Open Access. Powered by Scholars. Published by Universities.^{®}
 Discipline
 Publication

 Mechanical and Aerospace Engineering Faculty Research & Creative Works (7)
 Electrical and Computer Engineering Faculty Research & Creative Works (5)
 Engineering Management and Systems Engineering Faculty Research & Creative Works (2)
 Chemical and Biochemical Engineering Faculty Research & Creative Works (1)
 Computer Science Faculty Research & Creative Works (1)
Articles 1  16 of 16
FullText Articles in Engineering
High Bandwidth Control Of Precision Motion Instrumentation, Douglas A. Bristow, Jingyan Dong, Andrew G. Alleyne, Srinivasa M. Salapaka, Placid M. Ferreira
High Bandwidth Control Of Precision Motion Instrumentation, Douglas A. Bristow, Jingyan Dong, Andrew G. Alleyne, Srinivasa M. Salapaka, Placid M. Ferreira
Mechanical and Aerospace Engineering Faculty Research & Creative Works
This article presents a highbandwidth control design suitable for precision motion instrumentation. Iterative learning control (ILC), a feedforward technique that uses previous iterations of the desired trajectory, is used to leverage the repetition that occurs in many tasks, such as raster scanning in microscopy. Two ILC designs are presented. The first design uses the motion system dynamic model to maximize bandwidth. The second design uses a timevarying bandwidth that is particularly useful for nonsmooth trajectories such as raster scanning. Both designs are applied to a multiaxis piezoelectricactuated flexure system and evaluated on a nonsmooth trajectory. The ILC designs demonstrate significant ...
Weighting Matrix Design For Robust Monotonic Convergence In Norm Optimal Iterative Learning Control, Douglas A. Bristow
Weighting Matrix Design For Robust Monotonic Convergence In Norm Optimal Iterative Learning Control, Douglas A. Bristow
Mechanical and Aerospace Engineering Faculty Research & Creative Works
In this paper we examine the robustness of norm optimal ILC with quadratic cost criterion for discretetime, linear timeinvariant, singleinput singleoutput systems. A bounded multiplicative uncertainty model is used to describe the uncertain system and a sufficient condition for robust monotonic convergence is developed. We find that, for sufficiently large uncertainty, the performance weighting can not be selected arbitrarily large, and thus overall performance is limited. To maximize available performance, a timefrequency design methodology is presented to shape the weighting matrix based on the initial tracking error. The design is applied to a nanopositioning system and simulation results are presented.
Frequency Domain Analysis And Design Of Iterative Learning Control For Systems With Stochastic Disturbances, Douglas A. Bristow
Frequency Domain Analysis And Design Of Iterative Learning Control For Systems With Stochastic Disturbances, Douglas A. Bristow
Mechanical and Aerospace Engineering Faculty Research & Creative Works
In this work we examine the performance of iterative learning control (ILC) for systems with nonrepeating disturbances and random noise. Singleinput, single output linear timeinvariant systems and iterationinvariant learning filters are considered. We find that a tradeoff exists between the convergence rate and converged error spectrum. Optimal filter designs, which are dependant on the disturbance and noise spectra, are developed. We also present simple design guidelines for the case when explicit models of disturbance and noise spectra are not available. A numerical design example is presented.
Monotonic Convergence Of Iterative Learning Control For Uncertain Systems Using A TimeVarying Filter, Douglas A. Bristow, Andrew G. Alleyne
Monotonic Convergence Of Iterative Learning Control For Uncertain Systems Using A TimeVarying Filter, Douglas A. Bristow, Andrew G. Alleyne
Mechanical and Aerospace Engineering Faculty Research & Creative Works
Iterative learning control (ILC) is a learning technique used to improve the performance of systems that execute the same task multiple times. Learning transient behavior has emerged as an important topic in the design and analysis of ILC systems. In practice, the learning control is often lowpass filtered with a ldquoQfilterrdquo to prevent transient growth, at the cost of performance. In this note, we consider linear timeinvariant, discretetime, singleinput singleoutput systems, and convert frequencydomain uncertainty models to a timedomain representation for analysis. We then develop robust monotonic convergence conditions, which depend directly on the choice of the Qfilter and are ...
Reinforcement Learning Based OutputFeedback Control Of Nonlinear Nonstrict Feedback DiscreteTime Systems With Application To Engines, Peter Shih, Jonathan B. Vance, Brian C. Kaul, Jagannathan Sarangapani, J. A. Drallmeier
Reinforcement Learning Based OutputFeedback Control Of Nonlinear Nonstrict Feedback DiscreteTime Systems With Application To Engines, Peter Shih, Jonathan B. Vance, Brian C. Kaul, Jagannathan Sarangapani, J. A. Drallmeier
Electrical and Computer Engineering Faculty Research & Creative Works
A novel reinforcementlearning based outputadaptive neural network (NN) controller, also referred as the adaptivecritic NN controller, is developed to track a desired trajectory for a class of complex nonlinear discretetime systems in the presence of bounded and unknown disturbances. The controller includes an observer for estimating states and the outputs, critic, and two action NNs for generating virtual, and actual control inputs. The critic approximates certain strategic utility function and the action NNs are used to minimize both the strategic utility function and their outputs. All NN weights adapt online towards minimization of a performance index, utilizing gradientdescent based rule ...
