Applied Soft Computing Technologies: The Challenge of Complexity

We discuss the role of Zadeh’s paradigm of computing with words on the semantic web We describe the three important steps in using computing with words. We focuson the retranslation step, selecting a term from our prescribedvocabulary to express information represented using fuzzy sets. A number of criteria of concern in this retranslation processare introduced. Some of these criteria can be seen to correspondto a desire to accurately reflect the given information. Othercriteria may correspond to a desire, on the part provider ofthe information, to give a particular perception or “spin”.We discuss some methods for combining these criteria to evaluatepotential retranslations.

In this study, we present a general overview of Granular Computing being regarded as a coherent conceptual and algorithmic platform supporting the design of intelligent systems. Fundamental formalisms of Granular Computing are identified and discussed. The linkages between Granular Computing and Computational Intelligence are revealed as well.

It is of utmost important to maintain perfect condition of complex welded structures such as pressure vessels, load bearing structural members and power plants. The commonly used approach is non-destructive evaluation (NDE) of such welded structures. This paper presents an application of artificial neural networks (ANN) for weld data, extracted from reported radiographic images. Linear Vector Quantization based supervised neural network classifier is implemented in Parallel Processing Environment on PARAM 10000. Single Architecture Single Processor and Single Architecture Multiple Processor based parallel neuro classifier are developed for the weld defect classification. The results obtained for various statistical evaluation methods showed promising future of Single Architecture Single Processor based parallel neuro classifier in the problem domain.

Most of the classical clustering algorithms are strongly dependent on, and sensitive to, parameters such as number of expected clusters and resolution level. To overcome this drawback, a Genetic Programming framework, capable of performing an automatic data clustering, is presented. Moreover, a novel way of representing clusters which provides intelligible information on patterns is introduced together with an innovative clustering process. The effectiveness of the implemented partitioning system is estimated on a medical domain by means of evaluation indices.

This paper describes a novel adaptive network, which agglomerates a procedure based on the fuzzy min-max clustering method, a supervised ART (Adaptive Resonance Theory) neural network, and a constructive conflict-resolving algorithm, for pattern classification. The proposed classifier is a fusion of the ordering algorithm, Fuzzy ARTMAP (FAM) and the Dynamic Decay Adjustment (DDA) algorithm. The network, called Ordered FAMDDA, inherits the benefits of the trio,

viz

. an ability to identify a fixed order of training pattern presentation for good generalisation; stable and incrementally learning architecture; and dynamic width adjustment of the weights of hidden nodes of conflicting classes. Classification performance of the Ordered FAMDDA is assessed using two benchmark datasets. The performances are analysed and compared with those from FAM and Ordered FAM. The results indicate that the Ordered FAMDDA classifier performs at least as good as the mentioned networks. The proposed Ordered FAMDDA network is then applied to a condition monitoring problem in a power generation station. The process under scrutiny is the Circulating Water (CW) system, with prime attention to condition monitoring of the heat transfer efficiency of the condensers. The results and their implications are analysed and discussed.

In this paper, we examine the performance of fuzzy clustering algorithms as the major technique in pattern recognition. Both possibilistic and probabilistic approaches are explored. While the Possibilistic C-Means (PCM) has been shown to be advantageous over Fuzzy C-Means (FCM) in noisy environments, it has been reported that the PCM has an undesirable tendency to produce coincident clusters. Recently, an extension of the PCM has been presented by Timm et al., by introducing a repulsion term. This approach combines the partitioning property of the FCM with the robust noise insensibility of the PCM. We illustrate the advantages of both the possibilistic and probabilistic families of algorithms with several examples and discuss the PCM with cluster repulsion. We provide a cluster valid-ity function evaluation algorithm to solve the problem of parameter optimization. The algorithm is especially useful for the unsupervised case, when labeled data is unavailable.

This paper undertakes the problem of Design Centering and Tolerancing (DCT). This subject plays important role in design of analog circuits. Practically it is impossible to obtain the optimum solution by mathematically described computational methods, and the problem is numbered to NP-diffcult category. Optimum nominal values and tolerances of analog circuit parameters such as resistance, capacitance and inductance, are determined based on evolutionary strategy. The approach is verified on benchmark example and significant improvement, when compared with other design methods, can be observed.

The global optimization strategy of Simulated Annealing is applied to the optimization of knot parameters of NURBS for curve fitting, the objective being the reduction of fitting error to obtain a smooth curve. This is accomplished by using a unit weight vector and a fixed number of control points calculated using the least squares technique, while the sum of squared errors is taken as the objective function.

