Deep convolutional neural network for multi-level non-invasive tunnel lining assessment

In recent years, great attention has focused on the development of automated procedures for infrastructures control. Many efforts have aimed at greater speed and reliability compared to traditional methods of assessing structural conditions. The paper proposes a multi-level strategy, designed and implemented on the basis of periodic structural monitoring oriented to a cost- and time-efficient tunnel control plan. Such strategy leverages the high capacity of convolutional neural networks to identify and classify potential critical situations. In a supervised learning framework, Ground Penetrating Radar (GPR) profiles and the revealed structural phenomena have been used as input and output to train and test such networks. Image-based analysis and integrative investigations involving video-endoscopy, core drilling, jacking and pull-out testing have been exploited to define the structural conditions linked to GPR profiles and to create the database. The degree of detail and accuracy achieved in identifying a structural condition is high. As a result, this strategy appears of value to infrastructure managers who need to reduce the amount and invasiveness of testing, and thus also to reduce the time and costs associated with inspections made by highly specialized technicians.     

基于卷积神经网络的光学遥感图像目标检测

针对光学遥感图像中的目标检测问题,提出了1种基于卷积神经网络模型的算法,对遥感目标检测任务进行端到端的训练和检测,根据输入的光学遥感图像,直接输出目标包围盒的回归结果和置信度。为训练和测试模型,建立了1个包含1万多个飞机、舰船目标以及广泛复杂背景的数据集。所提算法在其测试集上达到了超过90%的准确率和召回率,在GPU上的运行速度也接近实时,体现了算法准确、高效、鲁棒和易于训练的特点。     

利用深度卷积神经网络将耳语转换为正常语音

耳语是一种特殊发音方式,将耳语转换为正常语音是提升耳语质量和可懂度的关键方法。为了充分利用语音的频域和时域相关性实现耳语转换,提出了使用深度卷积神经网络(Deep Convolutional Neural Networks,DCNN)将耳语转换为正常语音。它的卷积层用来提取连续帧语音谱包络之间的频域与时域的相关特征,而全连接层用来拟合耳语在卷积层提取的特征和对应正常语音之间的映射关系。实验结果表明与深度神经网络(Deep Neural Networks,DNN)模型相比,DCNN模型获得的转换后语音的梅尔倒谱失真度(Cepstral Distance,CD)降低了4.64%,而语音质量感知评价(Perceptual Evaluation of Speech Quality,PESQ)、短时客观可懂度(Short-Time Objective Intelligibility,STOI)与平均主观意见分(Mean Opinion Score,MOS)分别提高了5.41%,5.77%,9.68%。     

Pruning deep convolutional neural networks for efficient edge computing in condition assessment of infrastructures

Health monitoring of civil infrastructures is a key application of Internet of things (IoT), while edge computing is an important component of IoT. In this context, swarms of autonomous inspection robots, which can replace current manual inspections, are examples of edge devices. Incorporation of pretrained deep learning algorithms into these robots for autonomous damage detection is a challenging problem since these devices are typically limited in computing and memory resources. This study introduces a solution based on network pruning using Taylor expansion to utilize pretrained deep convolutional neural networks for efficient edge computing and incorporation into inspection robots. Results from comprehensive experiments on two pretrained networks (i.e., VGG16 and ResNet18) and two types of prevalent surface defects (i.e., crack and corrosion) are presented and discussed in detail with respect to performance, memory demands, and the inference time for damage detection. It is shown that the proposed approach significantly enhances resource efficiency without decreasing damage detection performance.     

