Abstract
Currently, the digital construction of temperature fields based on existing data (temperature field reconstruction) primarily relies on "interpolation methods" and "thermal parameter inversion methods." Although continuous advancements have been made in these approaches, the accuracy of digitally reconstructed temperature fields still requires further improvement. This paper proposes a novel methodology that establishes a finite element method-based temperature field computation framework, incorporating automated real-time corrections using measured temperature data during the calculation process, thereby achieving precise numerical reconstruction of temperature fields. Additionally, a water-cooling pipe calculation method is presented that simultaneously satisfies three critical requirements: simplified mesh discretization, accurate computation of water temperature along the flow path, and precise grid-based calculation of temperature fields surrounding cooling pipes. By employing this method, accurate temperature field reconstruction can be achieved using limited temperature measurement points without requiring thermal parameter inversion, while the inversion accuracy improves progressively with increased measurement data. This methodology lays the foundation for automated temperature field reconstruction technology based on measurement point data.
Key Words
automated; finite element method; reconstruction; temperature field
Address
State Key Laboratory of Simulation and Regulation of Water Cycle in River Basin, China Institute of Water Resources and Hydropower Research, Beijing 100038, P.R. China.
Abstract
Bridge dynamic displacement reconstruction methods based on neural networks usually use single-input neural networks, and most of the hyperparameters are determined by experience, which seriously affect the reconstruction accuracy. In this paper, a reconstruction method for bridge displacement response induced by moving load is proposed by using a small number of sensors and a triple-input IPSO-BiLSTM network. Firstly, the input strain and acceleration data are normalized in advance for data fusion. Secondly, IPSO-BILSTM network model with three-time sequence responses as input is constructed, and IPSO algorithm is used to optimize the network hyperparameters. Finally, three-time sequence responses are input into the trained iterative particle swarm optimization (IPSO)-Bidirectional LSTM (IPSO-BiLSTM) neural network to reconstruct the bridge displacement response. The proposed IPSO-BiLSTM network realizes the data fusion of three-time sequence responses and automatically establishes the relationship between input response and output displacement. Numerical examples indicate that the reconstruction accuracy is sensitive to road roughness and measurement noise. Experimental studies reveal that the reconstruction accuracy is insensitive to vehicle velocity and weigh.
Address
(1) Wen-Yu He, Ao Gao, Yi-Fan Li:
Hefei University of Technology, Hefei, Anhui Province, 230009, China;
(2) Wen-Yu He:
Anhui Province Road and Bridge Inspection Engineering Research Center, Hefei, Anhui 230009, China;
(3) Dong-Yang Hu:
Kunming Survey, Design and Research Institute Co., Ltd. of CREEC, Kunming 650200, China.
Abstract
Defect detection plays a crucial role in ensuring the safety and longevity of structures, with defect region classification particularly beneficial for focusing efforts on potential defect areas. Traditional deep convolutional neural networks (DCNNs) based defect classification networks still have a high number of parameters and computational demands, making them unsuitable for embedded systems. This paper proposes the Adaptive Prior Activation-Based Binary Information Enhancement Network (AOIE-Net), which significantly reduces computational requirements by binarizing weights and activations. Specifically designed for steel defect detection, AOIE-Net optimizes the binary quantization process and enhances feature representation to improve the performance of BNNs in steel defect detection tasks. AOIE-Net introduces a Dual Batch Normalization-based Information Enhancement Block (DBN-IEB) and an Adaptive Binary Activation Independent Optimization (ABA-IO) method to reduce computational complexity while boosting classification accuracy. Experimental results demonstrate that AOIE-Net outperforms state-of-the-art binary neural network models on CIFAR-10, ImageNet, and the NEU-CLS steel defect dataset, achieving classification accuracy of 90.6%, 72.1%, and 99.4%, respectively. The proposed method offers an efficient, low-complexity solution for real-time defect classification in large-scale structural inspections and holds significant potential for practical applications.
Key Words
binary neural network; deep learning; enhanced binary information; image classification; steels defect
Address
College of Physics and Information Engineering, Fuzhou University, Fuzhou 350116, China.
Abstract
Crack detection is an important measure in the field of structural health monitoring. However, visual crack detection is labor-intensive, time-consuming, inefficient, and expensive. Although image-based detection and processing provides an efficient way for structural crack detection, its accuracy depends on image quality. For engineering structures, especially bridges, the change of light conditions and the difference of surface characteristics of structural components pose a major challenge to traditional crack detection methods. In this paper, a novel crack detection method based on convolutional neural networks is proposed. The development of this method is divided into the following stages. The initial automated crack classification is carried out by using MobileNetV3, and then the improved DeepLabv3+ network is used to segment the classified crack image semantically accurately. Finally, the real crack image is used for verification. To verify the proposed method, several conventional deep learning networks are trained and compared. The improved DeepLabV3+ integrates MobileNetV3 as its feature extraction backbone and incorporates the convolutional block attention module, which achieves 87.79% average intersection and 93.87% average pixel accuracy on public and real data sets. Compared with traditional models such as VGG16, the proposed method shortens the training time by more than 80% while maintaining high detection accuracy. In addition, the compact parameter configuration and moderate model size make it particularly suitable for deployment on mobile detection devices.
Key Words
attention mechanism; crack detection; improved DeepLabv3+; lightweight deep learning; semantic segmentation
Address
(1) Dong Yang:
Earthquake Engineering Research & Test Center (EERTC), Guangzhou University, Guangzhou, China;
(2) Dong Yang:
Key Laboratory of Earthquake Resistance, Earthquake Mitigation and Structural Safety, Ministry of Education, Guangzhou University, Guangzhou, China;
(3) Yuan-Yuan Cai, En-Dian Xu:
Department of Civil Engineering, Hefei University of Technology, Hefei, Anhui Province, China;
(4) Jing Zhang:
Department of Mechanics and Construction Engineering, Jinan University, Guangzhou, China;
(5) Ye Yuan, Yan-Jia Wang:
Department of Civil Engineering, The University of Hong Kong, Pokfulam Road, Hong Kong, China.
