Damage Localization and Sensor Layout Optimization for In-Service Reinforced Concrete Columns Using Deep Learning and...

Area of Science:

• Structural Engineering
• Materials Science
• Artificial Intelligence

Background:

• Reinforced concrete (RC) columns are critical load-bearing elements in structures.
• Regular health assessment is vital for extending the service life and performance of structures.
• In-service RC column health monitoring presents significant challenges.

Purpose of the Study:

• To detect and locate acoustic emission (AE) sources in in-service RC columns.
• To determine the optimal sensor layout for effective health monitoring.
• To evaluate the performance of different deep learning models for AE source localization.

Main Methods:

• Implementation of deep learning algorithms combined with acoustic emission (AE) technology.
• Data cleaning using k-means clustering and voting selection to improve data quality.
• Comparison of Back Propagation (BP), Radial Basis Function (RBF), and Support Vector Regression (SVR) models for AE source localization.
• Evaluation of different sensor layout schemes (linear, hybrid linear-volumetric).

Main Results:

• The k-means clustering and voting selection data cleaning method significantly improved data quality.
• The BP model demonstrated superior AE source localization performance compared to RBF and SVR models, showing reduced Mean Absolute Error (MAE) and Root Mean Square Error (RMSE), and increased R-squared (R²).
• Linear sensor arrangements are effective for shallow concrete matrix damage, while hybrid linear-volumetric arrangements are better for deep matrix damage.

Conclusions:

• The Back Propagation (BP) deep learning model is highly effective for AE source localization in in-service RC columns.
• Optimized sensor layout strategies, particularly the hybrid linear-volumetric scheme, enhance the accuracy of damage detection in both shallow and deep concrete matrices.
• This combined approach offers significant application value for the structural health monitoring of in-service RC columns.