Search

Raster map should be corrected after scanned because of the errors caused by paper map deformation. In the paper, the deficiency of the polynomial fitting method is analyzed. The paper introduces an ANN (Artificial Neural Network) correcting method that utilizes the advantage of its function approximation ability. In the paper, two types of ANNs, BP and GRNN, are designed for the correcting. The comparing experiment is done with the same data by the polynomial fitting and ANN methods, utilizing the MALAB. The experiment results show that the ANN methods, especially the GRNN method, performances far better than the polynomial fitting method does. [ABSTRACT FROM AUTHOR]

This paper analyses the chaotic characteristics of violent abnormal vibration signals of a large rolling mill, and studies phase space reconstruction techniques of the signals. On this basis, the vibration model of wavelet neural networks and the model of backpropagation neural networks are set up, respectively, through inversion methods. The properties of these two models are tested and compared with each other. The result shows that the wavelet neural networks have an advantage over the backpropagation neural networks in rapid convergence and high accuracy. [ABSTRACT FROM AUTHOR]

In this paper, a Genetic-Algorithm-based Artificial Neural Network (GAANN) model for short-term traffic flow forecasting is proposed. GAANN can integrate capabilities of approximation of Artificial Neural Networks (ANN) and of global optimization of Genetic Algorithms (GA) so that the hybrid model can enhance capability of generalization and prediction accuracy, theoretically. With this model, both the number of hidden nodes and connection weights matrix in ANN are optimized using genetic operation. The real data sets are applied to the introduced method and the results are discussed and compared with the traditional Back Propagation (BP) neural network, showing the feasibility and validity of the proposed approach. [ABSTRACT FROM AUTHOR]

The objective of the present paper is to develop an artificial intelligent system for heatsink parameter design and performance evaluation. The system is combined both the artificial neural network and the computational fluid dynamics (CFD) techniques. The algorithm of the artificial neural network used in this paper is the Back-propagation Neural Network (BNN), which makes the intelligent design and performance evaluation for the extruded heatsink. The related data of the thermal resistance, pressure drop, velocity drop, and average heatsink temperature rise on the different heatsink designs and operating conditions are obtained by the results of the CFD simulation and the data will be sent to the BNN. After well trained, the BNN model is able to accurate estimate the corresponding output values under different heatsink design and operating conditions. [ABSTRACT FROM AUTHOR]

Automotive Magnetorheological (MR) fluid shock absorbers have been previously characterized by a series of nonlinear differential equations, which have some difficulties in developing control systems. This paper presents a recurrent neural network with 3 input neurons, 1 output neuron and 5 recurrent neurons in the hidden layer to simulate behavior of MR fluid shock absorbers to develop control algorithms for suspension systems. A recursive prediction error algorithm has been applied to train the recurrent neural network using test data from lab where the MR fluid shock absorbers were tested by the MTS electro-hydraulic servo vibrator system. Training of neural network model has been done by means of the recursive prediction error algorithm presented in this paper and data generated from test in laboratory. In comparison with experimental results of MR fluid shock absorbers, the neural network models are reasonably accurate to depict performances of MR fluid shock absorber over a wide range of operating conditions. [ABSTRACT FROM AUTHOR]

The purpose of this paper is to develop an on-line soft-sensor for dynamic estimation of the particle size distribution of hydrocyclones overflow in consecutive grinding process. The hybrid model based soft-sensor is based on the following model structures: 1. a neural net-based dynamic model of state space description for hydrocyclone with a neural net-based model for classifier and a population balance model for ball mill and sump, 2. an ANFIS-based model mainly for abnormal operating conditions, 3. a fuzzy logic coordinator for the final predictive result according to output values of aforementioned models. The fact that the soft-sensor performs well in particle size estimation demonstrates that the proposed hybrid intelligent soft-sensor model is effective for dynamic estimation of particle size distribution. [ABSTRACT FROM AUTHOR]

The ability of high tolerance for learning-by-example makes neural networks flexible and powerful in resolving various application problems. However, while being applied in real world, the time required to induce models from large data sets should be considered. In this paper, we apply QSS (Query-based learning with Selective-attention and Self-regulation) to back-propagation neural networks for resolving the data classification problem in biomedical applications. Results show that the proposed method can significantly reduce the training set cardinality. Additionally, the quality of training results can be ensured. It provides a powerful tool to help physicians analyze, model and make sense of complex clinical data for disease diagnosis. [ABSTRACT FROM AUTHOR]

