Experimental Study on Monitoring Damage Progression of Basalt-FRP Reinforced Concrete Slabs Using Acoustic Emission and Machine Learning

• Published in: Sensors (Basel, Switzerland) Vol. 23; no. 20; p. 8356
• Main Authors: Zhang, Tonghao, Mahdi, Mohammad, Issa, Mohsen, Xu, Chenxi, Ozevin, Didem
• Format: Journal Article
• Language: English
• Published: Basel MDPI AG 10.10.2023; MDPI
• Subjects: acoustic emission; Acoustic emission testing; Basalt; basalt-FRP; bridge deck; Cameras; Carbon; Cluster analysis; Concrete; Concrete slabs; Corrosion; Crack initiation; Energy; Failure; Machine learning; Mechanical properties; Methods; Polymers; Reinforced concrete; Sensors; SHAP analysis; Steel; supervised machine learning; Tensile strength; unsupervised machine learning
• ISSN: 1424-8220; 1424-8220
• DOI: 10.3390/s23208356

Basalt fiber-reinforced polymer (BFRP) reinforced concrete is a new alternative to conventional steel-reinforced concrete due to its high tensile strength and corrosion resistance characteristics. However, as BFRP is a brittle material, unexpected failure of concrete structures reinforced with BFRP may occur. In this study, the damage initiation and progression of BFRP-reinforced concrete slabs were monitored using the acoustic emission (AE) method as a structural health monitoring (SHM) solution. Two simply supported slabs were instrumented with an array of AE sensors in addition to a high-resolution camera, strain, and displacement sensors and then loaded until failure. The dominant damage mechanism was concrete cracking due to the over-reinforced design and adequate BFRP bar-concrete bonding. The AE method was evaluated in terms of identifying the damage initiation, progression from tensile to shear cracks, and the evolution of crack width. Unsupervised machine learning was applied to the AE data obtained from the first slab testing to develop the clusters of the damage mechanisms. The cluster results were validated using the k-means supervised learning model applied to the data obtained from the second slab. The accuracy of the K-NN model trained on the first slab was 99.2% in predicting three clusters (tensile crack, shear crack, and noise). Due to the limitation of a single indicator to characterize complex damage properties, a Statistical SHapley Additive exPlanation (SHAP) analysis was conducted to quantify the contribution of each AE feature to crack width. Based on the SHAP analysis, the AE duration had the highest correlation with the crack width. The cumulative duration of the AE sensor near the crack had close to 100% accuracy to track the crack width. It was concluded that the AE sensors positioned at the mid-span of slabs can be used as an effective SHM solution to monitor the initiation of tensile cracks, sudden changes in structural response due to major damage, damage evolution from tensile to shear cracks, and the progression of crack width.