An ensemble learning approach to condition assessment of dissipative CLT connections based on piezoceramic sensor data

Abstract

Novel dissipative connections of cross-laminated timber (CLT) structures, showing great ductility and energy dissipation capability in addition to meeting the requirement of strength and stiffness, have gained increasing attraction in CLT shear walls. As important elements of CLT structures, dissipative connections function as the main part to resist seismic loading. It often displays different mechanical behaviors and suffers stiffness degradation when subjected to external loads. Variations of mechanical behaviors would further affect structural performance during service. The traditional method of analyzing mechanical behaviors of CLT connections relies on destructive tests, which are time-consuming and irreversible. Thus, using the non-destructive method to identify the mechanical behavior of dissipative CLT connections is of great significance. An ensemble learning-aided prediction algorithm combined with piezoceramic-enabled sensing technology is proposed in this study to identify mechanical behaviors of dissipative CLT connections under external loading. The main contribution of this study is to build a bridge between the damage-sensitive feature extracted from non-destructive sensing signals and mechanical behaviors obtained by the destructive loading test through the proposed method, which can avoid the requirement of time-consuming destructive tests and the baseline signal in traditional piezoceramic-enabled sensing method. An energy dissipative hold-down connection model is established in numerical simulations to investigate stress wave propagation mechanisms via different loading conditions. In experimental studies, a reusable piezoceramic-based sensor is firstly designed, and its performance is validated through a preliminary test. Then, a loading test was conducted on a CLT joint with a dissipative hold-down connector to further verify the numerical findings and generate an intact-to-failure piezoceramic-based sensing signal dataset through all loading cases. Consequently, fuzzy entropy is extracted from the signal dataset and fed to an ensemble learning classifier to predict loading behaviors. Prediction results imply the great potential of the proposed method for predicting dissipative CLT joint life-cycle performance in field applications.