scispace - formally typeset
Search or ask a question

Showing papers in "Journal of Intelligent and Fuzzy Systems in 1994"


Journal ArticleDOI
TL;DR: An efficient method for estimating cluster centers of numerical data that can be used to determine the number of clusters and their initial values for initializing iterative optimization-based clustering algorithms such as fuzzy C-means is presented.
Abstract: We present an efficient method for estimating cluster centers of numerical data. This method can be used to determine the number of clusters and their initial values for initializing iterative optimization-based clustering algorithms such as fuzzy C-means. Here we use the cluster estimation method as the basis of a fast and robust algorithm for identifying fuzzy models. A benchmark problem involving the prediction of a chaotic time series shows this model identification method compares favorably with other, more computationally intensive methods. We also illustrate an application of this method in modeling the relationship between automobile trips and demographic factors.

2,815 citations


Journal ArticleDOI
TL;DR: This work develops, based upon the mountain clustering method, a procedure for learning fuzzy systems models from data, and uses a back propagation algorithm to tune the model.
Abstract: We develop, based upon the mountain clustering method, a procedure for learning fuzzy systems models from data. First we discuss the mountain clustering method. We then show how it could be used to obtain the structure of fuzzy systems models. The initial estimates of this model are obtained from the cluster centers. We then use a back propagation algorithm to tune the model.

670 citations


Journal ArticleDOI
TL;DR: A methodology is suggested here for the development of fuzzy systems models that is a combination of the AI-expert systems approach, with its heavy dependence on the use of expert knowledge, and the neural network-type systems building, and its emphasis on learning from data observations.
Abstract: A methodology is suggested here for the development of fuzzy systems models that is a combination of the AI-expert systems approach, with its heavy dependence on the use of expert knowledge, and the neural network-type systems building, with its emphasis on learning from data observations. We use expert-provided information in the form of template linguistic values to induce potential elemental rules for the knowledge base of the system model. We then introduce input-output observations into a simple learning mechanism to obtain weights characterizing the effect of each of the potential elemental rules on the overall systems model. The development of the learning mechanism is based on a representation of systems models combining fuzzy logic and Dempster-Shafer theory, which we previously introduced.

59 citations


Journal ArticleDOI
Wonseek Yu1, Zeungnam Bien1
TL;DR: An effective method for obtaining a final conclusion from such inconsistent if-then rules of fuzzy logic controller FLC is proposed and a “minimum distance inference method” is applied for FLC.
Abstract: As major functional units, the fuzzy logic controller FLC includes a fuzzy rule base and an inference engine. In constructing a fuzzy rule base, however, uncertainties and imprecision in the information about the controlled plant or in the extracted knowledge about actions of operators may result in inconsistent rules. But conventional inference methods for FLC often fail to handle such inconsistencies. In this article is proposed an effective method for obtaining a final conclusion from such inconsistent if-then rules. Also, as an alternative to conventional methods, a “minimum distance inference method” is applied for FLC. In the method, a metric is introduced to represent the distance between two fuzzy sets, and a new measure of certainty is used as a weight in the optimization to find a conclusion fuzzy set. The usefulness of the proposed methodologies is shown via a computer simulation of controlling a realistic overhead crane system.

50 citations


Journal ArticleDOI
TL;DR: Computer simulation shows that neuro-fuzzy identification is very effective in modeling fuzzy systems the fuzzy rules of which cannot be obtained easily and the neuro- fuzzy controller gives very effective control results by a learning process.
Abstract: A neuro-fuzzy identifier for fuzzy modeling of a system is explained, and a control structure using this neurofuzzy identifier is proposed. The neuro-fuzzy identifier contains not only an adaptive clustering process for determining center points of the input and virtual output membership functions but also an adaptive process for deciding the shapes of the input membership functions. Moreover, linguistic fuzzy rules of a system can be obtained from the proposed neuro-fuzzy identifier, which can learn the initial implication fuzzy control rules of the system and then compensate for the error of the initial fuzzy control rules by a feedback control structure that maintains the stability of the system. Computer simulation shows that neuro-fuzzy identification is very effective in modeling fuzzy systems the fuzzy rules of which cannot be obtained easily and the neuro-fuzzy controller gives very effective control results by a learning process.

