Study on stress intensity factor identification of a multi-cracked elastomer based on digital speckle pattern combined with transfer learning

• An image generation algorithm based on affine transformation and sub-pixel shift is proposed to generate simulated speckle images. • A deep convolutional neural network (DCNN) was established and trained to identify fracture parameters of multi-cracked elastomers. • A tensile test of PMMA plates with multiple interacting cracks is conducted to verify the efficiency and accuracy of the method. In multi-cracked elastomers, the fracture parameter at each crack tip is significantly influenced by neighboring cracks, complicating their identification using conventional optical methods such as digital image correlation (DIC). This study presents a novel approach that combines digital speckle pattern reconstruction with transfer learning to accurately and efficiently identify fracture parameters of multi-cracked elastomers. Finite element models were established to determine numerical displacement fields and calculate mixed-mode stress intensity factors (SIFs) of two-dimensional (2D) multi-cracked plates under various loading conditions. Synthetic deformed speckle images were generated from reference images by applying numerical displacements. A deep convolutional neural network (DCNN) with transfer learning was then trained using a dataset composed of numerous pairs of reference and deformed speckle images as inputs and the corresponding mixed-mode SIFs at different crack tips as outputs. Tensile and four-point bending tests were performed on 2D polymethyl methacrylate (PMMA) plates with randomly distributed cracks to evaluate the method’s practical performance. Results demonstrated that the proposed method combining digital speckle pattern reconstruction with deep learning can be employed to identify the mixed-mode fracture parameters of multi-cracked elastomer with high accuracy and efficiency.