Issues on representational capabilities of artificial neural networks and their implementation

This work is an organized review on the representational capabilities of artificial neural networks and the questions that arise in their implementation. It covers the Kolmogorov's superposition theorem and different statements regarding how it could be related to the representational power of neural networks. Generalization capability of neural networks is then considered and methods of improving this capability are discussed. Some theorems and statements concerning the bound on the number of hidden layers, form of the activation function, and time complexity of training of neural networks are other subjects of this article. © 1995 John Wiley & Sons, Inc.     

Contingency evaluation and monitorization using artificial neural networks

In this paper, different neural network-based solutions to the contingency analysis problem are presented. Contingency analysis is examined from two perspectives: as a functional approximation problem obtaining a numerical evaluation and ranking contingencies; and as a graphical monitoring problem, obtaining an easy visualization system of the relative severity of the contingencies. For the functional evaluation problem, we analyze the use of different supervised feed-forward artificial neural networks (multilayer perceptron and radial basis function networks). The proposed systems produce a very accurate evaluation and ranking, and so present a high applicability. For the graphical monitoring problem, unsupervised artificial neural networks such as self-organizing maps by Kohonen have been used. This solution allows both a rapid, easy and simultaneous visualization of the severity level of the complete contingency set. The proposed solutions avoid the main drawbacks of previous neural network approaches to this problem, which are explicitly analyzed here.     

Prediction of urban stormwater quality using artificial neural networks

There are a vast number of complex, interrelated processes influencing urban stormwater quality. However, the lack of measured fundamental variables prevents the construction of process-based models. Furthermore, hybrid models such as the buildup-washoff models are generally crude simplifications of reality. This has created the need for statistical models, capable of making use of the readily accessible data. In this paper, artificial neural networks (ANN) were used to predict stormwater quality at urbanized catchments located throughout the United States. Five constituents were analysed: chemical oxygen demand (COD), lead (Pb), suspended solids (SS), total Kjeldhal nitrogen (TKN) and total phosphorus (TP). Multiple linear regression equations were initially constructed upon logarithmically transformed data. Input variables were primarily selected using a stepwise regression approach, combined with process knowledge. Variables found significant in the regression models were then used to construct ANN models. Other important network parameters such as learning rate, momentum and the number of hidden nodes were optimized using a trial and error approach. The final ANN models were then compared with the multiple linear regression models. In summary, ANN models were generally less accurate than the regression models and more time consuming to construct. This infers that ANN models are not more applicable than regression models when predicting urban stormwater quality.     

Evolutionary ensemble of diverse artificial neural networks using speciation

Recently, many researchers have designed neural network architectures with evolutionary algorithms but most of them have used only the fittest solution of the last generation. To better exploit information, an ensemble of individuals is a more promising choice because information that is derived from combining a set of classifiers might produce higher accuracy than merely using the information from the best classifier among them. One of the major factors for optimum accuracy is the diversity of the classifier set. In this paper, we present a method of generating diverse evolutionary neural networks through fitness sharing and then combining these networks by the behavior knowledge space method. Fitness sharing that shares resources if the distance between the individuals is smaller than the sharing radius is a representative speciation method, which produces diverse results than standard evolutionary algorithms that converge to only one solution. Especially, the proposed method calculates the distance between the individuals using average output, Pearson correlation and modified Kullback–Leibler entropy to enhance fitness sharing performance. In experiments with Australian credit card assessment, breast cancer, and diabetes in the UCI database, the proposed method performed better than not only the non-speciation method but also better than previously published methods.     

Evaluating of traumatic brain injuries using artificial neural networks

The purpose of this study is to develop a diagnostic system to detect the severity of traumatic brain injuries using artificial neural networks. Three layered back propagation neural network with an input layer of 10 nodes whose output providing the inputs to a hidden layer was used. Thirty-two patients with traumatic brain injuries in different age and gender were taken in the study. Electroencephalography, Trauma and Glasgow coma scores were used for evaluating the data. The results obtained from the system were compared with the findings of neurologists. We found a significant relationship between the findings of neurologists and systems output for normal, mild, moderate and severe electroencephalography tracing data. Getting this system in routine use will lead to make a rapid decision for the degree of trauma with electroencephalography and revised trauma score.     