31 citations


Journal ArticleDOI
TL;DR: An innovative scheme for adaptive feedthrough cancellation in dynamic current mirrors is presented, which aids in reducing the most significant error source in chaotic current-mode discrete maps.
Abstract: This article presents architectures and circuit techniques to design chaotic piecewise-linear discrete maps in CMOS VLSI Maps are presented for white and 1/fγ noise generation, as well as to display bidimensional strange attractors The proposed circuits use only MOS transistors and thus are suitable for implementation in standard CMOS technologies containing only one poly layer A family of CMOS building blocks for piece-wise-linear function interpolation in current-domain is presented based upon the use of a current switch Novel circuit strategies are given for improved current-switch performance as well for high-resolution current comparison These strategies can be used for the design of chaotic current-mode circuits, as well for the representation problem via basis functions in feedforward neural networks and fuzzy inference engines Also, an innovative scheme for adaptive feedthrough cancellation in dynamic current mirrors is presented, which aids in reducing the most significant error source in chaotic current-mode discrete maps The article includes experimental results from monolithic n-well CMOS 16 μm prototypes of the building blocks as well as complete systems; in particular, a chaotic neuron according to the Aihara's model and a broadband white noise generator

24 citations


Journal ArticleDOI
TL;DR: A new neural network for implementing fuzzy systems is proposed, and it is proved that it can represent any continuous function over a compact set and yield better performance than previously proposed methods for extracting fuzzy systems and neural networks from input-output data.
Abstract: We propose a new neural network for implementing fuzzy systems, and we prove that it can represent any continuous function over a compact set. We propose and test a method for building a fuzzy neural system from input-output data. We analyze the output data using fuzzy c-means to obtain the number of rules and to set some of the initial weights in the network. Then, we use this fuzzy neural network to identify the input variables and to determine the number of input membership functions. We show that the resulting model is simpler and yields better performance than previously proposed methods for extracting fuzzy systems and neural networks from input-output data.

21 citations


Journal ArticleDOI
TL;DR: This article considers the problem of object identification when there is imprecise and perhaps conflicting information from information sources of unequal reliability and proposes a type II fuzzy set in which the memberships are themselves fuzzy sets.
Abstract: In this article we consider the problem of object identification when there is imprecise and perhaps conflicting information from information sources of unequal reliability. Data is represented by fuzzy sets, and the decision set in which the value of this imprecise information is modified by its importance in the context of the specific problem at hand is a type II fuzzy set in which the memberships are themselves fuzzy sets. Three criteria are proposed for making the classification based on the information contained in this type II fuzzy set. A simple example illustrates the procedure.

11 citations


Journal ArticleDOI
TL;DR: In this article, attributes are considered to be fuzzy sets: certain rules and possible rules are constructed and how much a fuzzy diagnosis is definable in terms of fuzzy attributes is studied.
Abstract: In this article we consider attributes to be fuzzy sets. Knowledge acquisition takes place by looking at examples. In each example, attributes as well as a corresponding decision is made available. The decision may be a fuzzy diagnosis. Based on these examples two sets of fuzzy rules are constructed: certain rules and possible rules. Corresponding measures of how much we believe these rules are also constructed. The concept of how much a fuzzy diagnosis is definable in terms of fuzzy attributes is studied. Finally, classifications and some of their properties are analyzed.

11 citations


Journal ArticleDOI
TL;DR: A feedback-linearization/fuzzy logic control scheme for a robotic manipulator with link flexibility is designed and a linguistic fuzzy controller that consists of 33 condition-action rules is used to command the rigid modes to track the desired trajectories while maintaining the residual vibration as small as possible.
Abstract: A feedback-linearization/fuzzy logic control scheme for a robotic manipulator with link flexibility is designed. This control scheme is composed of a feedback-linearization inner-loop control and a fuzzy linguistic outer-loop control. A reduced-order computed torque ROCT control is first used to linearize the whole system to a Newton's-law-like system, then a linguistic fuzzy controller that consists of 33 condition-action rules is used to command the rigid modes to track the desired trajectories while maintaining the residual vibration as small as possible.