Many algorithms for multiobjective optimization have been proposed in the last years. In the recent past a great importance have the MOEAs able to solve problems with more than two objectives and with a large number of decision vectors (space dimensions). The diffculties occur when problems with more than three objectives (higher dimensional problems) are considered. In this paper, a new algorithm for multiobjective optimization called Multiobjective Adaptive Representation Evolutionary Algorithm (MAREA) is proposed. MAREA combines an evolution strategy and an steady-state algorithm. The performance of the MAREA algorithm is assessed by using several well-known test functions having more than two objectives. MAREA is compared with the best present day algorithms: SPEA2, PESA and NSGA II. Results show that MAREA has a very good convergence.

Real optimization problems often involve not one, but multiple objectives, usually in conflict. In single-objective optimization there exists a global optimum, while in the multi-objective case no optimal solution is clearly defined, but rather a set of solutions, called Pareto-optimal front. Thus, the goal of multiobjective strategies is to generate a set of non-dominated solutions as an approximation to this front. This paper presents a novel adaptation of some of these metaheuristics to solve the multi-objective Graph Partitioning problem.

We present the results of our investigation into the use of Genetic Algorithms (GA) for identifying near optimal design parameters of Diagnostic Systems that are based on Artificial Neural Networks (ANNs) for condition monitoring of mechanical systems. ANNs have been widely used for health diagnosis of mechanical bearing using features extracted from vibration and acoustic emission signals. However, different sensors and the corresponding features exhibit varied response to different faults. Moreover, a number of different features can be used as inputs to a classifier ANN. Identification of the most useful features is important for an efficient classification as opposed to using all features from all channels, leading to very high computational cost and is, consequently, not desirable. Furthermore, determining the ANN structure is a fundamental design issue and can be critical for the classification performance. We show that GA can be used to select a smaller subset of features that together form a genetically fit family for successful fault identification and classification tasks. At the same time, an appropriate structure of the ANN, in terms of the number of nodes in the hidden layer, can be determined, resulting in improved performance.

Based on the idea of SoC (System-on-Chip), an embedded-link, selfadaptive medical advisory system is proposed for home health monitoring, which can be merged into the embedded platform of mobile or wearable medical sensors and run independently. Meanwhile, if necessary, it will up link to health centers and self-calibrate its knowledge base. To provide effective medical advices, the methods of soft computing, including temporal fuzzy variables and weighted medical rules, are applied to the proposed medical advisory system. The elementary components, including medical knowledge base, inference machine and shell, are addressed in detail. Lastly, the way to construct such embedded-link, self-adaptive medical advisory system and integrate it into a mobile cardiovascular monitoring device is introduced too.

In traditional Failure Mode and Effect Analysis (FMEA), the Risk Priority Number (RPN) ranking system is used to evaluate the risk level of failures, to rank failures, and to prioritize actions. This approach is simple but it suffers from several weaknesses. In an attempt to overcome the weaknesses associated with the traditional RPN ranking system, several fuzzy inference techniques for RPN determination are investigated in this paper. A generic Fuzzy RPN approach is described, and its performance is evaluated using a case study relating to a semiconductor manufacturing process. In addition, enhancements for the fuzzy RPN approach are proposed by refining the weights of the fuzzy production rules.

This paper presents a Fault Detection and Isolation (FDI) approach based on the use of Hybrid Dynamic Bayesian Networks (HDBN). The peculiarity of the proposed approach is that an analytical dynamic model of the process to be monitored is not required. Instead it is hypothesized that input/output measures performed on the considered process during different working conditions, including faults, are available. In the paper the proposed FDI approach is described and the performances are evaluated on synthetic and real data supplied by a standard benchmark consisting of an hydraulic actuators available in literature. The goodness of the proposed approach is assessed by using appropriate performance indices. An intercomparison between the BN approach and an other approach, namely a Multilayer Perceptron (MLP) neural network is given. Results show that the BN approach outperforms the MLP approach in some indices but it requires a high design and computational effort.

Foraging theory originated in attempts to address puzzling findings that arose in ethological studies of food seeking and prey selection among animals. The potential utilization of biomimicry of social foraging strategies to develop advanced controllers and cooperative control strategies for autonomous vehicles is an emergent research topic. The activity of foraging can be focused as an optimization process. In this paper, a bacterial foraging approach for path planning of mobile robots is presented. Two cases study of static environment with obstacles are presented and evaluated. Simulation results show the performance of the bacterial foraging in different environments in the planned trajectories.

This article reports the results of a thorough empirical examination concerning the parameter settings in optimum tracking with (1 +λ) evolution strategies. The investigated scenario is derived from real-world applications where the evaluation function is very expensive concerning the computation time. This is modeled by a strong dependence of the dynamics on the number of evaluations. This model enables the derivation of optimal population sizes that might serve as recommendations for general applications. Both random and linear dynamics and two modes for correlating population size and strength of dynamics are examined. The results show for mere tracking problems that plus selection outperforms comma selection and that self-adaptive evolution strategies are able to deliver a close to optimal performance.

