 month prediction accuracy of 80% and T+3 month prediction accuracy of 60%.
- Sensor categorization based on properties, with data managed on a log scale to efficiently integrate both old and new sensor data.
- Principal Component Analysis (PCA) to identify the most critical sensors (from a total of 63 high-frequency sensors) directly linked to machine failure, allowing for focused monitoring and maintenance.
This solution enabled the refinery to implement a new predictive maintenance schedule, significantly reducing the likelihood and impact of future breakdowns.
Outcome
Leveraging advanced anomaly detection and time-series forecasting techniques, the solution accurately predicted potential breakdown events, allowing the refinery to adjust maintenance schedules proactively. The introduction of a predictive maintenance model led to a 17% reduction in cooling tower breakdowns over six months.
