AI-Based Wood Failure Detection and Shear Strength Prediction in Adhesive Bonding of Cross-Laminated Timber

Abstract

Artificial intelligence (AI) can offer a faster and more consistent alternative to manual assessment of bonding quality in cross-laminated timber (CLT), particularly when thermally modified wood (TMW) introduces additional variability in bondline behavior. The authors developed three models and evaluated them to automate the prediction of two key quality indicators in CLT bonding: Wood Failure Percentage (WFP) and shear strength. A Convolutional Neural Network (CNN) was trained to quantify WFP from bondline images, while an Artificial Neural Network (ANN) and a hybrid CNN–Long Short Term Memory (LSTM) model were developed to predict shear strength using manufacturing parameters such as pressure, wood type, adhesive penetration, and bondline thickness. The CNN showed strong agreement with manual WFP measurements, achieving R² values of 0.92 for unmodified wood and 1.0 for TMW, demonstrating gains in speed and consistency in quality control. For shear strength prediction, the ANN performed best for unmodified CLT (R² = 0.90), whereas the CNN–LSTM model was superior for TMW (R² = 0.97), reflecting its ability to capture nonlinear interactions associated with thermal modification. Attention-based feature analysis indicated that pressure and penetration were the dominant predictors, with their relative importance differing between wood types. Overall, the models show that AI can reliably predict bonding quality in CLT, reducing dependence on manual inspection and offer a practical pathway toward real time, automated quality control in industrial manufacturing.