In this paper we describe a particular warning system for the purpose of monitoring ground deformation at Stromboli volcano (Aeolian Islands, Sicily). The paper, after a short introduction about the features of the recording system, describes the solutions adopted for modeling both short and long range ground deformation episodes. The guide lines for implementing a fuzzy rule base for the aims of the warning system are also outlined.

In this work we present a genetic algorithm for the job shop problem with recirculation. The genetic algorithm includes a local search procedure that is implemented as a genetic operator. This strategy differs from the memetic algorithm because it is not guaranteed that the local minimum is achieved in each iteration.

This study has developed and evaluated a multiobjective metaheuristic for redistricting to draw territory lines for geographical zones for the purpose of space control. The proposed multiobjective metaheuristic is briefly explained to shows its components and functionality. The redistricting problem definition is discussed, followed by the metaheuristic and the multiobjective decision rules. Then, an experiment is conducted in Geographic Information System (GIS) to shows its performances especially in term of its quality of result. The focus of the experiment is on the performance analysis of the coverage of the approximately non-dominated solution set and the number of objectives defined. The result of the experiment has demonstrated an improvement.

This paper approaches a computational solving of the industrial layout problem considering a set of hard-constraints not handled in previous works. The intention here has been to provide a new consistent benchmark problem that may help researchers to test their proposed algorithms. In order to handle the experiments, the computational optimization tool AVOLI (Visual Environment for Industrial

Layout

Optimization) is utilized. By using this computational tool, the provided hypothetic new facility layout problem mimicking a real one is solved in two steps: first a constructiveheuristic- based initial solution is generated and then Tabu Search (TS) heuristic is used to improve it.

Empathy is a distributed environment for the generation of emotions and other related affective phenomena like moods and temperaments. Empathy has been conceived as an object-oriented reusable framework entirely written in Java and realized for the purpose of studying the direct influences of emotions on behaviors and on decision-making processes of autonomous agents, interacting in complex or real environments. It allows for the realization of custom emotional agents, usable in several different domains, from the educational applications (e.g. entertainment, video games, intelligent tutoring systems.) to control systems in autonomous robots.

Time-series prediction can be interpreted in a way that is suitable for artificial intelligence learning. Two effective learning methods, Artificial Neural Networks and Support Vector Machines, are used to provide accurate non-linear models of the problem. In spite of the effectiveness of these methods we have to solve two problems. Firstly, time-series can have noise and a high dimensional embedding space. Secondly, the learning depends on several hyper-parameters that need to be set properly. To handle these problems we apply noise and dimension reduction tech- niques and model selection to get suitable hyper-parameters. Then, we introduce a meta-heuristic to refine the predictions of the selected models. Our experiments show improvements in the quality of predictions of a real-life problem compared to two ‘benchmark’ algorithms.

The

Evolutionary Algorithms

(EA), see [1] and [2], are stochastic techniques able to find the optimal solution to a given problem. The concept of

optimal solution

depends on the specific application, it could be the search of the global minimum of a complicated function. These algorithms are based on

Darwin

theories about

natural selection

. Natural selection allows to survive only best individuals (that is individuals more suitable to fit environment changes); in this way there is a generalized improvement of the entire population. Only the most performing individuals can transfer their genotype to the descendants.In the EA the parameter measuring individuals performance (in literature known as individuals

fitness

) is called

fitness function

. Time goes on by discrete steps. Starting by an initial population randomly generated, the process of evolution takes place. The most used operators that allow to obtain the new generation are:

Reproduction, Recombination, Mutation

and

Selection

. Let’s to consider more formally these statements. Given a generic fitness function

F

defined in a

N

-dimensional parameters space,

Y

, and with values in an

M

-dimensional space

Z

:

Nonlinear complexities and unknown uncertainty of models for dynamical systems are difficult problems in identification tasks. Fuzzy systems, especially Takagi-Sugeno (TS) fuzzy systems, viewed as nonlinear systems are potential candidates for identification and control of general nonlinear systems. A method of nonlinear identification in open-loop based on TS fuzzy system is evaluated in this paper. The contribution of this paper is the proposition of an optimization approach to automatically build a TS fuzzy system based on a set of input-output data of a process. The proposed scheme is based in particle swarm optimization with operators based on Gaussian and Cauchy distributions for the antecedent part design, while least mean squares technique is utilized for consequent part of production rules of a TS fuzzy system. Experimental example using a nonlinear yo-yo motion control system is analyzed by proposed approach.