Prediction of vehicle reliability performance using artificial neural networks

Product development is an important but also dynamic, lengthy and risky phase in the life of a new product. The optimisation of the product development phase through extensive knowledge of the involved procedures is believed to reduce the risks and improve the final product quality. Artificial intelligence and expert systems have been used successfully in optimising the development phase of some new products as it will be demonstrated by the first sections of this publication. This paper presents the first module of an expert system, a neural network architecture that could predict the reliability performance of a vehicle at later stages of its life by using only information from a first inspection after the vehicle’s prototype production. The paper demonstrates how a tool like neural networks can be designed and optimised for use in reliability performance predictions. Also, this paper presents an optimisation methodology that enabled the neural network to deal with the limited amount of available training data, common during new product development, and to finally achieve acceptable prediction performance with small error. A case example is presented to demonstrate the methodology.     

Classification of remotely-sensed image data using artificial neural networks

Abstract

Abstract. Artificial neural networks have been used recently for speech and character recognition. Their application for the classification of remotely-sensed images is reported in this Letter. Remotely sensed image data are usually large in size and spectral overlaps among classes of ground objects are common. This results in low convergence performance of the Back-Propagation Algorithm in a neural network classifier. A Blocked Back-Propagation (BB-P) algorithm was proposed arid described in this Letter. It improved convergence performance and classification accuracy.     

Prediction of flyrock launch velocity using artificial neural networks

Due to a potential to cause damage to machinery and structures and cause injuries to personnel, flyrock is the most dangerous adverse effect of blasting operations. Because of that, it is of primary importance to predict flyrock events and maximum range of flyrock fragments in order to define safety limits and secure the perimeter. There are various models for flyrock range prediction, and most of them rely on proper calculations of flyrock launch velocity. However, a unique and universally applicable model of launch velocity prediction still does not exist. Work presented in this article is a concept of adaptive system application for the prediction of flyrock launch velocities. It shows the principles of input data selection, acquisition and processing and presents the principles of design, training, validation and verification of applied artificial neural network.     

Prediction of sulphur dioxide concentration using artificial neural networks

A three-layer neural network model with a hidden recurrent layer is used to predict sulphur dioxide concentration and the predicted values are compared with the measured concentrations at three sites in Delhi. The Levenberg–Marquardt algorithm is used to train the network. The neural network is used to simulate the behaviour of the system. A multivariate regression model is also used for comparison with the results obtained by using the neural network model. The study results indicate that the neural network is able to give better predictions with less residual mean square error than those given by multivariate regression models.     

Simultaneous process mean and variance monitoring using artificial neural networks

Control chart patterns (CCPs) can be employed to determine the behavior of a process. Hence, CCP recognition is an important issue for an effective process-monitoring system. Artificial neural networks (ANNs) have been applied to CCP recognition tasks and promising results have been obtained. It is well known that mean and variance control charts are usually implemented together and that these two charts are not independent of each other, especially for the individual measurements and moving range (X–R_m) charts. CCPs on the mean and variance charts can be associated independently with different assignable causes when corresponding process knowledge is available. However, ANN-based CCP recognition models for process mean and variance have mostly been developed separately in the literature with the other parameter assumed to be under control. Little attention has been given to the use of ANNs for monitoring the process mean and variance simultaneously. This study presents a real-time ANN-based model for the simultaneous recognition of both mean and variance CCPs. Three most common CCP types, namely shift, trend, and cycle, for both mean and variance are addressed in this work. Both direct data and selected statistical features extracted from the process are employed as the inputs of ANNs. The numerical results obtained using extensive simulation indicate that the proposed model can effectively recognize not only single mean or variance CCPs but also mixed CCPs in which mean and variance CCPs exist concurrently. Empirical comparisons show that the proposed model performs better than existing approaches in detecting mean and variance shifts, while also providing the capability of CCP recognition that is very useful for bringing the process back to the in-control condition. A demonstrative example is provided.     

