Neural Networks in Structural Engineering

In the past few years literature on computational civil engineering has concentrated primarily on artificial intelligence (Al) applications involving expert system technology. This article discusses a different Al approach involving neural networks. Unlike their expert system counterparts, neural networks can be trained based on observed information. These systems exhibit a learning and memory capability similar to that of the human brain, a fact due to their simplified modeling of the brain's biological function. This article presents an introduction to neural network technology as it applies to structural engineering applications. Differing network types are discussed. A back-propagation learning algorithm is presented. The article concludes with a demonstration of the potential of the neural network approach. The demonstration involves three structural engineering problems. The first problem involves pattern recognition; the second, a simple concrete beam design; and the third, a rectangular plate analysis. The pattern recognition problem demonstrates a solution which would otherwise be difficult to code in a conventional program. The concrete beam problem indicates that typical design decisions can be made by neural networks. The last problem demonstrates that numerically complex solutions can be estimated almost instantaneously with a neural network.     

An expert-simulation system for construction productivity forecasting

Knowledge based and neural network systems provide an interesting new tool for dealing with uncertainty in decision making. The authors discuss the different sources of uncertainty for a construction planning expert system and show how the results (output) from the expert system can be used to estimate operation or project durations.     

Enhancing the performance of a neural network with entity embeddings: an application to real estate valuation

In contrast to the brilliant success of deep learning approach in dealing with unstructured data such as image and natural language, it has not shown noticeable achievements in handling structured data, that is, tabular format data. Categorical data types form a considerable portion of structured data, and a neural network, the most universal implementation algorithm for deep learning, is inefficient in processing these data types. This is a reason for the poor performance of the neural network applied to the structured data. In this study, a neural network is used to estimate land prices in the Gyunggi province, South Korea. To enhance the performance of the network when most input variables are categorical, the architecture of the neural network is specified using the entity embedding layers, a technique to reveal the continuity inherent in categorical data. This study demonstrates that information in the categorical data can be efficiently extracted by the entity embedding technique. The network architecture proposed in this study can be applied in valuation practices where categorical data are abundant. In addition, the interpretation of the resultant embedding layers can enhance the explainability of the deep learning approach, promoting its rapid adoption in the real estate industry.

     

The application of neural networks to rock engineering systems (RES)

This paper proposes a new approach for applying neural networks in Rock Engineering Systems (RES) based on the learning abilities of neural networks. By considering the analysis of the coding methods for the interaction matrix in RES and the learning processes of neural networks such as the Back Propagation (BP) method, neural networks can provide a useful mapping from system inputs to system outputs for rock engineering, so that the influence of inputs on outputs can be obtained. Then the results of the neural network analysis can be presented in a similar way to the global interaction matrix used in RES to present the fully-coupled system results. The neural network procedures are explained first, with illustrative demonstrations for simultaneous equations. Then, the link with the RES type of analysis is explained, together with some demonstration examples for rock engineering data sets. The specific analysis procedure is presented and then wider rock engineering examples are given relating to the characteristics of rock masses and engineering parameters. The main presentation tools used in this neural network approach are the Relative Strength Effect (RSE) and the Global Relative Strength Effect (GRSE) matrix. There is discussion of the value of this approach and an indication of the likely areas of future development.     

Estimating wall deflections in deep excavations using Bayesian neural networks

In this study, a neural network algorithm has been used to model the soil-structure interaction behavior of deep excavations in clays. The hybrid evolutionary Bayesian back-propagation (EBBP) neural network was used in this study and utilizes the genetic algorithms and gradient descent method to determine the optimal parameters within a Bayesian framework to regularize the complexity of learning and to statistically reflect the uncertainty in data. The EBBP analysis was carried out on an extensive database of braced excavation performance from finite element analyses. Additional parametric studies indicate that the model gives logical and consistent trends. Back-analyses of some instrumented case histories from the literature also indicate that the trained neural network model gives reasonable predictions in comparison to the actual measured values. The trained model can serve as a simple and reliable prediction tool to enable estimates of maximum wall deflection for preliminary design of braced excavations in clay. The model is able to take into consideration various factors such as the wall stiffness, support stiffness, the in-situ stress state, non-homogeneous soil conditions, and the variation of soil properties with depth. An added advantage of this approach is that it provides meaningful error bars for the model predictions.     

