Deep Neural Networks for Rapid Fault Detection in Marine Hydrokinetic Turbines

By Sean Passmore
Slide 1: Title slide for Deep Neural Networks for Rapid Fault Detection in Marine Hydrokinetic Turbines by Sean Passmore

Slide-1

Presented by Sean Passmore

Slide 2: Problem statement slide describing MHK cost challenges and maintenance issues

Slide-2

The Problem

  • MHK has a high Levelized Cost of Energy (LCOE)
  • LCOE for MHK is approximately 18% Operations and Maintenance (O&M)
  • O&M cost reduction with Neural Networks can increase MHK viability
Slide 3: Framework diagram showing system architecture with video source reference

Slide-3

The Framework

Source: https://www.youtube.com/watch?v=R2hO--TIjjA

The framework diagram illustrates the system architecture for implementing deep neural networks in marine hydrokinetic turbine fault detection.

Slide 4: Solution comparison between traditional methods and deep learning approaches

Slide-4

The Solution

Traditional Methods

  • Use hand-designed features
  • Require significant domain knowledge
  • Generalize poorly to other domains

Deep Learning

  • Automatic feature extraction
  • Independent of need for prior knowledge
  • Higher generalization ability
Slide 5: Model architecture diagram showing how the deep learning model works

Slide-5

How the Model Works

This slide presents a visual diagram of the deep learning model architecture, showing the data flow and processing stages for fault detection in marine hydrokinetic turbines.

Slide 6: Experimental design methodology explaining training parameters and optimization approach

Slide-6

Experimental Design

  • Important to find both the best training sequence length
  • Important to find best training fault
  • Find the best combination of both
Slide 7: Results section showing experimental findings and performance metrics

Slide-7

Results

This slide presents the experimental results and performance metrics of the deep neural network model for fault detection in marine hydrokinetic turbines.

Slide 8: Conclusions summarizing key findings about optimal parameters and model performance

Slide-8

Conclusions

  • 1 second of data provides the best accuracy
  • 5 degrees is the best fault to train on
  • Bi-LSTM creates a robust feature space resilient to noise
Slide 9: Applications and future research directions with CNN approach and software reference

Slide-9

Applications and Further Research

Reference: https://softwarerecs.stackexchange.com/questions/28169/drawing-convolutional-neural-networks

Last slide: Contains plain text stating 'End of presentation. Click the right arrow to return to beginning of slide show.'

End of Presentation

Click the right arrow to return to the beginning of the slide show.

For a downloadable version of this presentation, email: I-SENSE@FAU.

Additional Information
The Institute for Sensing and Embedded Network Systems Engineering (I-SENSE) was established in early 2015 to coordinate university-wide activities in the Sensing and Smart Systems pillar of FAU’s Strategic Plan for the Race to Excellence.
Address
Florida Atlantic University
777 Glades Road
Boca Raton, FL 33431
i-sense@fau.edu