Showing papers in "Intelligent Automation and Soft Computing in 2003"

Fuzzy Control for Pneumatic Muscle Tracking Via Evolutionary Tuning

Abstract: This paper studies the evolutionazy design of a fuzzy P+ID controller for an actual pneumatic muscle actuator system. The control of pneumatic muscles is a challenging problem because of their high degree of nonlineazity, time-varying parameters, and uncertainty. A fuzzy P+ID controller is constructed using an incremental fuzzy logic controller in place of the proportional term in a conventional PID controller. Several controller parameters are optimized via an evolutionary algorithm. The optimization is performed using a recurrent neuro-fuzzy dynamic model of the muscle rather than the muscle itself. Control results are presented, where the control objective is to force muscle length to follow a reference signal under a load. After evolutionary design, excellent tracking performance is obtained with the real muscle without the need for further tuning of controller pazameters. The tracking performance is compared to that of another fuzzy controller.

Dynamic neural networks for modeling and control of nonlinear systems

Abstract: This paper considers the design of a dynamic neural network (DNN) for modeling of a class of nonlinear systems for the purpose of real-time control. The primary contribution of the paper is in developing a DNN estimator with a stable training technique for on-line modeling of unknown (black box) dynamic nonlinear systems. The DNN acts as a generic model of the system, which can be trained on-line and, hence, can be utilized for the implementation of an adaptive model-based control strategy. The training of the network is based on a novel scheme that arranges the outputs of the hidden layer of the DNN into a set of basis functions. This allows for the derivation of a stable rule for the training of the DNN's weights and does not require random initialization of the weights.

Detection of Welding Flaws with MLP Neural Network and Case Based Reasoning

Abstract: The correct detection of welding flaws is important to the successful development of an automated weld inspection system. As a continuation of our previous efforts, this study investigates the performance of multi-layer perceptron (MLP) neural networks and case based reasoning (CBR) individually as well as their combined use. It is found that better performance is attained by all the methods tested in this study than that was obtained by the fuzzy clustering methods employed before. For each method, the effect of using different parameters is also investigated and discussed. An improvement of CBR performance is not guaranteed when the MLP NN based attribute weighting is used. In addition, none of the three combination-of-multiple-classifiers methods (majority voting, Borda count, and arithmetic averaging) tested improve the performance of the best individual classifiers.

Haptic Feedback in Teleoperation of Multifingered Robot Hands

Abstract: This paper presents a teleoperated robot hand system with haptic tactile feedback. This system enables an operator to control amulti-fmgered robot hand while feeling interactions of the robot hand with the remote environrent. We have designed a fingertip tactile sensor to capture high resolution and quality tactile images between the fmgertip and the environrent based on the optical total reflection principle. In order to feed back the tactile information to the operator, an electrotactile glove and a haptic display with 24 pins driven by structured DC solenoids have been developed. The haptic glove generates excitations when atouch/contact between the fingertips and the environment occurs. The tactile display can present contact points on the fmger tips. We have integrated the tactile sensor, the haptic glove and the tactile display into afive-fingered hand system and examined their performance and effectiveness by experiments.

Task Allocation and Communication Methodologies for Multi-Robot Systems

Abstract: Two of the most important aspects in the design of multi-robot systems aze the allocation of tasks among the robots and the robots’ communication in a productive and efficient manner. Task allocation methodologies must ensure that not only the global mission is achieved, but also that the tasks are well distributed among the robots. An effective task allocation approach considers the available resources, the entities to optimize (time, energy, quality), the capabilities of the deployable robots, and appropriately allocates the tasks accordingly. The communication capabilities allow the robots to implicitly or explicitly communicate their status and the needed information regazding each other, the environment, and the tasks. This paper provides a survey of task allocation and communication methodologies for multi-robot systems.

V-Lab®– A Distributed Intelligent Discrete-Event Environment for Autonomous Agents Simulationi

Abstract: This paper presents a fusion between discrete-event systems specification (DENS) and intelligent tools from soft computing. DENS provides a robust and generic environrnent for modeling and simulation applications employing single workstation, distributed, and real-time platforms. Soft computing is a consortium of tools for natural intelligence stemming from approximate reasoning (fuzzy logic), learning (neural network or stochastic learning automaton), optimization (genetic algorithms and genetic programming), etc. The outcome of this fusion is what is called “Intelligent DENS,” called IDEVS here. IDEVS is an element of a virtual laboratory, called V-Lab®, which is based on distributed multi-physics, multi-dynamic modeling techniques for multiple platforms. This paper will introduce IDEVS and V-Lab® and a theme example for amultiagent simulation of a number of robotic agents with a slew of dynamic models and multiple computer work stations.

Modeling of Supercritical Fluid Extraction by Neural Networks

Abstract: Modeling of the relationship between the pressure and yield of biomaterials is an essential issue in supercritical fluid extraction. In this paper, neural networks are proposed for modeling of supercritical fluid extraction. First a three-layer neural network with a fast learning algorithm is used, and its performance is compared to a conventional model of the Peng-Robinson equation of state. A novel hybrid model combining both a neural network and the Peng-Robinson equation is then proposed. With the learning capacity, the proposed models generally perform better than the conventional model that needs to select its parameters by trial and error. The effectiveness of the proposed approaches is demonstrated by simulation and comparison studies.

