Pixelated liquid-crystal light valve for neural network application

The liquid-crystal light valve (LCLV) is a useful component for performing integration, thresholding, and gain functions in optical neural networks. Integration of the neural activation channels is implemented by pixelation of the LCLV, with use of a structured metallic layer between the photoconductor and the liquid-crystal layer. Measurements are presented for this type of valve, examples of which were prepared for two specific neural network implementations. The valve fabrication and measurement were carried out at the State Optical Institute, St. Petersburg, Russia, and the modeling and system applications were investigated at the Institute of Microtechnology, Neuchatel, Switzerland.     

Remote sensing image segmentation using a Kalman filter-trained neural network

This article describes the application of a neural network to the segmentation of remote sensing images of multispectral SPOT and fully polarimetric SAR data. The structure of the network is a modified multilayer perceptron and is trained by the Kalman filter theory. The internal activity of the network is a nonlinear function, while the function at output layer is linearized through the use of a polynomial basis function, thus allowing us employ the theory of Kalman filtering as the learning rule. The network is therefore called the dynamic learning (DL) neural network. It is found that, when applied to SPOT and SAR data, the DL neural network gives a good segmentation results, while the learning rate is very promising compared to the standard backpropagation network and other fast-learning networks. In particular, for polarimetric SAR data, optimum polarizations for discriminating between different terrains are automatically built in through the use of the Kalman filter technique. The suitability and effectiveness of the proposed DL neural network to the segmentation of remote sensing images is demonstrated. © 1996 John Wiley & Sons, Inc.     

Use of neural networks and expert systems to control a gas/solid sorption chilling machine: Utilisation des réseaux neuronaux et des systèmes experts pour réguler une machine frigorifique à sorption gaz/solide

This works focuses on using neural networks and expert systems to control a gas/solid sorption chilling machine. In such systems, the cold production changes cyclically with time due to the batchwise operation of the gas/solid reactors. The accurate simulation of the dynamic performance of the chilling machine has proven to be difficult for standard computers when using deterministic models. Additionally, some model parameters dynamically change with the reaction advancement. A new modelling approach is presented here to simulate the performance of such systems using neural networks. The backpropagation learning rule and the sigmoid transfer function have been applied in feedforward, full connected, single hidden layer neural networks. Overall control of this system is divided in three blocks: control of the machine stages, prediction of the machine performance and fault diagnosis.     

J-Resistance curve testing of HY80 steel using SE(B) specimens and normalization method

The normalization method is adopted for standard and nonstandard specimens in this paper to develop J-R curves for HY80 steel directly from load versus load-line displacement records without use of automatic crack length measurement. The standard specimens usually contain high crack-tip constraints, while the nonstandard specimens involve low crack-tip constraints. To obtain J-R curves with different constraints, a series of single edge notched bend (SE(B)) specimens with different crack lengths for an HY80 steel are tested in accordance with ASTM standard E1820. The normalization method is then used for determining crack extension and J-R curves for these SE(B) specimens.To validate the normalization method, the J-R curves determined using the normalization method are compared with those obtained by the elastic unloading compliance method for the SE(B) specimens. The comparison shows that good agreements exist between the two methods, and the normalization method is a viable tool to be used to determine J-R curves of the HY80 steel for the standard as well as nonstandard SE(B) specimens. In the J-integral calculations, the resistance curve test method, the basic test method and the modified basic test method specified in ASTM E1820 are evaluated. The results indicate that the modified basic method can be equivalent to the resistance curve method.     

Weight and neuron output resolution requirements in optoelectronic neural networks

We demonstrate that the resolution requirements of the optoelectronic devices used in the communication links of an analog multiperceptron neural network, trained with the standard backpropagation algorithm, can be simultaneously reduced to 8 bits (receiver) and 4 bits (transmitter), respectively, without any significant effect on the network's learning or generalization performances. In addition, we also show that a simple modification to the sigmoidal function, used within each neuron architecture, permits the resolution requirements of the optoelectronic receiver to be further reduced to 4 bits without any additional effect on network performance other than a reduction in learning rate. Both of these limited device resolution performances, however, can be achieved only provided that the weight-storage and the weight-updating procedures are maintained at 14 bits or greater.     

