DINGO’s Trakka Predictive Analytics solution utilizes artificial intelligence and machine learning to predict impending equipment failures with confidence, allowing miners to proactively perform corrective maintenance actions to minimize downtime and optimize asset life.
Our Anomaly Detection and Remaining Useful Life (RUL) models are built by uniting failure data from actual equipment, DINGO’s industry expertise and data science to address common component-specific failure modes, such as final drive gear teeth wear.
Trakka’s Predictive Analytics solution, powered by our proprietary machine learning library, can predict the time until asset/component failure with a high degree of accuracy. As a result, our customers are reaping the benefits of these Remaining Useful Life (RUL) models as they:
|Unexpected Failures and downtime||Component Life by acting earlier|
|Repair costs as scheduling is optimized||Confidence in planning component replacements|
|Loss of wasted potential in capital||Equipment availability and reliability|
|Unnecessary maintenance activities||Budgeting and the bottom line|
|Personnel and process risk by creating a safe and more controlled environment||Business related processes such as procurement, logistics, and management|
Below are some case studies that show the practical application of predictive analytics.
Case Study: Final Drives
A DINGO customer was experiencing degrading final drive life on Caterpillar 789 C&D trucks, causing high cost and downtime impacts to the operation.
The average life of final drives had decreased from;
- 19,092 hours (2014 to 2016) to
- 13,229 hours (2016 to 2018).
- 151 final drives were replaced during that period.
Cost impact: US$ 1.7 MM (parts)
Availability impact: More than 3,000 downtime hours.
An unplanned failure in the pit incurs higher downtime/repair costs than scheduling repairs in the workshop.
Identifying Failure Modes & Developing Predictive Model
DINGO worked closely with the customer to understand the details of the problem, what the failure modes were and what condition monitoring data could be used to develop a predictive model
The most common failure mode was found to be:
- Broken/Worn teeth in the central gear
- Worn Copper Washers
Condition Monitoring Data
The only condition monitoring data available for these components was oil analysis
DINGO’s data science team used the failure data and oil analysis history to;
- Identify the most correlated oil analysis indicators to failure.
- Develop proprietary machine learning models to predict future failures
Accurately Predicting Failure
The outputs of these models are integrated into DINGO’s flagship Trakka™ software application.
Customers can quickly see the current age and Remaining Useful Life (RUL) for each component and plan maintenance activities accordingly.
Historical predictions from each model are also available for the customer.
Case Study: Gearbox and Pinion Bearings
A DINGO customer was experiencing unexpected failures in their Ball Mill drive gearboxes and pinion bearings. There was a large amount of condition monitoring data available (vibration, pressure & temperature sensors) but only a few failure records.
The challenge was to isolate the different failure mechanisms (no failure classifications provided) and develop a predictive model that would allow component repair/replacement to be scheduled during quarterly plant shutdowns.
DINGO’s data science team developed an anomaly detection algorithm that determined the indicators whose degradation factors correlated best with gearbox failures, separately to pinion bearings.
Failure Probability Modeling
DINGO created a predictive model from the anomaly detection algorithm’s outputs that could predict the probability of failure over time for each gearbox and pinion bearing.
The model was backtested against previous failure data to determine the appropriate confidence level of the model.
Backtesting results show 85% accuracy in predictions, correctly identifying past failure events separately to repair events.