A CNN-Based Crack Analysis Model for Metal Specimens

In the field of engineering, the combination of deep learning and traditional simulation has become a research hotspot. In healthy structure monitoring problems, strain sensors are spaced and spatially arranged to collect strains. However, methods such as strain gauges can only capture single-point strains, and in many cases these more discrete strain information is insufficient to determine the damage of the target. In this paper, we combine the strain field with deep learning methods to propose a novel damage detection method based on strain image inversion to perform crack information of workpieces. The method defines the type and location of cracks on the specimen as labels, a large number of simulated strain images as datasets, and uses CNN networks to train the model. The results show that the obtained crack inversion model for metal specimens can achieve 93.99% accuracy.