Case Study: Hydro Utility achieves one-month advance warning on rare turbine failures with SparkCognition SparkPredict

Predicting Rare Failures in Hydro Turbines

Hydro Utility faced a costly unplanned outage with an estimated $1.5 million impact and limited historical failure data to guide maintenance decisions. To address the challenge, the utility evaluated SparkCognition’s SparkPredict® product to see whether machine learning could better predict rare hydro turbine failures and reduce false positives from its existing monitoring approach.

SparkCognition used two years of turbine sensor data to build an unsupervised machine learning model and identify the most important operating tags, including generator speed, temperature, oil level, vibration, and shaft gap measurements. The solution detected the turbine failure more than one month in advance and surfaced key leading indicators, including nine sensors with notable changes, helping the utility improve root cause analysis and consider broader deployment across additional turbines.