Automatic Friction Identification and Compensation with a Self-Adapting Dual Relay

Abstract: The application of a dual relay feedback approach towards the identification of frictional effects in servo-mechanisms will be presented in this paper. The friction model, thus identified, will consist of Coulomb (including possibly directional Coulomb or load bias) and viscous friction components, both of which can be automatically extracted from suitably designed relay experiments. At the same time, from the same experiments, the dynamical model of the servomechanical system can be obtained from the experiments. These models will be directly useful in the design of the feedback controller and the friction compensator. Results from simulation and experiments are presented to illustrate the practical appeal of the proposed method.

A New Fuzzy Gain Scheduling Scheme for the PID Controllers

Abstract: In this paper, a new fuzzy gain scheduling scheme for the PID controller have been proposed. Fuzzy IF-THEN rules aze used on-line to adjust the parameters of the PID controller based on the system error and its derivative. In terms of settling time, one percent peak overshoot and the integral of the time multiplied by the absolute error, the simulation results clearly indicate that the performance of the proposed scheme is better than the Ziegler-Nichols PID controller [1] and the fuzzy gain scheduling scheme of PID controller [2]. For illustration and compazison, numerical examples, using third and fourth order plants, are presented.

Virtual Creatures Controlled by Developmental and Evolutionary CPM Neural Networks

Abstract: In this paper, a system where virtual creatures called bugs navigating a grid-based environment, which is controlled by developmental and evolutionary CPM neural networks, is presented. Each bug is born with a certain amount of energy that decreases in the navigation and increases only when the bug gets food. The bug can accumulate experience, i.e. training instances, in its life, which is used to incrementally tune its CPM network to improve the chance of making good decisions in later navigation. If two bugs meet then they may fight each other or produce an offspring, which is detemvned by their gender. The controlling organ, i.e. the CPM neural network, of the offspring is inherited from its parents in a specific way that the experience, i.e. the training instances, of its parents instead of the knowledge, i.e. the architectures or the weights, of them is genetically transmitted. Simulations show that the CPM networks are valuable to the longevity of the bugs, which exhibits not only the impor...

A Novel Method for Decentralized Robust Exponential Stabilization of Large-Scale Systems

Abstract: A novel approach to the design of decentralized controllers for large-scale systems by dynamic/static output state feedback is presented. A new formulation of the interaction which introduces some degrees of freedom into the design procedure is offered. Sufficient conditions for exponential stability with desirable rate of decay and maximal robustness to unstructured uncertainties in the controller and plant parameters are established. The derived conditions are generic, applicable to nonsquare and nonminimum-phase systems, and independent of the number of system states, inputs and outputs. Based on minimal sensitivity design of isolated subsystems, an analytical method for the satisfaction of the aforementioned sufficient conditions is presented. To this end, through eigenstructure assignment, compact-form sufficient conditions for minimal sensitivity are derived. Illustrative examples are presented to demonstrate the effectiveness of the proposed methodology. Genetic algorithm is employed in th...

A Sensor-Based Robot Controller for Autonomous Exploration of Unknown Environments

Abstract: This paper describes an oscillator-based controller that was implemented on a two-wheeled, differential drive robot. The controller is loosely based on central pattern generator circuits seen in many animals, and was developed to operate in unknown or changing environments. A unique feature is that the controller adjusts its oscillator parameters based on the pattern of sensory feedback. The controller's performance is evaluated in three different test arenas, each with three variable lighting patterns. The goal of the robot was to seek out areas of bright light and “collect” as much light as possible during a trial. The results indicate that the performance of the self-adjusting oscillator-based controller exceeds that of afixed-oscillator controller, with the performance difference increasing as the complexity of the environment increases.

H∞ Controller Design of Mixed Eigenstructure Assignment and H∞ Filter for Flexible Structure Vibration Control

Abstract: This paper presents a new H∞ controller design approach of mixed eigenstructure assignment and Ĩ filter for flexible structure vibration control problem. The new approach not only realizes the desired system dynamic response in the time domain through the choice of a state feedback gain using eigenstructure assignment, it also guarantees the performance specifications in the frequency domain by constructing an H∞ filter for the generalized plant to estimate the system state variables. The design approach is straightforward and more effective in tackling the spillover problem for vibration control of flexible structures. The performance of the new approach is further validated via numerical simulation of the vibration control of a cantilever beam bonded with piezoelectric actuator and sensor.

Online Relay Automatic Tuning of Multi-Loop Pi Controllers

Abstract: In this paper, an approach for online relay automatic tuning of multi-loop PI controllers is described for multivariable processes. In the absence of significant inter-loop cross-coupling, only one closed-loop experiment may be necessary to tune the controllers to achieve specified closed-loop characteristics. When the inter-loop interaction is significant, multiple relay experiments can be conducted to derive a process model, based on which both the decoupler and the multi-loop controllers can be designed. Simulation examples (on two-inputs-two-outputs (TITO) processes) and areal-time experiment are provided to illustrate the practical appeal of the proposed method. The performance is compared to multi-loop control design based on the Biggest Log Modulus (BLT) method.

