Structural Engineering Applications with Augmented Neural Networks

This paper presents an augmented neural network (ANN), a novel neural network architecture, and examines its efficiency and accuracy for structural engineering applications. The proposed architecture is that of a standard backpropagation neural network with augmented neurons, that is, logarithm neurons and exponent neurons are added to the network's input and output layers. The principles of augmented neural networks are (1) the augmented neurons are highly sensitive in the boundary domain, thereby facilitating construction of accurate mapping in the model's boundary domain, and (2) the network denotes each input variable with multiple input neurons, thus allowing a highly interactive function on hidden neurons to be easily formed. Therefore, the hidden neurons can more easily construct an accurate network output for a highly interactive mapping model. Experimental results demonstrate that the network's logarithm and exponent neurons provide a markedly enhanced network architecture capable of improving the network's performance for structural engineering applications.     

Structural Health Monitoring via Measured Ritz Vectors Utilizing Artificial Neural Networks

Abstract:   A pattern recognition approach for structural health monitoring (SHM) is presented that uses damage-induced changes in Ritz vectors as the features to characterize the damage patterns defined by the corresponding locations and severity of damage. Unlike most other pattern recognition methods, an artificial neural network (ANN) technique is employed as a tool for systematically identifying the damage pattern corresponding to an observed feature. An important aspect of using an ANN is its design but this is usually skipped in the literature on ANN-based SHM. The design of an ANN has significant effects on both the training and performance of the ANN. As the multi-layer perceptron ANN model is adopted in this work, ANN design refers to the selection of the number of hidden layers and the number of neurons in each hidden layer. A design method based on a Bayesian probabilistic approach for model selection is proposed. The combination of the pattern recognition method and the Bayesian ANN design method forms a practical SHM methodology. A truss model is employed to demonstrate the proposed methodology.     

动量BP算法及其在桩基沉降研究中的应用

在说明动量BP算法及其程序实现的基础上 ,阐述了网络训练中的主要参数 (初始权值与阈值、隐层单元数、输出单元数、归一化方法、传递函数 )对训练过程的影响 ,并将动量BP算法应用于桩基沉降的研究。研究表明 ,使用一定的训练参数 ,得到的预测沉降与最终沉降量具有良好的一致性     

应用自递归神经网络(SRNN)预测结构响应

本文提出了一种新的自递归神经网络结构.这种网络结构由全递归网络改造而成,只有一个隐层,而且隐单元仅存在自递归.研究了这种网络的学习算法.为了保证快速学习收敛,应用Lyapunov函数得到一种自适应学习率方法.用这种方法对一两层建筑结构响应进行在线预测.计算机仿真结果表明,这种网络学习算法是有效的,并且是可行的     

基于CPTu数据的黏性土不排水抗剪强度#br# 神经网络预测

利用人工神经网络模型,建立基于孔压静力触探（CPTu）现场测试数据的黏性土不排水抗剪强度的预测方法。为建立和验证人工神经网络模型,在3个场地开展CPTu和十字板剪切现场测试,共取得33个测孔的CPTu试验数据和相对应的不排水抗剪强度实测值。通过对比分析不同输入向量、不同网络隐层数、不同神经元数及不同改进算法对人工神经网络模型性能的影响,确定人工神经网络模型的具体形式。通过对训练组数据开展机器学习,所建立的人工神经网络模型能够有效地基于CPTu获得的端阻力和孔隙水压力现场测试数据对黏土不排水抗剪强度进行预测,预测结果与十字板剪切试验实测结果非常接近。与传统用于估算不排水强度的经验关系相比,采用人工神经网络模型预测结果与实测结果相关性显著提高、误差明显降低。     

Application of artificial neural networks to fracture analysis at the Äspö HRL, Sweden: fracture sets classification

This study investigates the potential of artificial neural networks (ANNs) to recognize, classify and predict patterns of different fracture sets in the top 450 m in crystalline rocks at the Äspö Hard Rock Laboratory (HRL), Southeastern Sweden. ANNs are computer systems composed of a number of processing elements that are interconnected in a particular topology which is problem dependent. ANNs have the ability to learn from examples using different learning algorithms; these involve incremental adjustment of a set of parameters to minimize the error between the desired output and the actual network output. Six fracture-sets with particular ranges of strike and dip have been distinguished. A series of trials were carried out using backpropagation (BP) neural networks for supervised classification, and the BP networks recognized different fracture sets accurately. Self-organizing neural networks have been used for data clustering analysis with supervised learning algorithms; (competitive learning and learning vector quantization), and unsupervised learning algorithms; (self-organizing maps). The self-organizing networks adapted successfully to different fracture clusters (sets). A set of trials has been carried out to investigate the effect of changing the network's topologies on the performance of the BP networks. Using two hidden layers with tan-sigmoid and linear transfer functions was beneficial for the performance of BP classification. ANNs improved fracture sets classification that was based on Kamb contouring method with constraint on areas between fracture clusters.     