A unified convolutional neural network integrated with conditional random field for pipe defect segmentation

Semantic segmentation of closed‐circuit television (CCTV) images can facilitate automatic severity assessment of sewer pipe defects by assigning defect labels to each pixel on the image, from which defect types, locations, and geometric information can be obtained. In this study, a unified neural network, namely DilaSeg‐CRF, is proposed by fully integrating a deep convolutional neural network (CNN) with dense conditional random field (CRF) for improving the segmentation accuracy. First, DilaSeg is constructed with dilated convolution and multiscale techniques for producing feature maps with high resolution. The steps of the dense CRF inference algorithm are converted into CNN operations, which are then formulated as recurrent neural network (RNN) layers. The DilaSeg‐CRF is proposed by integrating DilaSeg with the RNN layers. Images containing three common types of sewer defects are collected from CCTV inspection videos and are annotated with ground truth labels, after which the proposed models are trained and evaluated. Experiments demonstrate that the end‐to‐end trainable DilaSeg‐CRF can improve the segmentation significantly, with an increase of 32% and 20% in mean intersection over union (mIoU) values compared with fully convolutional network (FCN‐8s) and DilaSeg, respectively. Our proposed DilaSeg‐CRF also achieves faster inference speed than FCN and eliminates the manual postprocessing for refining the segmentation results.     

基于目标检测卷积神经网络的图像型火灾探测算法

摘 要：针对传统图像型火灾探测算法误差率高、延迟探测、计算量大等问题,提出了基于目标检测卷积神经网络（Faster-RCNN、R-FCN、SSD和YOLO v3）的图像型火灾探测算法。通过对比实验表明,基于目标检测卷积神经网络的探测算法准确性较高。其中,YOLO v3探测算法的平均精度为84.5%,探测速度为28帧/s,具有更高的稳定性,更适用于图像型火灾探测系统的开发。     

Transfer learning and deep convolutional neural networks for safety guardrail detection in 2D images

Safety has been a concern for the construction industry for decades. Unsafe conditions and behaviors are considered as the major causes of construction accidents. The current safety inspection of conditions and behaviors heavily rely on human efforts which are limited onsite. To improve the safety performance of the industry, a more efficient approach to identify the unsafe conditions on site is required to supplement the current manual inspection practice. A promising way to supplement the current manual safety inspection is automated and intelligent monitoring/inspection through information and sensing technologies, including localization techniques, environment monitoring, image processing and etc. To assess the potential benefits of contemporary technologies for onsite safety inspection, the authors focused on real-time guardrail detection, as unprotected edges are the ones cause for workers falling from heights.In this paper, the authors developed a safety guardrail detection model based on convolutional neural network (CNN). An augmented data set is generated with the addition of background image to guardrail 3D models and used as training set. Transfer learning is utilized and the Visual Geometry Group architecture with 16 layers (VGG-16) model is adopted to construct the basic features extraction for the neural network. In the CNN implementation, 4000 augmented images were used to train the proposed model, while another 2000 images collected from real construction jobsites and 2000 images from Google were used to validate the proposed model. The proposed CNN-based guardrail detection model obtained a high accuracy of 96.5%. In addition, this study indicates that the synthetic images generated by augment technology can be used to create a large training dataset, and CNN-based image detection algorithm is a promising approach in construction jobsite safety monitoring.     

Hybrid random aggregation model and Bayesian optimization-based convolutional neural network for estimating the concrete compressive strength

Numerous experimental studies have shown the type and gradation of coarse aggregates effect on the mechanical properties of concrete. The type and gradation of coarse aggregates have not been taken into account in the available machine learning prediction models. In this study, a two-dimensional concrete microscopic image was generated by using a random aggregate model (RAM), and the coarse aggregate and other concrete ingredients were represented innovatively using polygons and trichromatic chromaticity values in the RAM images. The RAM image set was created by applying this method to represent 1110 sets of different concrete mixes. Then based on the Bayesian optimization algorithm and the image set, a compressive strength prediction model considering the effect of coarse aggregate types and gradations was developed utilizing a convolutional neural network (CNN) model. Meanwhile, an artificial neural network (ANN) compressive strength prediction model was developed using 1110 sets of mix ratio data. The results show that the proposed RAM image generation method has the capability to represent different concrete mix ratios collected in this study. The prediction performance of the CNN compressive strength model considering aggregate types and gradations is better than that of the ANN model. The method can provide a new perspective for predicting other concrete mechanical properties and technically support performance-based intelligent concrete mix design.     

