top of page

60,000+

wind turbines in the US

10%

avg annual increase in wind capacity

500+

wind manufacturing facilities in US

clients_ThermoFisherLogo

EXECUTIVE SUMMARY

TensorIoT and AWS teamed up to provide SentientScience with a machine learning model that can predict wind turbine failure, and reduce costly downtime for wind farms.

GOAL

How do you find patterns in unpredictable wind and weather data, to forecast when a wind turbine might need maintenance?

RESULTS

The solution provided a machine learning model that used various public and private datasets on environment, weather, and terrain to forecast when a wind turbine needs preventative maintenance. Fixing problems before they happen helps prevent costly downtime for energy producing wind farms.

Background

With energy demands growing worldwide, a push towards using renewable over non-renewable is changing the energy market. The main challenge with renewable energy is the maintenance of the equipment. SentientScience, a provider of a digital platform that predicts short and long-term failure rates of mechanical systems, teamed up with TensorIoT and AWS to develop a solution that would reduce the cost of maintenance for wind farms by predicting wind turbine failure.

The Challenge

Wind turbines are complex mechanical machines, made up of many unique, interwoven parts. Adding to the complexity, wind turbines do not exist in a controlled environment, and need to be able to withstand fluctuations in weather and environment. When a wind turbine is in need of repair, energy cannot be produced until the turbine is fixed. Predicting when a turbine will fail is the key to reducing the cost of downtime and maintenance. Traditional statistical models, however, struggle to account for the erratic behavior of wind. Machine learning (ML) algorithms have the power to derive forecasts from historical data.

The Solution

To get started, our ML team set up a data pipeline to access and store the data they would be collecting. Next, they visualized and analyzed the data using Jupyter Notebook, an open-source web application. Looking at the data from every angle, the ML team searched for correlating features that might point to relationships within the data. Then, using Amazon SageMaker, machine learning models were built and trained. In just 3 months, TensorIoT and AWS were able to provide SentientScience with a solution that allows them to engage with Wind Energy customers at 10x scale and is projected to their shorten sales cycle by 50%.

bottom of page