(Return to ICNN97 Homepage) (Return to ICNN'97 Agenda)
ICNN'97 FINAL ABSTRACTS
ICNN97 Applications Session: AP1A Paper Number: 506 Oral
Adaptive critic design in learning to play game of Go.
Roanak Zaman, Danil Prokhorov, Donald C. Wunsch II
Keywords: Machine learning game go
Abstract: This paper examines the performance of an HDP-type adaptive critic design (ACD) of the game Go. The game Go is an ideal problem domain for exploring machine learning; it has simple rules but requires complex strategies to play well. All current commercial Go programs are knowledge based implementations; they utilize input feature and pattern matching along with minimax type search techniques. But the extremely high branching factor puts a limit on their capabilities, and they are very weak compared to the relative strengths of other game program like chess. In this paper, the Go-playing ACD consists of a critic network and an action network. The HDP type critic network learns to predict cumulative utility function of the current board position from training games, and, the action network chooses a next move which maximizes critics' next step cost-to-go. After about 6000 different training games against a public domain program, WALLY [1], the network (playing WHITE) began to win in some of the games and showed slow but steady improvements on test games.
_____
ICNN97 Applications Session: AP1B Paper Number: 88 Oral
Reduced-Order Functional Link Neural Network for HVAC Thermal System Identification and Modeling
Mo-yuen Chow and Jason Teeter
Keywords: System Identification Modeling Thermal System Intelligent Control
Abstract:
The use of computers for direct digital control highlights the recent trend toward more effective and efficient HVAC control methodologies. Researchers in the HVAC field have stressed the importance of self-learning in building control systems and the integration of optimal control and other advanced techniques into the formulation of such systems.
This paper describes a functional link neural network approach to perform the HVAC thermal system identification and modeling. Artificial neural networks are used to emulate the plant dynamics in order to estimate future plant outputs and obtain plant input/output sensitivity information for on-line neural control adaptation. Methodologies to appropriately reduce the inputs, thus the complexity, of the functional link network in order to speed up the training will be presented. This paper will also analyze and compare the performance and complexity between the functional link network and conventional network approaches for the HVAC thermal system identification and modeling.
Keywords : Neural network, system identification, modeling, functional link, HVAC, thermal system, intelligent control.
_____
ICNN97 Applications Session: AP1C Paper Number: 240 Oral
Cascade-CMAC Neural Network Applications on the Color scanner to printer calibration
King-Lung Huang, Shu-Cheng Hsieh, and Hsin-Chia Fu
Keywords: color calibration learning
Abstract:
This paper presents an application of using a Cascade-CMAC (Cerebellar Model Articulation Controller) neural network to solve some color calibration problems , which include color differences induced from gamuts mis-match and the non- linear transformation characteristics between color scanning input devices and color printing output devices. For this purpose, we proposed a scalable learning architecture "Cascade-CMAC " to implement an adaptive color calibration system. By analyzing the preliminary learning situation, the scalable architecture can dynamically create a new learning unit to better represent a finer color resolution, so that the learning capacity as well as the color details of the system can be greatly improved. From the experimental results, the proposed Cascade-CMAC architecture can improve the rate of convergence and also can adjust the learning architecture effectively.
The learning speed can be 2~ 4 times faster than the conventional CMAC. The effectiveness of this neural network has been tested by observing the differences between the calibrated and the un-calibrated output on a number of known samples. By using Macbeth Color-Checker which contains 24 color ptaches as benchmark, the average color differences between the original and the calibrated print-out is improved from 15 \delta E_ab to \delta E_ab under the 3 \delta E_ab convergent criterion for training. The calibration performance is somewhat significant.
_____
ICNN97 Applications Session: AP1D Paper Number: 662 Oral
Using neural networks to automatically refine expert system knoweledge-bases: Experiments in the NYNEX MAX domain
David W. Opitz, Mark W. Craven, Jude W. Shavlik
Keywords: machine learning Fault diagnosing
Abstract:
In this paper we describe our study of applying knowledge-based neural networks to the problem of diagnosing faults in local telephone loops. Currently, NYNEX uses an expert system called MAX to aid human experts in diagnosing these faults; however, having an effective learning algorithm in place of MAX would allow easy portability between different maintenance centers, and easy updating when the phone equipment changes. We find that (i) machine learning algorithms have better accuracy than MAX, (ii) neural networks perform better than decision trees, (iii) neural network ensembles perform better than standard neural networks, (iv) knowledge-based neural networks perform better than standard neural networks, and (v) an ensemble of knowledge-based neural networks performs the best.
