Neural network control

This paper proposes a new adaptive neural network control to stabilize a quadrotor helicopter against modeling error and considerable wind disturbance. The new method is compared to both deadzone and e-modification adaptive techniques and through simulation demonstrates a clear improvement in terms of achieving a desired attitude and reducing weight drift.

This paper discusses the use of a real-time digital control environment with a hardware-in-the-loop (HIL) magnetic levitation (maglev) device for modeling and controls education, with emphasis on neural network (NN) feedforward control. Many educational advantages are realized for the students if a single environment is used for simulation, hardware implementation, and verification as compared with multienvironment settings. This real-time environment requires two personal computers (host and target) employed to control an HIL system. It requires software tools by MathWorks, Inc., a C++ compiler, an off-the-shelf data acquisition card, and the HIL (a nonlinear, open-loop, unstable, and time-varying, custom-built maglev device) to be controlled. This environment provides for experimentation, such as data collection for system identification using NN's and their implementation as static nonlinear feedforward controllers. In addition, this environment was used to implement and demonstrate NNs with real-time dynamic weight tuning and controller performance comparison under various inputs or changes in the HIL device dynamics, as shown in the presented examples. The educational features of this environment were verified in a classroom setting in a graduate-level NN class. The presented environment is applicable to senior or graduate level (introductory intelligent and digital control) or a general introductory course on NNs with applications.

This paper presents system modeling, analysis, and simulation of an electric vehicle (EV) with two independent rear wheel drives. The traction control system is designed to guarantee the EV dynamics and stability when there are no differential gears. Using two in-wheel electric motors makes it possible to have torque and speed control in each wheel. This control level improves EV stability and safety. The proposed traction control system uses the vehicle speed, which is different from wheel speed characterized by a slip in the driving mode, as an input. In this case, a generalized neural network algorithm is proposed to estimate the vehicle speed. The analysis and simulations lead to the conclusion that the proposed system is feasible. Simulation results on a test vehicle propelled by two 37-kW induction motors showed that the proposed control approach operates satisfactorily.

This paper presents a novel approach for process control that uses neural networks to model the steadystate inverse of a process which is then coupled with a simple reference system synthesis to generate a multivariable controller. The control strategy is applied to dynamic simulations of two methanol-water distillation columns that express distinctly different behaviour from each other (one simulates a lab column, while the second simulates an industrial-scale high-purity column). A steady-state process simulation package was used to generate all the neural network training data. An efficient training algorithm based on a nonlinear least-squares technique was used to train the networks. The neural network modelbased controllers show robust performance for both setpoints and disturbances, and performed better than conventional feedback proportional-integral (PI) controllers with feedforward features.

In the field of Automation, Fuzzy Control Fuzzy control has significant merits which are utilized in intelligent controllers, especially for vibration control systems. This paper is concerned with the application aspects of the developed MR damper for landing gear system, to attenuate the sustained vibrations during the landing phase. Also a comparative study is made on the responses obtained from the MR damper landing gear by utilizing PID and Fuzzy PID controllers.Theory is a well-known technique to acquire the desired response of different non-linear systems.

This paper describes the models of a wind power system, such as the turbine, generator, power electronics converters and controllers, with the aim to control the generation of wind power in order to maximize the generated power with the lowest possible impact in the grid voltage and frequency during normal operation and under the occurrence of faults. The presented work considers a wind power system equipped with the doubly-fed induction generator and a vector-controlled converter connected between the rotor and the grid. The paper presents comparative results between proportional-integral controllers and neural networks based controllers, showing that better dynamic characteristics can be obtained using neural networks based controllers.

A pilot scaled chemical reactor is constructed and commissioned to study various conventional and advanced control strategies. One of the approaches is the use of neural network inverse model based controller to control the temperature of the chemical reactor. Neural network control was chosen due to its capabilities to overcome the hassle in periodically tuning the conventional controller to obtain a good process response for a certain set point. Tests in a form of load disturbance and set point tracking are carried out to evaluate the neural network controller. The neural network controller exhibits satisfactory performance. Simulation work prior to the planned online implementation is vital to predict the control strategy performance and behavior. With the expected performance in hand, one has the prior knowledge of the control strategy

This paper concerns the application of a neural network control strategy to the distributed collector field of a solar power plant. The neural network is trained based on measured data from the plant providing a way of scheduling between a set of PID controllers, a priori tuned in different operating points by means of Takahashi rules. The present work consists in a hybrid scheme combining the potentialities of neural networks for approximation purposes with the well-know theory and widespread industrial application of PID techniques.