Fuzzy controlled backpropagation neural network for tool condition monitoring in face milling

The performance of a fuzzy controlled backpropagation neural network has been studied to predict the tool wear in a face milling process based on simple process parameters and sensor signal features. The results show the potentiality of the method in comparison to the standard backpropagation neural network and one of its variants. The speed of convergence, accuracy of prediction and total time of system development make fuzzy controlled backpropagation an attractive technique amenable for online tool condition monitoring.     

A Deep Collocation Method for the Bending Analysis of Kirchhoff Plate

In this paper, a deep collocation method (DCM) for thin plate bending problems is proposed. This method takes advantage of computational graphs and backpropagation algorithms involved in deep learning. Besides, the proposed DCM is based on a feedforward deep neural network (DNN) and differs from most previous applications of deep learning for mechanical problems. First, batches of randomly distributed collocation points are initially generated inside the domain and along the boundaries. A loss function is built with the aim that the governing partial differential equations (PDEs) of Kirchhoff plate bending problems, and the boundary/initial conditions are minimised at those collocation points. A combination of optimizers is adopted in the backpropagation process to minimize the loss function so as to obtain the optimal hyperparameters. In Kirchhoff plate bending problems, the C¹ continuity requirement poses significant difficulties in traditional mesh-based methods. This can be solved by the proposed DCM, which uses a deep neural network to approximate the continuous transversal deflection, and is proved to be suitable to the bending analysis of Kirchhoff plate of various geometries.     

Incorporating pattern recognition capability in a flexible vibratory bowl feeder using a neural network

Programmable parts feeders that can orientate most of the parts of one or more part families, with short changeover times from one part to the next, are highly sought after in batch production. This study investigates a suitable neural-network-based pattern recognition algorithm for the recognition of parts in a programmable vibratory bowl feeder. Three fibre-optic sensors were mounted on a vibratory bowl feeder to scan the surface of each feeding part. The scanned signatures were used as the input for the different neural network models. The performances of ARTMAP, ART2 and backpropagation neural network models were compared. The results showed that, among the three models, ARTMAP is deemed to be superior, based on the criteria of learning speed, high generalization and flexibility. The better performance obtained with the ARTMAP neural network is mainly the result of its online training and supervised learning capabilities.     

Magnetic hysteresis modeling via feed-forward neural networks

A general neural approach to magnetic hysteresis modeling is proposed. The general memory storage properties of systems with rate independent hysteresis are outlined. Thus, it is shown how it is possible to build a neural hysteresis model based on feed-forward neural networks (NN's) which fulfills these properties. The identification of the model consists in training the NN's by usual training algorithms such as backpropagation. Finally, the proposed neural model has been tested by comparing its predictions with experimental data     

旋转机械故障诊断中的主从混合神经网络模型研究

针对ＢＰ网络在旋转机械故障诊断应用中的不足，借助Ｈｏｐｆｉｅｌｄ网络的优良特性，建立了以反馈式Ｈｏｐｆｉｅｌｄ网络为主控网络、前馈式ＢＰ网络为从网络的主从混合神经网络模型。通过这个网络模型的设计、动力学行为分析、学习算法的描述和测试以及它在旋转机械故障诊断中的应用，结果表明：该网络模型具有收敛速度快、稳定性好、最小系统误差等优点，是一种实现旋转机械故障诊断的优良网络模型。     

Proper and effective training of a pattern recognizer for cyclic data

A new approach to training backpropagation neural networks for identifying cyclic patterns on control charts is presented. The objectives of this research are to show that building an effective cyclic-pattern-recognition neural network requires proper training strategies and to demonstrate how these strategies, namely, incremental and decremental training, should be applied and how the performance can be improved with additional statistics. A series of experiments were conducted to study the effect of the number of output pattern classes and the effect of noise on network training and performance. Experiments show that reducing the number of output pattern classes to a small number, e.g., four or fewer, does not guarantee effective learning, and that the noise added to the training data should be maintained at a reasonable level to achieve a balanced performance. Further incorporation of harmonic amplitude statistics (HAS) also proved that the proper use of statistics adopted from Fourier analysis can improve the performance of a cyclic-pattern-recognition neural network. This study offers valuable insights as to how to construct and train a back-propagation neural network properly and effectively for detecting cyclic patterns.     

