Showing papers in "Control and Intelligent Systems in 2006"

An improved robust adaptive fuzzy controller for MIMO systems

Abstract: This paper addresses robust fuzzy adaptive control for nonlinear multi-input multi-output systems in the presence of parametric uncertainties and external disturbances. A universal approximator is used to approximate the plant model. The boundedness of the variables involved and convergence towards zero of the tracking error are guaranteed by a supervisory control. The effect of both the approximation errors and the external disturbances is attenuated to a prescribed level via an H∞ supervisor. Sufficient conditions are derived for robust stabilization in the sense of Lyapunov asymptotic stability. The effectiveness of the proposed controller design methodology is demonstrated by numerical simulations of a two-link robot.

Case-based plan recognition with novel input

Abstract: Our research investigates a case-based approach to plan recognition using incomplete incrementally learned plan libraries. To learn plan libraries, one must be able to process novel input. Retrieval based on similarities among concrete planning situations rather than among planning actions enables recognition despite the occurrence of newly observed planning actions and states. In addition, we explore the benefits of predictions using a measure that we call abstract similarity. Abstract similarity is used when a concrete state maps to no known abstract state. Instead a search is performed for nearby abstract states based on a nearest neighbour technique. Such a retrieval scheme enables accurate prediction in light of extremely novel observed situations. The properties of retrieval in abstract state-spaces are investigated in three standard planning domains. We first determine optimal radii to use that determines a spherical sub-hyperspace that limits the search. Experimental results then show that significant improvements in the recognition process are obtained using abstract similarity.

Multi-model modelling and predictive control based on local model networks

Abstract: This paper proposes multi-model modelling and predictive control based on Local Model Networks (LMN). An LMN modelling method using Satisfying Fuzzy c-Mean (SFCM) clustering algorithm is introduced. SFCM is designed to determine a satisfactory number of local models, and an identification algorithm based on weighted performance index is used to generate multiple models with good trade-off between global fitting and local interpretation. Considering that each local model is valid only in each local regime, different predictive controllers are designed for different local models with different local constraints, and Multi-model Predictive Control with Local Constraints (MMPCLC) is presented using Parallel Distribution Compensation (PDC) method. The presented modelling and controller design procedures are demonstrated on a Multi-Inputs Multi-Outputs (MIMO) simulated pH neutralization process.

Kalman filtering with statistical state constraints

Abstract: For linear dynamic systems with white process and measurement noise, the Kalman filter is known to be the minimum variance linear state estimator. In the case that the random quantities are Gaussian, then the Kalman filter is the minimim variance state estimator. However, in the application of Kalman filters known signal information is often either ignored or dealt with heuristically. For instance, state variable constraints (which may be based on physical considerations) are often neglected because they do not fit easily into the structure of the optimal filter. Previous work by the authors demonstrated an analytic method of incorporating deterministic state equality constraints in the Kalman filter. This paper extends that work to develop the properties of Kalman filters in the presence of statistical state constraints. That is, given a linear system such that the expected values of the state variables satisfy some linear equality, we can constrain the Kalman filter estimates to satisfy those constraints. This results in a family of constrained filters with each member parameterized by a weighting matrix. This paper derives several interesting properties of the constrained Kalman filters.

Robust output feedback control for discrete time-delay uncertain systems

Abstract: This paper studies the problem of robust output feedback control of discrete time-delay systems with parameter uncertainties. The parameter uncertainties under consideration are time varying and norm bounded. A sufficient condition for the existence of a dynamic output-feedback controller is presented in terms of linear matrix inequalities (LMIs). It is also shown that the set of dynamic output-feedback controller can be implicitly parameterized by the solutions (X, Y, R, S) to the LMIs. An example is given to demonstrate the potential of the proposed techniques.

Robust output feedback control for discrete time-delay uncertain systems

Abstract: This paper studies the problem of robust output feedback control of discrete time-delay systems with parameter uncertainties. The parameter uncertainties under consideration are time varying and no...

Maximizing performance and robustness of PI and PID controllers by global optimization

Abstract: This paper reports a new algorithm for global constrained optimization and shows its application to the design of PI and PID controllers. The algorithm is described in detail, and the features that make it suited for controller design are emphasized. Various design criteria and constraints are considered, for some plant models currently used as benchmarks. The numerical results show that the algorithm performs very well in all the tested cases: its flexibility and ease of use make it a valuable alternative to classical design procedures.