一种基于神经网络的材料成分模型

为研究如何利用神经网络预测材料化合物构成，建立了一个4层前向型网络，这种网络通过改变神经元非线性变换函数的参数，使连接权调整线性化，从而可提高学习速度，减少计算量，并避免了BP网络存在的易陷入局部极小和收敛速度慢的问题，以CaO-Al2O3-SiO2系统为例进行的仿真研究结果表明，该网络可成功包含材料化合物的构成信息。     

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.     

基于数据挖掘技术的黄土分类问题研究

依据数据挖掘技术,采用分类回归树决策树和概率神经网络对黄土的分类规则进行挖掘。利用主成分分析法对数据进行了清洗和降维处理,以处理后的新变量作为挖掘对象,使挖掘出的分类模型和规则得到了简化,提高了计算精度;同时归纳出了影响黄土分类的因素,所挖掘出的分类规则可用于黄土地层的智能划分。研究结果表明,挖掘出的知识具有良好的实用性。     

深基坑工程土层参数反分析方法探讨研究

深基坑工程受时空效应、地质条件以及土体不确定性的影响较大,其土层参数难以确定.根据现场实测数据利用基于FLAC正分析的BP神经网络反分析方法,对工程中的关键土层进行参数反分析,算例结果显示:以实测地下连续墙最大水平位移反分析的⑥1层土体参数与实验值吻合较好,说明在深基坑工程中,基于FLAC正分析的BP神经网络反分析方法...     

人工神经网络技术在高强混凝土中的应用--配合比设计初探

探讨了在多因素影响下，人工神经网络技术在混凝土配合比设计方面的实现手段．采用以正交设计试验作为学习样本模拟真实系统的方法，来模拟完全试验；同时，以部分试验数据为研究对象，通过自组织神经网络分类计算，构成学习样本来模拟真实系统，也得到了较为满意的结果．此项研究除提供了人工智能在混凝土配合比设计中的应用方法外，还在具体研究问题的背景下，为神经网络理论在确立学习样本的方法上寻求了一个可行的途径     

桩基极限承载力与沉降量的神经网络预测

利用BP神经网络较强的高次非一性映射能力和学习功能，建立了基于人工神经网络的单桩极限承载力与沉降量的预测模型。该模型依据现场实测资料建模，避免了计算过程中各种人为因素的影响。通过静载荷试验成果的学习与预测检验，证明其预测精度良好、适用性强，具有较大的工程实用价值。     

Estimation of air losses in compressed air tunneling using neural network

This paper explores the capabilities of neural networks to predict the air losses in compressed air tunneling. Field data from the Feldmoching tunnel in Munich were used in this study. In this project, compressed air was used to retain the groundwater and shotcrete was used as temporary support. The final permanent lining was installed in free air. The tunnel passed through variable ground conditions ranging from coarse gravel to sand and clay. Grouting, an additional layer of shotcrete and a layer of mortar were occasionally used to control the air losses. A back-propagation feed forward neural network was trained and used to predict the air losses from the Feldmoching tunnel. The results of the prediction of the air losses from the tunnel using a neural network were compared with the field measurements. Data from different tunnel lengths were used for training. In each case, the trained network was used to predict the air losses during the excavation of the rest of the tunnel. It is shown that, not only can a neural network learn the relationship between appropriate soil and tunnel parameters and air losses, it can also generalize the learning to predict air losses for very different geological and geometric conditions. It is also shown that data from a very short length (50 m in one case) of the tunnel (five data point only, in this case) may contain enough information for the neural network to learn and predict the air losses in the remaining (585 m) length of the tunnel with a good degree of accuracy. This can be of considerable value to tunnel engineers in control of tunneling operations and help them in preparation for possible changes in air losses with tunnel advance, with changes in ground conditions and tunnel geometry and with time.     