In this paper, a neuro-fuzzy approach is presented in order to guide a mobile robot. This task could be carried out specifying a set of fuzzy rules taking into account the di erent situations found by the mobile robot. This set of fuzzy rules could be optimised according to di erent criteria. However, the approach shown in this paper, is able to extract a set of fuzzy rules set from a set of trajectories provided by a human. These trajectories guide the mobile robot towards the target in di erent cases. Thus, it has been possible to obtain the rules and membership functions automatically, whereas other approaches need a previous deÿnition of the rules and membership functions. In order to verify that the obtained behaviour is satisfactory, the neuro-fuzzy approach has been implemented in two mobile robots.

This paper proposes a neural network control voltage tracking scheme of a DC-DC Flyback converter. In this technique, a back propagation learning algorithm is employed. The controller is designed to improve performance of the Flyback converter during transient and steady state operations. Furthermore, to investigate the effectiveness of the proposed controller, some operations such as starting-up and reference voltage variations are verified. The numerical simulation results show that the proposed controller has a better performance compare to the conventional PI-Controller method.

The main problem of vehicle vibration comes from road roughness. For that reason, it is necessary to control vibration of vehicle's suspension by using a robust artificial neural network control system scheme. Neural network based robust control system is designed to control vibration of vehicle's suspensions for full suspension system. Moreover, the full vehicle system has seven degrees of freedom on the vertical direction of vehicle's chassis, on the angular variation around X-axis and on the angular variation around Y-axis. The proposed control system is consisted of a robust controller, a neural controller, a model neural network of vehicle's suspension system. On the other hand, standard PID controller is also used to control whole vehicle's suspension system for comparison.

A new hybrid particle swarm optimization (PSO) that incorporates a wavelet-theory-based mutation operation is proposed. It applies the wavelet theory to enhance the PSO in exploring the solution space more effectively for a better solution. A suite of benchmark test functions and three industrial applications (solving the load flow problems, modeling the development of fluid dispensing for electronic packaging, and designing a neuralnetwork-based controller) are employed to evaluate the performance and the applicability of the proposed method. Experimental results empirically show that the proposed method significantly outperforms the existing methods in terms of convergence speed, solution quality, and solution stability. Index Terms?Load flow problem, modeling, mutation operation, neural network control, particle swarm optimization, wavelet theory. I. INTRODUCTION P ARTICLE swarm optimization (PSO) is a recently proposed population-based stochastic optimization algorithm which is inspired by the social behaviors of animals like fish schooling and bird flocking . Comparing with other population-based stochastic optimization methods, such as evolutionary algorithms, the PSO has comparable or even superior search performance for many hard optimization problems with faster and more stable convergence rates . The PSO has been used in different industrial areas, such as power systems [1], [18]?[21], parameter learning of neural networks (NNs) [16], [22], control [23], [24], prediction [25], and modeling [26], . However, observations reveal that the PSO sharply converges in the early stages of the searching process, but saturates or even terminates in the later stages. It behaves like the traditional local searching methods that trap in the local optima. As a result, it is hard to obtain any significant Manuscript improvements by examining neighboring solutions in the later stages of the search.Vaessens et al. and Reeves [14] put these searching methods into the context of local search or neighborhood search. Recently, different hybrid PSOs have been proposed to overcome the drawback of trapping in the local optima. The hybrid PSO was first proposed in 1998 , in which a standard selection mechanism is integrated with the PSO. A new hybrid gradient descent PSO (HGPSO), which is integrated with gradient information to achieve faster convergence without getting trapped in the local minima, is proposed by Noel and Jannett . However, the computational demand of the HGPSO is increased by the process of the gradient descent. Juang proposed a hybrid PSO algorithm named HGAPSO, which incorporates a genetic algorithm?s (GA?s) evolutionary operations of crossover, mutations, and reproduction. Ahmed et al. proposed a hybrid PSO named HPSOM, in which a constant mutating space is used in mutations. In the HGAPSO and the HPSOM, the solution space can be explored by performing mutation operations on particles along the search, and premature convergence is more likely to be avoided. However, the mutating space is kept unchanged all the time throughout the search, and the space for the permutation of particles in the PSO is also fixed. It can be further improved by varying the mutating space along the search. For GAs, the solution space is more likely to be explored in the early stage of the search by setting a larger mutating space, and it is more likely to be fine-tuned for a better solution in the later stage of the search by setting a smaller mutating space based on the properties of the wavelet [2]. This idea can be applied to introduce the hybrid PSO with the GA?s mutation. In this paper, a mutation with a dynamic mutating space by incorporating a wavelet function is proposed. The wavelet is a tool to model seismic signals by combining the dilations and the translations of a simple oscillatory function (the mother wavelet) of finite duration. The PSO?s mutating space is dynamically varying along the search based on the properties of the wavelet function. The resulting mutation operation aids the hybrid PSO to perform more efficiently and provides faster convergence than the PSO with constriction and inertia weight factors [9] and other hybrid PSOs [1], , [17], [28] in solving a suite of benchmark test functions. In addition, it achieves better and more stable solution quality. Application examples on solving some load flow problems [the multicontingency transient stability constrained optimal power flow (MC-TSCOPF) problem and the economic load dispatch with valve-points loading (ELD-VPL) problem], modeling the development of fluid dispensing for electronic packaging, and 1083-4419/$25.00 ? 2008 IEEE Authorized licensed use limited to: Hong Kong Polytechnic University. designing a NN-based controller (NN-BC) are employed to demonstrate that better performance can be achieved by the proposed hybrid PSO. This paper is organized as follows. Section II presents the operation of the hybrid PSO with a wavelet mutation (WM). Experimental studies and analysis are discussed in Section III. Eighteen standard benchmark test functions are given to evaluate the performance of the proposed method. Also, five