Soil dynamic properties determination: a neurofuzzy system approach

Abstract: Proper characterization of the dynamic behaviour of soil deposits is of utmost importance in earthquake ground-response analyses. Cone-tip penetration resistances, which are usually obtained in a typical geotechnical study for foundation design, can be used to evaluate the dynamic properties and thus to outline dynamically a given soil deposit. Because of the capacity of neurofuzzy techniques to combine the representational aspect of fuzzy models and the learning mechanisms of neural networks, in this paper a neurofuzzy system is proposed to map the values obtained from cone penetration tests into the dynamic properties of Mexico City clays. Utilizing this methodology, it is possible to achieve profies of shear modulus and damping ratios versus shear strain curves, from which soil deposits can be characterized dynamically at a reduced cost.

Robust PID/backstepping control design for permanent magnet synchronous motor drive

Abstract: This paper presents a new control design procedure for permanent magnet synchronous motor (PMSM) speed drive in the case of unknown load torque. The control law is based on the combination of nonlinear proportional integral derivative (PID) regulators and the backstepping methodology. More precisely, we determine the controllers imposing the current-speed tracking in two recursive steps and by using appropriate PID gains that are nonlinear functions of the system state. Moreover, a frequency domain analysis is done to derive under what conditions PID controllers fulfil the robust mixed sensitivity performance condition when applied to PMSM. A comparative study between the proposed PID/backstepping approach and the feedback linearizing control is made by realistic simulation including load torque change, parametric variations, and measurement noise. The results of current-speed tracking show the effectiveness of the proposed method in the presence of strong disturbances. Finally, a virtual experimental setup is simulated to show the feasibility of our control algorithm in practice.

Neurofuzzy network-based adaptive nonlinear PI controllers

Abstract: PI controllers are still among the most popular industrial controllers, as they are relatively easy to install and reasonably robust, and can remove steady-state errors of type 0 processes. However, if the system is highly nonlinear, the performance of the PI controllers may deteriorate rapidly. A popular approach to control nonlinear systems is to switch between several linear PI controllers using fuzzy logic based on the Takagi-Sugeno model. In this paper, B-spline neurofuzzy networks are used to implement such a nonlinear controller. Neurofuzzy networks are chosen to implement the nonlinear PI controller for their abilities to approximate nonlinear functions and be trained online using experimental data. Design guidelines and training of the proposed controller are presented. The implementation of the proposed controller is presented, and its performance is demonstrated using a two-tank water level control rig and a continuous stirred-tank reactor process, and compared with PI controllers tuned using existing techniques.

Hybrid learning algorithm for interval type-2 fuzzy logic systems

Abstract: A new learning methodology based on a hybrid algorithm for interval type-2 fuzzy logic systems (FLS) parameter estimation is presented. The new proposal consists of recursive least-squares (RLS) with back-propagation (BP) and square-root filter (REFIL) with BP combinations. A system with the proposed methodology was simulated to test its capability for output surface temperature prediction of a transfer bar at a hot strip mill (HSM) scale breaker (SB) entry zone for three different types of coils. The inputs of the system are the surface temperature of the transfer bar and the traveling time required to reach the SB entry zone. The inputs are modelled as interval singleton, type-1, or type-2 fuzzy sets depending on whether the noise was taken into account or not. The following combinations were simulated: (a) interval singleton type-2 FLS (type-2 SFLS), (b) interval type-1 non-singleton type-2 FLS (type-2 NSFLS-1) and (c) interval type-2 non-singleton type-2 FLS (type-2 NSFLS-2), where the bold face terms indicate the kind of input. Modelling results show the feasibility of the proposed methodology to be implemented into an intelligent process controller for the HSM industry.