11 citations


Journal ArticleDOI
TL;DR: A fuzzy-algebra analysis is proposed in this article that has the potential to appropriately reflect the information available and portray uncertainties well, especially for abnormal environments.
Abstract: Many safety analyses depend on uncertain inputs and on mathematical models chosen from various alternatives, but give fixed results implying no uncertainty. Conventional uncertainty analyses help, but are also based on assumptions and models, the accuracy of which may be difficult to assure. Some of the models and assumptions that on cursory examination seem reasonable can be misleading. As a result, quantitative assessments, even those accompanied by uncertainty measures, can give unwarranted impressions of accuracy. Because analysis results can be a major contributor to a safety-measure decision process, risk management depends on relating uncertainty to only the information available. The uncertainties due to abnormal environments are even more challenging than those in normal-environment safety assessments, and therefore require an even more cautious approach. A fuzzy-algebra analysis is proposed in this article that has the potential to appropriately reflect the information available and portray uncertainties well, especially for abnormal environments.

Journal ArticleDOI
Kwang-Choon Kim1, Jong-Hwan Kim1
TL;DR: A multicriteria fuzzy control MFC based on fuzzy measures and fuzzy integrals that has excellent control performance compared to conventional fuzzy control by computer simulations as well as via experiments performed on a DC servomotor angular position control.
Abstract: In this article, we propose a multicriteria fuzzy control MFC based on fuzzy measures and fuzzy integrals. The basic idea underlying this approach is based on analyzing the source of attributes of the system output response and applying the fuzzy measure and integral theory to the existing fuzzy control. With this scheme, we can tune the three attributes of rise time, overshoot, and settling time of the output response. We demonstrate that MFC has excellent control performance compared to conventional fuzzy control by computer simulations as well as via experiments performed on a DC servomotor angular position control. Moreover, our scheme can be easily implemented in practice simply by adding weight terms to existing fuzzy rules and assigning the weight to each rule by employing the fuzzy measure and integral.

Journal ArticleDOI
TL;DR: The effectiveness of the FCNN is illustrated by applying the model to a number of test data sets, analyzing the hardware complexity of the architecture, and comparing the performance to that of fuzzy c-means FCM algorithm.
Abstract: This article presents a fuzzy clustering neural network FCNN model that uses Gaussian nonlinearity. A learning algorithm, based on direct fuzzy competition between the nodes, is introduced. The connecting weights, which are adaptively updated in batch mode, converge towards values that are representative of the clustering structure of the input patterns. Mapping the proposed algorithm onto the corresponding architecture with three types of processing cells, it is feasible to implement the FCNN in parallel. The effectiveness of the FCNN is illustrated by applying the model to a number of test data sets, analyzing the hardware complexity of the architecture, and comparing the performance to that of fuzzy c-means FCM algorithm.

Journal ArticleDOI
TL;DR: By using fuzzy inference to determine the switching scheme of the VSC, the original robustness and fast response time of theVSC can be retained while reducing the control chattering.
Abstract: Today's technology requires controls capable of handling highly nonlinear, time-varying, uncertain systems. Variable structure control VSC is one such control method. Variable structure control is invariant to system perturbations and external disturbances; however, a high-frequency control chattering exists that renders the VSC impractical for most applications. Fuzzy inference can be used to reduce the chattering. By using fuzzy inference to determine the switching scheme of the VSC, the original robustness and fast response time of the VSC can be retained while reducing the control chattering. Optimization of the fuzzy parameters using genetic algorithms will also produce a system with improved response time and accuracy.

