Structural Health Monitoring Based on Data Science Techniques

Foreword
Preface
Acknowledgements
Contents
Vibration-Based Structural Damage Detection Using Sparse Bayesian Learning Techniques
	1 Sparse Bayesian Learning for Structural Damage Detection
	2 Bayesian Probabilistic Framework
		2.1 Structural Model Class
		2.2 Bayesian Model Updating Framework
		2.3 Likelihood Function for Structural Modal Parameters
		2.4 Prior PDF of the Damage Indexes
		2.5 Posterior PDF of the Damage Indexes
	3 Bayesian Inference Using the EM Algorithm
		3.1 Posterior Sampling
		3.2 Likelihood Sampling
		3.3 Summary
	4 Bayesian Inference Based on the Laplace Approximation
	5 Bayesian Inference Based on the VBI-DRAM Algorithm
		5.1 VBI
		5.2 DRAM Algorithm
		5.3 Summary
	6 Numerical Example
		6.1 Model Description
		6.2 Damage Identification
	7 Experimental Study
		7.1 Model Description
		7.2 Damage Identification
	8 Comparison and Discussions
	9 Conclusions
	References
Bayesian Deep Learning for Vibration-Based Bridge Damage Detection
	1 Background
		1.1 Structural Health Monitoring
		1.2 Machine Learning
	2 Method
		2.1 Assumptions
		2.2 The Damage Detection Algorithm
		2.3 Validation
		2.4 Determining the Sequence Error Threshold
	3 Experiments
		3.1 Z24 Benchmark
		3.2 Training
	4 Results
	5 Discussion
		5.1 Damage Sensitivity
	6 Conclusion
	References
Diagnosis, Prognosis, and Maintenance Decision Making for Civil Infrastructure: Bayesian Data Analytics and Machine Learning
	1 Introduction
	2 Summary of Data Sources
		2.1 Physics-Based Simulation Data
		2.2 Inspection Data
		2.3 Human Errors
		2.4 SHM Data
	3 Damage Diagnostics Using Bayesian Data Analysis and Machine Learning
		3.1 Surrogate Modeling for Physics-Based Models Using Machine Learning
		3.2 Damage Diagnostics Using Recursive Bayesian Updating
		3.3 Sensor Placement Optimization for Damage Diagnostics Using Machine Learning
	4 Failure Prognostics Using Bayesian Data Analysis and Machine Learning
		4.1 Failure Prognostics Based on Inspection Data
		4.2 Mapping Inspection Data into Continuous Degradation Model for Failure Prognostics
		4.3 Integrated Failure Diagnostics and Prognostics Using Dynamic Bayesian Networks
	5 Optimization of Maintenance Strategy
	6 Conclusion
	References
Real-Time Machine Learning for High-Rate Structural Health Monitoring
	1 Introduction
	2 Opportunities in the Input Space
	3 Deep Learning Strategy for HRSHM
		3.1 Physics-Informed Input Space
		3.2 Numerical Demonstration
	4 Path to Rapid State Estimation
		4.1 Methodology
		4.2 Numerical Demonstration
		4.3 Discussion
	5 Conclusion
	References
Development and Validation of a Data-Based SHM Method for Railway Bridges
	1 Introduction
	2 Non-destructive Evaluation, Physics-Based and Data-Based Methods in SHM and Damage Detection in Bridges
	3 An Unsupervised Data-Based Approach to SHM Based on Feedforward ANN
	4 A Numerical Case Study
	5 A Field Case Study
	6 Conclusions
	References
Real-Time Unsupervised Detection of Early Damage in Railway Bridges Using Traffic-Induced Responses
	1 Introduction
	2 SHM Procedure for Early Damage Detection
	3 Data Acquisition from a Bowstring-Arch Railway Bridge
		3.1 The Railway Bridge Over the Sado River
		3.2 SHM Monitoring System
		3.3 Baseline and Damage Scenarios Simulation
	4 Strategy for Early Damage Detection Using Train-Induced Responses
		4.1 Features Extraction Based on CWT and PCA
		4.2 Features Modelling Based on PCA
		4.3 Features Fusion
		4.4 Clustering-Based Classification
	5 Conclusions
	References
Fault Diagnosis in Structural Health Monitoring Systems Using Signal Processing and Machine Learning Techniques
	1 Introduction
	2 Fault Identification Through Signal Processing and Machine Learning
