SLIDE 1
Acoustic Based Condition Monitoring
Young Persons’ Lecture Competition
Kyle Saltmarsh
SLIDE 2 The Outline
- The Identity Crisis
- Acoustic Based Condition Monitoring
- The How
- Applications in Industry
- Conclusion
SLIDE 3
The Identity Crisis (About Me)
SLIDE 4 Acoustic Based Condition Monitoring
- Sounds tell stories
- You do this all the time
- On-site technicians do this
- Technology does this
SLIDE 5
An Example in Action
SLIDE 6
Problem Statement
How do you use acoustics to monitor equipment in noisy environments?
SLIDE 7 Beamforming
Robust Adaptive Beamforming - Jian Li & Petre Stoica
SLIDE 8
Beamforming Applications
SLIDE 9 Signal Processing
Fourier Techniques and Applications - Joseph Fourier
SLIDE 10 Extracting Acoustic Features
Healthy Bearing Unhealthy Bearing Feature #1 Feature #2
SLIDE 11 Model: Neural Network
Feature #1 Feature #2 Output #1 Equipment Type Output #2 Health Status
SLIDE 12
Problem Solved
SLIDE 13
Unearthed Mine Hackathon 2017
How can we more effectively use sound in mining?
SLIDE 14
Healthy Vs. Corroded Bearing
Industry Data: Woodside Fin Fans
SLIDE 15
Microphone Array
SLIDE 16
Beamforming
SLIDE 17
Signal Magnification
SLIDE 18
Time Series
SLIDE 19 Acoustic Feature Extraction
Feature #1 Feature #2
SLIDE 20 Model Performance: Confusion Matrix
Accuracy: 99.5% True Condition Normal Fin Fan Corroded bearing Fin Fan Prediction Normal Fin Fan 176 1 Corroded bearing Fin Fan 2 167
SLIDE 21 Conclusion
- Sounds tell stories
- Acoustics, signal processing and predictive
modelling
- Application in Industry (Rio Tinto)
- Easy Implementation
SLIDE 22
Thank You
Questions?
SLIDE 23 Contents
- The Identity Crisis
- Rio Tinto RailBam
- Beamforming
- Signal Processing
- Feature Extraction
- Predictive Model
- Unearthed 2017