In this paper, an adaptive cerebellar-model articulation computer (CMAC) neural network (NN) control system is developed for a linear piezoelectric ceramic motor (LPCM) that is driven by an LLCC-resonant inverter. The motor structure and LLCC-resonant driving circuit of an LPCM are introduced initially. The LLCC-resonant driving circuit is designed to operate at an optimal switching frequency such that the output voltage will not be influenced by the variation of quality factor. Since the dynamic characteristics and motor parameters of the LPCM are highly nonlinear and time varying, an adaptive CMAC NN control system is designed without mathematical dynamic model to control the position of the moving table of the LPCM drive system to achieve high-precision position control with robustness. In the proposed control scheme, the dynamic backpropagation algorithm is adopted to train the CMAC NN online. Moreover, to guarantee the convergence of output tracking error for periodic commands tracking, analytical methods based on a discrete-type Lyapunov function are utilized to determine the optimal learning-rate parameters of the CMAC NN. The effectiveness of the proposed driving circuit and control system is verified by experimental results in the presence of uncertainties, and the advantages of the proposed control system are indicated in comparison with a traditional integral-proportional position control system. Accurate tracking response and superior dynamic performance can be obtained due to the powerful online learning capability of the CMAC NN with optimal learning-rate parameters.

In this paper, a neuro-fuzzy approach is presented in order to guide a mobile robot. This task could be carried out specifying a set of fuzzy rules taking into account the di erent situations found by the mobile robot. This set of fuzzy rules could be optimised according to di erent criteria. However, the approach shown in this paper, is able to extract a set of fuzzy rules set from a set of trajectories provided by a human. These trajectories guide the mobile robot towards the target in di erent cases. Thus, it has been possible to obtain the rules and membership functions automatically, whereas other approaches need a previous deÿnition of the rules and membership functions. In order to verify that the obtained behaviour is satisfactory, the neuro-fuzzy approach has been implemented in two mobile robots.

We present a system for the animation of human hand that plays violin. Neural network controls the hand movement. We make use of an optimization method to generate the examples for the neural network training. The musical decision of which finger to use is automatically made by best first search. We will show that the movements of violinist's hands are physically and musically feasible, and that the musical decisions are consistent with those recommended in the violin pedagogy. A description of system, the results of the decisions, and the animations are presented.

