An artificial neural network approach to multiphase continua constitutive modeling

Constitutive equations describe intrinsic relationships among sets of material system parameters. This study utilizes artificial neural networks in place of a traditional micromechanical approach to calculate the global (macroscopic) elastic properties of composite materials given the local (microscopic) properties and local geometry. This approach is shown to be more computationally efficient than conventional numerical micromechanical approaches. An eight sub-celled representative volume element is used for the local geometry. Multi target artificial neural networks (MTANNs) and single target artificial neural networks are studied for applicability in predicting the global properties. The best performing MTANN achieves a precision of 9%. The single target artificial neural networks (STANNs) perform best and predicts the global properties within a target error of 5.3%. The computation time is 1.8 s for all six STANNs to predict six global properties for 19,683 different microstructures.     

An integrated approach for process monitoring using wavelet analysis and competitive neural network

A novel framework involving both a detection module and a classification module is proposed for the recognition of the six main types of process signals. In particular, a multi-scale wavelet filter is used for denoising and its performance is compared with that of single-scale linear filters. Moreover, two kinds of competitive neural networks, based on learning vector quantization (LVQ) and adaptive resonance theory (ART), are adopted for the task of pattern classification and benchmarking. Our results show that denoising through a wavelet filter is best for pattern classification, and the classification accuracy with respect to six predefined categories using a LVQ-X network is a little better than using an ART network. However, when an unexpected novel pattern occurs within the process, LVQ will force the novel pattern to be classified into one of those predefined categories that is most similar to the novel pattern. On the contrary, ART will automatically construct a new class when the similarity measured between the novel pattern and the most similar category is too small to be incorporated. Therefore, under the consideration of the stability–plasticity dilemma, our simplified ART network based on multi-scale wavelet denoising provides a more promising way to adapt unexpected novel patterns.     

An artificial neural network for modeling reliability, availability and maintainability of a repairable system

The paper explores the application of artificial neural networks to model the behaviour of a complex, repairable system. A composite measure of reliability, availability and maintainability parameters has been proposed for measuring the system performance. The artificial neural network has been trained using past data of a helicopter transportation facility. It is used to simulate behaviour of the facility under various constraints. The insights obtained from results of simulation are useful in formulating strategies for optimal operation of the system.     

A study of a storage reliability estimation problem

Storage reliability, which describes the failure or deterioration of items in a dormant state, is considered in this paper. The study presented here is focused on the estimation of the storage reliability after a certain amount of storage time. We start with simple non-parametric estimation of the current reliability and then study the problem of parametric estimation based on a simple Weibull distribution assumption. Both maximum likelihood estimation and graphical techniques are considered in this case. The study is useful for planning a storage environment and making a decision about the maximum length of storage. Furthermore, the information can be used in the design and improvement of products for which the storage is an important part of the product's life cycle. A numerical example is provided to enlighten the idea.     

On reaction network implementations of neural networks

This paper is concerned with the utilization of deterministically modelled chemical reaction networks for the implementation of (feed-forward) neural networks. We develop a general mathematical framework and prove that the ordinary differential equations (ODEs) associated with certain reaction network implementations of neural networks have desirable properties including (i) existence of unique positive fixed points that are smooth in the parameters of the model (necessary for gradient descent) and (ii) fast convergence to the fixed point regardless of initial condition (necessary for efficient implementation). We do so by first making a connection between neural networks and fixed points for systems of ODEs, and then by constructing reaction networks with the correct associated set of ODEs. We demonstrate the theory by constructing a reaction network that implements a neural network with a smoothed ReLU activation function, though we also demonstrate how to generalize the construction to allow for other activation functions (each with the desirable properties listed previously). As there are multiple types of ‘networks’ used in this paper, we also give a careful introduction to both reaction networks and neural networks, in order to disambiguate the overlapping vocabulary in the two settings and to clearly highlight the role of each network’s properties.     

An alternative to the analysis of variance for reliability data

Though the analysis of variance is a commonly applied method for testing for differences between means of several processes, it is based in part on the assumption that the processes give rise to output that is normally distributed on the measured variable. Reliability and life test studies frequently give birth to data that exhibit clear skew, and application of the analysis of variance is questionable in such cases. A method referred to as analysis of reciprocals, which is based on an assumed inverse Gaussian distribution, provides an alternative to the analysis of variance in these instances. With applications in a variety of functional areas, including reliability and life testing, the inverse Gaussian distribution is able to accommodate substantial skew. It is hoped that this exposition will increase awareness of both the inverse Gaussian distribution and data analysis methods that are based on this distribution.     

