Evaluating the bonding performance of wood/bamboo composites through deep learning: Delamination percentage

Scientifically, efficiently and accurately evaluating the adhesive bonding performance is a critical process in ensuring the quality of composite materials. The delamination rate (DP), as one of the key evaluation indicators of bonding performance, is mainly measured manually, which cannot meet the demands of industrial production. To address this issue, deep learning (DL) was employed for the detection of DP, and nine common application scenarios were simulated by combining epoxy resin, phenolic resin and methylene diphenyl diisocyanate resin in combination with bamboo and wood. Besides, an optimization approach integrating laser triangulation was adopted to improve the measurement accuracy of DL in detecting DP. The results showed that DL can served as a feasible approach for measuring DP across all scenarios. However, its measurement accuracy is substantially limited, with the maximum absolute error and relative error reaching up to 79.2 and 100 %, respectively. This limitation was primarily attributed to the variations among application scenarios, such as the inherent color of adhesive, sample, and defects, which constitute the key factors influencing measurement precision. Nevertheless, following the optimization through laser triangulation, the measurement accuracy of DL had been improved by a factor of 9.8, and with an average relative error and absolute error below 20 %, indicating that it was sufficiently accurate for industrial production. The automated measurement technology employed in this study has considerably enhanced the efficiency of measuring the DP index. This advancement provided robust theoretical and technical support for the integration of quality inspection and green processing in the field of composite materials.