This paper aims to study the short-term load forecasting of electricity by using an extended self-organizing map. We first adopt a traditional Kohonen self-organizing map (SOM) to learn time-series load data with weather information as parameters. Then, in order to improve the accuracy of the prediction, an extension of SOM algorithm based on error-correction learning rule is used, and the estimation of the peak load is achieved by averaging the output of all the neurons. Finally, as an implementation example, data of electricity demand from New York Independent System Operator (ISO) are used to verify the effectiveness of the learning and prediction for the proposed methods. [ABSTRACT FROM AUTHOR]

Based on similar day method and SVM, this paper proposes a new method for next day load forecasting. The new method uses the parameters of several similar days, instead of only selecting one similar day as in similar day method. The parameters of selected similar days are used as inputs to SVM for forecasting the loads of 24 points (one hour per point) of the next day. The method behaves the advantages of both similar day method and SVM method, Corresponding software was developed and used to forecast the next day load in a practical power system and the final forecasting error is low. [ABSTRACT FROM AUTHOR]

Epileptic seizure is a serious brain disease. The characteristic signature of epileptic seizure is interictal spikes and sharp waves. Development of a reliable method to detect spikes from EEG data is of major clinical and theoretical importance. In this paper, a new detection algorithm that combines the Empirical Mode Decomposition (EMD), Hilbert Transformation (HT) and Smoothed Nonlinear Energy Operator (SNEO) is proposed to detect spikes hidden in human EEG data. Finally, the EEG data generated by a nonlinear lumped-parameter cerebral cortex model and real EEG data from human are applied to test the performance of the new detection method. The results show that this method can efficiently detect the spikes hidden in EEG signals. [ABSTRACT FROM AUTHOR]

Recommendation is to offer information which fits user's interests and tastes to provide better services and to reduce information overload. It recently draws attention upon Internet users and information providers. Collaborative filtering is one of the widely used methods for recommendation. It recommends an item to a user based on the reference users' preferences for the target item or the target user's preferences for the reference items. In this paper, we propose a neural network based collaborative filtering method. Our method builds a model by learning correlation between users or items using a multi-layer perceptron. We also investigate integration of diverse information to solve the sparsity problem and selecting the reference users or items based on similarity to improve performance. We finally demonstrate that our method outperforms the existing methods through experiments using the EachMovie data. [ABSTRACT FROM AUTHOR]

Shape measurement is one of the critical problems in manufacturing robot systems. The point coordinates that we get change distinctly, because different objects to be processed own various shape forms. So it is difficult for traditional methods to get original accurate shape information. It always affects the processing results of manufacturing robot systems. According to the shipbuilding requirements, this paper proposes a dynamic and intelligent shape measurement method, which is based on the fuzzy reasoning (FR) and neural network (NN) method. FR is used to judge the relation of measured points. As the input of the NN, the fitted coordinate and the possibility of the rim point can be got. It has been demonstrated effective in Dalian Shipbuilding manufacturing robot system. [ABSTRACT FROM AUTHOR]

This paper discusses the trajectory following issue for redundant nonholonomic mobile modular manipulators. Dynamic model is established and an adaptive neural-network controller is developed to control the end-effector to follow a desired spacial trajectory. The proposed algorithm doesn't need any priori dynamics and provides a new solution for stabilization of redundant robotic self-motions. Simulation results for a real robot demonstrate the proposed algorithm is effective. [ABSTRACT FROM AUTHOR]

Movement state estimation plays an important role in navigating and movement controlling for wheeled mobile robots (WMRs), especially those in unknown environments such as planetary exploration. When exploring in unknown environments, mobile robot suffers from many kinds of abnormal movement state, such as baffled by an obstacle, slipping, among others. This paper employs neural network method to detect abnormal movement states. Specifically, it exploits the kinematics of the normal and abnormal movement states of the monitored robot. Several residuals are exploited and four probabilistic neural networks are used to classify the residuals. Simulation experiments show that the methods can detect and identify most abnormal movement states. [ABSTRACT FROM AUTHOR]

