Catálogo de publicaciones - libros

This paper is concerned with a multi–resolution tool for screening paper formation variations in various frequency regions on production line. A paper web is illuminated by two red diode lasers and the reflected light recorded as two time series of high resolution measurements constitute the input signal to the papermaking process monitoring system. The time series are divided into blocks and each block is analyzed separately. The task is treated as kernel based novelty detection applied to a multi–resolution time series representation obtained from the band-pass filtering of the Fourier power spectrum of the series. The frequency content of each frequency region is characterized by a feature vector, which is transformed using the canonical correlation analysis and then categorized into the or class by the novelty detector. The ratio of outlying data points, significantly exceeding the predetermined value, indicates abnormalities in the paper formation. The tools developed are used for online paper formation monitoring in a paper mill.

Collaborative design has been recognized an alternative environment for product design in which multidisciplinary participants are naturally involving. Reuse of product design information has long been recognized as one of core requirements for efficient collaborative product design. This paper addresses integration of multi-modal data using a graph for an assembly design information sharing and reuse in the collaborative environment. In the system, assembly product images obtained from multi-modal devices are utilized to share and to reuse design information. The proposed system conducts the segmentation of an assembly product image by using a labeling method and generates an attribute relation graph (ARG) that represents properties of segmented regions and their relationships. The generated ARG is extended by integrating corresponding part/assembly information. In this manner, the integration of multi-modal data has been realized to retrieve assembly design information using a product image.

The probabilistic filtering method filters out an unpromising candidate solution by conducting a simple preliminary evaluation before a complete evaluation in order to improve the efficiency of a local search. In this paper, we improve probabilistic filtering so that it can be applied in general to large-scaled optimization problems. As compared to the previous probabilistic filtering method, our enhanced version includes a scaling and truncation function to increase the discriminating power of probabilistic filtering and repair some defects of the previous bias function in adjusting the level of greediness. Experiments have shown that our method is more effective in improving the performance of a local search than the previous method. It has also been shown that the probabilistic filtering can be effective even when the preliminary evaluation heuristic is somewhat inaccurate, and the lesser the cost of preliminary evaluation, the greater is its effectiveness.

A successful Case-Based Reasoning (CBR) system highly depends on how to design an accurate and efficient case retrieval mechanism. In this research we propose a Weighted Feature C-means clustering algorithm (WF-C-means) to group all prior cases in the case base into several clusters. In WF-C-means, the weight of each feature is automatically adjusted based on the importance of the feature to clustering quality. After executing WF-C-means, the dissimilarity definition adopted by K-Nearest Neighbor (KNN) search method to retrieve similar prior cases for a new case becomes refined and objective because the weights of all features adjusted by WF-C-means can be involved in the dissimilarity definition. On the other hand, based on the clustering result of WF-C-means, this research proposes a cluster-based case indexing scheme and its corresponding case retrieval strategy to help KNN retrieving the similar prior cases efficiently. Through our experiments, the efforts of this research are useful for real world CBR systems.

This work presents forecast models for the natural inflow in the Basin of Iguaçu River, incorporating rainfall information, based on artificial neural networks. Two types of rainfall data are available: measurements taken from stations distributed along the basin and ten-day rainfall forecasts using the ETA model developed by CPTEC (Brazilian Weather Forecating Center). The neural nework model also employs observed inflows measured by stations along the Iguaçu River, as well as historical data of the natural inflows to be predicted. Initially, we applied preprocessing methods on the various series, filling missing data and correcting outliers. This was followed by methods for selecting the most relevant variables for the forecast model. The results obtained demonstrate the potential of using artificial neural networks in this problem, which is highly non-linear and very complex, providing forecasts with good accuracy that can be used in planning the hydroelectrical operation of the Basin.

The technology of data mining is more important in recent years, and it is generally applied to commercial forecast and decision supports. Sequential pattern mining algorithms in the field of data mining play one of the important roles. Many of sequential pattern mining algorithms were proposed to improve the efficiency of data mining or save the utility rate of memory. So, our major study tries to improve the efficiency of sequential pattern mining algorithms.

We propose a new algorithm - GRS (A radational eduction Approach for Mining equential Patterns) which is an efficient algorithm of mining sequential patterns. GRS algorithm uses gradational reduction mechanism to reduce the length of transactions and uses GraDec function to avoid generating large number of infrequent sequential patterns; and it is very suitable to mine the transactions of databases whose record lengths are very long. The GRS algorithm only generates some sequences which are very possible to be frequent. So, the GRS algorithm can decrease a large number of infrequent sequences and increase the utility rate of memory.

Measuring structural similarity between XML documents has become a key component in various applications, including XML data mining, schema matching, web service discovery, among others. The paper presents a novel structural similarity measure between XML documents using kernel methods. Results on preliminary simulations show that this outperforms conventional ones.

Using the neural network model for oriented principal component analysis (OPCA), we propose a solution to the data least squares (DLS) problem, in which the error is assumed to lie in the data matrix only. In this paper, We applied this neural network model to channel equalization. Simulations show that DLS outperforms ordinary least squares in channel equalization problems.

We develop novelty detection techniques for the analysis of data from a large-vehicle engine turbocharger in order to illustrate how abnormal events of operational significance may be identified with respect to a model of normality. Results are validated using polynomial function modelling and reduced dimensionality visualisation techniques to show that system operation can be automatically classified into one of three distinct state spaces, each corresponding to a unique set of running conditions.

This classification is used to develop a regression algorithm that is able to predict the dynamical operating parameters of the turbocharger and allow the automatic detection of periods of abnormal operation. Visualisation of system trajectories in high-dimensional space are communicated to the user using parameterised projection techniques, allowing ease of interpretation of changes in system behaviour.

A novel equalization scheme for 16 QAM signals through a wireless ATM communication channel using Reduced-Complex Bilinear Recurrent Neural Network (R-CBLRNN) is proposed in this paper. The 16 QAM signals from a wireless ATM communication channel have severe nonlinearity and intersymbol interference due to multiple propagation paths in the channel. The R-CBLRNN equalizer is compared with the conventional equalizers including a Volterra filter equalizer, a decision feedback equalizer (DFE), and a multilayer perceptron type neural network (MLPNN) equalizer. The results show that the R-CBLRNN equalizer for 16 QAM signals gives very favorable results in both of the Mean Square Error(MSE) and the Symbol Error Rate (SER) criteria over conventional equalizers.