When working close to their nominal operating point, induction motors are highly-efficient. However, at light load, efficiency is greatly reduced if the magnetic flux is maintained at nominal value. It is necessary to adjust the flux according to the load level in order to improve the efficiency. This paper proposes a new approach based on Artificial Neural Network (ANN) to operate the machine at optimum condition. The artificial neural network is used to predict the applied voltage and frequency as the load torque and motor speed changes. The training and test patterns required to develop the ANN model were collected by conducting experiments on a 5 H.P Induction motor. A comparison of the experimental data and the ANN-based simulation results has shown a very good concordance between them.

Active control of sound and vibration has been the subject of a lot of research in recent years, and examples of applications are now numerous. However, few practical implementations of nonlinear active controllers have been realized. Nonlinear active controllers may be required in cases where the actuators used in active control systems exhibit nonlinear characteristics, or in cases when the structure to be controlled exhibits a nonlinear behavior. A multilayer perceptron neuralnetwork based control structure was previously introduced as a nonlinear active controller, with a training algorithm based on an extended backpropagation scheme. This paper introduces new heuristical training algorithms for the same neural-network control structure. The objective is to develop new algorithms with faster convergence speed (by using nonlinear recursive-leastsquares algorithms) and/or lower computational loads (by using an alternative approach to compute the instantaneous gradient of the cost function). Experimental results of active sound control using a nonlinear actuator with linear and nonlinear controllers are presented. The results show that some of the new algorithms can greatly improve the learning rate of the neural-network control structure, and that for the considered experimental setup a neural-network controller can outperform linear controllers.

This paper presents system modeling, analysis, and simulation of an electric vehicle (EV) with two independent rear wheel drives. The traction control system is designed to guarantee the EV dynamics and stability when there are no differential gears. Using two in-wheel electric motors makes it possible to have torque and speed control in each wheel. This control level improves EV stability and safety. The proposed traction control system uses the vehicle speed, which is different from wheel speed characterized by a slip in the driving mode, as an input. In this case, a generalized neural network algorithm is proposed to estimate the vehicle speed. The analysis and simulations lead to the conclusion that the proposed system is feasible. Simulation results on a test vehicle propelled by two 37-kW induction motors showed that the proposed control approach operates satisfactorily.

This report constitutes an expanded version of a presentation given by the author at the 1993 European Control Conference (short course on "Neural Nets for Control"). The first part places neurocontrol techniques in a general learning control framework. The second part of the report, which is essentially independent of the first, briefly surveys several basic theoretical results regarding neural networks.

This paper describes the implementation of a neural network control system for guiding a wheelchair, using an architecture based on a digital signal processor (DSP). We use a recurrent radial basis neural network as a system controller and a Kalman ÿlter as identiÿer, which acts as propagator of the control errors to the neurocontroller. The hub of the algorithm architecture is a DSP of the company Texas Instruments (TMS320C31). The board has complete autonomy of action and is speciÿcally designed for executing control algorithms in real time. Various practical tests have been carried out to check the correct functioning of the system when governing the output of the wheelchair.

This paper proposes a hybrid control scheme for the synchronization of two chaotic Duffing oscillator system, subject to uncertainties and external disturbances. The novelty of this scheme is that the Linear Quadratic Regulation (LQR) control, Sliding Mode (SM) control and Gaussian Radial basis Function Neural Network (GRBFNN) control are combined to chaos synchronization with respect to external disturbances. By Lyapunov stability theory, SM control is presented to ensure the stability of the controlled system. GRBFNN control is trained during the control process. The learning algorithm of the GRBFNN is based on the minimization of a cost function which considers the sliding surface and control effort. Simulation results demonstrate the ability of the hybrid control scheme to synchronize the chaotic Duffing oscillator systems through the application of a single control signal.

In this paper, the locomotion of an autonomously navigated undersea vehicle that uses vorticity control propulsion is computationally simulated. The navigation procedure employs a set of vehicle geometric and state variables to predict the needed vehicle body deformations in order to pass through a set of predefined path-points. To simulate the movement of the vehicle, a two-dimensional unsteady potential flow solver was developed based on the unsteady panel method coupled with the vehicle dynamics. The developed flow solver was validated against published computational results of unsteady flow standard test cases. Then, another set of properly planned test cases to cover the range of possible conditions were processed with the simulation and the output data was subsequently used to train a Multi-Layer Perceptron (MLP) neural network. The trained network can predict what body deflection time history is necessary for the vehicle to pass through the given path-points. Several autonomous navigation test cases are presented to show the capabilities of the present method.

