Machine Learning Meets Compressed Sensing in Vibration-Based Monitoring

Area of Science:

• Engineering
• Computer Science
• Data Science

Background:

• Structural Health Monitoring (SHM) benefits from Artificial Intelligence (AI) but faces challenges in data management and network congestion.
• Continuous data collection in SHM strains memory resources and necessitates complex communication protocols.

Purpose of the Study:

• To develop a comprehensive framework for vibration-based structural diagnostics.
• To address data management challenges in SHM through compression and efficient AI models.
• To evaluate the impact of low-cost sensors and environmental factors on damage detection accuracy.

Main Methods:

• Implementation of data compression techniques to reduce data dimensionality.
• Development of neural network models for binary classification to detect structural damage.
• Inclusion of environmental factors and simulated MEMS sensor noise in the analysis.

Main Results:

• Data compression effectively reduced data management overhead.
• Neural network models achieved high classification scores (accuracy > 96%, precision > 95%) for damage detection.
• The framework demonstrated robustness against noise from low-cost sensors and environmental variations.

Conclusions:

• A combined approach of data compression, optimized machine learning, and environmental data enhances SHM.
• The proposed framework offers an efficient and accurate solution for vibration-based structural diagnostics.
• The study validates the framework's effectiveness using real-world data from the Z24 bridge case study.