_____
ICNN97 Applications Session: AP1E Paper Number: 144 Oral
Option pricing with genetic algorithms: an alternative to neural networks
Shu-Heng Chen and Woh-Chiang Lee
Keywords: Option pricing genetic algorithms artificial neural networks European call option
Abstract:
_____
ICNN97 Applications Session: AP2A Paper Number: 56 Oral
Ultrashort Laser Pulse Characterization by Neural Network
Martin Searcy, Donald Cooley, Rick Trebino, Marco Krumbugel
Keywords: Laser Pulse FROG Wavelet Transform
Abstract:
The electric field associated with spectrograms generated by Frequency- Resolved Optical Gating (FROG) of ultrashort laser pulses can be recovered through an iterative computational process. The process, however, is limited in application by its long compute time. Training a neural network to recognize features in the spectrograms, or FROG traces, gives a more direct, or instantaneous, solution of the electric fields.
This paper describes a study of an original method of compact FROG trace feature description for neural network training. The method consists of performing a wavelet transform on each trace, and then describing groups of meaningful wavelet coefficients in each wavelet order through statistical moments in three dimensions. Experimental results demonstrate that this approach of using a wavelet transform as a basis for training a neural network on large low-feature FROG images is quite successful in terms of standard recognition error estimates.
_____
ICNN97 Applications Session: AP2B Paper Number: 337 Oral
Surface identification by Acoustic reflection characteristics using time delay spectrometry and artificial neural network
Pubudu N Pathirana, Anthony Zaknich
Keywords: Surface identification Spectrometry
Abstract:
Abstract - The identification of surfaces using incident sound waves is associated with a variety of different applications including, sonar, seabed scanning and medical ultrasound imaging. The biologically innocuous nature, applicability, and simplicity involved in generation and measurement, makes sound inherently a more attractive agent for most applications. Time delay spectrometry can be employed as a way of isolating a desired reflected signal from other reflections dramatically increasing the signal to noise ratio of the receiver of a neural network based classification system. A surface classification system with the analysis of its performance will be introduced in this paper as a successful implementation of the proposed methodology.
_____
ICNN97 Applications Session: AP2C Paper Number: 648 Oral
BP Network for partial discharge analysis and dielectrics classification
Riccardo Bozzo, Luca Sciutto and Rodolfo Zunino
Keywords: Feature reduction Classification
Abstract:
A neural network (NN) approach is used to tackle the problem of discriminating different dielectric materials from the analysis of electrical data recorded during high voltage tests. While the general approach to the problem (Partial Discharge analysis) is well consolidated, there is neither agreement to the network architecture that best fits the problem nor to the inputs of such a NN.
In this work we shall demonstrate that the MLP structure can help to select a reduced set of the input features and to successfully classify between two similar materials.
_____
ICNN97 Applications Session: AP2D Paper Number: 644 Oral
A fuzzy-neural approach to real time plasma boundary reconstruction in tokamak reactors
Fransesco Carlo Morabito and M. Versaci
Keywords: fuzzy systems plasma shape identification
Abstract:
A novel plasma shape identification procedure based on both Neural Network (NN) and Fuzzy System (FS) approaches is presented. The derived processor takes firstly advantage of the Fuzzy Curves (FCs) concept for carrying out a guided dimensionality reduction of the available pattern of measurements (inputs of the reconstruction procedure). The mapping between the input pattern and the corresponding set of parameters describing the plasma (outputs of the procedure) is then approximated by a FS whose bank of rules is directly extracted from the relevant FCs. An extremely simplified NN is finally trained to learn the estimation error of the above mentioned fuzzy block. One of the most relevant consequences of the analysis carried out on the first phase is the possibility of using a very limited number of measurements for correcting the mapping. This may have an impact on both the rapidity of the identification in real time and the reduction of the number of sensors required to achieve a prescribed accuracy for control. As a by-product, the designed Fuzzy-Neural System reduces the extrapolation problems typically encountered by the researchers when using simple NN models.