Comparison of optical performance monitoring techniques using artificial neural networks

In this paper, we make an overview of three techniques that have used artificial neural networks (ANNs) to model impairments in optical fiber. A comparison between a linear partial least squares regression algorithm and ANN is also shown. We demonstrate that nonlinear modeling is required for multi-impairment monitoring in optical fiber when using Parametric Asynchronous Eye Diagram (PAED). Results demonstrating the accuracy of PAED are also shown. A comparison between PAED and Synchronous Eye Diagrams is also demonstrated, for NRZ, RZ and QPSK modulated signals. We show that PAED can provide comprehensible diagrams for QPSK modulated signals, under a certain range of chromatic dispersion.     

Prediction of building energy consumption by using artificial neural networks

In this study, the main objective is to predict buildings energy needs benefitting from orientation, insulation thickness and transparency ratio by using artificial neural networks. A backpropagation neural network has been preferred and the data have been presented to network by being normalized. The numerical applications were carried out with finite difference approach for brick walls with and without insulation of transient state one-dimensional heat conduction. Three different building samples with different form factors (FF) were selected. For each building samples 0–2.5–5–10–15 cm insulations are assumed to be applied. Orientation angles of the samples varied from 0° to 80° and the transparency ratios were chosen as 15–20–25%. A computer program written in FORTRAN was used for the calculations of energy demand and ANN toolbox of MATLAB is used for predictions. As a conclusion; when the calculated values compared with the outputs of the network, it is proven that ANN gives satisfactory results with deviation of 3.43% and successful prediction rate of 94.8–98.5%.     

Implementations of artificial neural networks using current-modepulse width modulation technique

The use of a current-mode pulse width modulation (CM-PWM) technique to implement analog artificial neural networks (ANNs) is presented. This technique can be used to efficiently implement the weighted summation operation (WSO) that are required in the realization of a general ANN. The sigmoidal transformation is inherently performed by the nonlinear transconductance amplifier, which is a key component in the current integrator used in the realization of WSO. The CM-PWM implementation results in a minimum silicon area, and therefore is suitable for very large scale neural systems. Other pronounced features of the CM-PWM implementation are its easy programmability, electronically adjustable gains of neurons, and modular structures. In this paper, all the current-mode CMOS circuits (building blocks) required for the realization of CM-PWM ANNs are presented and simulated. Four modules for modular design of ANNs are introduced. Also, it is shown that the CM-PWM technique is an efficient method for implementing discrete-time cellular neural networks (DT-CNNs). Two application examples are given: a winner-take-all circuit and a connected component detector.     

应用人工神经元网络进行pH中和过程故障诊断

应用NeurOn-Line神经元网络应用系统开发技术和G2实时智能专家系统开发技术,开发了一套pH中和过程的故障诊断系统。先简单描述了该pH中和过程及其建模,然后详细论述了该故障诊断系统在NeurOn-Line和G2软件平台上的设计和编程开发情况。共进行了pH中和过程的正常运行模式,pH传感器测量值偏高、pH传感器测量值偏低和碱液浓度变稀三种故障模式的仿真和诊断。仿真结果表明该故障诊断系统能快速准确诊断出pH中和过程的正常运行和故障模式。     

Probabilistic identification of subsurface gypsum geohazards using artificial neural networks

Subsurface gypsum dissolution hazards imply risks to the construction and operation of new transport infrastructure including subsidence, cavity collapse and cavity flooding. This is a concern in Abu Dhabi, United Arab Emirates, where gypsum geohazards are observed and an extensive transportation network is planned. This paper proposes an artificial neural network (ANN)-based approach for the prediction of underground gypsum. Moreover, the approach is developed to provide the expected probability of gypsum presence and to generate gypsum hazard maps. Such maps provide both a general planning instrument and an input for the decision support systems. An application to Masdar City, Abu Dhabi, is discussed at the site of a planned metro line. Twenty-one boreholes are used to train and validate the ANN that is used to produce a 3D geological model identifying the expected presence of gypsum. Most significantly, the application illustrates how gypsum hazard maps can be obtained at any required depth providing planners and designers with essential information for risk assessment and management.

