Machine Learning-Assisted Improved Anomaly Detection for Structural Health Monitoring.

Area of Science:

• Civil Engineering
• Data Science
• Machine Learning

Background:

• Civil engineering infrastructure, particularly bridges, is vital for global supply chains and economic activity.
• Structural health monitoring (SHM) is crucial for managing infrastructure costs and lifespan.
• SHM data often contains anomalies that require automated classification for accurate condition assessment.

Purpose of the Study:

• To develop an automated, interpretable framework for classifying anomalies in SHM data.
• To accurately evaluate the current condition of infrastructure systems in a timely and cost-effective manner.

Main Methods:

• A recursive decision tree framework was developed for multiclass classification of acceleration data from a real bridge.
• Random forest classifiers were used as decision nodes, recursively invoked with synthetically augmented training data.
• Statistical features, chosen for interpretability, were used to define feature vectors for classifier training.

Main Results:

• The proposed framework achieved 98% accuracy in classifying non-anomalous time-series data.
• The use of interpretable statistical features enhanced the understandability of the classification models.

Conclusions:

• The developed machine learning framework provides an accurate and interpretable method for anomaly detection in bridge SHM.
• This approach can significantly improve the efficiency and reliability of infrastructure health assessment.