Fiber optic gyros (FOG) is the important sensor for measuring the heading of mobile robot. Combined with measured data of E-Core RD1100 interferometric FOG made by American KVH company, the paper analyses the common calibration for the heading errors of mobile robot caused by the drift of FOG, and uses the method of evolutionary neural networks prediction to compensate it. By the experiments of mobile robot prototype, the paper also proves this method can reduce the error influence of FOG on the heading of mobile robot and enhance the localization precision of mobile robot navigation. [ABSTRACT FROM AUTHOR]

Underactuation may arise when a manipulator includes at least one passive joint which is either a free joint where torque propagating along the axis of rotation is only due to friction, or a flexible joint which has passive elements introducing stiffness and damping. In this regard, free flying platform can be considered as linked to a fixed base by six DOF free joints with no friction. Dynamics of underactuated cooperating manipulators become a bigger challenge when underactuation is encountered by the combination of free, flexible and active joints based on a free-flying platform in no-gravity setting. We call this class of systems as "general underactuated systems". This paper demonstrates the theoretical background of an order of N algorithm for the dynamical modeling and neural network based control of general underactuated manipulators cooperating to perform a common task on a free-flying space platform. [ABSTRACT FROM AUTHOR]

SVMs have been very successful in pattern recognition and function approximation problems. In this paper, by collecting the human driving data, a data-driven lateral controller for autonomous vehicle based on SVMs is presented. Furthermore, according to the model of the vehicle, a simple method to improve the performance of this controller is introduced. The simulation results and experiments on the real vehicle show that the performance of the controller is satisfactory. [ABSTRACT FROM AUTHOR]

A support vector regression (SVR) model based batch to batch optimal control strategy is proposed in this paper. Because of model plant mismatches and unknown disturbances the control performance of optimal control profile calculated from empirical model is deteriorated. Due to the repetitive nature of batch processes, it is possible to improve the operation of the next batch using the information of the current and previous batch runs. A batch to batch optimal control strategy based on the linearization of the SVR model around the control profile is proposed in this paper. Applications to a simulated batch styrene polymerization reactor demonstrate that the proposed method can improve process performance from batch to batch in the presence of model plant mismatches and unknown disturbances. [ABSTRACT FROM AUTHOR]

Adaptive inverse control (AIC) uses three adaptive filters: plant model, controller and disturbance canceller. A kind of hybrid AIC system based on Least Squares Support Vector Machines (LS-SVMs) is proposed in this paper. It has a PID controller to compensate the control signal error. A kind of adaptive disturbance canceller based on LS-SVM is also proposed. It can optimally eliminate plant disturbance. Simulation example is presented to demonstrate that the proposed method works very well. [ABSTRACT FROM AUTHOR]

Cost prediction is very important for cost control, but the factors of influencing cost are many and complex. The factors affect each other, and the coupling phenomenon exists, so enterprise cost is difficult to be predicted correctly. On the basis of production cost composition model, the product cost prediction model based on neural network is established. A hybrid algorithm that trains neural network weight by real-coded adaptive mutation genetic algorithm is designed, and it overcomes the disadvantage that traditional neural network is easy to fall into local minima. Furthermore, the model is successfully applied to cost prediction in an ore dressing plant, and it improves the prediction accuracy. [ABSTRACT FROM AUTHOR]

The paper presents two methods for ABC classification of stock keeping units (SKUs), The first method is to apply principal components analysis (PCA) to classify inventory. The second method combines PCA with artificial neural networks (ANNs) with BP algorithm. The reliability of the models is tested by comparing their classification ability with a data set. The results show that the hybrid method could not only overcome the shortcomings of input limitation in ANNs, but also further improve the prediction accuracy. [ABSTRACT FROM AUTHOR]

The paper describes management-decision functions in a reservation allocation problem (RAP), and the purpose is to discuss how some recently developed techniques are combined with neural network to provide a firm's management with important tools for this storage control. The research herein is not sophisticated from the mathematical point of view , as a matter of fact, borrows heavily from work already published in neural-network-research literature. An optimization technique using linear programming is essential in solving the discrete requirements case, and it has been used to solve this problem mathematically. Consequently, a reservation distribution is defined in the domain of storage management. We propose an exact and a heuristic way that transforms the RAP into a linear programming model, and compute the optimal solution in some methods in neural network. [ABSTRACT FROM AUTHOR]