Online Reinforcement Learning Control Of Unknown Nonaffine Nonlinear Discrete Time Systems, Qinmin Yang, Jagannathan Sarangapani
Online Reinforcement Learning Control Of Unknown Nonaffine Nonlinear Discrete Time Systems, Qinmin Yang, Jagannathan Sarangapani
Electrical and Computer Engineering Faculty Research & Creative Works
In this paper, a novel neural network (NN) based online reinforcement learning controller is designed for nonaffine nonlinear discretetime systems with bounded disturbances. The nonaffine systems are represented by nonlinear auto regressive moving average with exogenous input (NARMAX) model with unknown nonlinear functions. An equivalent affinelike representation for the tracking error dynamics is developed first from the original nonaffine system. Subsequently, a reinforcement learningbased neural network (NN) controller is proposed for the affinelike nonlinear error dynamic system. The control scheme consists of two NNs. One NN is designated as the critic, which approximates a predefined longterm cost function, whereas an ...
Near Optimal Neural NetworkBased Output Feedback Control Of Affine Nonlinear DiscreteTime Systems, Qinmin Yang, Jagannathan Sarangapani
Near Optimal Neural NetworkBased Output Feedback Control Of Affine Nonlinear DiscreteTime Systems, Qinmin Yang, Jagannathan Sarangapani
Electrical and Computer Engineering Faculty Research & Creative Works
In this paper, a novel online reinforcement learning neural network (NN)based optimal output feedback controller, referred to as adaptive critic controller, is proposed for affine nonlinear discretetime systems, to deliver a desired tracking performance. The adaptive critic design consist of three entities, an observer to estimate the system states, an action network that produces optimal control input and a critic that evaluates the performance of the action network. The critic is termed adaptive as it adapts itself to output the optimal costtogo function which is based on the standard Bellman equation. By using the Lyapunov approach, the uniformly ultimate ...
Nonlinear Control Concepts For A Ua, Vijayakumar Janardhan, Derek Schmitz, S. N. Balakrishnan
Nonlinear Control Concepts For A Ua, Vijayakumar Janardhan, Derek Schmitz, S. N. Balakrishnan
Mechanical and Aerospace Engineering Faculty Research & Creative Works
A reconfigurable flight control method is developed to be implemented on an Unmanned Aircraft (UA), a thirty percent scale model of the Cessna 150. This paper presents the details of the UA platform, system identification, reconfigurable controller design, development, and implementation on the UA to analyze the performance metrics. A Crossbow Inertial Measurement Unit provides the roll, pitch, and yaw accelerations and rates along with the roll and pitch. The 100400 miniair data boom from SpaceAge Control provides the airspeed, altitude, angle of attack, and the side slip angles. System identification is accomplished by commanding preprogrammed inputs to the control ...
An Enhanced LeastSquares Approach For Reinforcement Learning, Hailin Li, Cihan H. Dagli
An Enhanced LeastSquares Approach For Reinforcement Learning, Hailin Li, Cihan H. Dagli
Engineering Management and Systems Engineering Faculty Research & Creative Works
This paper presents an enhanced leastsquares approach for solving reinforcement learning control problems. Modelfree leastsquares policy iteration (LSPI) method has been successfully used for this learning domain. Although LSPI is a promising algorithm that uses linear approximator architecture to achieve policy optimization in the spirit of Qlearning, it faces challenging issues in terms of the selection of basis functions and training samples. Inspired by orthogonal leastsquares regression (OLSR) method for selecting the centers of RBF neural network, we propose a new hybrid learning method. The suggested approach combines LSPI algorithm with OLSR strategy and uses simulation as a tool to ...
Adaptive Critic Based Neurocontroller For Autolanding Of Aircraft With Varying Glideslopes, Gaurav Saini, S. N. Balakrishnan
Adaptive Critic Based Neurocontroller For Autolanding Of Aircraft With Varying Glideslopes, Gaurav Saini, S. N. Balakrishnan
Mechanical and Aerospace Engineering Faculty Research & Creative Works
In this paper, adaptive critic based neural networks have been used to design a controller for a benchmark problem in aircraft autolanding. The adaptive critic control methodology comprises successive adaptations of two neural networks, namely `action' and `critic' networks until closed loop optimal control is achieved. The autolanding problem deals with longitudinal dynamics of an aircraft which is to be landed in a specified touchdown region in the presence of wind disturbances and gusts using elevator deflection as the control for glideslope and flare modes. The performance of the neurocontroller is compared to that of a conventional PID controller. Neurocontroller ...