Journal ArticleDOI
TL;DR: A model of a fully interconnected neural network composed of neurons having a refractory period and integrating the afferent signals is presented and its potentialities, as well as its possible applications, are discussed.
Abstract: A model of a fully interconnected neural network composed of neurons having a refractory period and integrating the afferent signals is presented. A theoretical analysis of its dynamics is carried out, considering the general case with periodic input and output sequences from which a teaching algorithm for storing periodic sequences in the network is derived. Finally, the performances of the network are illustrated through a selected example and its potentialities, as well as its possible applications, are discussed.

Journal ArticleDOI
TL;DR: It was observed that although the tuning rules were elicited from human experts, it was not essential in this study to emulate the human's information aggregation processes and it was shown that these rules significantly improved the performance of the tuning system.
Abstract: The electric circuit tuning process, which requires manual tuning of a set of trimmers by the human operators, was automated through the application of a fuzzy knowledge-based system. In a complex tuning process, multiple circuit specification criteria had to be simultaneously satisfied by several trimmers. The main objective of this study was to examine different tuning evidence aggregation methods in order to reduce the overall circuit tuning time. In the proposed fuzzy knowledge-based system, the effect of each trimmer on each tuning criterion was expressed by a grade of the fuzzy membership related to each circuit output. The overall effect of each trimmer on the circuit tuning performance was modeled by an aggregation of the grades used for trimmer selection. The model simulation results showed that the geometrical average operator was the best method for evidence aggregation. To compensate for the lack of fuzzy rules, some heuristic rules were also introduced to adjust the aggregated evidence values. It was shown that these rules significantly improved the performance of the tuning system. Finally, it was observed that although the tuning rules were elicited from human experts, it was not essential in this study to emulate the human's information aggregation processes. This was due to the fact that in the manual tuning process the human aggregation of evidence about circuit performance did not necessarily provide the best solution for the intended task.

Journal ArticleDOI
TL;DR: This method outperforms the traditional open-loop, single step learning procedure when applied to the texture recognition problem of 12 classes and to the image annotation problem of natural scenes.
Abstract: This article presents a novel approach to noise-tolerant symbolic learning. The approach is applicable to many existing learning programs, and it has been implemented for the AQ-14 rule learning and C4.5 decision tree learning programs. In this approach, the system 1 acquires initial concept descriptions from pre-classified attributional training data, 2 optimizes concept descriptions to improve their descriptiveness, 3 applies optimized concept descriptions to filtrate initial training data, and 4 repeats the learning process from filtered data. This method outperforms the traditional open-loop, single step learning procedure when applied to the texture recognition problem of 12 classes and to the image annotation problem of natural scenes.

Journal ArticleDOI
TL;DR: A neural network approach is presented for the adaptive control of real-time systems and an algorithm partially inverting the network is used to control the system where only controllable inputs are adjusted based on the gradient of a control error.
Abstract: A neural network approach is presented for the adaptive control of real-time systems. Forward modeling and the partial inversion algorithm are used to solve the one-to-many mapping problem in constructing a neural controller. Inputs are disaggregated into controllable and uncontrollable inputs, and an algorithm partially inverting the network is used to control the system where only controllable inputs are adjusted based on the gradient of a control error. The suggested neural network scheme is applied to a traffic signal control system. The results show the effectiveness of the approach and suggest the potential applications to the real-time systems such as manufacturing control system, process control system, and communication network system.

Journal ArticleDOI
TL;DR: This article summarizes several attempts made to apply fuzzy logic to structural engineering and particularly to bridge engineering and two fuzzy models previously proposed in the literature and a third model adopted by the authors are presented here.
Abstract: This article summarizes several attempts made to apply fuzzy logic to structural engineering and particularly to bridge engineering. Many of these attempts have been implemented in depth and are readily available for practical use. The major research activity has been in the areas of structural damage assessment, bridge condition evaluation, and structural rating. In the case of bridge rating, the inspection procedure currently being used suggested the need for a multi-attributive decision making model. Two fuzzy models previously proposed in the literature and a third model adopted by the authors are presented here. The later produces a priority setting obtained through the solution of an eigenvalue problem involving a closest discrete pairwise matrix indicative of the relative structural importance of the different elements of the bridge components considered.

