Application of Support Vector Machine for Structural Health Monitoring.pdf
- 1. Department of Civil Engineering SHARAD INSTITUTE OF TECHNOLOGY COLLEGE OF ENGINEERING YADRAV, KOLHAPUR, MAHARASHTRA-416121. Application of Support Vector Machine for Structural Health Monitoring Dr. Maloth Naresh
- 2. Flow of presentation 1 Introduction 2 Components of the ANN 3 4 Advantages of ANN References 6 5 Disadvantages of ANN Artificial neural network (ANN)
- 3. Introduction • Structural Health Monitoring (SHM) is the continual monitoring and evaluation of the health of civil, mechanical, and aeronautical structures. • This includes: Data Acquisition (via Sensors). Signal processing. Damage Detection/Classification. Decision-making for Maintenance. • Goals: Enhance safety. Reduce your maintenance expenditures. Increase dependability and lifetime.
- 4. Machine Learning in SHM • Traditional NDT methods (e.g., visual inspection, ultrasonic) are limited by: • Manual intervention • Subjectivity • Inability to detect early-stage damage · Machine Learning (ML) automates decision-making by learning from data. · ML enables: • Pattern recognition from sensor data • Classification and regression • Real-time damage prediction • Real-time damage prediction
- 5. Support vector machine (SVM) • Vapnik introduced SVM, a supervised learning method, in the 1990s. • Objective: Increase the margin between classes in the feature space. Works well for: • Binary classification (e.g., damaged versus undamaged). • Multi-class problems using One-vs-Rest or One-vs.-One techniques. • Nonlinear issues (using the kernel trick). • Kernel Functions • Linear Kernel • Polynomial Kernel • Radial Basis Function (RBF)
- 6. Types of SHM data for SVM • Time-domain: acceleration, displacement, strain, and velocity. • Frequency-domain: FFT spectra and modal frequencies. • Time-frequency domain: wavelet coefficients, scalograms. • Other: images, audio emissions, temperature, and humidity. • Sensors include accelerometers, strain gauges, piezoelectric sensors, and fibre optics.
- 7. Hyperparameters of SVM and Optimization techniques • Important parameters: C (Penalty parameter): Trade-off between margin and classification mistake. γ (Gamma in RBF) controls the form of the decision boundary. • Optimisation Techniques: To select the best hyperparameters Grid Search. Random Search. Genetic Algorithm Particle Swarm Optimisation Bayesian optimisation
- 8. Limitations and solutions • Limitations: Requires labelled data Parameter tweaking is complicated. Not suitable for enormous datasets. Solutions: Use cross-validation. Integrate with other algorithms (hybrid models). Feature Selection to Reduce Dimensionality
- 9. Conclusions • SVM provides high classification accuracy, robust generalisation, and interpretable results for SHM applications. • Especially useful for binary and multiclass categorisation of structural situations. • The future scope includes hybrid models, real-time monitoring systems, and interaction with digital twin technologies.
- 10. References 1.Vapnik, V. (1998). Statistical Learning Theory, Wiley. 2.Worden, K., Manson, G., & Fieller, N. R. (2007). "Damage detection using outlier analysis." Journal of Sound and Vibration. 3.Farrar, C. R., & Worden, K. (2007). "An introduction to structural health monitoring." Philosophical Transactions A. 4.Sohn, H., Farrar, C. R., et al. (2004). "A Review of Structural Health Monitoring Literature: 1996–2001."
- 11. Thank you