A nonlinear PID-based multiple controller incorporatinga a multilayered neural network learning submodel

Abstract: A new nonlinear minimum-variance adaptive proportional integral derivative (PID) based multiple controller, incorporating a multi- layered neural network learning submodel, is presented. The unknown non-linear plant is represented by an equivalent stochastic model consisting of a linear least-squares-based submodel plus a non- linear multi-layered back propagation (BP) neural network-based learning submodel. The proposed multiple controller methodology provides the designer with a choice of using either a conventional PID self-tuning controller, a PID structure-based pole-placement controller, or a newly proposed PID structure-based pole-zero placement controller through simple switching. The novel PID structure based pole-zero placement controller employs an adaptive mechanism, which ensures that the closed-loop poles and zeros are located at their prespecified positions. The switching decision between the different nonlinear fixed structure controllers is made manually in the present case but can be automated using fuzzy logic or stochastic learning automata techniques. Simulation results using a nonlinear plant model demonstrate the effectiveness of the proposed multiple controller with respect to tracking set-point changes. The aim is to achieve a desired speed of response while penalizing excessive control action, for applications in nonminimum phase and unstable systems.

Ram position control in plastic injection molding machines with higher-order iterative learning

Abstract: In plastic injection molding, the ram position plays an important role in production quality. This paper introduces a new method, which is a combination of the current cycle feedback control (a PI controller) and a feed-forward higher-order iterative learning control (ILC), to control the ram position in injection molding. The PI controller is used to stabilize the system, and the feed-forward higher-order ILC control is used to compensate for nonlinear/unknown dynamics and disturbances, thereby gaining the precision tracking to ram position. The simulation results indicate that the new method outperforms the conventional PI controller. In addition, it outperforms the conventional ILC in the convergence performance.

Supervised switching control of a deployable manipulator system

Abstract: This paper presents a supervisory system for controller switching as applied to ground-based and space-based deployable manipulator systems. First, a finite family of candidate controllers is established so that the manipulator system performs satisfactorily under the control of one of the controllers, in the possible presence of model uncertainties, unknown parameters, time variance, nonlinearities, and variable operating conditions. A supervisory unit in the system monitors the performance of the manipulator. Based on this, a decision-making logic unit selects an appropriate controller from the family, and activates it while deactivating the currently active controller, so as to ensure proper performance. Capabilities of instantaneous switching and gradual switching using a switching function and a switching compensator are incorporated into the system. The performance of the supervised switching control system is evaluated through computer simulation of an innovative manipulator system consisting of a combination of revolute and prismatic degrees of freedom and joint and link flexibilities.

Decentralized disturbance attenuation for large-scale nonlinear systems with delayed state interconnections

Abstract: The problem of decentralized disturbance attenuation is considered for a new class of large-scale nonlinear systems with delayed state interconnections. This class of large-scale time-delay systems broadens most existing classes of large-scale time-delay systems in that the uncertain interconnections are bounded by general nonlinear functions instead of linear or polynomial-type functions. It is shown that by decentralized memoryless state feedback control, the closed-loop system achieves internal global asymptotical stability in the sense of Lyapunov and external stability in the sense of L2 gain. Nonlinear Lyapunov-Krasovskii functionals are constructed that render the linear and polynomial-type growth conditions on the interconnections as special cases.

Delay-dependent robust H ∞ filtering for interval systems with state delays

Abstract: This paper deals with the problem of robust H∞ filtering for a class of linear continuous-time interval systems with delay dependence and structured uncertainties, which are not restricted to the matched uncertainty or the norm-bounded uncertainty. The problem aims at designing a stable linear filtering assuring asymptotic stability and a prescribed H∞ performance level for the filtering error system. A sufficient condition for the existence of such a filter is developed in terms of linear matrix inequalities. A numerical example demonstrates the validity of the newly developed theoretical results.

Comparing frequency domain, optimal, and asymptotic filtering: a tutorial

Abstract: This paper studies some connections between the main results of the Kalman-Bucy stochastic approach to filtering problems based mainly on linear stochastic estimation theory and emphasizing the optimality aspects of the achieved results and the classical deterministic frequency domain linear filters (such as Chebyshev, Butterworth, Bessel, etc.). A new non-stochastic but not necessarily deterministic (possibly nonlinear) alternative approach to signal filtering based mainly on concepts of signal power, signal energy, and an equivalence relation plays a dominant role in the presentation. Causality, error invariance, and especially error convergence properties are the most important and fundamental features of resulting filters. Therefore, it is natural to call them asymptotic filters. Although error convergence aspects are emphasized in the approach, it is shown that introducing the signal power as the quantitative measure of signal energy dissipation makes it possible to achieve reasonable results from the optimality point of view as well. The authors show that the notion of the asymptotic filter can be used as a proper tool for unifying stochastic and non-stochastic, linear and nonlinear approaches to signal filtering.