A novel intelligent strategy for improving measurement precision ofFOG

This paper discusses a neural network-based strategy for reducing the existing errors of fiber-optic gyroscope (FOG). A series-single-layer neural network, which is composed of two single-layer networks in series, is presented for eliminating random noises. This network has simpler architecture, faster learning speed, and better performance compared to conventional backpropagation (BP) networks. Accordingly after considering the characteristics of the power law noise in FOG, an advanced learning algorithm is proposed by using the increments of errors in energy function. Furthermore, a radial basis function (RBF) neural network-based method is also posed to evaluate and compensate the temperature drift of FOG. The orthogonal least squares (OLS) algorithm is applied due to its simplicity, high accuracy, and fast learning speed. The simulation results show that the series-single-layer network (SSLN) with the advanced learning algorithm provides a fast and effective way for eliminating different random noises including stable and unstable noises existing in FOG, and the RBF network-based method offers a powerful and successful procedure for evaluating and compensating the temperature drift     

Recurrent wavelet neural network controller with improved particle swarm optimisation for induction generator system

A recurrent wavelet neural network (RWNN) controller with improved particle swarm optimisation (IPSO) is proposed to control a three-phase induction generator (IG) system for stand-alone power application. First, the indirect field-oriented mechanism is implemented for the control of the IG. Then, an AC/DC power converter and a DC/AC power inverter are developed to convert the electric power generated by a three-phase IG from variable frequency and variable voltage to constant frequency and constant voltage. Moreover, two online trained RWNNs using backpropagation learning algorithm are introduced as the regulating controllers for both the DC-link voltage of the AC/DC power converter and the AC line voltage of the DC/AC power inverter. Furthermore, an IPSO is adopted to adjust the learning rates to further improve the online learning capability of the RWNN. Finally, some experimental results are provided to demonstrate the effectiveness of the proposed IG system.     

扩展双向联想记忆神经网络及其应用

提出了一种采用δ规则作为学习算法的扩展双向联想记忆神经网络模型，并从理论上证明了其稳定性。该模型克服了现有采用Ｈｅｂｂ规则作为学习算法的联想记忆神经网络对记忆模式有正交性要求和所模式吸引域小的不足。实验结果表明，其联想记忆能力优于目前现有的联想记忆网络。     

A neural network-based algorithm that searches for the measuring points of a rule surface

The purpose of this study is to use a proposed neural network-based algorithm to explore the determination of the recommended measuring points for a rule surface. The task of measuring a rule surface starts from the rule surface design blueprint. Mesh grid data on the designed rule surface were selected. The pattern recognition capability of the back-propagation neural network is explored in this article. The network learning was successfully performed by the learning and testing of the network, the support of a designated acceptable perpendicular error value, a learning model in which training examples were gradually added and the adjustment of the number of training examples according to the network structure.     

Variable Selection for the Prediction of C[0,1]-Valued Autoregressive Processes using Reproducing Kernel Hilbert Spaces

A model for the prediction of functional time series is introduced, where observations are assumed to be continuous random functions. We model the dependence of the data with a nonstandard autoregressive structure, motivated in terms of the reproducing kernel Hilbert space (RKHS) generated by the auto-covariance function of the data. The new approach helps to find relevant points of the curves in terms of prediction accuracy. This dimension reduction technique is particularly useful for applications, since the results are usually directly interpretable in terms of the original curves. An empirical study involving real and simulated data is included, which generates competitive results. Supplementary material includes R-code, tables, and mathematical comments.     