This paper presents an artificial neural network model for crop yield responding to soil parameters. The experimental data had been obtained via a precision agriculture experiment, which is carried out by PAC in a demo farm locating in Shunyi district, Beijing in 2000. The model has been established by training a back propagation neural network with 58 samples and tested with other 14 samples. The model consists of 6, 11 and 1 processing units in the input, hidden and output layers, and the step length is 0.05, the momentum coefficient is 0.5. The training was terminated after 20000 times and the convergence effect was very good. The training results are that the correlation coefficient is 0.916 and the average error value is 2.8×10-2. It shows that the model can precisely describe crop yield responding to soil parameters. [ABSTRACT FROM AUTHOR]

Electric arc furnace is commonly used in iron and steel industry to produce quality steel by melting iron and steel scraps using electric arc. It represents one of the most disturbing loads in the subtransmission or transmission electric power systems. Therefore, it is necessary to build a practical model to descript the behavior of electric arc furnace in the simulation of power system. The electrical fluctuations in the electric arc furnace have proven to be chaotic in nature. This paper deals with the problem of electric arc furnace modeling using the combination of chaos theory and neural network. The radial basis function neural network is used to predict the arc voltage of arc furnace with one-step and multi-step ahead. The results can be applied to simulate the EAF load in power system and estimate the future state of arc furnace for control purpose. [ABSTRACT FROM AUTHOR]

Simultaneous Multithreaded (SMT) processors improve the instruction throughput by allowing fetching and running instructions from several threads simultaneously at a single cycle. With the pipeline deepen and issue widths increase, the branch predictor plays a more important role in improving the performance of an SMT processor. Many predictors based on neural network, especially on perceptron, are proposed to provide a more accurate dynamic branch prediction than before in the literature. In this paper, we propose an effective method to improve the accuracy of a perceptron predictor through correlating data values in SMT processors. The key idea is using a dynamic bias input, which comes from some information independent on the branch histories (data values for example), to realize the objective of improving accuracy. The implementation of our method is simple, and the predicting latency is not lengthened. Execution-driven simulation results show that our method works successfully on improving the accuracy of a perceptron predictor and increasing the overall instruction throughput of SMT processors. [ABSTRACT FROM AUTHOR]

This paper presents a kind of model based on compensatory fuzzy neural network (CFNN) without normalization to predict product quality in the acetone refining process. Important technological influence factors are selected according to the analysis results of several variables selection methods. Using the selected factors as the input variables of the network, a product quality prediction model is constructed. Experiment results show that the trained model achieves good effects, and has more advantages in convergence speed and error precision compared with CFNN with normalization. [ABSTRACT FROM AUTHOR]

Chamber scheduling in etching tools is an important but difficult task in integrated circuit manufacturing. In order to effectively solve such combinatorial optimization problems in etching tools, this paper presents a novel chamber scheduling approach on the base of Adaptive Artificial Neural Networks (ANNs). Feed forward, multi-layered neural network meta-models were trained through the back-error-propagation (BEP) learning algorithm to provide a versatile job-shop scheduling analysis framework. At the same time, an adaptive selection mechanism has been extended into ANN. By testing the practical data set, the method is able to provide near-optimal solutions for practical chamber scheduling problems, and the results are superior to those generated by what have been reported in the neural network scheduling literature. [ABSTRACT FROM AUTHOR]

This paper presents some results on geometrical error compensation using multilayer neural networks (NNs). It is the objective to attain higher compensation performance with less or comparable memory, using this approach. There are three main contributions. First, multilayer NNs are used to approximate the components of geometrical errors. This results in a significantly less number of neurons compared to the use of radial basis functions (RBFs). Secondly, the direction of motion is considered in the compensator. This is important as the geometrical errors can be quite distinct depending on the direction of motion due to backlash and other nonlinearities in the servo systems. Thirdly, the Abbe error is explicitly addressed in the compensator. [ABSTRACT FROM AUTHOR]

