Machine learning based crack mode classification from unlabeled acoustic emission waveform features

• Published in: Cement and concrete research 2019-07, Vol.121, p.42-57
• Main Authors: Das, Avik Kumar, Suthar, Deepak, Leung, Christopher K.Y.
• Format: Article
• Language: English
• Subjects: Acoustic emission; Algorithms; Artificial intelligence; Audio frequencies; Cement reinforcements; Classification; Clustering; Crack mode classification; Cracking (fracturing); Damage detection; Fiber composites; Fracture mechanics; Gaussian mixture modeling (GMM); Hyperplanes; Machine learning; Reinforced concrete; Reinforcing steels; Smart structures; Steel fiber reinforced concretes; Steel fibers; Strain hardening; Strain hardening cementitious composites (SHCC); Stress waves; Structural damage; Structural health monitoring; Structural safety; Support vector machine (SVM); Support vector machines
• ISSN: 0008-8846; 1873-3948
• DOI: 10.1016/j.cemconres.2019.03.001

As the cracking mode (tensile or shear) of a crack is related to the underlying physical mechanisms, crack mode classification is a very useful method to identify the damage state of a structure for proper maintenance to enhance structural safety and durability. Acoustic Emission (AE) is a passive structural health monitoring technique based on the stress wave generated due to cracking in a structure. A framework has been designed in this study for automated probabilistic classification of the cracks in cementitious components based on the AE signals. With this approach, unlabeled hand designed waveform parameters, i.e. RA values (RA) and Average frequency (AF) are clustered using density dictated unsupervised clustering algorithm. Intersecting clusters in the data were then separated through a hyperplane created using Support Vector Machine (SVM) algorithm. Based on physical insight obtained from labeled data, unlabeled data was classified into events corresponding to different cracking modes. The framework was applied to the analysis of AE data from Steel Fiber Reinforced Concrete (SFRC) beam under bending and Strain Hardening Cementitious Composite (SHCC) samples under direct tension. The cracking modes obtained from the proposed machine learning approach are found to be in good agreement with expectations based on composite theory. With good qualitative prediction, the proposed approach shows promise for the prediction of damage state in structures based on unlabeled data obtained in the field.