Artificial neural network optimized by differential evolution for predicting diameters of jet grouted columns

A novel and effective artificial neural network (ANN) optimized using differential evolution (DE) is first introduced to provide a robust and reliable forecasting of jet grouted column diameters. The proposed computational method adopts the DE algorithm to tackle the difficulties in the training and performance of neural networks and optimize the four quintessential hyper-parameters (i.e. the epoch size, the number of neurons in a hidden layer, the number of hidden layers, and the regularization parameter) that govern the neural network efficacy. This approach is further enhanced by a stochastic gradient optimization algorithm to allow ‘expensive’ computation efforts. The ANN-DE is first trained using a prepared jet grouting dataset, then verified and compared with the prevalent machine learning tools, i.e. neural networks and support vector machine (SVM). The results show that, the ANN-DE outperforms the existing methods for predicting the diameter of jet grouting columns since it well balances training efficiency and model performance. Specifically, the ANN-DE achieved root mean square error (RMSE) values of 0.90603 and 0.92813 for the training and testing phases, respectively. The corresponding values were 0.8905 and 0.9006 for the optimized ANN, then, 0.87569 and 0.89968 for the optimized SVM, respectively. The proposed paradigm is bound to be useful for solving various geotechnical engineering problems regardless of multi-dimension and nonlinearity.     

利用人工神经网络预测火灾下管桁架的温度

3层反向传播(BP)神经网络已用于预测火灾下平面管桁架钢的极限温度。网络模型的输入参数有直径比(β)、墙宽厚比(τ)、径厚比(γ)和荷载比,输出参数有极限温度。利用有限元软件ABAQUS建立神经网络模型。105组数据用于建立BP神经网络,15组数据用于测试和验证BP网络。建立BP网络的过程中,选用Levenberg-Marquardt反向传播算法。隐藏层选用tansig函数,输出层选用purelin函数。分析结果表明,使用BP网络模型预测的极限温度是准确有效的。     

Evolving neural network using a genetic algorithm for predicting the deformation modulus of rock masses

The determination of deformation modulus of rock masses is one of the most difficult tasks in the field of rock mechanics. Due to the high cost and measurement difficulties of in situ tests in modulus determination, the predictive models using regression based statistical methods, back propagation neural networks (BPNN) and fuzzy systems are recently employed for the indirect estimation of the modulus. Among these methods, the BPNN has been reported to be very useful in modeling the rock material behavior, such as deformation modulus, by many researchers. Despite its extensive applications, design and structural optimization of BPNN are still done via a time-consuming reiterative trial-and-error approach. This research focuses on the efficiency of the genetic algorithm (GA) in design and optimizing the BPNN structure and its application to predict the deformation modulus of rock masses. GA is utilized to find the optimal number of neurons in hidden layer, learning rates and momentum coefficients of hidden and output layers of network. Then the result is compared with that of trial-and-error procedure. For the purpose, a database including 120 data sets was employed from four dam sites and power house locations in Iran. Taking advantages of performance criteria such as MSE, MAE, r, proved that the GA-ANN model gives superior predictions over the trial-and-error model.     

BP神经网络算法对建筑保温材料性能的预测

针对建筑保温材料性能表征十分复杂、困难的情况,利用人工神经网络BP算法,建立了复合保温材料性能预测模型,模型由3层神经元组成,分别为输入层、隐含层和输出层。以炉渣复合材料性能与成分的关系为研究对象,采取108组实验数据对神经网络进行8 000次训练,神经网络输出值的平方平均误差为0.000 12。然后,选用18组实验数据对训练成熟的试验神经网络模型进行检测,并把检测样本的神经网络输出值和试验值进行比较。结果表明:所建立的网络能反映炉渣复合保温材料与材料性能之间的关系,为实验设计提供了新的思想,节省了时间和劳动力。     