In metal forming processes, automatic selection of forming tools is heavily depended on the estimation of forming forces. Due to complex relationships between processes parameters like die angle, co-efficient of friction, velocity of dies, and temperature of billet for forming products with sound quality and forming forces related, there is a need to develop approximate models to estimate the forming forces without complex mathematical models or time-consuming simulation techniques. In this paper, an Artificial Neural Networks (ANNs) model has been developed for rapid predication of the forming forces based on process parameters. The results obtained are found to correlate well with the finite element simulation data in case of hot extrusion. [ABSTRACT FROM AUTHOR]

Among breast abnormalities, spiculated lesions are one of the most difficult type of tumor to detect. In this paper, we apply a feature extraction method to generate four feature images for a single mammogram, and then partition every feature image into a series of small square blocks. The four average feature values of each block are considered as an instance describing the block. Finally we use an artificial neural network ensemble method to detect the spiculated lesions. Experiments show that the accuracy of this method is well on digital mammograms. [ABSTRACT FROM AUTHOR]

In this paper, the learning algorithm of networks is discussed. The programming example of 3-layer BP networks is given with Visual C++6.0 program langue. Based on this model, a lung cancer intelligent diagnosis system is successfully implemented. Furthermore, the paper introduces network's structure design, preferences and the source of sample datum in factual applications. The ameliorative arithmetic is applied to the study of networks and BP dynamic evolving process is designed. The experiments indicate cell images are recognized and classified by the trained neural network. The study illustrates the system has feasibility and clinical value in lung cancer diagnosis. [ABSTRACT FROM AUTHOR]

This paper proposes a novel visual automatic incident detection method on freeway based on RBF and SOFM neural networks. Two stages are involved. First, get the freeway traffic flow model based on the RBF neural networks and use the model to obtain the output prediction. The residuals will be gotten from the comparison between the actual and prediction. Second, use a SOFM neural networks to classify the residuals to detect the incident. Because the SOFM has the character of topological ordering, the winning neuron's running trajectory on SOFM neuron array corresponds to the actual traffic state on freeway. We can observe the trajectory to detect the incident and achieve the visual traffic incident detection. [ABSTRACT FROM AUTHOR]

Biomedical image is often complex. Using SVM for pixel-based segmentation may achieve good results, but training by conventional way always leads to high time cost. In this paper, a novel and real-time training strategy is presented. First, the mean-shift procedures are used to find local modes in RGB 3D histogram. Second, pure samples are selected by the divided modes. Third, the training set is constructed by uniform sampling from the pure samples, so its size can be reduced sharply. In the no-niose case, hard margin criterion replaces soft margin criterion for classification. This strategy constructs an unsupervised support vector classifier. Experimental results demonstrate that the new classifier can achieve accurate results, is more robust to change of the color and faster than watershed algorithm. The new method is suitable to segment blood and bone marrow microscopic images. [ABSTRACT FROM AUTHOR]

Based on the nonlinear processing ability of neural networks, a new method of estimation of latency change in evoked potentials (EPs) is proposed in this paper. Neural networks are utilized as filters before DLMS algorithm in EP latency change estimation in order to suppressing impulsive background noises. The new latency change estimation method shows robust performance under non-Gaussian α-stable noise conditions. [ABSTRACT FROM AUTHOR]

In this paper, a predictive control strategy based on neuro-fuzzy (NF) model of the plant is applied to Continuous Stirred Tank Reactor (CSTR). An optimizer algorithm based on evolutionary programming technique (EP) uses the identifier-predicted outputs and determines input sequence in a time window. Using the proposed neuro-fuzzy predictive controller, the performance of Ph tracking problem in a CSTR process is investigated. [ABSTRACT FROM AUTHOR]

Rough Set Theory (RST) is a powerful mathematics tool, which can deal with fuzzy and uncertain knowledge, and radial basis function (RBF) neural network has the ability to approach any nonlinear function precisely. According to the non-linear relation characteristics of ship power load, a short-term load prediction method based on RST and RBF neural network is presented in this paper. Using RST on the advantage of data analysis, the important input nodes can be selected, followed by a second stage selecting the important centers and leaning the weights of hidden nodes. The experimental results proved that this method could achieve greater predictive accuracy and generalization ability. [ABSTRACT FROM AUTHOR]