_____
ICNN97 Applications Session: AP2E Paper Number: 256 Oral
A hopfield neural network for flow field computation based on particle image velocimetry/particle tracking velocimetry image sequences
Knaak, M; Rothluebbers, C; Orglmeister, R
Keywords: Hopfield network particle tracking Velocimetry flow fields
Abstract:
A new application of a Hopfield network for the detection of particle pairs in Particle Image Velocimetry/ Particle Tracking Velocimetry (PIV/ PTV) is described. PIV/PTV are the most advanced techniques for the examination of flow fields. Our aims are to apply these techniques to fluid mechanics and the investigation of hydraulic turbomachinery. To obtain correct particle correspondences in subsequent images, a specific cost function is defined and then mapped onto a two-dimensional Hopfield network. First investigations show better performance than conventional techniques for PTV/PIV. In comparison to conventional nearest neighbor techniques, the number of correct particle pairs detected significantly increases, whereas the number of mismatches decreases.
_____
ICNN97 Applications Session: AP2F Paper Number: 573 Oral
An artificial neuron network for naval theater ballistic missile defense program
Tai-Ching Chu, Harold Szu
Keywords: Ballistic Missile
This paper discusses the possible relationship between an Artificial Neural Network (ANN) and the Naval Theater Ballistic Missile Defense (TBMD) program, and potential applications of ANN to TBMD system. A successful marriage between ANN and TBMD program would result in more effective and efficient warfare planning, training, and operational effectiveness.
_____
ICNN97 Applications Session: AP3A Paper Number: 469 Oral
Using a neural network to Diagnosing anterior wall myocardial infarction
Ning Ouyang, Mitsuru Ikeda, Kazunobu Yamauchi
Keywords: Medical diagnosis ECG data
Abstract:
The purpose of this study is to evaluate the usefulness of a feed-forward neural network for diagnosing anterior wall myocardial Infarction (AI).
Data used in the study are 165 ECG records diagnosed as old AI by the commercially available computer-assisted ECG interpretation system, but only 80 of 165 cases have been proved to suffer from AI, the other 85 cases have been proved without AI. The training set is composed of 40 ECG randomly selected from the 80 AI cases and 42 ECG randomly selected from the 85 cases without AI. The performance of the network was tested with the remaining 83 ECG data; 40 with AI and 43 without it. The testing data had not been exposed to the training network. The network correctly diagnosed 34 of the 40 cases with AI and 39 of the 43 cases without AI.
The sensitivity and the specificity were 85% and 90.7%, respectively, and the diagnostic accuracy rate was 87.9%. The good diagnostic accuracy rate revealed the network has the potential to improve computer-assisted interpretation of ECG.
_____
ICNN97 Applications Session: AP3B Paper Number: 39 Oral
Application of Fuzzy Neural Network to ECG Diagnosis
Xie Zhi-xing Xie Han-zhong Ning Xing-bao
Keywords: ECG Signals Fuzzy Neural Network Diagnosis System Fuzzy Model
Abstract:
_____
ICNN97 Applications Session: AP3C Paper Number: 175 Oral
Noninvasive diagnosis of delayed gastric emptying from the cutaneous electrogastrogram using neural networks
Zhiyue Lin Richard W. McCallum, Jian De Z Chen
Keywords: Gastric Emptying Delayed Emptying Electrogastrogram
Abstract:
The currently established gastric emptying test requires the patient to take a radio-active test meal and to stay under a gamma camera for acquiring abdominal images for 2 hours. It is invasive and expensive. Since the electrogastrogram (EGG) is a cutaneous recording of gastric myoelectrical activity which modulates gastric motor activity, we hypothesized that delayed gastric emptying might be predicted from the EGG using a neural network approach. In this study, simultaneous recordings of the EGG and the emptying rate of the stomach by means of the established method were made in 152 patients with suspected gastric motility disorders. A multilayer feedforward neural network approach for the diagnosis of delayed gastric emptying from the noninvasive EGG was developed. Using 5 spectral parameters of the EGG as inputs, a correct classification of 85% was achieved with an optimized three-layer network. This study indicates that the neural network approach is a potentially useful tool for the noninvasive diagnosis of delayed gastric emptying.
