|
|
| Academic Year |
2024/2025 |
| Name |
Dott. - MI (1383) Ingegneria Meccanica / Mechanical Engineering |
| Programme Year |
1 |
| ID Code |
062883 |
| Course Title |
METHODS FOR HEALTH MONITORING AND PROGNOSIS OF ENGINEERING SYSTEMS SUBJECT TO DEGRADATION |
| Course Type |
MONO-DISCIPLINARY COURSE |
| Credits (CFU / ECTS) |
5.0 |
| Course Description |
In the last decades, many different methods have emerged for the health monitoring of engineering systems, including structural, mechanical, aerospace, energy and bio-inspired applications, just to mention a few. In most of the cases, this requires the acquisition of multiple sensor data, coupled with signal processing techniques for capturing any feature of interest, in the hope that the occurring degradation mechanism is correctly identified, then enabling system prognosis. Within this framework, the course introduces different methods for the analysis of real-time data aiming the diagnosis and prognosis of systems subject to degradation in a realistic environment, including (i) outlier analysis for novelty detection, (ii) data normalisation for suppression of confounding influences, (iii) machine learning for diagnosis and surrogate modelling, (iv) sequential Monte-Carlo methods for state identification and prognosis. Follows the course programme:
- Introduction and Design for SHM systems: Basic principles and definitions, Hardware components of the CMS / SHM and sensors, Design process, Motivation for SHM, (NDE vs SHM), Model-based vs Data-based SHM, Statistical pattern recognition paradigm, Historical overview Feature selection and extraction: basic statistics, statistical moments/distributions, central limit theorem, principal component analysis
- Application of feature extraction methods Novelty detection: features in the context of detection theory, feature classification, environmental/operational effects on SHM
- Application of methods for Novelty Detection, Outlier Analysis, ROC curves
- Methods for classification: Support Vector Machines and Artificial Neural Networks Methods for damage identification
- Artificial Neural Networks for localization and assessment
- Gaussian process for damage assessment Bayesian Inference as a mean to solve inverse problems in SHM (fusion of prior-actual information and sensor-model information) Introduction to Monte-Carlo sampling theory
- Metropolis-Hastings MCMC algorithm for Bayesian model updating.
- Application of MH-MCMC to anomaly identification. Sequential Monte-Carlo sampling theory: the particle filter.
- Application of SMC sampling algorithm to anomaly identification and prognosis. |
| Scientific-Disciplinary Sector (SSD)
|
--
|
|
Alphabetical group
|
Name
|
Teaching Assignment Details
|
|
From (included)
|
To (excluded)
|
|
A
|
ZZZZ
|
Sbarufatti Claudio, Cadini Francesco
|
|
|
