• Structural Monitoring and Maintenance
• /
• 제9권3호
• /
• pp.289-303
• /
• 2022

In this study, the impact of assigned pixel labels on the accuracy of crack image identification of steel structures is examined by using an atrous separable convolution neural network (ASCNN). Firstly, images containing fatigue cracks collected from steel structures are classified into four datasets by assigning different pixel labels based on image features. Secondly, the DeepLab v3+ algorithm is used to determine optimal parameters of the ASCNN model by maximizing the average mean-intersection-over-union (mIoU) metric of the datasets. Thirdly, the ASCNN model is trained for various image sizes and hyper-parameters, such as the learning rule, learning rate, and epoch. The optimal parameters of the ASCNN model are determined based on the average mIoU metric. Finally, the trained ASCNN model is evaluated by using 10% untrained images. The result shows that the ASCNN model can segment cracks and other objects in the captured images with an average mIoU of 0.716.

• Steel and Composite Structures
• /
• 제45권4호
• /
• pp.501-511
• /
• 2022

In this study, a two-dimensional model of the contact problem has been examined using the finite element method (FEM) based software ANSYS and based on the multilayer perceptron (MLP), an artificial neural network (ANN). For this purpose, a functionally graded (FG) half-infinite layer (HIL) with a crack pressed by means of two rigid blocks has been solved using FEM. Mass forces and friction are neglected in the solution. Since the problem is analyzed for the plane state, the thickness along the z-axis direction is taken as a unit. To check the accuracy of the contact problem model the results are compared with a study in the literature. In addition, ANSYS and MLP results are compared using Root Mean Square Error (RMSE) and coefficient of determination (R2), and good agreement is found. Numerical solutions are made by considering different values of external load, the width of blocks, crack depth, and material properties. The stresses on the contact surfaces between the blocks and the FG HIL are examined for these values, and the results are presented. Consequently, it is concluded that the considered non-dimensional quantities have a noteworthy influence on the contact stress distributions, and also, FEM and ANN can be efficient alternative methods to time-consuming analytical solutions if used correctly.

• 한국재료학회지
• /
• 제23권2호
• /
• pp.149-153
• /
• 2013

Delamination crack detection is very important for improving the structural reliability of laminated composite structures. This requires real-time delamination detection technologies. For composite laminates that are reinforced with carbon fiber, an electrical potential method uses carbon fiber for reinforcements and sensors at the same time. The use of carbon fiber for sensors does not need to consider the strength reduction of smart structures induced by imbedding sensors into the structures. With carbon fiber reinforced (CF/) epoxy matrix composites, it had been proved that the delamination crack was detected experimentally. In the present study, therefore, similar experiments were conducted to prove the applicability of the method for delamination crack detection of CF/polyetherethereketone matrix composite laminates. Mode I and mode II delamination tests with artificial cracks were conducted, and three point bending tests without artificial cracks were conducted. This study experimentally proves the applicability of the method for detection of delamination cracks. CF/polyetherethereketone material has strong electric resistance anisotropy. For CF/polyetherethereketone matrix composites, a carbon fiber network is constructed, and the network is broken by propagation of delamination cracks. This causes a change in the electric resistance of CF/polyetherethereketone matrix composites. Using three point bending specimens, delamination cracks generated without artificial initial cracks is proved to be detectable using the electric potential method: This method successfully detected delamination cracks.

• Steel and Composite Structures
• /
• 제50권1호
• /
• pp.77-87
• /
• 2024

In this study, the two-dimensional crack problem was investigated by using the finite element method (FEM)-based ANSYS package program and the artificial neural network (ANN)-based multilayer perceptron (MLP) method. For this purpose, a half-infinite functionally graded (FG) layer with a crack pressed through two rigid blocks was analyzed using FEM and ANN. Mass forces and friction were neglected in the solution. To control the validity of the crack problem model exercised, the acquired results were compared with a study in the literature. In addition, FEM and ANN results were checked using Root Mean Square Error (RMSE) and coefficient of determination (R²), and a well agreement was found. Numerical solutions were made considering different geometric parameters and material properties. The stress intensity factor (SIF) was examined for these values, and the results were presented. Consequently, it is concluded that the considered non-dimensional quantities have a noteworthy influence on the SIF. Also FEM and ANN can be logical alternative methods to time-consuming analytical solutions if used correctly.

• Smart Structures and Systems
• /
• 제29권1호
• /
• pp.195-206
• /
• 2022

Crack detection is essential for inspection of existing structures and crack segmentation based on deep learning is a significant solution. However, datasets are usually one of the key issues. When building a new dataset for deep learning, laborious and time-consuming annotation of a large number of crack images is an obstacle. The aim of this study is to develop an approach that can automatically select a small portion of the most informative crack images from a large pool in order to annotate them, not to label all crack images. An active learning method with difficulty learning mechanism for crack segmentation tasks is proposed. Experiments are carried out on a crack image dataset of a steel box girder, which contains 500 images of 320×320 size for training, 100 for validation, and 190 for testing. In active learning experiments, the 500 images for training are acted as unlabeled image. The acquisition function in our method is compared with traditional acquisition functions, i.e., Query-By-Committee (QBC), Entropy, and Core-set. Further, comparisons are made on four common segmentation networks: U-Net, DeepLabV3, Feature Pyramid Network (FPN), and PSPNet. The results show that when training occurs with 200 (40%) of the most informative crack images that are selected by our method, the four segmentation networks can achieve 92%-95% of the obtained performance when training takes place with 500 (100%) crack images. The acquisition function in our method shows more accurate measurements of informativeness for unlabeled crack images compared to the four traditional acquisition functions at most active learning stages. Our method can select the most informative images for annotation from many unlabeled crack images automatically and accurately. Additionally, the dataset built after selecting 40% of all crack images can support crack segmentation networks that perform more than 92% when all the images are used.

