To be 100% CO2 free in 2050, as the EU climate goal states, 15 to 50% of our energy needs to come from wind energy! 🔋 The culprit holding us back is wind turbine maintenance ✋ & that we're solving with @TaruccaTech in the AI for Wind Energy challenge!

The context

To achieve the EU-climate goal of 100% CO2-free energy production in 2050, 15% to 50% of Europe’s electricity is to come from wind energy. 

  • This means 50.000+ new offshore turbines needed for 22x increase in European offshore wind power capacity. 
  • Blades are the last frontier for predictive maintenance, causing major reduced power output and efficiency as wind parks age. The result is immense costs of inadequate detection of blade damage. 
  • The periodic and visual damage using specialists or drones on-site occurs only at intervals, requires stoppage and does not monitor inside the blade for structural health monitoring (SHM) predictive maintenance. Damage identification based on ambient vibration data is one of the most effective SHM technologies that allows the continuous acquisition of the structural response. 
  • The lack of an – in all weather conditions reliable – SHM is one of the reasons hindering the widespread adoption of these techniques for industrial applications.

🎈 Our partner in this challenge Tarucca is creating a system to monitor wind turbines & predict damage, minimising down time!

The challenge

Extend the lifetime of wind turbine blades by identifying damage based on ambient vibration data 💪🏽

This challenge can be broken down in the following steps:

  1. Preprocessing of raw vibration data (e.g. with high low pass filters used by the industry)
  2. Discover which ML model is best at extracting damage sensitive features from the influence of environmental and operational variables.
  3. Replicate and build upon the state-of-the-art damage detection algorithm.

Damage is represented as the modal parameters (frequency, shape and damping) shifts on the wind turbine blade. The difficulty for the algorithm is that the environment and operational variables (EOVs) impact these parameters (5-10% variations in frequency), so the solution needs to be:

  • Robust to nonstationary conditions. Nonstationarity primarily stems from constantly-changing wind and loading conditions (gusts, turbulence, etc.).
  • Localize damage. Damage localisation is challenging for operational wind turbines, since this is a function of excitation frequency and number of sensors, as well as data-acquisition rates and sensor placement.

Who we’re looking for

We expect some experience with programming languages and an interest in machine learning. Anyone with the right motivation and ‘proof’ of understanding of the core concepts found in the application form can sign up!

During this challenge you will use / learn these tech skills:
🤖 Convolutional Neural Networks
📈 XGBoost with SHAP
🪅 outlier analysis

You will work in a team of 35 FruitPunchers, together with experts from the field, like: CTO Jesse van Kempen, CEO Hans van Beek, COO Tim Ewart & Mark van der Bruggen!

You can join as a contributor (12 hours per week commitment for 2 months), coach (2-4 hours per week, only for experienced ML professionals) and teacher (give one relevant ML / domain masterclass).

During the challenge we will arrange for masterclasses on relevant topics like the use of the cloud computation resources Microsoft, IBM & the HPC Lab have made available for us.

Challenge format

The challenge will run from the 17th of November until the 16th of February, & you will collaborate with a diverse team of over 30 international data specialists and domain experts in subteams, all tackling this problem from different angles.

Some important dates:

  • Info session                                – Tuesday 26th of October – 17:00 CET
  • Application deadline                  – 15th of November
  • Kickoff date & time                    – 15th of October 19:00 CEST
  • Midterm date & time                 12th of Januari 19:00 CEST
  • Final presentation date & time – 16th of February 19:00 CEST

Apply to join the Challenge

Fill in the Google Form behind this button to become one of the 50 engineers collaborating with Tarucca to develop ML models that’ll enable wind energy as a clean energy alternative!

Background info

🥥 Rather than working together to solve what’s clearly an industry-wide problem, everyone appears to be scrambling to figure it out on their own. Trade secrets don’t often pass between corporations, yet that may be exactly what is needed to rid wind power of some of these efficiency problems.

🍉 It is crucial that a developed SHM is reliable and has a low amount of false alarms. Wrong interpretations of data can lead to unnecessary inspections driving up the cost. Differentiating between novelties caused due to the influence of Environmental and Operational Variables (EOVs) or damage in the structure can help reduce false alarms.

Partners in this challenge