Integrating schedulability analysis with UML-RT

Abstract: The use of object-oriented techniques and methodologies to design real-time control systems appears to be necessary in order to deal with the increasing complexity of such systems. Recently, many object-oriented methods have been used for the modelling and designing of real-time control systems. We believe an approach that integrates the advancements in both object modelling and design methods, and real-time scheduling theory is a key to the successful use of object-oriented technology for real-time software. However, past approaches to integrate the two either have restricted the object models, or did not allow sophisticated schedulability analysis techniques. In this paper, we show the manner in which schedulability analysis can be integrated with object-oriented design; we develop the schedulability and feasibility analysis method for external messages that may suffer release jitter due to being dispatched by a tick-driven scheduler in real-time control system; and we also develop the schedulability method for sporadic activities, where messages arrive sporadically and are then executed periodically for some bounded time. This method can be used to cope with timing constraints in complex real-time control systems.

A partial model matching design of robust 2-DOF PID controller for time-delay systems

Abstract: This paper presents a design method of robust PID controller with two degrees of freedom (2-DOF PID controller) for SISO plants with a time-delay and parametric uncertainty based on the partial model matching (PMM) approach It is assumed that the adjustable parameters of the 2-DOF PID controller are chosen so as to minimize the two performance indices for step reference and disturbance responses, which are maximized by the plant parameters belonging to a given bounded set Thus, the design problem is formulated as a multi-objective minimax optimization problem, which is solved by new genetic algorithm (GA) A novel feature of the present paper lies in the exact robust stability check for time-delay systems without using approximant Numerical examples show the effectiveness of the present approach

Learning-enhanced generalized predictive control for systems with periodic disturbances

Abstract: In this paper, a generalized predictive control (GPC) algorithm is developed for linear discrete systems that is capable of effectively addressing the presence of disturbances that are periodic in nature. First, a GPC algorithm is derived based on the assumption of the presence of a periodic disturbance. Next, a feedforward disturbance learning scheme is proposed for estimating the periodic disturbance. Finally, two sufficient conditions are provided to guarantee the convergence of the disturbance learning law.

Design of multiobjective H ∞ lopp-shaping SISO control systems by fuzzy logic

Abstract: The H∞ loop-shaping design method is an effective way to design robust controllers. One of the main limitations in the design of H∞ loop-shaping controllers is that the design method concerns frequency domain specifications that do not precisely represent the underlying time domain performance requirements. Even if the time response constraints were not present, the approach is based on trial-and-error selection of the frequency-dependent weights that are used in the design. Selection of these weights becomes more difficult if time response constraints must be satisfied. This paper presents a systematic way of selecting the loop parameters for the frequency-dependent weights, and develops a fuzzy algorithm that selects these parameters so that the resulting H∞ loop-shaping controller yields a closed-loop system that satisfies the time response constraints for a linear single-input single-output (SISO) system. It is seen that although most of the time response characteristics of the system can be improved by changing any of the loop parameters, the frequency domain constraints normally set limits on some of these loop parameters in such a way that the latter cannot be changed further to improve the time response characteristics of the system.

Robust mixed H 2 / H ∞ control for uncertain singular systems with state delay

Abstract: This paper considers the problem of the mixed H2/H∞ control of uncertain continuous singular systems with state time delay. We present a kind of H2/H∞ mixed controller that can minimize the H2 performance measure satisfying a prescribed H∞ norm bounded on the resulting closed-loop systems, and guarantee that the systems is regular, impulse free, and stable. The sufficient conditions for the existence of controller are presented using a set of linear matrix inequalities (LMIs), and all solutions including controller gain and upper bound of performance measure are also obtained simultaneously. To illustrate the effectiveness of the given scheme, two numerical examples are given.

Robust pid model following control

Abstract: The paper deals with properties of the so-called model-following control systems that contain a nominal model of the controlled plant and two PID controllers. An approach to design robust controllers for a SISO process from the viewpoint of its input sensitivity is presented. The proposed structures have been tested by simulation and experiment on linear and nonlinear control plants with perturbed parameters. Results of tests lend support to the view that the proposed structure significantly extends the range of applicability, performance, and robustness of the control system, and thereby eliminates to a large extent the need to employ an adaptation mechanism. Therefore, the structure may find wide application to robust control of plants with time-varying parameters.