_____
ICNN97 Applications Session: AP3D Paper Number: 625 Oral
Predictive Medicine: Initial symptoms may determine outcome in clinically treated depressions
Joannes S. Luciano, Michael A. Cohen, Jacqueline A. Samson
Keywords: Prediction Nonlinear modeling
Abstract:
_____
ICNN97 Applications Session: AP3E Paper Number: 553 Oral
Locating Anatomical Landmarks for prosthetics design using ensemble neural networks
Daniel Jimenez, Thomas Darm, Bill Rogers, Nicolas Walsh
Keywords: Prosthetics Design Anatomical landmarks
Abstract:
Computer aided design of a prosthesis for a below-the-knee (trans-tibial) amputee begins with a digitized representation of the shape of the residual limb. Certain anatomical landmarks must be located on this shape to identify optimal areas for load and pressure relief. A method of locating the midpoint of the patellar tendon, the distal end of the tibia and the head of the fibula is presented. The method involves training ensembles of neural networks on shapes for which the markers have been located manually; the neural networks are then used to find the landmarks for arbitrary shapes. Experimental results show that the method is at least as accurate as a trained prosthetist.
_____
ICNN97 Applications Session: AP3F Paper Number: 214 Oral
Motif neural network design for large-scale protein family identification
Cathy H. Wu, Sheng Zhao, Kevin Simmons , and Sailaja Shivakumar
Keywords: molecular databases motif identification protein family identification
Abstract:
This paper describes an application of artificial neural networks for the sequence analysis and management of the large and rapidly growing molecular databases. The neural network system, based on a motif identification neural design (MOTIFIND) that incorporates both global and motif sequence information, has been implemented for large- scale protein family identification. More than nine hundred backpropagation networks were trained, one for each protein family. The protein families were defined collectively by the ProSite and PIR databases. As a part of an integrated protein family identification system, the neural networks were used as filters to quickly detect potential new members in comprehensive searches against the two major protein sequence databases, SwissProt and PIR. The integrated system identified a large number of false negative members missed by both ProSite and PIR. The speed, sensitivity, general applicability, together with its capability to learn existing protein classification schemes, make the MOTIFIND neural network system an ideal tool for the full-scale protein family classification effort. The system is available from our WWW on-line server (http://diana.uthct.edu) for identification of query sequences.
_____
ICNN97 Applications Session: AP4A Paper Number: 638 Oral
Identifying Disordered Regions in Proteins from Amino Acid sequence
P. Romero, Z. Obradovic, C. Kissinger, J. E. Willafranca, A. K. Dunker
Keywords: prediction rule-based prediction
Abstract:
A rule-based and several neural network predictors are developed for identifying disordered regions in proteins. The rule-based predictor, which relied on the observation that disordered regions contain few aromatic amino acids, was suitable only for very long disordered regions, whereas the neural network predictors were developed separately for short-, medium-, and long-disordered regions (S-, M-, and LDRs, respectively). The out-of-sample prediction accuracies on a residue-by-residue basis ranged from 69 to 74% for the neural network predictors when applied to the same length class, but fell to 59 to 67% when applied to different length classes.
Testing the rule-based predictor on a residue-by-residue basis using out-of-sample LDRs gave a success rate of 70%.
Application of both the rule-based and LDR neural network predictors to large databases of protein sequences provide strong evidence that disordered regions are very common in nature. These results are consistent with our recent proposal that disordered regions are crucial for the evolution of molecular recognition.
_____
ICNN97 Applications Session: AP4B Paper Number: 195 Oral
Categorizing Web Documents Using competitive learning: An Ingredient of a personal adaptive agent
I. Khan, D. Blight, R. D. McLeod, and H. C. Card
Keywords: competitive learning Web Documents document clustering
Abstract:
This paper describes the application of competitive learning to categorize Web documents, which is one component of our Personal Adaptive Web (PAW) agent.
The PAW agent achieves its delegated task by performing the following seven subtasks:
1) Monitor the user while she is browsing the Web 2) Determine which of the visited documents are relevant 3) Textually analyze these relevant documents and reduce each into a document vector 4) Classify the document vectors into categories using unsupervised competitive learning 5) Scan the Web for new similar documents 6) Produce document vectors for the new documents and classify them using the trained neural network 7) Decide whether to suggest the new documents to the user
We discuss the framework of the PAW agent, the implementation of the document clustering (subtasks 3 to 6), and current results. This work therefore suggests a new application, rather than an improved learning algorithm.