	3 Fault Diagnosis Based on Signal Processing and Machine Learning Techniques
	4 Validation Tests
		4.1 Description of the Machine Learning Algorithms
		4.2 Description of the Structure Used for Validation
		4.3 Data Preparation and Determination of the Machine Learning Algorithm Parameters
		4.4 Validation Tests and Results
	5 Summary and conclusions
	References
A Self-adaptive Hybrid Model/data-Driven Approach to SHM Based on Model Order Reduction and Deep Learning
	1 Introduction
	2 Monitoring Procedure
	3 Numerical Modeling
		3.1 Full-Order Model
		3.2 Dataset Assembly
		3.3 Parametric Model Order Reduction for Dataset Generation
	4 Deep Learning
		4.1 Damage Detection and Localization
		4.2 Damage Quantification
		4.3 Transfer Learning
	5 Results
	6 Conclusions
	References
Predictive Monitoring of Large-Scale Engineering Assets Using Machine Learning Techniques and Reduced-Order Modeling
	1 Introduction
	2 A Model for Sensor Measurements
		2.1 Governing Equation
		2.2 Parametric Discrete Problem
	3 Techniques of Reduced-Order Modeling
		3.1 Proper Orthogonal Decomposition
		3.2 Reduced-Order Solutions
	4 Automatic Anomaly Detection with Unbalanced Datasets
		4.1 Local Semi-supervised Method
		4.2 Damage-Sensitive Features
	5 Finding Optimal Sensor Locations Using Gaussian Processes
		5.1 (Sparse) Gaussian Process Regression
		5.2 Variational Approximation for Systematic Sensor Placement
	6 Numerical Example
	7 Conclusion
	References
Unsupervised Data-Driven Methods for Damage Identification in Discontinuous Media
	1 Introduction
		1.1 Motivation and Background
		1.2 Prior Work
		1.3 Research Aim and Scope
	2 Methods
		2.1 Supervised Gaussian Process Regression
		2.2 Unsupervised Manifold Learning
		2.3 Experiments
	3 Results and Discussion
		3.1 Experimental Results
		3.2 Demonstration with Synthetic Structure
		3.3 Sensitivity to Segmentation
		3.4 Sensitivity to Noise
		3.5 Sensitivity to Library Size
		3.6 Manifold Learning
	4 Conclusions and Future Directions
		4.1 Summary
		4.2 Future Directions
	References
Applications of Deep Learning in Intelligent Construction
	1 Introduction
	2 Related Deep Learning Algorithms
	3 Structural Safety Monitoring
		3.1 Bolted Joints
		3.2 Structural Displacement
		3.3 Structural Surface Quality
	4 Worker Safety Management
	5 Construction Machinery Management
	6 Conclusion
	References
Integrated SHM Systems: Damage Detection Through Unsupervised Learning and Data Fusion
	1 Introduction
	2 Feature Extraction
	3 Data Normalization, Cleansing and Fusion
		3.1 Statistical Models for Data Normalization
		3.2 Data Cleansing and Residual Analysis
		3.3 Data Fusion
	4 Damage Detection Using Control Charts
	5 Conclusions and Final Remarks
	References
Environmental Influence on Modal Parameters: Linear and Nonlinear Methods for Its Compensation in the Context of Structural Health Monitoring
	1 Introduction
	2 Methods for EOV Influence Compensation
		2.1 Multiple Linear Regression
		2.2 Principal Component Analysis
		2.3 Second Order Blind Identification
		2.4 Kernel PCA
	3 Applications
		3.1 Pre-stressed Steel Cable
		3.2 Hospital’s Buildings
	4 Conclusions
	References
Vibration-Based Damage Feature for Long-Term Structural Health Monitoring Under Realistic Environmental and Operational Variability
	1 Introduction
	2 Case Study Description
	3 Methodology
		3.1 Pre-processing and Model Order Selection
		3.2 Damage Feature and Outlier Detection
	4 Results and Discussion
	5 Conclusions
	Bibliography
On Explicit and Implicit Procedures to Mitigate Environmental and Operational Variabilities in Data-Driven Structural Health Monitoring
	1 Introduction
	2 Problem Statement
		2.1 Explicit Procedures
		2.2 Implicit Procedures
	3 Multivariate Nonlinear Approach as an Explicit Procedure
		3.1 Main Definition
