14.3 Pattern Recognition and Anomaly Detection in SHM Data

Pattern recognition in SHM data analysis helps identify meaningful patterns and trends in structural health monitoring datasets. It extracts features that characterize the state of monitored structures and classifies data into predefined categories based on learned patterns.

Anomaly detection focuses on identifying data points that deviate significantly from the norm. This helps detect unusual behavior in monitored structures, such as cracks or corrosion, and distinguishes between normal conditions and potential damage or degradation.

Pattern Recognition in SHM Data Analysis

Pattern recognition and anomaly detection

Top images from around the web for Pattern recognition and anomaly detection

• 

  Frontiers | Vision-Based Bridge Deformation Monitoring View original

  Is this image relevant?

• 

  Frontiers | Structural Health Monitoring of a Cable-Stayed Bridge Using Regularly Conducted ... View original

  Is this image relevant?

• 

  Frontiers | Diagnostic Testing of a Vertical Lift Truss Bridge for Model Verification and ... View original

  Is this image relevant?

• 

  Frontiers | Vision-Based Bridge Deformation Monitoring View original

  Is this image relevant?

• 

  Frontiers | Structural Health Monitoring of a Cable-Stayed Bridge Using Regularly Conducted ... View original

  Is this image relevant?

1 of 3

Top images from around the web for Pattern recognition and anomaly detection

• 

  Frontiers | Vision-Based Bridge Deformation Monitoring View original

  Is this image relevant?

• 

  Frontiers | Structural Health Monitoring of a Cable-Stayed Bridge Using Regularly Conducted ... View original

  Is this image relevant?

• 

  Frontiers | Diagnostic Testing of a Vertical Lift Truss Bridge for Model Verification and ... View original

  Is this image relevant?

• 

  Frontiers | Vision-Based Bridge Deformation Monitoring View original

  Is this image relevant?

• 

  Frontiers | Structural Health Monitoring of a Cable-Stayed Bridge Using Regularly Conducted ... View original

  Is this image relevant?

1 of 3

• Pattern recognition in SHM data analysis involves identifying meaningful patterns, trends, or relationships within SHM datasets to extract features that characterize the state or behavior of the monitored structure (bridges, buildings, aircraft) and classify data into predefined categories or classes based on learned patterns
• Anomaly detection in SHM data analysis focuses on identifying data points or patterns that deviate significantly from the norm to detect unusual or unexpected behavior in the monitored structure (cracks, corrosion, fatigue) and distinguish between normal operational conditions and potential damage or degradation

Algorithms for pattern recognition

• Supervised learning algorithms for pattern recognition require labeled training data with known class or category assignments, train the model using labeled data to learn the mapping between input features and output classes, and use the trained model to predict the class or category of new, unseen SHM data points
  • Support Vector Machines (SVM)
  • Decision Trees
  • Random Forests
  • k-Nearest Neighbors (k-NN)
• Unsupervised learning algorithms for pattern recognition do not require labeled training data, discover inherent structures, groupings, or patterns within the SHM dataset, and group similar data points together based on their intrinsic characteristics or similarities
  • k-means clustering
  • Hierarchical clustering
  • Principal Component Analysis (PCA)

Anomaly Detection in SHM Data Streams

Methods for anomaly detection

• Statistical anomaly detection methods assume that normal data follows a specific statistical distribution (Gaussian), identify data points that deviate significantly from the expected distribution using techniques such as z-score, Mahalanobis distance, or Gaussian Mixture Models (GMM)
• Machine learning-based anomaly detection methods learn the normal behavior of the monitored structure from historical data, identify data points that differ significantly from the learned normal patterns using algorithms like One-Class SVM, Isolation Forest, or Autoencoders
• Threshold-based anomaly detection methods define thresholds or limits for specific SHM parameters or features, flag data points that exceed the predefined thresholds as anomalies, and require domain knowledge and expert input to set appropriate thresholds

Evaluation of detection techniques

• Performance metrics for pattern recognition include:
  1. Accuracy: Percentage of correctly classified instances
  2. Precision: Proportion of true positive predictions among all positive predictions
  3. Recall (Sensitivity): Proportion of true positive predictions among all actual positive instances
  4. F1-score: Harmonic mean of precision and recall, providing a balanced measure of performance
• Performance metrics for anomaly detection include:
  • True Positive Rate (TPR): Proportion of correctly identified anomalies among all actual anomalies
  • False Positive Rate (FPR): Proportion of normal instances incorrectly identified as anomalies
  • Area Under the Receiver Operating Characteristic (ROC) Curve (AUC-ROC): Measures the trade-off between TPR and FPR at different threshold settings
• Validation strategies ensure the robustness and generalization ability of the developed models:
  • Hold-out validation: Split the dataset into separate training and testing sets
  • k-fold cross-validation: Divide the dataset into $k$ equal-sized folds, use each fold as a testing set while training on the remaining folds
  • Leave-one-out cross-validation: Use each single instance as a testing set while training on the remaining instances