A hybrid of type-1 and type-2 fuzzy model is proposed, which is applied in controlling the surface roughness of mechanical workpiece in metal cutting manufacturing. There are dozens of factors that affect the quality of surface roughness. The factors can be divided into two groups that are controlled and uncontrolled factors, e.g. feed rate can be setup. Therefore it is controlled factor while tool wear is an example of uncontrolled factor. There are two kinds of factors respectively correspond to type-1 and type-2 solutions because type-1 is suitable for controlled factors and type-2 fuzzy logic can handle uncontrolled or uncertain inputs. The proposed study will use genetic algorithm to identify the significant factors during the cutting process and a mathematical model that can predict the surface roughness under process variations. A fuzzy set based model for metal cutting operations can be used to reliably predict surface roughness under variations so that a continuous control of surface roughness can be affirmed. Two main factors (feed rate and tool wear) which affect the quality of surface roughness are investigated and simulated. The result of simulation shows that hybrid fuzzy logic system has improved precision of output.

Soft Computing has become popular in Steel Industry for its applications in the areas of reduction in defects, prediction of properties, classification of the products and many others. In recent times, the prediction of properties of steel strip is an area of increased interest mainly because of its prospective benefits of reduction in testing cost, better control on properties, reduction of inventory, increase in yield, and improvement in delivery compliance. Prediction of mechanical properties is a complicated task, as it depends on the chemical composition of the steel, and a number of processing parameters. In general, a high degree of nonlinearity exists between the property and the factors influencing it. In the past only Artificial Neural Network (ANN) was used, sometimes along with the variable reduction technique such as principle components / factor analysis. However, Multivariate Adaptive Regression Splines (MARS) has never been used despite some of its known advantages over the ANN. In this work two predictive models have been developed - one based on ANN, and another, MARS. This paper discusses on the model development and the comparative performance analysis of these two. The analysis shows that the results from both the models are comparable. However, shorter training time and automatic selection of important predictor variables, give MARS an edge over ANN.

This paper presents the designing steps and simulation results of a pulse classification system for the ECDM process using artificial neural networks (ANN). An Electro Discharge Machining (EDM) machine was modified by incorporating an electrolyte system and by modifying the control system. Gap voltage and working current waveforms were obtained. By observing the waveforms, pulses were classified into five groups. A feed forward neural network was trained to classify pulses. Various neural network architectures were considered by changing the number of neurons in the hidden layer. The trained neural networks were simulated. A quantitative analysis was performed to evaluate various neural network architectures.

Satellite image processing is a complex task that has received considerable attention from many researchers. In this paper, an interactive image query system for satellite imagery searching and retrieval is proposed. Like most image retrieval systems, extraction of image features is the most important step that has a great impact on the retrieval performance. Thus, a new technique that fuses color and texture features for segmentation is introduced. Applicability of the proposed technique is assessed using a database containing multispectral satellite imagery. The experiments demonstrate that the proposed segmentation technique is able to improve quality of the segmentation results as well as the retrieval performance.

In this paper, a boosted Fuzzy Min-Max Neural Network (FMM) is proposed. While FMM is a learning algorithm which is able to learn new classes and to refine existing classes incrementally, boosting is a general method for improving accuracy of any learning algorithm. In this work, AdaBoost is applied to improve the performance of FMM when its classification results deteriorate from a perfect score. Two benchmark databases are used to assess the applicability of boosted FMM, and the results are compared with those from other approaches. In addition, a medical diagnosis task is employed to assess the effectiveness of boosted FMM in a real application. All the experimental results consistently demonstrate that the performance of FMM can be considerably improved when boosting is deployed.

In this paper we proposed a genetic algorithm to minimize a nonconvex and nonlinear cost function based on statistical estimators for solving blind source separation-independent component analysis problem. In this way a novel method for blindly separating unobservable independent component signals from their linear and non linear (using mapping functions) mixtures is devised. The GA presented in this work is able to extract independent components with faster rate than the previous independent component analysis algorithms based on Higher Order Statistics (HOS) as input space dimension increases showing significant accuracy and robustness.

We present a new digital watermarking in which redundant wavelet transform (RDWT) is applied for watermark embedding and independent component analysis (ICA) is used to extract the watermark. By using RDWT, a large logo watermark can be embedded into transform coefficients. The advantage of using large logo is the ability to carry redundant information about copyright, increasing the robustness. In the embedding procedure, the watermark is embedded in RDWT domain. However, in the extraction procedure, the watermark is directly extracted from the watermarked image in spatial domain by an ICA-based detector. The experiment shows that the proposed scheme produces less image distortion than conventional DWT and is robust against Jpeg/Jpeg2000/SPIHT image compression and other attacks.