_____
ICNN97 Applications Session: AP4C Paper Number: 193 Oral
A Comparison of backpropagation and statistical classifiers for bird identification
A. L. McIlraith and H. C. Card
Keywords: Backpropagation methods Statistical classifiers Bird identification
Abstract:
We compare neural networks and statistical methods used to identify birds by their songs. Six birds native to Manitoba were chosen which exhibited overlapping characteristics in terms of frequency content, song components and length of songs. Songs from multiple individuals in each species were employed. These songs were analyzed using backpropagation learning in two-layer perceptrons, as well as methods from multivariate statistics including quadratic discriminant analysis. Preprocessing methods included linear predictive coding and windowed Fourier transforms. Generalization performance ranged from 82% to 93% correct identification, with the lower figures corresponding to smaller networks that employed more preprocessing for dimensionality reduction.
Computational requirements were significantly reduced in the later case.
_____
ICNN97 Applications Session: AP4D Paper Number: 38 Oral
The Backpropagation Algorithm Applied to Protective Relaying
Denis V. Coury David C. Jorge
Keywords: Distance Protection Pattern Classifier Protective Relaying
Abstract:
Distance relays have attracted considerable attention for the protection of transmission lines. They are usually designed on the basis of fixed settings. Therefore, the reach of such relays is affected by the changing network conditions. The implementation of a pattern recognizer for power system diagnosis can provide great advances in the protection field. This paper demonstrates the use of an Artificial Neural Network as a pattern classifier for a distance relay operation. The backpropagation algorithm is utilized for the learning process. The scheme utilizes the magnitudes of three phase voltage and current phasors as inputs. An improved performance with the use of an Artificial Neural Network approach is experienced once the relay can operate correctly, keeping the reach when faced with different fault conditions as well as network configuation changes.
Keywords: Distance Protection, Backpropagation, Relaying, Power Systems, Neural Networks
_____
ICNN97 Applications Session: AP5A Paper Number: 216 Oral
On-line prediction of polymer product quality in an industrial reactor using recurrent neural networks
Randall S. Barton and David M. Himmelblau
Keywords: Recurrent Neural Networks Prediction Product quality
Abstract:
In this paper, Internally Recurrent Neural Networks (IRN) are used to predict a key polymer product quality variable
from an industrial polymerization reactor. IRN are selected as the modeling tools for two reasons: 1) over the wide range of operating regions required to make multiple polymer grades, the process is highly nonlinear, and 2) the finishing of the polymer product after it leaves the reactor imparts significant dynamics to the process by ''mixing'' effects. IRN are shown to be very effective tools for predicting key polymer quality variables from secondary measurements taken around the reactor.
_____
ICNN97 Applications Session: AP5B Paper Number: 502 Oral
Neural network for wind power generation with compressing function
Shuhui Li, Don C. Wunsch, Edgar O'Hair, Michael G. Giesselmann
Keywords: wind power
Abstract:
The power generated by electric wind turbines changes rapidly because of the continuous fluctuation of wind speed and direction. It is important for the power industry to have the capability to estimate this changing power. In this paper, the characteristics of wind power generation are studied and a neural network is used to estimate it. We use real wind farm data to demonstrate a neural network solution for this problem, and show that the network can estimate power even in changing wind conditions.
_____
ICNN97 Applications Session: AP5C Paper Number: 405 Oral
Applying the connectionist inductive learning and logic programming system to power system diagnosis
Artur S. d'Avila Garcez, Gerson Zaverucha and Victor Navarro A. L. da Silva
Keywords: Hybrid Systems Machine learning power systems
Abstract:
The Connectionist Inductive Learning and Logic Programming System, C-IL2P, integrates the symbolic and connectionist paradigms of Artificial Intelligence through neural networks that perform massively parallel Logic Programming and inductive learning from examples and background knowledge. This work presents an extension of C-IL2P that allows the implementation of Extended Logic Programs in Neural Networks. This extension makes C-IL2P applicable to problems where the background knowledge is represented in a Default Logic. As a case example, we have applied the system for fault diagnosis of a simplified power system generation plant, obtaining good preliminary results.
Keywords: Hybrid Systems, Machine Learning, Neural Networks, Extended Logic Programming, Power Systems.