The paper adopts rough sets theory and fuzzy ARTMAP method to explore the online adaptive contingency classification in power system transient stability control. On the basis of contingency vector space model, the rough sets theory is applied to generalize the information system comprised by contingency samples set, and compute the best reducing properties set. So dimension of contingency feature space is reduced greatly, and disturbance in contingency classification is decreased too, which improves the efficiency of classification. In addition, using the advantage of adaptive classification and incremental learning of Fuzzy ARTMAP neural network, the online adaptive classification of contingency is achieved. [ABSTRACT FROM AUTHOR]

Current Intrusion Detection Systems (IDS) examine all data features to detect intrusion or misuse patterns. Some of the features may be redundant or contribute little (if anything) to the detection process. The purpose of this study is to identify important input features in building an IDS that is computationally efficient and effective. This paper proposes an IDS model based on general and enhanced Flexible Neural Tree (FNT). Based on the pre-defined instruction/operator sets, a flexible neural tree model can be created and evolved. This framework allows input variables selection, over-layer connections and different activation functions for the various nodes involved. The FNT structure is developed using an evolutionary algorithm and the parameters are optimized by particle swarm optimization algorithm. Empirical results indicate that the proposed method is efficient. [ABSTRACT FROM AUTHOR]

This paper presents a methodology for the actuator fault detection in the satellite's attitude control system (ACS) by using a dynamic neural network based observer. In this methodology, a neural network is used to model a nonlinear dynamical system. After training, the neural network, it can give very accurate estimation of the attitude positions of the satellite. The difference between the actual and the estimated outputs is used as a residual error for fault detection. The simulation results show advantages of this method as compared to the method based on a generalized Luenberger linear observer. [ABSTRACT FROM AUTHOR]

Amplifier is a very important device in the missile control system and its working state directly decides the missile's flying stability and hitting precision. In the past, some traditional methods were applied to pick up the features from the signal which was disturbed by noises, but the transient signal can't be identified. In this paper, an adaptive wavelet packet neural network (WPNN) method is presented for diagnosing amplifier fault based on pattern recognition at the first time. This method consists of two stages: firstly, the wavelet packet decomposition is used for feature extraction and secondly, a feed-forward neural network is utilized for pattern classification. Moreover, during the learning phase of WPNN, the wavelet packet entropy and the weights are updated adaptively to minimize the learning error. The experimental results show that the adaptive WPNN method is effective in detecting and diagnosing the amplifier fault of the missile. [ABSTRACT FROM AUTHOR]

In this paper, a fault diagnosis and isolation (FDI) strategy based on a Dynamically Driven Recurrent Neural Network (DDRNN) architecture is proposed for use in situations when there are thruster/actuator failures in the satellite's attitude control system. The proposed architecture is motivated from the following facts: (1) the satellite's attitude dynamics is highly complex and nonlinear, (2) the large volume of data that is generated in the attitude control system has to be monitored in real-time by the ground station operators which is a highly labor-intensive and time-consuming task, and (3) dynamically driven recurrent neural networks (DDRNN) have been shown to have the ability to learn, recognize and generate complex temporal patterns. To improve the FDI performance accuracy, the proposed architecture is designed to consist of two DDRNNs. The first DDRNN determines and diagnoses the presence of a faulty thruster. The second DDRNN then identifies which thruster is faulty, i.e. it isolates the location of the fault. Extensive simulation results are shown that demonstrate and verify that the proposed two DDRNNs scheme is more efficient and robust as compared to a scheme that is based on a single feed-forward back-propagation neural network or a single DDRNN scheme, especially in the presence of external disturbances and noise. [ABSTRACT FROM AUTHOR]

This paper presents a new method that applies fuzzy logic, rough set theory and SOFM neural network to rotating machinery fault diagnosis. In this method, firstly, relationships between the fault causations and fault symptoms are established by fuzzy logics. Then the Rough Set Theory (RST) is applied to obtain a minimal sufficient subset of features, which is helpful to simplify the structure of neural network. Next, the 2-dimension output mapping of the standard fault samples (training samples) is obtained by a self-organizing neural network. Finally, we input some simulation samples (testing samples) and gain the reasonable conclusions by comparison between the two output mappings. Experimental results have demonstrated the effectiveness of this method and its nice prospect of applying to rotating machinery fault diagnosis. [ABSTRACT FROM AUTHOR]