The performance of public-key cryptosystems is primarily determined by the implementation efficiency of the modular multiplication and exponentiation. As the operands, i.e. the plain text of a message or the cipher (possibly a partially ciphered) text are usually large (1024 bits or more), and in order to improve time requirements of the encryption/decryption operations, it is essential to attempt to minimise the number of modular multiplications performed. In this paper, we exploit the ant colony’s principles to engineer a minimal addition chain that allows one to compute modular exponentiations very efficiently. The obtained results are compared to existing heuristics as well as to genetically evolved addition chains, i.e. using genetic algorithms.

The curse of dimensionality is a well known but not entirely wellunderstood phenomena. Too much data, in terms of the number of input variables, is not always a good thing. This is especially true when the problem involves unsupervised learning or supervised learning with unbalanced data (many negative observations but minimal positive observations). This paper addresses two issues involving high dimensional data: The first issue explores the behavior of kernels in high dimensional data. It is shown that variance, especially when contributed by meaningless noisy variables, confounds learning methods. The second part of this paper illustrates methods to overcome dimensionality problems with unsupervised learning utilizing subspace models. The modeling approach involves novelty detection with the one-class SVM.

Medical semeiotics often deals with patient databases and would greatly benefit from efficient clustering techniques. In this paper a new evolutionary algorithm for data clustering, the Self–sizing Genome Genetic Algorithm, is introduced. It does not use a priori information about the number of clusters. Recombination takes place through a brand–new operator, i.e.,

gene–pooling

, and fitness is based on simultaneously maximizing intra–cluster homogeneity and inter–cluster separability. This algorithm is applied to clustering in dermatological semeiotics. Moreover, a Pathology Addressing Index is defined to quantify utility of the clusters making up a proposed solution in unambiguously addressing towards pathologies.

The computational identification of genes in DNA sequences has become an issue of crucial importance due to the large number of DNA molecules being currently sequenced. We present a novel neural network based multi-classifier system, MultiNNProm, for the identification of promoter regions in E.Coli

1

DNA sequences. The DNA sequences were encoded using four different encoding methods and were used to train four different neural networks. The classification results of these neural networks were then aggregated using a variation of the LOP method. The aggregating weights used within the modified LOP aggregating algorithm were obtained through a genetic algorithm. We show that the use of different neural networks, trained on the same set of data, could provide slightly varying results if the data were differently encoded. We also show that the combination of more neural classifiers provides us with better accuracy than the individual networks.

Drug discovery (DD) research has evolved to the point of critical dependence on computerized systems, databases and newer disciplines. Such disciplines include but are not limited to bioinformatics, chemoinformatics and soft computing. Their applications range from sequence analysis methods for finding biological targets to design of combinatorial libraries in lead compound optimisation. This paper presents a brief overview of classical techniques in DD with their limitations, and outlines current SC based techniques in this area.

The use of ontology in order to provide a mechanism to enable machine reasoning has continuously increased during the last few years.This paper suggests an automated method for document classification using an ontology, which expresses terminology information and vocabulary contained in Web documents by way of a hierarchical structure. Ontology-based document classification involves determining document features that represent the Web documents most accurately, and classifying them into the most appropriate categories after analyzing their contents by using at least two pre-defined categories per given document features. In this paper, Web documents are classified in real time not with experimental data or a learning process, but by similar calculations between the terminology information extracted from Web texts and ontology categories. This results in a more accurate document classification since the meanings and relationships unique to each document are determined.

Spam mail recognition is a new growing field which brings together the topic of natural language processing and machine learning as it is in essence a two class classification of natural language texts. An important feature of spam recognition is that it is a cost-sensitive classification: misclassification of a nonspam mail as spam is generally a more severe error than misclassifying a spam mail as non-spam. In order to be compared, the methods applied to this field should be all evaluated with the same corpus and within the same cost-sensitive framework. In this paper, the performances of Support Vector Machines (SVM), Neural Networks (NN) and Naϊve Bayes (NB) techniques are compared using a publicly available corpus (LINGSPAM) for different cost scenarios. The training time complexities of the methods are also evaluated. The results show that NN has significantly better performance than the two other, having acceptable training times. NB gives better results than SVM when the cost is extremely high while in all other cases SVM outperforms NB.

We propose a text summarization system known as

MySum

in finding the significance of sentences in order to produce a summary based on asymmetric word similarity and topic similarity. We use mass assignment theory to compute similarity between words based on the basis of their contexts. The algorithm is incremental so that words or documents can be added or subtracted without massive re-computation. Words are considered similar if they appear in similar contexts, however, these words do not have to be synonyms. We also compute the similarity of a sentence to the topic using frequency of overlapping words. We compare the summaries produced with the ones by humans and other system known as TF.ISF (

term frequency-inverse sentence frequency

). Our method generates summaries that are up to 60% similar to the manually created summaries taken from DUC 2002 test collection.