In this paper a novel method is presented to design a sliding mode spatial control for a large Pressurized Heavy Water Reactor (PHWR) using a new formulation of Multirate Output Feedback (MROF). In the new formulation of MROF, the outputs of the system should be some of the states of the system or system should be in special form. The non-linear model of PHWR including xenon and iodine dynamics is characterized by 70 state variables and 14 inputs and outputs each. Linear nodal model is obtained by linearizing the non-linear dynamic equations of the reactor about the full power operating point. The 14 outputs of the PHWR are the power levels in 14 zones, also these are the 14 states of the system. The PHWR model is perfectly suitable for the application of this new formulation in which the states of the system can be computed using reduced system matrix inversion. The PHWR is an ill-conditioned system and the computation of the states of the system using existing MROF require the larger matrix inversions which sometimes may not possible. The proposed approach avoids this difficulty and produces the similar result as it is produced by the existing technique. The proposed control method does not require state information of the system for feedback purposes and hence may be easier to implement. From simulation of the non-linear model of the reactor in representative transients, the proposed control scheme is found to be superior to other methods.

This paper presents system modeling, analysis, and simulation of an electric vehicle (EV) with two independent rear wheel drives. The traction control system is designed to guarantee the EV dynamics and stability when there are no differential gears. Using two in-wheel electric motors makes it possible to have torque and speed control in each wheel. This control level improves EV stability and safety. The proposed traction control system uses the vehicle speed, which is different from wheel speed characterized by a slip in the driving mode, as an input. In this case, a generalized neural network algorithm is proposed to estimate the vehicle speed. The analysis and simulations lead to the conclusion that the proposed system is feasible. Simulation results on a test vehicle propelled by two 37-kW induction motors showed that the proposed control approach operates satisfactorily.

Neural control is a branch of the general field of intelligent control, which is based on the concept of artificial intelligence. This work presents the spacecraft orbit determination, dimensioning of the renewable power system, and mathematical modeling of spacecraft power system which are required for simulating the system. The complete system is simulated using MATLAB-SIMULINK. The NN controller out perform PID in the extreme range of non-linearity. Well trained neural controller can operate at different conditions of load current at different orbital periods without any tuning such in case of PID controller. So an artificial neural network (ANN) based model has been developed for the optimum operation of spacecraft power system. An ANN is trained using a back propagation with Levenberg-Marquardt algorithm. The best validation performance is obtained for mean square error is equal to 9.9962 × 10 -11 at epoch 637. The regression between the network output and the corresponding target is equal to 100% which means a high accuracy. NNC architecture gives satisfactory results with small number of neurons, hence better in terms of memory and time are required for NNC implementation. The results indicate that the proposed control unit using ANN can be successfully used for controlling the spacecraft power system in low earth orbit (LEO). Therefore, this technique is going to be a very useful tool for the interested designers in space field.

In this paper a new layered architecture for the implementation of intelligent distributed control systems is proposed. The proposed architecture distinguishes four layers in a distributed control system. Upper layer consists of a digital control layer, where high level decisions are taken. This level is implemented by means of intelligent agents that carry out the discrete control functions, system supervision as well as diagnosis and fault tolerance. Third layer deals with numeric values, performs analog operations and implement analog control loops. It is also in carry of the conversion from numerical variables values to evaluated expressions. This layer has been implemented by means of neural networks. Communications network appears in the second layer. Selected network was CAN (Controller Area Network) because of its particular characteristics. Finally, every node should implement a hardware interface with the process to control. Some interesting characteristics provided by this architecture are its low-cost implementation, easy distribution through a communication network, distributed execution in virtual generic nodes -with no hardware dependency-, bounded response time and fault tolerance mechanisms.

This paper presents a novel approach for process control that uses neural networks to model the steadystate inverse of a process which is then coupled with a simple reference system synthesis to generate a multivariable controller. The control strategy is applied to dynamic simulations of two methanol-water distillation columns that express distinctly different behaviour from each other (one simulates a lab column, while the second simulates an industrial-scale high-purity column). A steady-state process simulation package was used to generate all the neural network training data. An efficient training algorithm based on a nonlinear least-squares technique was used to train the networks. The neural network modelbased controllers show robust performance for both setpoints and disturbances, and performed better than conventional feedback proportional-integral (PI) controllers with feedforward features.