     

Automated thermal fuse inspection using machine vision and artificial neural networks

Machine vision is an excellent tool for inspecting a variety of items such as textiles, fruit, printed circuit boards, electrical components, labels, integrated circuits, machine tools, etc. This paper presents an intelligent system that incorporates machine vision with artificial intelligent networks to automatically inspect thermal fuses. An effective inspection flow is proposed to detect four commonly seen defects, including black-dot, small-head, bur, and flake during the production of thermal fuses. Backpropagation neural networks and learning vector quantization performance is compared in detecting the bur defect because of its illegibility. Different numbers of defective samples were screened out from a production line in a case study company and used to demonstrate the efficacy of the proposed system. Currently, the proposed inspection system is operating at the case study company, replacing four to six human inspectors. The system not only ensures the quality of the thermal fuses produced, but also reduced the cost of manual visual inspection.     

Detection and characterization of defects using GMR probes and artificial neural networks

This work presents an eddy-current testing system based on a giant magnetoresistive (GMR) sensing device. Non-destructive tests in aluminum plates are applied in order to extract information about possible defects: cracks, holes and other mechanical damages. Eddy-current testing (ECT) presents major benefits such as low cost, high checking speed, robustness and high sensitivity to large classes of defects. Coil based architecture probes or coil-magnetoresistive probes are usually used in ECT. In our application the GMR sensor is used to detect a magnetic field component parallel to a plate surface, when an excitation field perpendicular to the plate is imposed. A neural network processing architecture, including a multilayer perceptron and a competitive neural network, is used to classify defects using the output amplitude of the eddy-current probe (ECP) and its operation frequency. The crack detection, classification and estimation of the geometrical characteristics, for different classes of defects, are described in the paper.     

Real-time assessment of mental workload using psychophysiological measures and artificial neural networks

The functional state of the human operator is critical to optimal system performance. Degraded states of operator functioning can lead to errors and overall suboptimal system performance. Accurate assessment of operator functional state is crucial to the successful implementation of an adaptive aiding system. One method of determining operators' functional state is by monitoring their physiology. In the present study, artificial neural networks using physiological signals were used to continuously monitor, in real time, the functional state of 7 participants while they performed the Multi-Attribute Task Battery with two levels of task difficulty. Six channels of brain electrical activity and eye, heart and respiration measures were evaluated on line. The accuracy of the classifier was determined to test its utility as an on-line measure of operator state. The mean classification accuracies were 85%, 82%, and 86% for the baseline, low task difficulty, and high task difficulty conditions, respectively. The high levels of accuracy suggest that these procedures can be used to provide accurate estimates of operator functional state that can be used to provide adaptive aiding. The relative contribution of each of the 43 psychophysiological features was also determined. Actual or potential applications of this research include test and evaluation and adaptive aiding implementation.     

Classification of modulation signals using statistical signal characterization and artificial neural networks

Modulation recognition systems have to be able to correctly classify the incoming signal modulation scheme in the presence of noise. A new method for classification of analogue and digital modulated signals at low signal-to-noise ratio (SNR) is introduced in this paper. This method uses the statistical signal characterization (SSC) to extract parameters to classify the different modulation signals. The SSC technique produces a set of four numerical parameters for a specific modulated signal. Subsequent comparison of these parameters to those of other waveforms provides the basis for our classification system. The results of SSC technique are applied to an artificial neural network (ANN) to have a robust classification system in the presence of noise down to SNR of 3 dB. No a priori information is required by this technique about the set of input waveforms. The input to the classification system can be analogue or digital signals or a combination of both. The proposed technique shows a 100% efficiency of classification of analogue signals or digital signals at SNR of 7 dB. This classification efficiency reduces to 83% and 86% for analogue or digital signals at SNR of 3 dB. The SSC technique shows better classification results in comparison with other techniques with an important advantage over other methods, which is the simplicity of the neural network needed with this technique due to the small number of features used in the classification.     

Analysis of international debt problem using artificial neural networks and statistical methods

It is known from the scientific researches that artificial neural networks are alternatives of statistical methods such as regression analysis and classification in recent years. Since multi-layer backpropagation neural network models are nonlinear, it is expected that the neural network models should make better classifications and predictions. The studies on this subject support that idea. In this study, a macro-economic problem on rescheduling or non-rescheduling of the countries’ international debts is taken into account. Among the statistical methods, logistic and probit regression, and the different neural network backpropagation algorithms are applied and comparisons are made. Evaluations and suggestions are made depending on the results and different neural network architecture.