Predictive Iterative Learning Control

Abstract: There are applications in which the control system is commanded to execute identical and repetitive tasks. These include pick-and-place robot control, moulding control, and many semiconductor processes. The iterative learning control (ILC) method first introduced by Arimoto et al. (1984) is based on the previous control history and a learning mechanism. The monograph by Moore (1992) contains more detail on the background. A recent book (Chen and Wen, 1999) surveys the literature on ILC up to 1998.

The Application of Wavelets Transforms and Neural Networks to Speech Classification

Abstract: This paper proposes a hybrid wavelet–neural network approach to classify speech for multifunctional control applications. The classification of the consonants (b,d,g) is the focus of this work. MultiResolution Wavelet Analysis (MRWA) was used to extract utterance features while a modular Artificial Neural Network (ANN) was used for classification. The performance of the proposed method was compared to that of the cepstrum method. The results show that MRWA are superior to the cepstrum in at least two points: higher recognition rate and consistent output demonstrating higher reliability.

Adaptive Neural Network-Based Predictive Control for Nonlinear Dynamical Systems

Abstract: In the paper, we propose a predictive control scheme using a neural network-based prediction model for nonlinear processes. To identify the system dynamics, we approximate the nonlinear function with an affine function of some of its arguments and construct a special type of prediction model using three-layered feedforward neural networks. Using some available input-output data pairs of the plant, we estimate the weights of neural networks by the Gauss-Newton based Levenberg-Marquard method. To cope with load disturbances and reduce the effect of unmodelled dynamics in the control system, we implement an on-line adaptation algorithm. Comparative simulations are given to show superiority of the proposed predictive control method to the adaptive GPC algorithm for some processes.

Adaptive Intelligent Control of Economical Flows by Stochastic Dynamics

Abstract: Economical flows are studied by means of a particle model. Simulation methods are applied to the problem and a theoretical framework is constructed. Adaptive methods for the analysis, the evaluation and the control of economical flows are deduced.

An Algorithm For Siting Nonlinearly-Designed Robust Excitation Controllers in Power Systems

Abstract: In this paper, the problem of siting controllers developed based on a nonlineaz power system model is addressed. Emphasis is placed on improving transient response of the power system using limited number of controllers located at well selected sites. A performance index that measures quantitatively transient response of each generator and consequently effects of controllers is introduced. The nonlineazly-designed, decentralized, robust excitation controller is chosen as the class of controllers considered for siting, although all kinds of nonlinear controllers aze admissible. Numerical results of simulations validate the theoretical development in the paper.

Singular Perturbation Method in Control of Multiple Flexible-Joint Robots

Abstract: The trajectory-tracking control of a typical industrial robot with six revolute joints is studied, where the joint flexibility is taken into account. An explicit form of the control law is obtained by using the composite control technique. The stability of system is discussed. The global convergence of tracking error can be reached by tuning the controller parameters. This method has a unique advantage in handling a multiple flexible joint robot as well as hybrid rigid/flexible joint robot. Numerical simulation on a Fanuc-s430iw heavy-duty robot arm has been performed to verify the control algorithm. Satisfactory tracking performance has been achieved.

A Line Balancing and Productivity Improvement Method for Surface Mounting Machines

Abstract: In this paper, we propose a hierarchical optimization technique of the PCB assembly line including multi-head non-identical surface mounting machines. The optimization problem related to the PCB assembly line is generally divided into three sub-problems: a line balancing problem, a reel assignment problem and apick-and-place sequencing problem. Since each problem is known to be combinatorial and NP-hard, we propose ameta-heuristic algorithm such as genetic algorithm. Computer simulation results show that the proposed algorithm is superior to the heuristic algorithm that is commonly used in industry.

Fuzzy Data Window Memory and Its Application to System Identification by GA

Abstract: In this paper, an approach for storing and retrieving the past data by means of a novel Fuzzy Data Window Memory (FDWM) is reported The data, which is selected for memorization, is based on the highest firing strength of the fuzzy rule The size of the proposed FDWM is much smaller than traditional window memory with no degradation in performance A computer simulation study of one of the applications of the proposed FDWM is reported which uses the FDWM to reduce the training data set that will be used in the evaluation module in system identification by Genetic Algorithms (GA)

Hybrid Intelligent Control for Ship Steering

Abstract: This paper is concerned with the application of hybrid intelligent control techniques for improving the performance of ship steering. Hybrid intelligent controllers can make full use of the advantages of a variety of intelligent algorithms. In this paper, optimization with genetic algorithms is used in off-line learning periods and reinforcement learning and neural fuzzy control are integrated in on-line learning periods. This combination overcomes the need for measurement data as is the case in conventional hybrid intelligent algorithms. According to a new definition of fitness function, it is shown that the optimized result obtained is more suitable to the actual situation. Similarly, simulation results show that our hybrid intelligent controller can effectively improve the ship steering performance in cases where additional sea state disturbances are present. We feel that our hybrid intelligent controller is a promising alternative to conventional autopilots.