Identification of Grout Sleeve Joint Defect in Prefabricated Structures Using Deep Learning

Abstract

A grout sleeve connection is a typical kind of joint in prefabricated structures. However, for construction and manufacturing reasons, defects in this kind of joint are usually inevitable. The joint quality of a prefabricated structure has a significant influence on its overall performance and can lead to structural failure. Due to the complexity of various types of materials used in grout sleeve connections, traditional non-destructive testing methods, such as Acoustic Emission (AE), Ultrasonic Testing (UT), Guided Wave Testing (GW), are facing great challenges. The recent development of deep learning technology provides a new opportunity to solve this problem. Deep learning can learn the inherent rules and abstract hierarchies of sample data, and it has a powerful ability to extract the intrinsic features of training data in complex classification tasks. This paper illustrates a deep learning framework for the identification of joint defects in prefabricated structures. In this method, defect features are extracted from the acceleration time history response of a prefabricated structure using a convolutional neural network. The proposed method is validated by vibration experiments on a half-scaled, two-floor prefabricated frame structure with column rebars spliced by different defective grout sleeves.

Deep transfer learning for connection defect identification in prefabricated structures

“…The sizes of the C1, C2, and C3 filters and SS1, SS2, and SS3 pooling kernels depend on the size of input sample. The activation function is between the convolution and mean pooling layers, using the Tanh function, 28…”

“…As shown in Figure 16, there was no overfitting during training. 28 According to Figure 17, a CNN trained with sufficient training samples from the source domain can accurately identify defect types.…”

“…The activation function is between the convolution and mean pooling layers, using the Tanh function, 28 y Tanh = e x À e Àx e x + e Àx ; ð1Þ…”

“…As shown in Figure 16, there was no overfitting during training. 28 According to Figure 17, a 1 and 3 and between types 2 and 4. Hence, the impact of through hole defects in the bed mortar layer on the structural acceleration response was less than that of incomplete fill of grout sleeve defects.…”

Structural health monitoring of a historic building during uplifting process: system design and data analysis

“…Neural networks have strong fitting and generalization abilities; they can acquire knowledge by training with adequate datasets. [40][41][42] Therefore, a reliable structural state can be obtained through the neural networks and reflected in the data of the hidden layer, realizing the processing of multi-source monitoring data and the assessment of structural safety levels. 43 Before using neural network training, it is necessary to select the appropriate monitoring data and preprocess the data.…”