Locating Crack Tip in Mode-I Fracture Tests of Engineered Wood by Acoustic Emission and Machine Learning

Abstract

As an eco-friendly construction material, engineered wood is becoming increasingly popular as it is environmentally friendly, less labor-intensive and more cost-effective compared with concrete and steel. On the other hand, compared with solid wood, layered engineered wood has more consistent mechanical properties, can minimize the influence of flaws like knots and get rid of the limitation of dimensions. Yet there is limited number of research focusing on the fracture behavior of layered engineered wood. This article investigates the fracture behavior of the laminated veneer lumber (LVL) via acoustic emission in mode-I fracture tests. The fracture test is conducted on an LVL sample with the help of six AE sensors to collect the AE waveforms. The difference of time of arrival is used to locate the crack tips. Since wood is highly heterogeneous, instead of using the wave velocity for AE source localization, this work leverages the benefits of probabilistic machine learning methods. Moreover, to capture the outliers more easily and being less sensitive to noise, a new type of probabilistic machine learning method, i.e., student-t process regression (STPR), is firstly proposed to address the AE source localization in mode-I fracture tests. As a comparison, the Gaussian process regression (GPR) is also implemented. Because not all AE events are related with the crack tip propagation, the Gaussian mixture models is implemented to identify AE events relevant with the crack tip. The results suggest STPR can outperform GPR both in pencil lead break test and in the mode-I fracture tests, only with the cost of having one additional parameter, i.e., the degree of freedom.