Predicting Failures in Hydro Turbines: Easy as H2O

demo

From the massive blades spinning on top of hillsides to solar panels decorating rooftops, advances in technology are helping to create a cleaner, more sustainable source of energy. Hydropower is no different. Assets to harness such power are large, complex, and expensive, but with a predictive, machine learning-based approach, these assets can be monitored in such a way that these sources of fear can remain solely sources of energy.

By harnessing the power of water through dams, natural kinetic energy is transformed into electricity at an extremely efficient rate. Hydro turbines are massive, have an efficiency of up to 90%, and generate hundreds of megawatts of power. A single megawatt can power approximately 250 homes; hundreds of megawatts can power far more.

Read our hydro turbine case study.

However, with an asset of this magnitude, it also means that unexpected downtime hits hard. It is estimated that a single day of downtime can equate to upwards of $50,000. With particularly severe failures, repair time can last for months, and costs from downtime alone can be catastrophic. Even worse, there is potential to disrupt the energy grid affecting those relying on the utility.

Needless to say, utility companies relying on hydropower would prefer to avoid unexpected failures in their assets. However, in order to avoid unplanned downtime, it’s necessary for the utility to be able to see past what’s happening in real time and to be able to predict when and where an asset might fail.

This is where machine learning comes in. The data put out by the assets can be monitored and anomalies flagged. There have been very few instances of hydro failures, leaving little basis for comparison and ruling out traditional, statistics-based analysis of the data due to multiple errors and false positives. SparkCognition’s machine learning based approach, SparkPredict, functions differently.

In one case study, SparkPredict measured variables such as generator speed, power output, temperature, oil levels, and vibrations and was able to accurately identify a turbine failure with one month’s notice. Even more impressive, the software was able to conclude this with little failure data for the asset it was working on. Instead, the program used what is called unsupervised learning, which generates a solution without a known desired outcome.

In implementing machine learning, a utility company is able to predict turbine failures with enough time to prevent expensive repairs and downed grids—and of course, they can still save the planet.

To get more specifics on how SparkCognition is revolutionizing maintenance for hydro turbines, download our case study for a major utility.

Latest blogs

SparkCognition is committed to compliance with applicable privacy laws, including GDPR, and we provide related assurances in our contractual commitments. Click here to review our Cookie & Privacy Policy.