_____
ICNN97 Applications Session: AP5D Paper Number: 398 Oral
Modeling of the Proofing Process in a Continuos Bread-making Industrial Plant using neural networks
A. Ferroni, A. Ficola, M. La gava, M. Fravolini
Keywords: Model identification System identification
Abstract:
_____
ICNN97 Applications Session: AP6A Paper Number: 249 Oral
A Neural network Driven Solution to a Channel Assignment Problem in Wireless Telephony
D. Tissainayagam, D. Everitt, M. Palaniswami
Keywords: wireless telephony channel assignment algorithms
Abstract:
The introduction of artificial neural networks to real-time channel assignment techniques in cellular mobile radio systems has met with varying degrees of success. We have translated a dynamic channel assignment algorithm into a combinatorial optimization problem which can be solved on a suitably modified Hopfield neural network.
Here we prove that the conventional and our neural network based approaches are equivalent.
This neural architecture can be further extended to include a more powerful algorithm that incorporates channel rearrangements between cells.
Extensive simulations show that our approach outperforms some other neural model s considered elsewhere in the literature.
_____
ICNN97 Applications Session: AP6B Paper Number: 219 Oral
Signal estimation with neural networks for multipath mobile communication
Thomas L. Hemminger
Keywords: Signal estimation multipath fading wireless communication
Abstract:
This paper deals with signal estimation in multipath fading channels.
Multipath fading is a significant problem in wireless communications systems and may occur whenever there is more than one path from a transmitter to the intended receiver. One method of addressing this task is to implement a RAKE receiver. However, this requires estimates of tap weights based on characteristics of the channel model. Within limited environments the multipath characteristics of a channel may exhibit little variation over time. Under these conditions a neural network can be employed to learn the tap weights. This paper illustrates the use of neural networks in solving this problem and presents results from simulations.
Keywords: Mobile Communications, Fading Channels, RAKE Reciever
_____
ICNN97 Applications Session: AP6C Paper Number: 237 Oral
A decision feedback recurrent neural equalizer for digital communication
Sunghwan Ong, Sooyong Choi, Cheolwoo Yoo, and Daesik Hong
Keywords: Digital communication Equalization recurrent Neural equalizer
Abstract:
In this paper, we introduce an adaptive Decision Feedback Recurrent Neural Equalizer (DFRNE) whose small size and high performance makes it suitable for high-speed channel equalization. By evaluating its per-formance through computer simulations for various channels, the DFRNE has comparable performance with traditional equalizers when the channel interferences are mild. And it outperforms them when the channel*s trans-fer function has spectral nulls or when severe nonlinear distortion is present. In addition, the DFRNE, being es-sentially an IIR filter, is shown to outperform multi-layer perceptron equalizers in linear and non-linear channel equalization cases.
_____
ICNN97 Applications Session: AP6D Paper Number: 177 Oral
A Complex Pi-sigma network and its application to equalization of nonlinear satellite channels
Y. Shin, K.-S. Jin and B.-M. Yoon
Keywords: Satellite Communication Pi-sigma network
Abstract:
Digital satellite communication channels have a nonlinearity with memory due to saturation characteristics of the high power amplifier in the satellite and transmitter/receiver linear filters used in the overall system. In this paper, we propose a network structure and a learning algorithm for complex pi-sigma network (CPSN) and exploit CPSN in the problem of equalization of nonlinear satellite channels. The proposed CPSN is a complex-valued extension of real-valued pi-sigma network (PSN) that is a higher-order feedforward network with fast learning while greatly reducing network complexity by utilizing efficient form of polynomials for many input variables. The performance of the proposed CPSN is demonstrated by computer simulation on the equalization of complex-valued QPSK input symbols distorted by a nonlinear channel modeled as a Volterra series and additive noise. The results indicate that the CPSN shows good equalization performance, fast convergence, and a lot less computations as compared to conventional higher-order neural networks such as Volterra filters.
_____
ICNN97 Applications Session: AP6E Paper Number: 85 Oral
Neural Network-Based Real-Time Dynamic Nonhierarchical Network Routing for B-ISDN
Hua Nan, Zemin Liu
Keywords: Network Routing Hopfield Network Optimization
Abstract:
_____
ICNN97 Applications Session: APP2 Paper Number: 322 Poster
Defect Prediction for reactive Ion Etching using neural networks
D. Stokes, G. May, V. Chen and Y. T. Lin
Keywords: defect prediction reactive ion etching particle monitoring
Abstract:
As geometries in integrated circuits continue to decrease, the elimination of submicron defects caused by particles generated within semiconductor processes becomes more and more critical. These particles can cause surface defects which lead to reduced yield. While 100% inspection of processed wafers during fabrication provides the most accurate means for detecting these anomalies, it is also very time-consuming and costly. This cost can be mitigated through the use of automated in-situ particle monitoring systems (ISPMs) which provided real-time estimates of particle counts in process chambers for different categories of particle sizes. However, the challenge is to correlate ISPM measurements with actual surface defects. In this paper, neural network models are used to estimate the number of particles that are deposited on a semiconductor wafer based on ISPM data collected during processing in a reactive ion etching (RIE) chamber. This particle prediction methodology can lead to reduced testing costs and more accurate defect detection.