A deep feed-forward neural network for damage detection in functionally graded carbon nanotube-reinforced composite plates using modal kinetic energy

This paper proposes a new Deep Feed-forward Neural Network (DFNN) approach for damage detection in functionally graded carbon nanotube-reinforced composite (FG-CNTRC) plates. In the proposed approach, the DFNN model is developed based on a data set containing 20 000 samples of damage scenarios, obtained via finite element (FE) simulation, of the FG-CNTRC plates. The elemental modal kinetic energy (MKE) values, calculated from natural frequencies and translational nodal displacements of the structures, are utilized as input of the DFNN model while the damage locations and corresponding severities are considered as output. The state-of-the art Exponential Linear Units (ELU) activation function and the Adamax algorithm are employed to train the DFNN model. Additionally, in order to enhance the performance of the DFNN model, the mini-batch and early-stopping techniques are applied to the training process. A trial-and-error procedure is implemented to determine suitable parameters of the network such as the number of hidden layers and the number of neurons in each layer. The accuracy and capability of the proposed DFNN model are illustrated through two distinct configurations of the CNT-fibers constituting the FG-CNTRC plates including uniform distribution (UD) and functionally graded-V distribution (FG-VD). Furthermore, the performance and stability of the DFNN model with the consideration of noise effects on the input data are also investigated. Obtained results indicate that the proposed DFNN model is able to give sufficiently accurate damage detection outcomes for the FG-CNTRC plates for both cases of noise-free and noise-influenced data.     

Prediction of effective thermal conductivity of moist porous materials using artificial neural network approach

An artificial neural networks (ANNs) approach is presented for the prediction of effective thermal conductivity of porous systems filled with different liquids. ANN models are based on feedforward backpropagation network with training functions: Levenberg–Marquardt (LM), conjugate gradient with Fletcher–Reeves updates (CGF), one-step secant (OSS), conjugates gradient with Powell–Beale restarts (CGB), Broyden, Fletcher, Goldfrab and Shanno (BFGS) quasi-Newton (BFG), conjugates gradient with Polak–Ribiere updates (CGP). Training algorithm for neurons and hidden layers for different feedforward backpropagation networks at the uniform threshold function TANSIG-PURELIN are used and run for 1000 epochs. The complex structure encountered in moist porous materials, along with the differences in thermal conductivity of the constituents makes it difficult to predict the effective thermal conductivity accurately. For this reason, artificial neural networks (ANNs) have been utilized in this field. The resultant predictions of effective thermal conductivity (ETC) of moist porous materials by the different models of ANN agree well with the available experimental data.     

基于OIF-Elman神经网络的燃气日负荷预测

与传统的Elman神经网络相比,采用具有输出一输入反馈机制的改进Elman(即OIF-Elman)神经网络对燃气日负荷进行预测,不仅计入了隐层节点的反馈,而且考虑输出层节点的反馈,以便从有限的训练样本中获得更多的信息.预测结果表明,在样本较少时,无论在训练速度上,还是在预测准确度上,OIF-Elman网络明显优于Elman网络.     

Position-Invariant Neural Network for Digital Pavement Crack Analysis

Abstract:   This article presents an integrated neural network-based crack imaging system to classify crack types of digital pavement images. This system includes three neural networks: (1) image-based neural network, (2) histogram-based neural network, and (3) proximity-based neural network. These three neural networks were developed to classify various crack types based on the subimages (crack tiles) rather than crack pixels in digital pavement images. These spatial neural networks were trained using artificially generated data following the Federal Highway Administration (FHWA) guidelines. The optimal architecture of each neural network was determined based on the testing results from different sets of the number of hidden units, learning coefficients, and the number of training epochs. To validate the system, actual pavement pictures taken from pavements as well as the computer-generated data were used. The proximity value is determined by computing relative distribution of crack tiles within the image. The proximity-based neural network effectively searches the patterns of various crack types in both horizontal and vertical directions while maintaining its position invariance. The final result indicates that the proximity-based neural network produced the best result with the accuracy of 95.2% despite its simplest neural network structure with the least computing requirement.     