桩基极限承载力与沉降量的神经网络预测

利用BP神经网络较强的高次非线性映射能力和学习功能 ,建立了基于人工神经网络的单桩极限承载力与沉降量的预测模型。该模型依据现场实测资料建模 ,避免了计算过程中各种人为因素的影响。通过静载荷试验成果的学习与预测检验 ,证明其预测精度良好、适用性强 ,具有较大的工程实用价值     

Neural network analysis of construction safety management systems: a case study in Singapore

A neural network analysis was conducted on a quantitative occupational safety and health management system (OSHMS) audit with accident data obtained from the Singapore construction industry. The analysis is meant to investigate, through a case study, how neural network methodology can be used to understand the relationship between OSHMS elements and safety performance, and identify the critical OSHMS elements that have significant influence on the occurrence and severity of accidents in Singapore. Based on the analysis, the model may be used to predict the severity of accidents with adequate accuracy. More importantly, it was identified that the three most significant OSHMS elements in the case study are: incident investigation and analysis, emergency preparedness, and group meetings. The findings imply that learning from incidents, having well-prepared consequence mitigation strategies and open communication can reduce the severity and likelihood of accidents on construction worksites in Singapore. It was also demonstrated that a neural network approach is feasible for analysing empirical OSHMS data to derive meaningful insights on how to improve safety performance.     

Enhancing Neural Network Traffic Incident-Detection Algorithms Using Wavelets

Researchers have presented freeway traffic incident-detection algorithms by combining the adaptive learning capability of neural networks with imprecision modeling capability of fuzzy logic. In this article it is shown that the performance of a fuzzy neural network algorithm can be improved through preprocessing of data using a wavelet-based feature-extraction model. In particular, the discrete wavelet transform (DWT) denoising and feature-extraction model proposed by Samant and Adeli (2000) is combined with the fuzzy neural network approach presented by Hsiao et al. (1994). It is shown that substantial improvement can be achieved using the data filtered by DWT. Use of the wavelet theory to denoise the traffic data increases the incident-detection rate, reduces the false-alarm rate and the incident-detection time, and improves the convergence of the neural network training algorithm substantially.     

Spatial choice behaviour: logit models and neural network analysis

Neural networks are becoming popular analysis tools in spatial research, as is witnessed by various applications in recent years. The performance of neural network analysis needs to be carefully judged, however, since the theoretical underpinning of neuro-computing is still weakly enveloped. In the present paper we will use the logit model as a benchmark for evaluating the result of neural network models, based on an empirical case study from Italy. The present paper aims to assess the foreseeable impact of the high-speed train in Italy, by investigating competition effects between rail and road transport modes. Two statistical models will then be compared, viz. the traditional logit model and a new technique for information processing, viz. the feedforward neural network model. In the study two different cases – corresponding to a different set of attributes – are investigated, namely by using only ‘time’ attributes and by using both ‘time’ and ‘cost’ attributes. From an economic viewpoint, both models appear to highlight the advantage of introducing the high-speed train system in that they show high probabilities of choosing the improved rail transport mode. The feedforward neural net model seems to provide reasonable predictions compared to those obtained by means of a logit model. An important lesson however, is that it is important to define properly the neural network architecture and to train sufficiently the network during the learning phase. Received: June 1996 / Accepted: February 1997     

