Fracture mode identification in cementitious materials using supervised pattern recognition of acoustic emission features

Cracking in concrete as a ubiquitous cementitious material in civil structures has been a worldwide critical issue in the field of engineering. Acoustic emission (AE) has demonstrated promising outcomes in research and laboratory experiments for monitoring these structures that led to plethora of reports, articles and recommendations for concrete structures. Many of these studies focus on cracking mode detection to estimate the significance of damage because in general, shear-like phenomena indicate severe damage and occur after tensile (flexural) cracking. The distinctive signs of the cracking modes are embedded in some AE parameters like the RA-value and average frequency (AF). Signals emitted from shear fracture exhibit higher RA-values with smaller AF than tensile ones. However, there are no universally fixed boundaries for classification of these features due to the parameters like member geometry, material properties sensor location and response. In addition, although AE consists of a random set of data, the role of uncertainty is not fully taken into account in data processing. To overcome these deficiencies, this article proposes a pattern classifier technique titled support vector machines. Small-scale fracture experiments were carried out to impose controlled cracking modes, record AE data for each cracking mode, and evaluate the performance of classifiers. The results show that the classification boundaries for AE features and their associate uncertainties could be successfully estimated. The effect of sensor distance as an imperative parameter in variation of classification boundaries could be quantified. Furthermore, the adequacy of other feature sets (i.e., other than RA and AF) for classification was also examined.