Neural network based non-invasive method to detect anemia from images of eye conjunctiva

Detection of anemia can be done by examining the hemoglobin concentration level in the blood using complete blood count, which is an invasive, time-consuming, and costly technique. Preliminary methods for detecting anemia include examining the color of the palpebral conjunctiva, which is a non-invasive method, but color perception may vary from person to person. This study aims to develop a computerized non-invasive technique for anemia detection. We propose a novel machine learning model using the artificial neural network to detect anemic patients from the images of eye conjunctiva. Since limited and small dataset has been used in the earlier approaches, this may cause over fitting of the model. We have improved the number of available training images using image augmentation techniques. To standardize a non-invasive method, we have used computer vision algorithms for preprocessing and feature extraction. This article derives the backpropagation rules mathematically for adjusting the weights for the proposed neural network model. After hyper parameter tuning and using the mathematically derived backpropagation rules, the model was able to achieve the best accuracy of 97.00% with sensitivity 99.21% and specificity 95.42% on the created dataset.     

Robust recurrent neural network modeling for software fault detection and correction prediction

Software fault detection and correction processes are related although different, and they should be studied together. A practical approach is to apply software reliability growth models to model fault detection, and fault correction process is assumed to be a delayed process. On the other hand, the artificial neural networks model, as a data-driven approach, tries to model these two processes together with no assumptions. Specifically, feedforward backpropagation networks have shown their advantages over analytical models in fault number predictions. In this paper, the following approach is explored. First, recurrent neural networks are applied to model these two processes together. Within this framework, a systematic networks configuration approach is developed with genetic algorithm according to the prediction performance. In order to provide robust predictions, an extra factor characterizing the dispersion of prediction repetitions is incorporated into the performance function. Comparisons with feedforward neural networks and analytical models are developed with respect to a real data set.     

AN UNSUPERVISED NEURAL NETWORK APPROACH TO TOOL WEAR IDENTIFICATION

An unsupervised neural network approach is proposed for tool wear identification. Conventional pattern recognition approaches to automating the wear monitoring task are non-adaptive and require expensive or inaccessible information. Rangwala's application of the supervised backpropagation neural network to tool wear identification in a turning operation represented a pioneering effort to integrate sensor signals (cutting force and acoustic emission) and to employ a neural network in the classification of those signals. However, backpropagation also requires expensive training information and cannot remain adaptive after training. The unsupervised adaptive resonance network exhibited the ability to classify sensor signals into fresh and worn classes, to remain adaptive, and to utilize considerably less training information.     

Convergence of higher-order two-state neural networks with modified updating

The Hopfield network is a standard tool for maximizing aquadratic objective function over the discrete set {− 1,1} ⁿ . It is well-known that if a Hopfield network is operated in anasynchronous mode, then the state vector of the network converges to a local maximum of the objective function; if the network is operated in asynchronous mode, then the state vector either converges to a local maximum, or else goes into a limit cycle of length two. In this paper, we examine the behaviour ofhigher-order neural networks, that is, networks used for maximizing objective functions that are not necessarily quadratic. It is shown that one can assume, without loss of generality, that the objective function to be maximized ismultilinear. Three methods are given for updating the state vector of the neural network, called the asynchronous, the best neighbour and the gradient rules, respectively. For Hopfield networks with a quadratic objective function, the asynchronous rule proposed here reduces to the standard asynchronous updating, while the gradient rule reduces to synchronous updating; the best neighbour rule does not appear to have been considered previously. It is shown that both the asynchronous updating rule and the best neighbour rule converge to a local maximum of the objective function within a finite number of time steps. Moreover, under certain conditions, under the best neighbour rule, each global maximum has a nonzero radius of direct attraction; in general, this may not be true of the asynchronous rule. However, the behaviour of the gradient updating rule is not well-understood. For this purpose, amodified gradient updating rule is presented, that incorporates bothtemporal as well as spatial correlations among the neurons. For the modified updating rule, it is shown that, after a finite number of time steps, the network state vector goes into a limit cycle of lengthm, wherem is the degree of the objective function. Ifm = 2, i.e., for quadratic objective functions, the modified updating rule reduces to the synchronous updating rule for Hopfield networks. Hence the results presented here are “true” generalizations of previously known results.