Innovative piled raft foundations design using artificial neural network

Studying the piled raft behavior has been the subject of many types of research in the field of geotechnical engineering. Several studies have been conducted to understand the behavior of these types of foundations, which are often used for uniform loading on the raft and piles with the same length, while generally the transition load from the upper structure to the foundation is non-uniform and the choice of uniform length for piles in the above model will not be optimally economic and practical. The most common method in identifying the behavior of piled rafts is the use of theoretical relationships and software analyses. More precise identification of this type of foundation behavior can be very difficult due to several influential parameters and interaction of set behavior, and it will be done by doing time-consuming computer analyses or costly full-scale physical modeling. In the meantime, the technique of artificial neural networks can be used to achieve this goal with minimum time consumption, in which data from physical and numerical modeling can be used for network learning. One of the advantages of this method is the speed and simplicity of using it. In this paper, a model is presented based on multi-layer perceptron artificial neural network. In this model pile diameter, pile length, and pile spacing is considered as an input parameter that can be used to estimate maximum settlement, maximum differential settlement, and maximum raft moment. By this model, we can create an extensive domain of results for optimum system selection in the desired piled raft foundation. Results of neural network indicate its proper ability in identifying the piled raft behavior. The presented procedure provides an interesting solution and economically enhancing the design of the piled raft foundation system. This innovative design method reduces the time spent on software analyses.     

Machine learning methods to assist energy system optimization

This study evaluates the potential of supervised and transfer learning techniques to assist energy system optimization. A surrogate model is developed with the support of a supervised learning technique( by using artificial neural network) in order to bypass computationally intensive Actual Engineering Model( AEM). Eight different neural network architectures are considered in the process of developing the surrogate model. Subsequently,a hybrid optimization algorithm( HOA) is developed combining Surrogate and AEM in order to speed up the optimization process while maintaining the accuracy. Pareto optimization is conducted considering Net Present Value and Grid Integration level as the objective functions. Transfer learning is used to adapt the surrogate model( trained using supervised learning technique) for different scenarios where solar energy potential,wind speed and energy demand are notably different. Results reveal that the surrogate model can reach to Pareto solutions with a higher accuracy when grid interactions are above 10%( with reasonable differences in the decision space variables). HOA can reach to Pareto solutions( similar to the solutions obtained using AEM) around 17 times faster than AEM. The Surrogate Models developed using Transfer Learning( SMTL) shows a similar capability. SMTL combined with the optimization algorithm can predict Pareto fronts efficiently even when there are significant changes in the initial conditions.Therefore,STML can be used along with the HOA,which reduces the computational time required for energy system optimization by 84%. Such a significant reduction in computational time enables the approach to be used for energy system optimization at regional or national scale.     

Neural Networks for Evaluating CPT Calibration Chamber Test Data

Abstract: The feasibility of using neural network models for evaluating CPT calibration chamber test data is investigated. The backpropagation neural network algorithm was used to analyze the data. After learning from a set of randomly selected patterns, the neural network model was able to produce reasonably accurate predictions for patterns not included in the training set. The neural network performance was found to be simpler and more effective than regression analysis for modeling the CPT test data. Correlations between the cone measurements and the engineering properties of sand can be developed using the generalization capabilities of the neural network.     

Deep convolutional neural networks for uncertainty propagation in random fields

The development of a reliable and robust surrogate model is often constrained by the dimensionality of the problem. For a system with high‐dimensional inputs/outputs (I/O), conventional approaches usually use a low‐dimensional manifold to describe the high‐dimensional system, where the I/O data are first reduced to more manageable dimensions and then the condensed representation is used for surrogate modeling. In this study, a new solution scheme for this type of problem based on a deep learning approach is presented. The proposed surrogate is based on a particular network architecture, that is, convolutional neural networks. The surrogate architecture is designed in a hierarchical style containing three different levels of model structures, advancing the efficiency and effectiveness of the model in the aspect of training. To assess the model performance, uncertainty quantification is carried out in a continuum mechanics benchmark problem. Numerical results suggest the proposed model is capable of directly inferring a wide variety of I/O mapping relationships. Uncertainty analysis results obtained via the proposed surrogate have successfully characterized the statistical properties of the output fields compared to the Monte Carlo estimates.