An artificial neural network with all its elements is a rather complex structure, not easily constructed and/or trained to perform a particular task. Consequently, several researchers used genetic algorithms to evolve partial aspects of neural networks, such as the weights, the thresholds, and the network architecture. Indeed, over the last decade many systems have been developed that perform total network induction. In this work it is shown how the chromosomes of Gene Expression Programming can be modified so that a complete neural network, including the architecture, the weights and thresholds, could be totally encoded in a linear chromosome. It is also shown how this chromosomal organization allows the training/adaptation of the network using the evolutionary mechanisms of selection and modification, thus providing an approach to the automatic design of neural networks. The workings and performance of this new algorithm are tested on the 6-multiplexer and on the classical exclusive-or problems.

The effects of various neighborhood models on the particle swarm algorithm were investigated in this paper. We also gave some additional insight into the PSO neighborhood model selection topic. Our experiment results testified that the gbest model converges quickly on problem solutions but has a weakness for becoming trapped in local optima, while the lbest model converges slowly on problem solutions but is able to “flow around” local optima, as the individuals explore different regions. The gbest model is recommended strongly for unimodal objective functions, while a variable neighborhood model is recommended for multimodal objective functions.

The multiobjective 0/1 knapsack problem is a generalization of the well known 0/1 knapsack problem in which multiple knapsacks are considered. A new evolutionary algorithm for solving multiobjective 0/1 knapsack problem is proposed in this paper. This algorithm used a ε-dominance relation for direct comparison of two solutions. This algorithm try to improve another algorithm which also uses an ε domination relation between solutions. In this new algorithm the value of ε is adaptive (can be changed) depending on the solutions quality improvement. Several numerical experiments are performed using the best recent algorithms proposed for this problem. Experimental results clearly show that the proposed algorithm outperforms the existing evolutionary approaches for this problem.

In the last years, the guideline for injection control systems of diesel engines is the realization of a micro-controller, which is able in real-time to find, through an optimization procedure oriented to decrease polluting gases emitted by the engines and the engine fuel consumptions, the “optimal” injection strategy related to driver load demand. The “optimal” injection strategy is associated with the optimal tradeoff among the following conflicting objectives: maximize torque, minimize fuel consumptions, reduce noise and polluting emissions (NOx and soot).

In this paper a new autotuning fuzzy

PID

control method for

SISO

and

MIMO

systems is proposed. The fuzzy autotune procedure adjusts on-line the parameters of a conventional PID controller located in the forward loop of the process. Fuzzy rules, employed to determine the set of PID gains, are based on the representation of human expertise on how can be the behaviour of gain and phase margins of a control system to efficiently compensating the system errors. The proposed control scheme offers advantages over the conventional fuzzy controller such as:

i

) a systematic design is attained in both SISO and MIMO cases;

ii

) it is necessary only one rule base for all loops;

iii

) the tuning mechanism is simple and control operators can easily understand how it works; and

iv

) it is completely autotuned, requiring only one relay feedback experiment per loop. Simulation examples and a practical essay are assessing the effectiveness of the proposed control algorithms.

In this paper, the controller’s tuning and performance of a speed controller prototype for switched reluctance motor is presented using significant experimental tests. The system uses an on-line learning mechanism to acquire and modify, if needed, the “good” fuzzy control rules. Experimental essays are analyzed and discussed in order to reveal some advantages of having a learning speed controller for the SR machine, and also the drawbacks that the use of using these controllers can introduce to the drive system and possible ways to overcome them.

We describe in this paper a new approach for pattern recognition using modular neural networks with a fuzzy logic method for response integration. We proposed a new architecture for modular neural networks for achieving pattern recognition in the particular case of human faces and fingerprints. Also, the method for achieving response integration is based on the fuzzy Sugeno integral with some modifications. Response integration is required to combine the outputs of all the modules in the modular network. We have applied the new approach for fingerprint and face recognition with a real database from students of our institution.

There exists no standard method for obtaining a nonlinear input-output model using external dynamic approach. In this work, we are using an evolutionary optimization method for estimating the parameters of an NFIR model using the Wiener model structure. Specifically we are using a Breeder Genetic Algorithm (BGA) with fuzzy recombination for performing the optimization work. We selected the BGA since it uses real parameters (it does not require any string coding), which can be manipulated directly by the recombination and mutation operators. For training the system we used amplitude modulated pseudo random binary signal (APRBS). The adaptive system was tested using sinusoidal signals.

