Evaluation of Early Concrete Damage Caused by Chloride-Induced Steel Corrosion Using a Deep Learning Approach Based...

Area of Science:

• Materials Science and Engineering
• Civil Engineering
• Artificial Intelligence in Engineering

Background:

• Corrosion of steel reinforcement (rebar) is a major cause of concrete degradation, impacting structural integrity.
• Early detection of corrosion-induced damage is crucial for timely maintenance and preventing catastrophic failures.
• Conventional methods for assessing concrete health often lack sensitivity for early-stage damage.

Purpose of the Study:

• To investigate the feasibility of using ultrasonic pulse wave (UPW) measurements for early detection of concrete damage caused by rebar corrosion.
• To develop and evaluate deep learning models, specifically recurrent neural networks (RNNs), for classifying corrosion-induced damage.
• To compare the performance of RNN models against traditional ultrasonic testing parameters.

Main Methods:

• Concrete cube specimens with embedded steel rebars were subjected to accelerated corrosion using an impressed current technique in a 3% NaCl solution.
• Ultrasonic pulse waves were measured using 50 kHz P-wave transducers in a through-transmission setup before and after corrosion.
• Three RNN models (LSTM, GRU, BiLSTM) were trained for damage classification and compared with ultrasonic pulse velocity and signal consistency.

Main Results:

• Deep learning models, particularly RNNs, demonstrated superior performance in detecting corrosion-induced concrete damage compared to conventional ultrasonic parameters.
• The bidirectional long short-term memory (BiLSTM) RNN model achieved the highest accuracy (74%) and Cohen's kappa coefficient (0.48).
• The study successfully classified concrete damage levels based on UPW data using deep learning.

Conclusions:

• Ultrasonic pulse wave analysis combined with deep learning, specifically RNNs, shows significant potential for the early and accurate detection of concrete damage due to steel corrosion.
• The BiLSTM model is identified as a promising approach for enhanced structural health monitoring systems.
• This research paves the way for non-destructive evaluation techniques in civil infrastructure maintenance.