Optimal fusion distributed filter for discrete multichannel ARMA signals

Abstract: Based on the multi-sensor optimal information fusion criterion weighted by matrices in the linear minimum variance sense, an optimal information fusion distributed Kalman filter is given for discrete multi-channel ARMA (autoregressive moving average) signals with correlated noises. When all subsystems have the steady-state filters, a steady-state information fusion filter is also given. It has the reduced computation burden compared with the optimal fusion filter. The precision of the fusion filter is higher than that of any local filter, but is lower than that of the centralized filter. The filtering error cross-covariance matrix between any two subsystems is given for discrete multichannel ARMA signals. Applying it to a four-channel ARMA signal system with three sensors shows its effectiveness.

Stabilizing discrete finite-horizon H∞ control with guaranteed infinite norm bound

Abstract: Stabilizing properties based on monotonic behaviour of the solution of the Riccati equation for linear quadratic gaussion (LQG) optimal control have been well studied In this paper, these results are extended to H∞ optimal control for time-invariant discrete linear system The authors present sufficient conditions for monotonic nondecreasing behaviour of the Riccati difference equation (RDE) associated with H∞ optimal control subject only to the terminal constraints This allows for deriving linear time-invariant "frozen" optimal control with guaranteed infinite-horizon H∞-norm bound

Achieving or exceeding maximum available feedback

Abstract: A feedback system for control or electronics should have high loop gain, so that its output is close to its desired state, and the effects of changes in the system and of disturbances are minimised. Bode proposed a method for single loop feedback systems to obtain the maximum available feedback, defined as the largest possible loop gain over a bandwidth pertinent to the system, with appropriate gain and phase margins. The method uses asymptotic approximations, and this paper describes some novel adjustments to the asymptotes, so that the final system often exceeds the maximum available feedback. The implementation of the method requires the cascading of a series of lead-lag element. This paper describes a new way to determine how many elements should be used.

Auto-tuning of adaptive control with dead zone

Abstract: It is useful for system engineers to initialize tuning parameters of adaptive control automatically. Adaptation gain and width of dead zone are tuning parameters of an estimation law using a dead zone. A control law has additional tuning parameters. This paper presents a new variable dead zone and a new automatic initialization method for adaptive control using a variable dead zone with a few tuning parameters that are determined by previously measured input and output data. The dead zone width in the estimation law is changed depending on the input signal by a rational function initialized using previously measured input and output signals. The control law used here is an H∞ control one referring to a complementary sensitivity function against unstructured uncertainty. We apply the proposed method to an electro-hydraulic servo system and verify the effects by experimentation.

Model-based fault detection and identification: oriented sensor selection approach

Abstract: This paper deals with continuous linear time-invariant systems and is focused on the model-based failure detection and identification problem, and more precisely, on the so-called residual generation problem. Our purpose is to present some new results concerning the practical situation arising when the problem cannot be solved using conventional tools for a given system. Thus, the residual generation must be done by choosing a new measured output matrix, that is, we are concerned with the development of sensor location methodologies that are oriented to guarantee the solution of the failure detection problem. Solvability conditions are presented and procedures for sensor location are developed. When the solvability conditions hold, the minimality of the associated residual generator is ensured. A geometric approach and the notions of fundamental and extended residual generation problems as stated by Massoumia are followed.

Caustic dilution control : An industrial application

Abstract: Caustic solution is commonly used in pulp-and-paper mills for various processes, such as bleaching and pH control. It is usually purchased as caustic concentrate, which is diluted on site to the desired concentration using common mill water and is stored in a reservoir for mill utilization. The diluted caustic has to meet the desired specifications as variations in the caustic concentration have adverse effects on the process, such as poor bleaching and washing. Because direct measurement of caustic concentration is not available, the concentration is indirectly controlled by maintaining a certain mixing temperature that causes significant nonlinearity in the process. Another challenge in the process was the level limitations imposed by the storage tank system. A multistructured control strategy combining an adaptive PID controller and optimal averaging level control was successfully implemented in this industrial caustic dilution system. The overall process operation was significantly improved.