Chloride-Induced Damage Monitoring of Reinforced Concrete Using Ultrasonic Pulse Wave-Based Machine Learning

This research aims to utilize the machine learning approach for detecting the early stages of concrete corrosion within reinforced concrete based on ultrasonic pulse wave data. The study was conducted by performing accelerated corrosion on specimens with different rebar sizes and mix design strength, with each combination subjected to three levels of corrosion. The collected ultrasonic pulse data is then preprocessed, which includes smoothing, normalization, resampling, and shortening of the full-length signal to be used as the input for machine learning model training. Machine learning models developed in this study were based on the support vector machine (SVM), K-nearest neighbor (KNN), and subspace-KNN ensemble methods. We found that the machine learning models have a respectable performance for classifying concrete with a corrosion level of less and more than 3%. The SVM has the best-performing accuracy of 71.8%, F1-score of 0.715, and Cohen’s kappa of 0.430.