Using Deep Learning Techniques for Sandwich Panels with Truss Core Damage Detection

[摘要] In the previous study, the accuracy of damage detection for the sandwich structures with truss core (SPTC) was affected by the selected damage index, Other than this, human subjective judgment could also not directly determine the degree and the location of damage for SPTC. In this paper, the deep learning method is applied to identify the damage for SPTC, and the dataset of the training deep learning model is obtained based on the dynamic method. This paper adopts to the Caffe, which is a deep learning open source framework, object detection model Faster R-CNN is utilized to study the lattice sandwich plate. The damage data set, the optimal hyperparameters for training the deep learning model, and the optimal ratios of the test set and training set for damage dataset are also studied. It is difficult to detect the damage of SPTC applying to the deep learning algorithms, so the good results cannot be gotten. In this paper, the method of Faster R-CNN has used extracts the deep features of the defective target by ZF that is a kind of Convolutional Neural Network (CNN), the method effectively solves the problem that the traditional algorithm cannot effectively detect the damage. As to the damage of SPTC that the traditional algorithms could also identify, the deep learning algorithm is excelled, the experimental mean average precision(mAP) can be raised to 90%. At the same time, the deep learning method can effectively identify locations and size of the damage in SPTC, the method is proven that the accuracy is higher and the speed is faster for damage detection. In the future, a real-time damage monitoring system is possible, and the theory is worth exploring further.

[发布机构] Key Laboratory for Mechanics in Fluid Solid Coupling Systems, Institute of Mechanics, Chinese Academy of Sciences, Beijing, China^1;School of Engineering Science, University of Chinese Academy of Sciences, Beijing, China^2

[效力级别] 计算机科学

[关键词] Convolutional neural network;damage dection;Damage monitoring;Hyperparameters;Learning methods;Learning techniques;Open source frameworks;Truss core