This paper concerns the application of a neural network control strategy to the distributed collector field of a solar power plant. The neural network is trained based on measured data from the plant providing a way of scheduling between a set of PID controllers, a priori tuned in different operating points by means of Takahashi rules. The present work consists in a hybrid scheme combining the potentialities of neural networks for approximation purposes with the well-know theory and widespread industrial application of PID techniques.

This paper describes the models of a wind power system, such as the turbine, generator, power electronics converters and controllers, with the aim to control the generation of wind power in order to maximize the generated power with the lowest possible impact in the grid voltage and frequency during normal operation and under the occurrence of faults. The presented work considers a wind power system equipped with the doubly-fed induction generator and a vector-controlled converter connected between the rotor and the grid. The paper presents comparative results between proportional-integral controllers and neural networks based controllers, showing that better dynamic characteristics can be obtained using neural networks based controllers.

This paper describes the implementation of a neural network control system for guiding a wheelchair, using an architecture based on a digital signal processor (DSP). We use a recurrent radial basis neural network as a system controller and a Kalman ÿlter as identiÿer, which acts as propagator of the control errors to the neurocontroller. The hub of the algorithm architecture is a DSP of the company Texas Instruments (TMS320C31). The board has complete autonomy of action and is speciÿcally designed for executing control algorithms in real time. Various practical tests have been carried out to check the correct functioning of the system when governing the output of the wheelchair.

This paper proposes a neural network control voltage tracking scheme of a DC-DC Flyback converter. In this technique, a back propagation learning algorithm is employed. The controller is designed to improve performance of the Flyback converter during transient and steady state operations. Furthermore, to investigate the effectiveness of the proposed controller, some operations such as starting-up and reference voltage variations are verified. The numerical simulation results show that the proposed controller has a better performance compare to the conventional PI-Controller method.

This paper concerns the application of a neural network control strategy to the distributed collector field of a solar power plant. The neural network is trained based on measured data from the plant providing a way of scheduling between a set of PID controllers, a priori tuned in different operating points by means of Takahashi rules. The present work consists in a hybrid scheme combining the potentialities of neural networks for approximation purposes with the well-know theory and widespread industrial application of PID techniques.

In this study, power spectrum of the EEG data and the heartbeat data obtained from 250 patients has been applied to the designed Neural network system. A backpropagation artificial neural network has been developed which contains 53 nodes in the input layer, 27 nodes in the hidden and 1 node in the output layer. In the artificial neural network inputs, the power spectral density values corresponding 1-50 Hz frequency interval of the EEG slices which has 10 seconds of time interval, the ratio of the total of the PSD values of current EEG slice to the total PSD values of EEG slice of pre-anesthesia, the ratio of the total PSD values of the EEG data to the total PSD values of the previous EEG data, and the previous anaesthetic gas ratio values have been applied and the network has been educated. The designed neural network system has been tested by using 10 data set obtained from 4 different patients. In the anesthetic gas prediction according to the anesthesia level, successful results have been obtained with the designed system. The system has been able to correctly purposeful responses in average accuracy of 94% of the cases. This method is also computationally fast and acceptable real-time clinical performance has been obtained.

A constrained penalty function method for exploratory adaptive-critic neural network (NN) control is presented. While constrained approximate dynamic programming has been effective to guarantee closed-loop system performance and stability objectives, in the presence of a change in the plant dynamics it may not have the necessary plasticity to explore and fully adapt to the new behaviors of the plant, if these violate the constraints. A generalized constrained approach is introduced to overcome these limitations. Through this methodology it is shown that NNs are not only capable to acquire new plasticity when necessary, but also can adjust their parametric structure reducing their hidden nodes and becoming more computationally efficient.

The design and development of the neural network (NN)-based controller performance for the activated sludge process in sequencing batch reactor (SBR) is presented in this paper. Here we give a comparative study of various neural network (NN)-based controllers such as the direct inverse control, internal model control (IMC) and hybrid NN control strategies to maintain the dissolved oxygen (DO) level of an activated sludge system by manipulating the air flow rate. The NN inverse model-based controller with the model-based scheme represents the controller, which relies solely upon the simple NN inverse model. In the IMC, both the forward and inverse models are used directly as elements within the feedback loop. The hybrid NN control consists of a basic NN controller in parallel with a proportional integral (PI) controller. Various simulation tests involving multiple set-point changes, disturbances rejection and noise effects were performed to review the performances of these various controllers. From the results it can be seen that hybrid controller gives the best results in tracking set-point changes under disturbances and noise effects.