基于卷积神经网络的故障诊断模型振动信号处理方法比较

针对利用卷积神经网络(convolutional neural networks, CNN)对滚动轴承进行故障诊断时可采用的振动信号处理方法较多的情况,设计了基于CNN的振动信号处理方法对比实验,采用不同的振动信号处理方法对滚动轴承在不同工况下的采样数据进行处理,再将动信号输入CNN故障诊断模型进行训练及测试,根据测试精度比较处理方法对故障诊断精度的影响。采用CNN中的AlexNet作为实验模型,选择模型中的最后3个全连接层,以达到快速训练的目的。对比不同信号处理方法对应的检测准确率可知,基于小波变换的滚动轴承故障诊断模型的检测准确率最高。     

Assessment of ROC curves for inspection of random fields

Inspection by non-destructive testing (NDT) techniques of existing structures is not perfect and it has become a common practice to model their reliability in terms of probability of detection (PoD), probability of false alarms (PFA) and receiver operating characteristic (ROC) curves. These results are generally the main inputs needed by owners of structures in order to achieve inspection, maintenance and repair plans (IMR). The assessment of PoD and PFA is even deduced from intercalibration of NDT tools or from the modelling of the noise and the signal. In this last case when the noise and the signal depend on the location on the structure PoD and PFA are spatially dependent. This paper presents how to define PoD and PFA when damage and detection are stochastic fields or spatially dependent. Corrosion of coastal structures in harbours is considered for illustration and ROC curves are deduced. Identification of probability density functions on polynomial chaos is shown to be more suitable than predefined probability distribution functions (pdf) in view of fitting noise and signal plus noise distributions.     

Postdisaster image‐based damage detection and repair cost estimation of reinforced concrete buildings using dual convolutional neural networks

Reinforced concrete (RC) buildings are commonly used around the world. With recent earthquakes worldwide, rapid structural damage inspection and repair cost evaluation are crucial for building owners and policy makers to make informed risk management decisions. To improve the efficiency of such inspection, advanced computer vision techniques based on convolutional neural networks have been adopted in recent research to rapidly quantify the damage state (DS) of structures. In this article, an advanced object detection neural network, named YOLOv2, is implemented, which achieves 98.2% and 84.5% average precision in training and testing, respectively. The proposed YOLOv2 is used in combination with the classification neural network, which improves the identification accuracy for critical DS of RC structures by 7.5%. The improved classification procedures allow engineers to rapidly and more accurately quantify the DSs of the structure, and also localize the critical damage features. The identified DS can then be integrated with the state‐of‐the‐art performance evaluation framework to quantify the financial losses of critical RC buildings. The results can be used by the building owners and decision makers to make informed risk management decisions immediately after the strong earthquake shaking. Hence, resources can be allocated rapidly to improve the resiliency of the community.     

A neural networks approach for cost flow forecasting

Artificial neural networks, which simulate neuronal systems of the brain, are useful methods that have attracted the attention of researchers in many disciplinary areas. They have many advantages over traditional methods in situations where the input-output relationship of the system under study is not explicitly known. This paper investigates the feasibility of using neural networks for predicting the cost flow of construction projects, explains the need for cost flow forecasting, and demonstrates the limitation of the existing models. It then introduces neural networks as an alternative approach to those mathematical and statistical methods. The method used in collecting data and modelling the cost flow is described. Results of the testing are presented and discussed.     

Impact of model uncertainty on soil quality standards for cadmium in rice paddy fields

At present, soil quality standards used for agriculture do not consider the influence of pH and CEC on the uptake of pollutants by crops. A database with 750 selected paired samples of cadmium (Cd) in soil and paddy rice was used to calibrate soil to plant transfer models using the soil metal content, pH, and CEC or soil Cd and Zn extracted by 0.01 M CaCl₂ as explanatory variables. The models were validated against a set of 2300 data points not used in the calibration. These models were then used inversely to derive soil quality standards for Japonica and Indica rice cultivars based on the food quality standards for rice. To account for model uncertainty, strict soil quality standards were derived considering a maximum probability that rice exceeds the food quality standard equal to 10 or 5%. Model derived soil standards based on Aqua Regia ranged from less than 0.3 mg kg^− 1 for Indica at pH 4.5 to more than 6 mg kg^− 1 for Japonica-type cultivars in clay soils at pH 7. Based on the CaCl₂ extract, standards ranged from 0.03 mg kg^− 1 Cd for Indica cultivars to 0.1 mg kg^− 1 Cd for Japonica cultivars. For both Japonica and Indica-type cultivars, the soil quality standards must be reduced by a factor of 2 to 3 to obtain the strict standards. The strong impact of pH and CEC on soil quality standards implies that it is essential to correct for soil type when deriving national or local standards. Validation on the remaining 2300 samples indicated that both types of models were able to accurately predict (> 92%) whether rice grown on a specific soil will meet the food quality standard used in Taiwan.     