Numerical implementation of a neural network based material model in finite element analysis

Neural network (NN) based constitutive models can capture non‐linear material behaviour. These models are versatile and have the capacity to continuously learn as additional material response data becomes available. NN constitutive models are increasingly used within the finite element (FE) method for the solution of boundary value problems. NN constitutive models, unlike commonly used plasticity models, do not require special integration procedures for implementation in FE analysis. NN constitutive model formulation does not use a material stiffness matrix concept in contrast to the elasto‐plastic matrix central to conventional plasticity based models. This paper addresses numerical implementation issues related to the use of NN constitutive models in FE analysis. A consistent material stiffness matrix is derived for the NN constitutive model that leads to efficient convergence of the FE Newton iterations. The proposed stiffness matrix is general and valid regardless of the material behaviour represented by the NN constitutive model. Two examples demonstrate the performance of the proposed NN constitutive model implementation. Copyright © 2004 John Wiley & Sons, Ltd.     

基于BP神经网络的ETO产品配置设计方法

按订单设计(engineering-to-order, ETO)的定制产品因产品族结构比较复杂,产品间结构差异较大,设计过程涉及个人经验和灵感,并大量应用人机交互处理,难以实现设计自动化、程序化。人工神经网络模仿人脑结构及智能行为,具有大规模并行处理、容错、自组织和自适应能力及联想功能,符合ETO配置设计的特点。通过对ETO定制产品需求的分析,构建并训练具有一定结构和功能的BP神经网络,训练好的网络蕴含着ETO配置设计规则和经验。实例证明了该方法的可行性。     

A neural network approach to microwave imaging

We present a neural network approach to microwave imaging for medical diagnosis. The problem is to reconstruct the complex permittivity of the biological tissues illuminated by the transverse magnetic (TM) incident waves. In order to avoid the inherent ill‐posedness of the inverse scattering problem, we introduce a stochastic process based on Markov random field and a priori knowledge. A coupled gradient neural network is proposed to deal with the mixed‐variable problem because the reconstructed dielectric permittivities are continuous complex variables and the line processes, which can preserve the edges of the reconstructed image, are binary variables. We report the numerical results of a simple human forearm model. We also point out the advantages and the limitations of this method. © 2001 John Wiley & Sons, Inc. Int J Imaging Syst Technol 11, 159–163, 2000     

Neural network approach to sensitivity and uncertainty analysis

Computer simulation of the dynamic evolution of complex systems has become a fundamental tool for many modern engineering activities. In particular, risk-informed design projects and safety analyses require that the system behavior be analyzed under several diverse conditions in the presence of substantial model and parameter uncertainty which must be accounted for. In this paper we investigate the capabilities of artificial neural networks of providing both a first-order sensitivity measure of the importance of the various parameters of a model and a fast, efficient tool for dynamic simulation, to be used in uncertainty analyses. The dynamic simulation of a steam generator is considered as a test-bed to show the potentialities of these tools and to point out the difficulties and crucial issues which typically arise when attempting to establish an efficient neural network structure for sensitivity and uncertainty analyses.     

一类神经网络对连续函数的逼近

借助卷积逼近的工具研究前向神经网络对连续函数的逼近,构造了具有nd个神经元的一类神经网络,并证得用它逼近[0,1]d上的连续函数f(X)时,偏差是O{ω{f,n-d1+2}+n-d1+2‖f‖∞}.其中ω(f,δ)表示f(X)在[0,1]d上的连续模,‖f‖∞表示|f(X)|的极大值.     

Autoprogressive training of neural network constitutive models

A new method, termed autoprogressive training, for training neural networks to learn complex stress–strain behaviour of materials using global load–deflection response measured in a structural test is described. The richness of the constitutive information that is generally implicitly contained in the results of structural tests may in many cases make it possible to train a neural network material model from only a small number of such tests, thus overcoming one of the perceived limitations of a neural network approach to modelling of material behaviour; namely, that a voluminous amount of material test data is required. The method uses the partially-trained neural network in a central way in an iterative non-linear finite element analysis of the test specimen in order to extract approximate, but gradually improving, stress–strain information with which to train the neural network. An example is presented in which a simple neural network constitutive model of a T300/976 graphite/epoxy unidirectional lamina is trained, using the load–deflection response recorded during a destructive compressive test of a [(±45)₆]_S laminated structural plate containing an open hole. The results of a subsequent forward analysis are also presented, in which the trained material model is used to simulate the response of a compressively loaded [(±30)₆]_S structural laminate containing an open hole. Avenues for further improvement of the neural network model are also suggested. The proposed autoprogressive algorithm appears to have wide application in the general area of Non-Destructive Evaluation (NDE) and damage detection. Most NDE experiments can be viewed as structural tests and the proposed methodology can be used to determine certain damage indices, similar to the way in which constitutive models are determined. © 1998 John Wiley & Sons, Ltd.     