Control over neuronal growth is a prerequisite for the creation of defined in vitro neuronal networks as assays for the elucidation of interneuronal communication. Neuronal growth has been directed by focusing a near-infrared laser beam at a nerve cell's leading edge ͓A. Ehrlicher, T. Betz, B. Stuhrmann, D. Koch, V. Milner, M. G. Raizen, and J. Käs, Proc. Natl. Acad. Sci. U.S.A. 99, 16024 ͑2002͔͒. The setup reported by Ehrlicher et al. was limited to local laser irradiation and relied on a great deal of subjective interaction since the laser beam could only be steered manually. To overcome the drawbacks of the reported setup, we developed and here present a fully automated low-contrast edge detection software package, which responds to detected cell morphological changes by rapidly actuating laser steering devices, such as acousto-optical deflectors or moving mirrors, thus enabling experiments with minimum human interference. The resulting radiation patterns can be arbitrary functions of space, time, and cell morphology, and are calculated by experiment specific feedback routines. Data processing is repeated on the order of 1 s allowing rapid reactions to morphological changes. The strengths of our program are the combination of real-time low contrast shape detection with complex feedback mechanisms, as well as easy adaptability due to a modular programming concept. In this article we demonstrate automated optical guidance; however, the software is easily adaptable to other problems requiring automated rapid responses of equipment to changes in the morphology of low contrast objects.

This paper describes the models of a wind power system, such as the turbine, generator, power electronics converters and controllers, with the aim to control the generation of wind power in order to maximize the generated power with the lowest possible impact in the grid voltage and frequency during normal operation and under the occurrence of faults. The presented work considers a wind power system equipped with the doubly-fed induction generator and a vector-controlled converter connected between the rotor and the grid. The paper presents comparative results between proportional-integral controllers and neural networks based controllers, showing that better dynamic characteristics can be obtained using neural networks based controllers.

This paper concerns the application of a neural network control strategy to the distributed collector field of a solar power plant. The neural network is trained based on measured data from the plant providing a way of scheduling between a set of PID controllers, a priori tuned in different operating points by means of Takahashi rules. The present work consists in a hybrid scheme combining the potentialities of neural networks for approximation purposes with the well-know theory and widespread industrial application of PID techniques.

In this paper, a reinforcement learning method is applied to coordinate a pair of horizontal hydraulic actuators engaged in the cooperative positioning of an object. The goal is to enable the actuators to discover how to intelligently select control actions that tend to reduce the interaction forces directed along the axis of motion, while maintaining the desired trajectory. First, a detailed and realistic dynamic model of the entire system is derived. A multi-layer reinforcement learning neural network control architecture is designed next to regulate the interaction force during positioning. To regulate the interaction force, the neural network measures the interaction force and proposes a modification to the a priori prescribed formation constrained position trajectory. Each actuator system is outfitted with such a neural controller so that a decentralized reinforcement learning control system results. Simulations demonstrate the efficacy of the approach towards reducing the interaction forces and minimizing the associated object internal force in a single degree of freedom.

Active Queue Management (AQM) has been widely used for congestion avoidance in Transmission Control Protocol (TCP) networks. Although numerous AQM schemes have been proposed to regulate a queue size close to a reference level, most of them are incapable of adequately adapting to TCP network dynamics due to TCP's non-linearity and time-varying stochastic properties. To alleviate these problems, we introduce an AQM technique based on a dynamic neural network using the Back-Propagation (BP) algorithm. The dynamic neural network is designed to perform as a robust adaptive feedback controller for TCP dynamics after an adequate training period. We evaluate the performances of the proposed neural network AQM approach using simulation experiments. The proposed approach yields superior performance with faster transient time, larger throughput, and higher link utilization compared to two existing schemes: Random Early Detection (RED) and Proportional-Integral (PI)-based AQM. The neural AQM outperformed PI control and RED, especially in transient state and TCP dynamics variation.