We describe in this paper the use of hierarchical genetic algorithms for fuzzy system optimization in intelligent control. In particular, we consider the problem of optimizing the number of rules and membership functions using an evolutionary approach. The hierarchical genetic algorithm enables the optimization of the fuzzy system design for a particular application. We illustrate the approach with the case of intelligent control in a medical application. Simulation results for this application show that we are able to find an optimal set of rules and membership functions for the fuzzy system.

Great amount of data under varying intrinsic features is thought of as high dimensional nonlinear manifold in the observation space. How to analyze the mapping relationship between the high dimensional manifold and the corresponding intrinsic low dimensional one quantitatively is important to machine learning and cognitive science. In this paper, we propose SVD (singular value decomposition) based magnification factors and spread direction for quantitative analyzing the relationship. The result of conducting experiments on several databases show the advantages of this proposed SVD-based approach in manifold learning.

This paper describes the design of multi-category support vector machines (SVMs) for classification of bags. To train and test the SVMs a collection of 120 images of different types of bags were used (backpacks, small shoulder bags, plastic flexible bags, and small briefcases). Tests were conducted to establish the best polynomial and Gaussian RBF (radial basis function) kernels. As it is well known that SVMs are sensitive to the number of features in pattern classification applications, the performance of the SVMs as a function of the number and type of features was also studied. Our goal here, in feature selection is to obtain a smaller set of features that accurately represent the original set. A Kfold cross validation procedure with three subsets was applied to assure reliability. In a kernel optimization experiment using nine popular shape features (area, bounding box ratio, major axis length, minor axis length, eccentricity, equivalent diameter, extent, roundness and convex perimeter), a classification rate of 95% was achieved using a polynomial kernel with degree six, and a classification rate of 90% was achieved using a RBF kernel with 27 sigma. To improve these results a feature selection procedure was performed. Using the optimal feature set, comprised of bounding box ratio, major axis length, extent and roundness, resulted in a classification rate of 96.25% using a polynomial kernel with degree of nine. The collinearity between the features was confirmed using principle component analysis, where a reduction to four components accounted for 99.3% of the variation for each of the bag types.

In this paper, we elaborate on the well-known relationship between Gaussian Processes (GP) and Support Vector Machines (SVM) under some convex assumptions for the SVM loss function for the binary classification. We also investigate the relationship between the Kernel method and Gaussian Processes. This paper concentrates mainly on the derivation of the error bar approximation for classification. An error bar formula is derived based on the convex loss function approximation.

To determine the optimal configuration of the kernel parameters, the physical characteristic of the mapped least squares (LS) support vector machine (SVM) is investigated by analyzing the frequency responses of the filters deduced from the support value and LS-SVM itself. This analysis of the mapped LS-SVM with Gaussian kernel illustrates that the optimal configuration of the kernel parameter exists and the regulation constant is directly determined by the frequency content of the image. The image regression estimation experiments demonstrate the effectiveness of the presented method.

SVMs have been used to classify the labeled data. While in many applications, to label data is not an easy job. In this paper, a new quadratic program model is presented so that SVMs can be used to classify the unlabeled data. Based on this model, a new semi-supervised SVM is also presented. The experiments show that the new semi-supervised SVM can be used to improve the correct rate of classifiers by introducing the unlabeled data.

In this paper, a Type-II fuzzy and neural network system is introduced to car auxiliary control system for inexperienced drivers. The approach applies Type-II fuzzy logic to generate the training data set for the neural network. The Type-II fuzzy logic is launched by the component called the fuzzy controller. With Type-II fuzzy logic and neural network system, the auxiliary system can manipulate more driving conditions so that enables the neural network to predict more reliably. The results of the approach is compared with Type-I fuzzy/neural system and discussed in detail.

In this paper, an automatic synthesis methodology based on evolutionary computation is applied to evolve neural controllers for a homogeneous team of miniature autonomous mobile robots. Both feed-forward and recurrent neural networks can be evolved with fixed or variable network topologies. The efficacy of the evolutionary methodology is demonstrated in the framework of a realistic case study on collective robotic inspection of regular structures, where the robots are only equipped with limited local on-board sensing and actuating capabilities. The neural controller solutions generated during evolutions are evaluated in a sensorbased embodied simulation environment with realistic noise. It is shown that the evolutionary algorithms are able to successfully synthesize a variety of novel neural controllers that could achieve performances comparable to a carefully hand-tuned, rule-based controller in terms of both average performance and robustness to noise.