Neural Modeling And Control Of A Distillation Column, James Edward Steck, K. Krishnamurthy, Bruce M. Mcmillin, Gary G. Leininger
Neural Modeling And Control Of A Distillation Column, James Edward Steck, K. Krishnamurthy, Bruce M. Mcmillin, Gary G. Leininger
Mechanical and Aerospace Engineering Faculty Research & Creative Works
Control of a ninestage threecomponent distillation column is considered. The control objective is achieved using a neural estimator and a neural controller. The neural estimator is trained to represent the chemical process accurately, and the neural controller is trained to give an input to the chemical process which will yield the desired output. Training of both the neural networks is accomplished using a recursive least squares training algorithm implemented on an Intel iPSC/2 multicomputer (hypercube). Simulated results are presented for a numerical example.
An Industrial Application To Neural Networks To Reusable Design, Donald C. Wunsch, R. Escobedo, T. P. Caudell, S. D. G. Smith, G. C. Johnson
An Industrial Application To Neural Networks To Reusable Design, Donald C. Wunsch, R. Escobedo, T. P. Caudell, S. D. G. Smith, G. C. Johnson
Electrical and Computer Engineering Faculty Research & Creative Works
Summary form only given, as follows. The feasibility of training an adaptive resonance theory (ART1) network to first cluster aircraft parts into families, and then to recall the most similar family when presented a new part has been demonstrated, ART1 networks were used to adaptively group similar input vectors. The inputs to the network were generated directly from computeraided designs of the parts and consist of binary vectors which represent bit maps of the features of the parts. This application, referred to as group technology, is of large practical value to industry, making it possible to avoid duplication of design ...
A Neural Architecture For Unsupervised Learning With Shift, Scale And Rotation Invariance, Efficient Software Simulation Heuristics, And Optoelectronic Implementation, Donald C. Wunsch, D. S. Newman, T. P. Caudell, R. A. Falk, C. David Capps
A Neural Architecture For Unsupervised Learning With Shift, Scale And Rotation Invariance, Efficient Software Simulation Heuristics, And Optoelectronic Implementation, Donald C. Wunsch, D. S. Newman, T. P. Caudell, R. A. Falk, C. David Capps
Electrical and Computer Engineering Faculty Research & Creative Works
A simple modification of the adaptive resonance theory (ART) neural network allows shift, scale and rotation invariant learning. The authors point out that this can be accomplished as a neural architecture by modifying the standard ART with hardwired interconnects that perform a FourierMellin transform, and show how to modify the heuristics for efficient simulation of ART architectures to accomplish the additional innovation. Finally, they discuss the implementation of this in optoelectronic hardware, using a modification of the Van der Lugt optical correlator
Intelligent Control Of A Robotic Arm Using Hierarchical Neural Network Systems, Xavier J. R. Avula, Luis C. Rabelo
Intelligent Control Of A Robotic Arm Using Hierarchical Neural Network Systems, Xavier J. R. Avula, Luis C. Rabelo
Chemical and Biochemical Engineering Faculty Research & Creative Works
Two artificial neural network systems are considered in a hierarchical fashion to plan the trajectory and control of a robotic arm. At the higher level of the hierarchy the neural system consists of four networks: a restricted Coulomb energy network to delineate the robot arm workspace; two standard backpropagation (BP) networks for coordinates transformation; and a fourth network which also uses BP and participates in the trajectory planning by cooperating with other knowledge sources. The control emulation process which is developed using a second neural system at a lower hierarchical level provides the correct sequence of control actions. An example ...
An Empirical Analysis Of Backpropagation Error Surface Initiation For Injection Molding Process Control, Alice E. Smith, Elaine R. Raterman, Cihan H. Dagli
An Empirical Analysis Of Backpropagation Error Surface Initiation For Injection Molding Process Control, Alice E. Smith, Elaine R. Raterman, Cihan H. Dagli
Engineering Management and Systems Engineering Faculty Research & Creative Works
Backpropagation neural networks are trained by adjusting initially random interconnecting weights according to the steepest local error surface gradient. The authors examine the practical implications of the arbitrary starting point on the error landscape of the ensuing trained network. The effects on network convergence and performance are tested empirically, varying parameters such as network size, training rate, transfer function and data representation. The data used are live process control data from an injection molding plant
Parallel Implementation Of A Recursive Least Squares Neural Network Training Method On The Intel Ipsc/2, James Edward Steck, Bruce M. Mcmillin, K. Krishnamurthy, M. Reza Ashouri, Gary G. Leininger
Parallel Implementation Of A Recursive Least Squares Neural Network Training Method On The Intel Ipsc/2, James Edward Steck, Bruce M. Mcmillin, K. Krishnamurthy, M. Reza Ashouri, Gary G. Leininger
Computer Science Faculty Research & Creative Works
An algorithm based on the MarquardtLevenberg leastsquare optimization method has been shown by S. Kollias and D. Anastassiou (IEEE Trans. on Circuits Syst. vol.36, no.8, p.1092101, Aug. 1989) to be a much more efficient training method than gradient descent, when applied to some small feedforward neural networks. Yet, for many applications, the increase in computational complexity of the method outweighs any gain in learning rate obtained over current training methods. However, the leastsquares method can be more efficiently implemented on parallel architectures than standard methods. This is demonstrated by comparing computation times and learning rates for the ...