• Smart Structures and Systems
• /
• 제29권1호
• /
• pp.221-235
• /
• 2022

Manual inspection of steel box girders on long span bridges is time-consuming and labor-intensive. The quality of inspection relies on the subjective judgements of the inspectors. This study proposes an automated approach to detect and segment cracks in high-resolution images. An end-to-end cascaded framework is proposed to first detect the existence of cracks using a deep convolutional neural network (CNN) and then segment the crack using a modified U-Net encoder-decoder architecture. A Naïve Bayes data fusion scheme is proposed to reduce the false positives and false negatives effectively. To generate the binary crack mask, first, the original images are divided into 448 × 448 overlapping image patches where these image patches are classified as cracks versus non-cracks using a deep CNN. Next, a modified U-Net is trained from scratch using only the crack patches for segmentation. A customized loss function that consists of binary cross entropy loss and the Dice loss is introduced to enhance the segmentation performance. Additionally, a Naïve Bayes fusion strategy is employed to integrate the crack score maps from different overlapping crack patches and to decide whether a pixel is crack or not. Comprehensive experiments have demonstrated that the proposed approach achieves an 81.71% mean intersection over union (mIoU) score across 5 different training/test splits, which is 7.29% higher than the baseline reference implemented with the original U-Net.

• Journal of Mechanical Science and Technology
• /
• 제17권11호
• /
• pp.1599-1607
• /
• 2003

The characteristics of elastic waves emanated from crack initiation in 6061 aluminum alloy subjected to fatigue loading are investigated through experiments. The objective of the study is to determine the differences in the properties of the signals generated from fatigue test and also to examine if the sources of the waves could be identified from the temporal and spectral characteristics of the acoustic emission (AE) waveforms. The signals are recorded using nonresonant, flat, broadband transducers attached to the surface of the alloy specimens. The time dependence and power spectra of the signals recorded during the tests were examined and classified according to their special features. Six distinct types of signals were observed. The waveforms and their power spectra were found to be dependent on the crack propagation stage and the type of fracture associated with the signals. The potential application of the approach in health monitoring of structural components using a network of surface mounted broadband sensors is discussed.

• 한국해양공학회지
• /
• 제10권1호
• /
• pp.65-74
• /
• 1996

Up to now, the existing crack growth modelling has used a mathematical approximation but an assumed function have a great influence on this method. Especially, crack growth behavior that shows very strong nonlinearity needed complicated function which has difficulty in setting parameter of it. The main characteristics of neural network modelling to engineering field are simple calculations and absence of assumed function. In this paper, after discussing learning and generalization of neural networks, we performed crack growth modelling on the basis of above learning algorithms. J'-da/dt relation predicted by neural networks shows that test condition with unlearned data is simulated well within estimated mean error(5%).

• 비파괴검사학회지
• /
• 제21권4호
• /
• pp.398-405
• /
• 2001

리벳이음은 응력집중 등에 의한 피로균열 발생 가능성이 높은 구조특성을 가지므로 구조물의 안전성 확보를 위하여 비파괴 평가 방법에 의하여 리벳홀 주위 균열에 대한 평가가 요구되고 있다. 유도 초음파의 일종인 Lamb파는 판형상의 구조물의 비파괴 평가에 적합하며, 신경회로망은 비파괴 평가 기술에서 결합의 크기 및 종류 인식에 관하여 가장 효율적인 기법으로 많은 연구자들에 의해 적용되어 왔다. 본 연구에서는 항공기의 스킨재료로 적용되는 A12024-T3판재에 대하여 유도초음파의 일종인 판파를 적용하여 리벳홀 주위 균열 신호를 검출하였으며, 또한 리벳홀 주위 균열의 크기 평가를 위하여 백프로퍼게이션 알고리즘을 적용한 신경회로망을 적용하였다. 이때, 초음파 트랜스듀서와 시험편 사이의 불균일 접촉에 의한 오차를 줄이기 위하여 초음파 파형에서 시간 및 주파수 성분의 특성을 추출하여 신경회로망에 적용하였다. 그리고 이들 판파신호에서 추출한 시간 및 주파수 성분의 특성은 균열 크기 결정에 유용하게 적용될 수 있음을 증명하였다.

• 한국공작기계학회논문집
• /
• 제12권2호
• /
• pp.65-70
• /
• 2003

In this study, researchers developed the estimative algorithm for artificial defect in semiconductor packages and performed it by pattern recognition technology. For this purpose, the estimative algorithm was included that researchers made software with MATLAB. The software consists of some procedures including ultrasonic image acquisition, equalization filtering, Self-Organizing Map and Backpropagation Neural Network. Self-organizing Map and Backpropagation Neural Network are belong to methods of Neural Networks. And the pattern recognition technology has applied to classify three kinds of detective patterns in semiconductor packages : Crack, Delamination and Normal. According to the results, we were confirmed that estimative algerian was provided the recognition rates of 75.7% (for Crack) and 83.4% (for Delamination) and 87.2 % (for Normal).