In this paper, neural network approaches are for the first time employed to identify acoustic fault sources of underwater vehicles. According to the characteristics of acoustic fault sources, a novel sources identification model based on modular structure variable radial basis function (SVRBF) network is proposed. Unsupervised algorithms for clustering and supervised learning are combined to train the network, especially, the number of hidden and output layer neurons can be modified on-line so that the network has the capability of incremental learning. The results of experiment show that the proposed network has better generalization performance than traditional BP network and RBF network, and is effective in learning new fault patterns. [ABSTRACT FROM AUTHOR]

Feature selection is a key issue to machine fault diagnosis. This paper presents a study that uses self-organizing maps to realize feature selection and reduce dimensionality of the raw feature space for machine faults classification. By means of evaluating the responses of every dimensional feature in SOM networks neurons weights to the input data, the feature sets having the main responses and being sensitive to pattern recognition are selected. Industrial gearbox vibration signals measured under different operating conditions are analyzed using the method. The experimental results indicate that the method selects sensitive feature sets effectively for gear faults classification and recognition, and has a good potential for application in practice. [ABSTRACT FROM AUTHOR]

In this paper a new type of fault detection (FD) problem is considered where the measured information is the stochastic distribution of the system output rather than its value. A multi-layer perceptron (MLP) neural network is adopted to approximate the probability density function (PDF) of the system outputs and nonlinear principal component analysis (NLPCA) is applied to reduce the model order for a lower-order model. For such a dynamic model in discrete-time context, where nonlinearities, uncertainties and time delays are included, the concerned FD problem is investigated. The measure of estimation errors, which is represented by the distances between two output PDFs, will be optimized to find the detection filter gain. Guaranteed cost detection filter are designed based on LMI formulations. [ABSTRACT FROM AUTHOR]

In the paper, a method of fault diagnosis based on integration of the variable precision rough set model (VPRSM) and neural networks is proposed. By VPRSM attribute reduction, redundant attributes are identified and removed. The reduction results are used as the inputs of neural network. With the proposed method an example for rotary machinery fault diagnosis is given. The diagnosis results show that the proposed approach has better learning efficiency and diagnosis accuracy. [ABSTRACT FROM AUTHOR]

An intelligent diagnostic system for a large rotor system based on radial basis function network, called rapid response intelligent diagnosis network (RRIDN), is proposed and introduced into practice. In this paper, the principles, model, net architecture, and fault feature selection of RRIDN are discussed in detail. Correct model architecture selection are emphasized in constructing a radial basis neural network of high performance. In order to reduce the amount of real training data, the counterexamples of real data are adopted. Some training and testing results of rapid response intelligent diagnosis networks are given. The practical effects in two chemical complexes are analyzed. Both of them indicate that RRIDN possesses good function. [ABSTRACT FROM AUTHOR]

Corona discharge (CD) and partial discharge (PD) indicate early stages of insulation problems in motors. Early detection of CD/PD will enable better coordination and planning of resources such as maintenance personnel, ordering of parts, etc. Most importantly, one can prevent catastrophic failures during normal operations. In this paper, we summarize the application of Support Vector Machine (SVM) to CD/PD monitoring. Hardware testbeds have been developed to emulate CD/PD behaviors and real-time experimental results showed the effectiveness of SVM for fault detection and classification. [ABSTRACT FROM AUTHOR]

In this paper we propose a novel clustering algorithm in mobile ad hoc network. By predicting the mobility of mobile nodes with the wavelet neural network, the resulting clustering algorithm can provide a generic and stable cluster structure for the upper-layer protocols. For this clustering scheme, we give analytical model and performance evaluation. [ABSTRACT FROM AUTHOR]

The research of applying Self-organizing Maps for intrusion detection is investigated in this paper. A novel approach is presented for enhancing SOM's abilities of identifying temporal network attacks, which combine with FIR filter. Meanwhile, we reconsider the heterogeneous dataset that composed of network connection's features, and select HVDM as the distance function determining the winning neuron during SOM's training and testing. In the end, KDD benchmark dataset is employed to validate the efficiency of our approach, and the results is detection rates of 96.5%, false positive rates of 6.2%, which accounts for good performance of our new approach in intrusion detection fields. [ABSTRACT FROM AUTHOR]