		3.2 Outlier Analysis
	4 Metric Learning Approach as an Implicit Procedure
		4.1 Main Definition
		4.2 Outlier Analysis
	5 Experimental Campaign of Wind Turbine Blade Monitoring
		5.1 The Test Rig, Data Collection and Artificial Damage
	6 Results and Discussions
		6.1 Application Example with an Explicit Procedure
		6.2 Application Example with an Implicit Procedure
	7 Comprehensive Discussion and Best Practices
	8 Conclusions
	References
Explainable Artificial Intelligence to Advance Structural Health Monitoring
	1 Introduction
	2 Machine Learning Algorithms for Structural Health Monitoring
		2.1 Artificial Intelligence and Machine Learning
		2.2 Machine Learning in Structural Health Monitoring
		2.3 Applications of Machine Learning Algorithms in Structural Health Monitoring
	3 Design of a Conceptual XAI Framework
		3.1 Overview and Existing XAI Approaches
		3.2 Conceptual XAI Framework
	4 Summary, Conclusions, and Future Work
	References
Physics-Informed Machine Learning for Structural Health Monitoring
	1 Introduction
	2 Grey-Box Models, Overview and Literature
		2.1 Grey-Box Models for SHM
	3 Be More Bayes
		3.1 Prior Mean Functions—Residual Modelling
		3.2 Physics-Derived Covariance Functions
	4 Constrained Gaussian Processes
	5 Gaussian Processes in a State-Space Approach
	6 Conclusions
	7 Gaussian Process Regression
	References
Interpretable Machine Learning for Function Approximation in Structural Health Monitoring
	1 Introduction
		1.1 Function Approximation as Domain Problem
		1.2 The Need for IML
		1.3 Overview of How We Address the Need
		1.4 Novelty and Structure of This Chapter
	2 Modeling Capability of Sigmoidal Neural Networks
	3 Procedure for IML
	4 Case Studies
	5 Justifications Beyond IML
	6 Future Work
	7 Summary
	References
Partially Supervised Learning for Data-Driven Structural Health Monitoring
	1 Limited Labels in Data-driven Performance and Health Monitoring
	2 Inspection Management: Active Learning
		2.1 Simulated Data
		2.2 Query Schemes: Uncertainty Sampling
		2.3 Query Schemes: Information-Theoretic Active Learning
		2.4 Application Examples
		2.5 Query Schemes: Value of Information
	3 Combining Labelled and Unlabelled Data: Semi-supervised Learning
		3.1 Entropy Minimisation
	4 Transfer Learning: Transferring Labels Between Similar Domains
		4.1 Concepts for Transfer
		4.2 Application Example
	5 A Brief Review of SHM Applications
	6 Concluding Remarks
	References
Population-Based Structural Health Monitoring
	1 Population-Based Structural Health Monitoring
	2 Homogeneous Populations
		2.1 The Concept of a Form
		2.2 Transfer Learning for Homogeneous Populations
	3 Heterogeneous Populations
		3.1 Transfer Learning for Heterogeneous Populations
		3.2 The Problem of Negative Transfer
		3.3 Abstract Representation Framework for Spaces of Structures
	4 Conclusions and Future Directions
	References
Machine Learning-Based Structural Damage Identification Within Three-Dimensional Point Clouds
	1 Introduction
	2 Background
	3 Dataset
	4 Data Classification Methodology
		4.1 Voxel Transformation
		4.2 3DFCN Model
		4.3 K-Means Clustering Algorithm
		4.4 3DCNN Model
	5 Results and Discussion
		5.1 3DFCN Model Training and Testing
		5.2 3DCNN Model Training and Testing
	6 Conclusions
	References
New Sensor Nodes, Cloud, and Data Analytics: Case Studies on Large Scale SHM Systems
	1 Introduction
	2 System Design
		2.1 Hardware
		2.2 Software
		2.3 Cloud
		2.4 Structural Diagnostics and Health Evaluation
	3 A Case Study on a Large Scale SHM System
		3.1 Long-Term Damage Detection Strategies on Bridge External Tendons
		3.2 Approach 1: Damage Detection Strategies and Results
		3.3 Approach 2: An Unsupervised Method with Auto Regressive Models and PCA
	4 Conclusions
	References