Accuracy of approximate methods of uncertainty propagation in seismic loss estimation

In this paper the efficacy of an approximate method of uncertainty propagation, known as the first-order second-moment (FOSM) method, for use in seismic loss estimation is investigated. The governing probabilistic equations which define the Pacific Earthquake Engineering Research (PEER)-based loss estimation methodology used are discussed, and the proposed locations to use the FOSM approximations identified. The justification for the use of these approximations is based on a significant reduction in computational time by not requiring direct numerical integration, and the fact that only the first two moments of the distribution are known. Via various examples it is shown that great care should be taken in the use of such approximations, particularly considering the large uncertainties that must be propagated in a seismic loss assessment. Finally, a complete loss assessment of a structure is considered to investigate in detail the location where significant approximation errors are incurred, where caution must be taken in the interpretation of the results, and the computational demand of the various alternatives.     

3D convolutional neural network-based one-stage model for real-time action detection in video of construction equipment

This study aims to propose a three-dimensional convolutional neural network (3D CNN)-based one-stage model for real-time action detection in video of construction equipment (ADVICE). The 3D CNN-based single-stream feature extraction network and detection network are designed with the implementation of the 3D attention module and feature pyramid network developed in this study to improve performance. For model evaluation, 130 videos were collected from YouTube including videos of four types of construction equipment at various construction sites. Trained on 520 clips and tested on 260 clips, ADVICE achieved precision and recall of 82.1% and 83.1%, respectively, with an inference speed of 36.6 frames per second. The evaluation results indicate that the proposed method can implement the 3D CNN-based one-stage model for real-time action detection of construction equipment in videos of diverse, variable, and complex construction sites. The proposed method paved the way to improving safety, productivity, and environmental management of construction projects.     

各向异性随机场下的边坡模糊随机可靠度分析

土性参数具有很大的空间变异性,且在水平方向和垂直方向上差异显著。基于随机变量模型的传统边坡模糊随机可靠度分析方法并未对此进行考虑。提出一种能合理考虑土性参数空间变异性的边坡模糊随机可靠度分析方法。首先,视黏聚力和内摩擦角的均值为正态模糊数,对其取不同的λ截集水平并在各截集水平上进行参数组合。其次,利用各向异性随机场模拟土性参数的空间变异性,将有限元法和Monte–Carlo模拟相结合,计算各参数组合对应的可靠度指标。再通过数学方法得到边坡在各截集水平上的可靠度指标。最后,运用加权平均法计算边坡的模糊随机可靠度指标。算例分析表明:与水平方向的空间变异性相比,垂直方向的空间变异性对边坡模糊随机可靠度的影响更为显著;不考虑土性参数的空间变异性在一般情况下会低估边坡的模糊随机可靠度指标,但在抗剪强度参数变异性较大时,反而可能会高估边坡的模糊随机可靠度指标;此外,黏聚力与内摩擦角之间的相关性对边坡失效概率的影响趋势基本不受土性参数空间变异性的干扰。     

Hybrid semantic segmentation for tunnel lining cracks based on Swin Transformer and convolutional neural network