_____
ICNN97 Applications Session: APP2 Paper Number: 543 Poster
Computer access authentication with neural network based keystroke identity verification
Daw-tung Lin
Keywords: COmputer access security idendity authentication backpropagation
Abstract:
This paper presents a novel application of neural net to user identity authentication on computer-access security system. Keystroke latency is measured for each user and form the patterns of keyboard dynamics. A three-layered backpropagation neural network with flexible number of input nodes was used to discriminate valid users and impostors according each individual's password keystroke pattern. System verification performance was improved by setting convergence criteria RMSE to a smaller threshold value during training procedure. The resulting system gave an 1.1 FAR (false alarm rate) in rejecting valid users and zero IPR (impostor pass rate) in accepting no impostors. The performance of the proposed identification method is superior to that of previous studies. A suitable network structure for this application was also discussed. Furthermore, the implementation of this approach requires no special hardware and is easy to be integrated with most computer systems.
_____
ICNN97 Applications Session: APP2 Paper Number: 235 Poster
An Artificial Neural Network based real-time fault locator for transmission line
Zhihong Chen, Jean-Claud Maun
Keywords: Fault locator Transmission line
Abstract:
This paper describes the application of an artificial neural network-based algorithm to the single-ended fault location of transmission lines using voltage and current data. From the fault location equations, similar to the conventional approach, this method selects phasors of prefault and superimposed voltages and currents from all phases of the transmission line as inputs of the artificial neural network. The outputs of the neural network are the fault position and the fault resistance. With its function approximation ability, the neural network is trained to map the nonlinear relationship existing in the fault location equations with the distributed parameter line model. It can get both fast speed and high accuracy. The influence of the remote-end infeed on neural network structure is studied. A comparison with the conventional method has been done. It is shown that the neural network-based method can adapt itself to big variations of source impedances at the remote terminal.
Finally, when the remote source impedances vary in small ranges, the structure of artificial neural network has been optimized by the pruning method.
_____
ICNN97 Applications Session: APP2 Paper Number: 589 Poster
Neural Networks on Fatigue damage prediction
Tiago A. Piedras Lopes
Keywords: fatigue damage Offshore structure
Abstract:
_____
ICNN97 Applications Session: APP2 Paper Number: 106 Poster
Prediction of Products Quality Parameters of a Crude Fractionation Section of an Oil Refinery Using Neural Networks
K. Bawazeer , Ali Zilouchian
Keywords: Inferential Analysis Prediction Backpropagation
Abstract:
Inferential analysis using neural network technology is proposed for an existing crude fractionation section of an oil refinery. Plant data for a three month operation period is analyzed in order to construct various neural network models using backpropagation algorithm. The proposed neural networks can predict various properties associated with crude oil productions. The simulation results for modeling Naphtha 95% cut point and Naphtha Reid vapor pressure properties are analyzed. The results of the proposed work can ultimately enhance the on-line prediction of crude oil product quality parameters for crude fractionation processes.
_____
ICNN97 Applications Session: APP2 Paper Number: 483 Poster
An on-line distribution feeder optimal reconfiguration algorithm for resistive loss reduction using a Multi-layer perceptron
E. Gauche J. Coelho, R. C. G. Teive
Keywords: Optimization network reconfiguration loss minimization
Abstract:
This paper presents an on-line distribution feeder optimal reconfiguration algorithm for resistive loss reduction. Artificial neural networks (ANN) were used to assure the application feasibility in real-time. The demand variation used during the ANN training is represented by samplings via Monte Carlo Simulation. A consolidated heuristic algorithm is utilized to obtain the demand topologies. An integer formulation 0-1 is used to guarantee the solution optimality from the initial solution supplied by the ANN. It is also presented the application results to a demonstrative test system, indicating to new applications in real systems where topological alteration are required.