Bayesian modeling of finger joints’ tensile and bending properties considering censored data

Abstract

Finger joints (FJs) are longitudinal interlocking connections between timber boards commonly used in engineered wood products such as glued laminated timber beams. The mechanical properties of FJs, particularly tensile strength, are vital for the performance of such products; however, practical constraints often lead to testing FJs primarily in bending rather than tension. Therefore, it is crucial to investigate the relationship between tensile and bending strengths. The present study investigates this relationship using paired FJs. It further evaluates the predictability of FJ tensile and bending properties based on the dynamic modulus of elasticity of the connected timber boards. Tensile and bending tests were conducted, and censored data resulting from failures outside the FJs were handled using a hierarchical Bayesian linear regression method. The study proposes models to estimate FJ tensile strength from bending strength and to predict tensile and bending properties based on the dynamic modulus of elasticity of the connected timber boards. Model predictions are compared with experimental outcomes, indicating the effective integration of censored data. Results further show that excluding censored data underestimates FJ tensile and bending strength. The proposed Bayesian modeling procedure offers a means to connect failure modes with data types and appropriately address censored data. Such models can help to enhance quality control/production efficiency and the structural reliability of engineered wood products.