In the field of tunnel lining crack identification, the semantic segmentation algorithms based on convolution neural network (CNN) are extensively used. Owing to the inherent locality of CNN, these algorithms cannot make full use of context semantic information, resulting in difficulty in capturing the global features of crack. Transformer-based networks can capture global semantic information, but this method also has the deficiencies of strong data dependence and easy loss of local features. In this paper, a hybrid semantic segmentation algorithm for tunnel lining crack, named SCDeepLab, is proposed by fusing Swin Transformer and CNN in the encoding and decoding framework of DeepLabv3+ to address the above issues. In SCDeepLab, a joint backbone network is introduced with CNN-based Inverse Residual Block and Swin Transformer Block. The former is used to extract the local detailed information of the crack to generate the shallow feature layer, whereas the latter is used to extract the global semantic information to obtain the deep feature layer. In addition, Efficient Channel Attention enhanced Feature Fusion Module is proposed to fuse the shallow and deep features to combine the advantages of the two types of features. Furthermore, the strategy of transfer learning is adopted to solve the data dependency of Swin Transformer. The results show that the mean intersection over union (mIoU) and mean pixel accuracy (mPA) of SCDeepLab on the data sets constructed in this paper are 77.41% and 84.42%, respectively, which have higher segmentation accuracy than previous CNN-based and transformer-based semantic segmentation algorithms.     

Pixel-level thin crack detection on road surface using convolutional neural network for severely imbalanced data

Because roads are the major backbone of the transportation network, research about crack detection on road surfaces has been popular in computer and infrastructure engineering. When training a convolutional neural network (CNN) for pixel-level road crack detection, three common challenges include (1) the data are severely imbalanced, (2) crack pixels can be easily confused with normal road texture and other visual noises, and (3) there are many unexplainable characteristics regarding the CNN itself. When it comes to very fine and thin cracks, these challenges are exaggerated and a new challenge is introduced, as there can be a discrepancy between the actual width and the annotated width of a crack. To tackle all these challenges of thin crack detection, this paper proposes a new variant of CNN named ThinCrack U-Net, designed to provide thin results upon pixel-level crack detection on road surfaces. The main contribution is to demystify how pixel-level thin crack detection results are affected by different loss functions as well as various combinations of the U-Net components. The experimental results show that ThinCrack U-Net yields a significant performance boost in CrackTree260, from 65.71% to 94.48% F-measure, compared to the existing variant of U-Net previously proposed in the context of pixel-level thin crack detection. Finally, this paper locates the source of undesirable result thickness and solves it with the balanced usage of downsampling/upsampling layers and atrous convolution. Unlike suggested by previous works, different loss functions show no significant impact on ThinCrack U-Net, whereas normalization layers are proved crucial in pixel-level thin crack detection.     

On the propagation of uncertainty in inflow turbulence to wind turbine loads

When stochastic simulation of inflow turbulence random fields is employed in the analysis or design of wind turbines in normal operating states, it is common to use well-established standard spectral models represented in terms of parameters that are usually treated as fixed or deterministic values. Studies have suggested, though, that many of these spectral parameters can exhibit some degree of variability. It is not unreasonable to expect, then, that derived flow fields based on simulation with such spectral models can be in turn highly variable for different realizations. Turbine load and performance variability would also be expected to result if response simulations are carried out with these variable flow fields. The aim here is to assess the extent of variability in derived inflow turbulence fields that arises from the noted variability in spectral model parameters. Simulation of these parameters as random variables forms the basis of this study. A commercial-sized 1.5 MW concept wind turbine is considered in the numerical studies. Variability in turbulence power spectra at field points on the rotor plane and in turbulence coherence functions for separations on the order of a rotor diameter and smaller is studied. Using time domain simulations, variability in various wind turbine response measures is also studied where the focus is on statistics such as response root-mean-square and 10-min extreme estimates. It is seen that while variability in inflow turbulence spectra can be great, the variability in turbine loads is generally considerably lower. One exception is for turbine yaw loads whose larger variability arises due to sensitivity to a coherence decay parameter that is itself highly variable. Finally, because reduced-order representations of turbulence random fields using empirical orthogonal decomposition techniques allow useful physical insights into spatial patterns of flow, variability in the energy distribution and the shapes of such empirical eigenmodes is studied and a simplified model is proposed that retains key variability sources in a limited number of modes and that accurately preserves overall inflow turbulence field uncertainty.