Autonomous mobile robots capable of intelligent behavior must operate with minimal human interaction, have the capability to utilize local resources, and routinely make closed-loop decisions in real-time based on sensor data inputs. One of the bottlenecks in achieving this is an often computationally intensive perception process. In this paper, we discuss a class of cognitive sensor devices capable of intelligent perception that can facilitate intelligent behavior. The primary emphasis is on achieving safe mobile guidance for planetary exploration by distributing some of the perception functionality to self-contained sensors. An example cognitive sensor, called the traversability sensor, is presented, which consists of a camera and embedded processor coupled with an intelligent visual perception algorithm. The sensor determines local terrain traversability in natural outdoor environments and, accordingly, directs movement of a mobile robot toward the safest visible terrain area in a self-contained fashion, placing minimal burden on the main processor. A cognitive sensor design approach is presented and a traversability sensor prototype is described.

Automated Guided Vehicles (AGV) are commonly used to transfer parts in flexible manufacturing systems (FMS). AGV dispatching deals with assigning vehicles to move parts based on the relationship between AGV and parts availability. The dispatching method influences the performance of the FMS and can lead to cost saving and improvement. The production environment is complex and the dispatching decision is sometimes vague and obscure. To deal with such complicated environments a fuzzy logic dispatching model has been developed.

An optimization approach based on Multi-Objective Evolutionary Algorithms (MOEA) - the Reduced Pareto Set Genetic Algorithm with Elitism (RPSGAe)-is used to design barrier screws for single screw extruders. A numerical modeling routine able to describe the thermo-mechanical experience suffered by the polymer inside the extruder is developed. A specific screw design methodology is proposed and used to optimize conventional and barrier screws. In this methodology, the fittest screw geometry is chosen after performing a sensitivity study of a set of “best” screws to limited changes in operating conditions.

This paper presents a fuzzy neural network approach to virtual product design. In the paper, a novel soft computing framework is developed for engineering design based on a hybrid intelligent system technique. First, a fuzzy neural network (FNN) model is proposed for supporting modeling, analysis and evaluation, and optimization tasks in the design process, which combines fuzzy logic with neural networks. The developed system provides a unified soft computing design framework with computational intelligence. The system has self-modifying and self-learning functions. Within the system, only one network is needed training for accomplishing the evaluation, rectification/modification and optimization tasks in the design process.

This paper presents the development of an Internet server controller based intelligent maintenance system for products. It also discusses on how to develop products and manufacturing systems using Internet-based intelligent technologies and how to ensure the product quality, coordinate the activities, reduce costs and change maintenance practice from breakdown reaction to breakdown prevention. In the paper, an integrated approach using hardware and software agents (watchdog agent) is proposed to develop the Internet server controller based integrated intelligent maintenance system. The effectiveness of the proposed scheme is verified by developing a real system for the washing machine.

Every individual intelligent technique has particular computational properties (e.g. ability to learn, explanation of decisions) that make it suited for particular problems and not for others. There is now a growing realization in the intelligent systems community that many complex problems require hybrid solutions. This paper presents a novel genetic (fuzzy) knowledge Petri net based approach for the integration of KBSs, fuzzy logic (FL) and GAs. Two genetic Petri net models are presented for integrating genetic models and knowledgebased models, including genetic knowledge Petri nets and genetic fuzzy knowledge Petri nets (GKPN, GFKPN). The GKPN and GFKPN models can be used for (fuzzy) knowledge representation and reasoning, especially for (fuzzy) knowledge base tuning and verification & validation. An application example for the proposed models in engineering design is given.

With the increase of global marketing competition and the dynamic change of the marketing environment, products produced by enterprises that are needed by consumers are more individualized and more flexible. It is an important research project for product consumers and enterprises of how to gain product information effectively. Intelligent customization technology of individual products in remote locations is one of the important sides of this project. In this paper, a customized motorcycle is taken as an example to study individual product specifications. A product customization agent is designed to match the contents and to calculate the methods of product value itemized. Using the agents, the traditional delivery of products through addresses can be replaced by product customization communization between consumers and enterprises. Furthermore a calculation method of individual motorcycle customization is proposed.

The main objective of this intelligent designing research is to provide an astute environment for designers. This type of environment is to be designed and processed so that it is compatible with a person’s thoughts. In this environment the designers’ thoughts are effectively taken to process functional requirements of products. The concept of designing and evaluating the schemes of the products can be achieved and furthermore optimized, the structure designing and its manufacturability analysis are carried out. This paper studies cell classes of the functions and structures. By the use of artificial intelligent technology the knowledge required and the innovative scheme expressed are developed. As an example, an intelligent design of a motorcycle is used to illustrate the application of the method to produce an innovative scheme.

In chaotic cryptosystems, it is recognized that using (very) high dimensional chaotic attractors for encrypting a given message may improve the privacy of chaotic encoding. In this paper, we study a kind of hyperchaotic systems using some classical methods. The results show that besides the high dimension, the sub-Nyquist sampling interval is also an important factor that can improve the security of the chaotic cryptosystems. We use the method of time series analysis to verify the result.