Journal ArticleDOI
TL;DR: A fuzzy environment is prepared to solve successfully the multicriteria load dispatch problem in power systems, viewed more realistically to contain some flexible or fuzzy operational constraints.
Abstract: The fuzzy linear programming method for solving optimization problems containing several objective functions is briefly discussed and adapted to deal with the various cases of fuzzy constraints. For each type of fuzzy constraints, proper piecewise linear membership functions are suggested, where some of them are decomposed into new simpler types in a manner to fit the method of solution based on the method used in Zimmerman 1978 and extended in Yang et al., 1991. Thus, a fuzzy environment is prepared to solve successfully the multicriteria load dispatch problem in power systems, viewed more realistically to contain some flexible or fuzzy operational constraints. A program is developed to test this algorithm applied to an economic-environmental load dispatch problem on a sample power system, where the total production cost and the total produced emission are to be minimized. The results of various cases of fuzzy and crisp modes of the load and the generation range of the units are analyzed in the case study. It will be seen that the method also can be applied efficiently when priorities are assigned to the objective functions.

Journal ArticleDOI
TL;DR: Two neurocontroller architectures: a the specialised learning, and b the emulator/controller are compared, which allow a neural network to learn the difficult task of balancing an inverted pendulum on a moving cart.
Abstract: The use of neural networks in control applications has made rapid progress in recent years. In this article, two neurocontroller architectures: a the specialised learning, and b the emulator/controller are compared. Both architectures allow a neural network to learn the difficult task of balancing an inverted pendulum on a moving cart. Computer simulation results show that learning is quite rapid and both architectures take a small number of trials 15--25 for balancing the inverted pendulum.

Journal ArticleDOI
TL;DR: A case-by-case comparison of two multiobjective decision methods for selecting a retaining wall type from a set of possible alternatives given a finite set of selection criteria of concern to the decision maker indicates superior performance of the fuzzy method for retaining wall selection.
Abstract: Multiobjective decisions are frequently required for planning and managing highway construction projects. One example is the selection of an earth retaining structure. This article presents a case-by-case comparison of two multiobjective decision methods for selecting a retaining wall type from a set of possible alternatives given a finite set of selection criteria of concern to the decision maker. The first method is a weighting technique. The second method is based on fuzzy logic. The methods are compared according to agreement of outcome, sensitivity to changes in the importance of the selection criteria, and sensitivity to changes in the degree to which each alternative satisfies the criteria. Results indicate superior performance of the fuzzy method for retaining wall selection.

Journal ArticleDOI
TL;DR: This article analyzes the process of executing the knowledge base KB of a fuzzy control system following the “Sup-*” compositional rule of inference, and two different methodologies are derived: rule by rule and global implication matrices.
Abstract: In this article we analyze the process of executing the knowledge base KB of a fuzzy control system following the “Sup-*” compositional rule of inference [Zadeh LA 1973: “Outline of a new approach to the analysis of complex systems and decision processes.” IEEE Trans. Syst. Man Cybern. 31: 28--44]. From the expressions obtained, two different methodologies are derived: rule by rule and global implication matrices. Two systolic architectures following these methodologies are proposed using a partitioning and projection technique. Both architectures permit executing any KB independently from the size parameters that characterize it, such as number of rules, number of antecedents and consequents, and discretization degree of the universes of discourse. The solutions also present great flexibility with respect to the operators implementing the inference process implication and conjunction functions and “*” t-norm. The selection of one of these solutions is conditioned by the particular features of the system to be controlled: the relationship between the parameters that characterize the KB and the most adequate reasoning mechanism to carry out the control process. The need to achieve a compromise between operational speed, flexibility of the reasoning mechanism, and complexity of the KB will determine the selection of one solution or the other for each specific case.