Supply chain relationship quality and performance in technological turbulence: an artificial neural network approach

A well-functioning supply chain management relationship cannot only develop seamless coordination with valuable members, but also improve operational efficiency to secure greater market share, increased profits and reduced costs. An accurate decision-making system considering multifactor relationship quality is highly desired. This study offers an alternative perspective and characterisation of the supply chain relationship quality and performance. A decision-making model is proposed with an artificial neural network approach for supply chain continuous performance improvement. Supply chain performance is analysed via a supervised learning back-propagation neural network. An ‘inverse’ neural network model is proposed to predict the supply chain relationship quality conditions. Optimal performance parameters can be obtained using the proposed neural network scheme, providing significant advantages in terms of improved relationship quality. This study demonstrates a new solution with the combination of qualitative and quantitative methods for performance improvement. The overall accuracy rate of the decision-making model is 88.703%. The results indicated that trust has the greatest influence on the supply chain performance. Relationship quality among supply chain partners impacts performance positively as the pace of technological turbulence increases.     

利用波束形成和神经网络进行语音增强

语音增强在语音信号处理的前端非常重要,直接影响后端语音识别等效果。目前用神经网络进行单通道语音分离对于解决鸡尾酒会问题取得了很大的进步,但是用于复杂混合语音时分离效果仍不令人满意。针对单通道情形下的不足,使用多通道结构形成4个方向的超指向波束,结合神经网络算法实现对于指定方向的目标语音增强。仿真和实验结果表明,该算法相较于超指向波束形成算法和谱减法在多种评价指标上均有了明显的提升。     

基于模拟退火改进的神经网络算法在颗粒碰撞阻尼研究中的应用

振动问题的普遍存在,使得对振动的研究极为必要。振动分析中建立的系统阻尼的目标函数往往比较复杂,影响因素众多,很难用传统数学建模的方法建立模型,传统的神经网络分析和模拟也很难得到建模问题的全局最优解。为此,将传统的反向传播算法（BP算法）神经网络模型结合模拟退火算法及最佳保留原则,提出一种改进的神经网络模型,得出了颗粒粒度、颗粒填充率和系统阻尼之间的关系。     

Satellite and satellite subsystems reliability: Statistical data analysis and modeling

Reliability has long been recognized as a critical attribute for space systems. Unfortunately, limited on-orbit failure data and statistical analyses of satellite reliability exist in the literature. To fill this gap, we recently conducted a nonparametric analysis of satellite reliability for 1584 Earth-orbiting satellites launched between January 1990 and October 2008. In this paper, we extend our statistical analysis of satellite reliability and investigate satellite subsystems reliability. Because our dataset is censored, we make extensive use of the Kaplan–Meier estimator for calculating the reliability functions. We derive confidence intervals for the nonparametric reliability results for each subsystem and conduct parametric fits with Weibull distributions using the maximum likelihood estimation (MLE) approach. We finally conduct a comparative analysis of subsystems failure, identifying the “culprit subsystems” that drive satellite unreliability. The results here presented should prove particularly useful to the space industry for example in redesigning subsystem test and screening programs, or providing an empirical basis for redundancy allocation.     

Fuzzy neural network based yield prediction model for semiconductor manufacturing system

Accurate die yield prediction is very useful for improving yield, decreasing cost and maintaining good relationships with customers in the semiconductor manufacturing industry. To improve prediction accuracy of die yield, a novel fuzzy neural networks based yield prediction model is proposed in which the impact factors of yield and critical electrical test parameters are considered simultaneously and are taken as independent variables. The mapping between these independent variables and yield is constructed in the fuzzy neural network (FNN). The lineal regression between FNN-based yield predicting output and actual yield demonstrates the effectiveness of the proposed approach by historical experimental data of semiconductor fabrication line in Shanghai. The comparison experiment verifies the proposed yield prediction method improves on three traditional yield prediction methods with respect to prediction accuracy.     

应用层组播路由的一种改进神经网络解法

研究了使用双层递归神经网络(DRNN)模型求解应用层组播路由的问题.对原有模型的神经元矩阵及能量函数进行改变,并引入了新的线性编程神经元,解决了原模型不能求解组播路由的缺陷.与启发式组播路由算法相比,该解决方案的计算复杂度低,速度较快,而且与其它的神经网络相比,由于引入了基尔霍夫限制条件,保障了解的质量,且所使用的神经元数量少,而在解的精确度上则与其它算法相当.     

六自由度运动平台控制系统的神经网络优化

阐述了六自由度运动平台的控制原理，并根据控制系统的特点，提出采用基于RBF和BP神经网络来改进常规PID控制器实现系统控制性能。在该控制系统结构中，提出了在RBF网络辨识Jacobian的基础上，将BP神经网络引入了平台控制系统中PID控制器的控制参数在线整定的算法，最后给出了在MATLAB下的具体仿真算法。