Asset Integrity Dashboard Version 2 & Machine Learning

As Robotics and Autonomous Systems (RAS) take more responsibility for Operations and Maintenance (O&M), the data collected must be displayed in a manner which is easily understandable and intuitive for remote human operators. As we progress the Asset Integrity Dashboard (AID) tool, the Smart Systems Group have the potential to use the AID to collect data from a number of inspection devices (cameras, thermography and FMCW radar) to allow for a holistic approach and enhanced operational overview of a wind farm for a remote human operator at the shoreline. The software allows for historical data to be compared and an overall improvement in the management of O&M of an offshore array.

Asset Integrity Dashboard Version 2 displaying the material properties for a wind turbine blade sample with a defect of a variation in the adhesive thickness.

The Smart Systems Group at Heriot-Watt University recently adapted their Asset Integrity Dashboard to display results from the patented Frequency Modulated Continuous Wave (FMCW) Radar which was utilised for non-destructive evaluation of samples which were manufactured to represent wind turbine blade defects. The millimetre-wave sensing mechanism was utilised to evaluate the capabilities of the radar on sandwich composite and monolithic composites for the detection of the following:

  • Internal Features – Air voids and adhesive thickness variation
  • Surface and Near Subsurface Defects – Uncured epoxy resin and composite density
  • Inclusions – Presence of anomalies, metals, planar inclusions (such as balsa)

As displayed in the interactive Asset Integrity Dashboard below and corresponding video. The advancement in this updated version includes a more effective understanding of the procedure of the investigation undertaken to verify the FMCW sensing capabilities for our external partner. We identify a consistent signal response to adhesive variations where an increase in thickness at the chord correlates to an increase in return signal amplitude and a decrease in thickness at the chord correlates to a decrease in return signal amplitude.

Machine Learning

Parallel work which was completed beyond the scope of the project included the combination of FMCW radar data alongside Machine Learning techniques. This was achieved by utilising a robotic arm manipulator to construct a library for machine learning to detect defects within the sample.

  • The utilisation of machine learning models on FMCW radar return signals collected from the different turbine blade samples demonstrate that we can classify blade types by composition, and diameter differentials of 3mm, with 98.5% classification accuracy.
  • The result also shows the capability of this novel approach as when trained with compressed or limited range of radar signal classification, the accuracy still reaches over 92.9%.

Try the Asset Integrity Dashboard Below: Controls: A=Left, D=Right, Mouse left click to interact.

Make sure to play the Asset Integrity Dashboard in full screen.

Leave a Reply

Your email address will not be published.