Journal ArticleDOI
TL;DR: A tree structured network for improving the performance of the feedforward neural network FN classifier by improving the recognition accuracy from 80% for the single FN to 97.7% using the same simple set of features.
Abstract: This article introduces a tree structured network for improving the performance of the feedforward neural network FN classifier. The building blocks of the tree are the feedforward neural network with backpropagation learning scheme and the simple logical OR neural network ORNN. The confusion matrix CM, resulting from some preliminary experiments, is used to divide the considered patterns into groups in several primary stages until no more grouping could be obtained. The proposed structure can be used for any pattern classification problem. In this article, the testing environment is the isolated handwritten Arabic character set, which is a problem of reasonable complexity. Two simple kinds of feature vectors are used to represent characters before the FN. The use of the proposed tree structure improved the recognition accuracy from 80% for the single FN to 97.7% using the same simple set of features. The results show that with the proposed method, better classification results could be obtained without having to introduce more complex features.

Journal ArticleDOI
TL;DR: The study is aimed at the development of neural networks useful in modelling processes of fuzzy decision-making with variety of logic-oriented neurons both aggregative and referential processing units.
Abstract: The study is aimed at the development of neural networks useful in modelling processes of fuzzy decision-making. A variety of logic-oriented neurons both aggregative and referential processing units makes it possible to directly treat the efficacies of the decision problem at hand and handle them within the topology of the network. The learning capabilities of the decision networks are thoroughly investigated. Several aspects of these architectures including dynamical characteristics of the decision processes and representing and processing of incomplete or uncertain information are also addressed.

Journal ArticleDOI
TL;DR: This article shows that Phong reflectance functions from three different directions can be learned through neural networks by treating the values of three image irradiances as inputs while treating the corresponding two surface gradient parameters as outputs.
Abstract: This article proposes a new idea to determine surface gradients uniquely by using neural networks that can learn any reflectance maps. The Phong illuminating function is used to represent the glossy surface, including Lambertian surfaces, and it includes three parameters that characterize the reflectance property of the object susrface. This article shows that Phong reflectance functions from three different directions can be learned through neural networks by treating the values of three image irradiances as inputs while treating the corresponding two surface gradient parameters as outputs. Computer simulation was demonstrated for three layered networks. Learning was done for a spherical object and it was repeated by the back-propagation algorithm. The desirable surface gradients could be recovered by neural networks when any triples of image irradiances were inputed as a test pattern. Neural networks have a great capacity to store the three reflectance maps, and this method offers the advantage that special analysis to solve the simultaneous equations is not required as in the conventional method.

Journal ArticleDOI
Ali Bahrami1
TL;DR: The proposed system tries to mimic the thinking process of human designers by using FAM to retrieve the desired parts based on how well a design solution satisfies marketing demands and user specifications.
Abstract: The objective of the work presented in this article is to develop and test a new method of automating design by associating needs and wants of the end users to functional requirements and functional requirements to physical structures. This new method of design uses fuzzy associative memory FAM to generate an innovative design's components based solely on the desired needs and wants of the customers. The proposed system tries to mimic the thinking process of human designers by using FAM to retrieve the desired parts based on how well a design solution satisfies marketing demands and user specifications. Once a design's components have been selected based on the perceived needs, a new design will be realized by synthesizing the generated parts. Examples are given to show the evolution of new products during the conceptual design.

Journal ArticleDOI
TL;DR: A framework for using artificial neural networks to design injection molded parts is presented and an intelligent hybrid system that integrates the knowledge-base expert system and the neural networks together into one combined system is presented.
Abstract: A framework for using artificial neural networks to design injection molded parts is presented. This framework can be achieved through two approaches. The first approach uses a mapping technique to represent the design knowledge into an artificial neural network. This mapping approach is used in a distributed way to encode the design knowledge. The second approach uses an intelligent hybrid system that integrates the knowledge-base expert system and the neural networks together into one combined system. This tightly coupled intelligent system takes the benefits of the expert system and the neural networks and combines them to eliminate the limitations of each. Working design examples of these two implemented approaches will be explored. The concepts to be incorporated in future implementations will be mentioned.