Early Gearbox Fault Detection
and Condition Monitoring

Challenge

Monitoring gearboxes on large compressors requires distinguishing genuine wear or misalignment from normal variations in load and vibration. Early signs of bearing or gear tooth damage are often subtle and can be masked by changing operating conditions.

The challenge was to detect developing mechanical issues early enough to prevent damage, while avoiding false alarms caused by normal process and load fluctuations that can reduce trust in condition monitoring systems. 

Solution

Historical gearbox and process sensor data were used to build anomaly detection models focused on mechanical displacement of the gearbox wheel shaft (measured in mm/s). The models were trained to reflect normal gearbox behavior across varying load and operating conditions.

New data was then evaluated against the established baseline to identify deviations from normal behavior. Recurring anomaly patterns were reviewed by operations and maintenance teams, who applied their domain expertise to distinguish genuine mechanical issues from normal process variability.

This approach supported earlier identification of developing gearbox issues and improved confidence in condition monitoring without relying on fixed thresholds.

Value Delivered

• Reduced energy consumption without compromising throughput
• Clear understanding of operating envelopes and trade-offs
• Decision support for operators to run closer to optimal conditions