A Multiwavelet Neural Network‐Based Response Surface Method for Structural Reliability Analysis

A new multiwavelet neural network‐based response surface method is proposed for efficient structural reliability assessment. Although multiwavelet network can be used for approximating nonlinear functions, its application has been limited to small dimension problems due to computational cost. The new method expands the application of multiwavelet network to moderate dimension by introducing a series of intermediate nodes, and the number of these intermediate nodes is determined by the multiwavelet theory. Thus, a multidimensional function learning problem is transformed into a one‐dimensional function learning problem. Four examples involving one stochastic finite element‐based reliability problem illustrate the effectiveness of the proposed method, which indicate that the new method is more efficient up to 10 random variables than the classical multilayer perceptron‐based response surface method.     

Identification of land use with water quality data in stormwater using a neural network

To control stormwater pollution effectively, development of innovative, land-use-related control strategies will be required. An approach that could differentiate land-use types from stormwater quality would be the first step to solving this problem. We propose a neural network approach to examine the relationship between stormwater water quality and various types of land use. The neural network model can be used to identify land-use types for future known and unknown cases. The neural model uses a Bayesian network and has 10 water quality input variables, four neurons in the hidden layer, and five land-use target variables (commercial, industrial, residential, transportation, and vacant). We obtained 92.3 percent of correct classification and 0.157 root-mean-squared error on test files. Based on the neural model, simulations were performed to predict the land-use type of a known data set, which was not used when developing the model. The simulation accurately described the behavior of the new data set. This study demonstrates that a neural network can be effectively used to produce land-use type classification with water quality data.     

Structural Damage Detection with Automatic Feature‐Extraction through Deep Learning

Structural damage detection is still a challenging problem owing to the difficulty of extracting damage‐sensitive and noise‐robust features from structure response. This article presents a novel damage detection approach to automatically extract features from low‐level sensor data through deep learning. A deep convolutional neural network is designed to learn features and identify damage locations, leading to an excellent localization accuracy on both noise‐free and noisy data set, in contrast to another detector using wavelet packet component energy as the input feature. Visualization of the features learned by hidden layers in the network is implemented to get a physical insight into how the network works. It is found the learned features evolve with the depth from rough filters to the concept of vibration mode, implying the good performance results from its ability to learn essential characteristics behind the data.     

基于神经网络的深基坑沉降预测模型比较

神经网络预测为深基坑预测提供了一种有效的路径。运用哪种模型较优,输入层、输出层、隐含层参数如何选取,对预测的结果都有一定的影响,本文结合实际轨道交通工程案例,以深基坑沉降监测数据为例,对常见的几种神经网络预测模型进行了对比分析,对几种模型的残差、均方根误差(RMSE)和绝对平均误差(MAE),收敛次数这几个方面进行对比,结果表明遗传算法神经网络对深基坑沉降监测数据预测较为有效,同时对模型参数的选取提出了建议。     

Prediction of blast-induced ground vibration using artificial neural network

An attempt has been made to evaluate and predict the blast-induced ground vibration and frequency by incorporating rock properties, blast design and explosive parameters using the artificial neural network (ANN) technique. A three-layer, feed-forward back-propagation neural network having 15 hidden neurons, 10 input parameters and two output parameters were trained using 154 experimental and monitored blast records from one of the major producing surface coal mines in India. Twenty new blast data sets were used for the validation and comparison of the peak particle velocity (PPV) and frequency by ANN and other predictors. To develop more confidence in the proposed method, same data sets have also been used for the prediction of PPV by commonly used vibration predictors as well as by multivariate regression analysis (MVRA). Results were compared based on correlation and mean absolute error (MAE) between monitored and predicted values of PPV and frequency.     

MIV-PSO-BP神经网络用户热负荷预测

提出利用MIV-PSO-BP神经网络预测用户热负荷。MIV-PSO-BP神经网络基于BP神经网络,利用PSO算法优化神经网络初始参数,采取MIV算法筛选与输出变量相关程度最大的输入变量。以绝对误差、均方根误差作为指标,评价MIV-PSO-BP神经网络的预测效果。结合箱线图,比较BP神经网络、MIV-PSO-BP神经网络的预测相对误差分散程度与异常点数量。与BP神经网络相比,MIV-PSO-BP神经网络的预测效果更理想。由BP神经网络、MIV-PSO-BP神经网络的预测结果相对误差箱线图可知,MIV-PSO-BP神经网络预测结果相对误差集中,异常点少;BP神经网络预测结果相对误差分散,异常点多。MIV-PSO-BP神经网络预测准确性、稳定性更高。