This paper proposes a neural network control voltage tracking scheme of a DC-DC Flyback converter. In this technique, a back propagation learning algorithm is employed. The controller is designed to improve performance of the Flyback converter during transient and steady state operations. Furthermore, to investigate the effectiveness of the proposed controller, some operations such as starting-up and reference voltage variations are verified. The numerical simulation results show that the proposed controller has a better performance compare to the conventional PI-Controller method.

Maximization of the catalyst efficiency in automotive fuel-injection engines requires the design of accurate control systems to keep the air-to-fuel ratio at the optimal stoichiometric value AF. Unfortunately, this task is complex since the air-to-fuel ratio is very sensitive to small perturbations of the engine parameters. Some mechanisms ruling the engine and the combustion process are in fact unknown and/or show hard nonlinearities. These difficulties limit the effectiveness of traditional control approaches. In this paper, we suggest a neural based solution to the air-to-fuel ratio control in fuel injection systems. An indirect control approach has been considered which requires a preliminary modeling of the engine dynamics. The model for the engine and the final controller are based on recurrent neural networks with external feedbacks. Requirements for feasible control actions and the static precision of control have been integrated in the controller design to guide learning toward an effective control solution.

We created a neural architecture that can use two different types of information encoding strategies depending on the environment. The goal of this research was to create a simulated agent that could react to two different overlapping chemicals having varying concentrations. The neural network controls the agent by encoding its sensory information as temporal coincidences in a low concentration environment, and as firing rates at high concentration. With such an architecture, we could study synchronization of firing in a simple manner and see its effect on the agent's behaviour.

An interesting alternative to electric actuators for medical purposes, particularly promising for rehabilitation, is a pneumatic artificial muscle (PAM) actuator because of its muscle-like properties such as tunable stiffness, high strength to weight ratio, structure flexibility, cleanliness, readily available and cheap power source, inherent safety and mobility assistance to humans performing tasks. However, some limitations still exist, such as the air compressibility and the lack of damping ability of the actuator bring the dynamic delay of the pressure response and cause the oscillatory motion. Then it is not easy to realize the performance of transient response of PAM manipulator due to the changes in the physical condition of patients as well as the various treatment methods.

We derive a linear neural network model of the chemotaxis control circuit in the nematode Caenorhabditis elegans and demonstrate that this model is capable of producing nematodelike chemotaxis. By expanding the analytic solution for the network output in time-derivatives of the network input, we extract simple computational rules that reveal how the model network controls chemotaxis. Based on these rules we find that optimized linear networks typically control chemotaxis by computing the first time-derivative of the chemical concentration and modulating the body turning rate in response to this derivative. We argue that this is consistent with behavioral studies and a plausible mechanism for at least one component of chemotaxis in real nematodes.

This report constitutes an expanded version of a presentation given by the author at the 1993 European Control Conference (short course on "Neural Nets for Control"). The first part places neurocontrol techniques in a general learning control framework. The second part of the report, which is essentially independent of the first, briefly surveys several basic theoretical results regarding neural networks.

In this paper, a simple neural network (NN) control scheme is developed for a class of discrete-time multi-input multi-output (MIMO) non-affine nonlinear systems with triangular form inputs and disturbances. The system studied is described by NARMAX (Nonlinear Auto Regressive Moving Average with eXogenous inputs) model. Firstly, by using implicit function theorem, the existence of the implicit desired feedback control (IDFC) is proved. Then single layer neural networks are used as the emulators of the desired controls. The stability of the closed-loop system is rigorously proved by using Lyapunov method. Because only input and output sequences are needed to construct the approximation based controls, the method proposed is very simple to be implemented in practical applications.

An undergraduate bioengineering laboratory course using small autonomous robots has been developed to demonstrate control theory, learning, and behavior. The lab consists of several modules that demonstrate concepts in classical control theory, fuzzy logic, neural network control, and genetic algorithms. The autonomous agents are easy-to-build, inexpensive kit robots. Each robot functions independently in a realworld environment. Students program and retrieve data wirelessly using handheld computers. The hands-on nature of the lab modules engages students in ways that lectures, readings and software simulations cannot. By interacting with these robots, students directly experience the effects of unexpected environmental factors on designs and deviations from software simulations. The robots are easily adapted for use in many different aspects of two-year college and K-12 STEM education. Students are motivated to understand engineering, math and science principles in order to control the robots. Examples of use of the robots and modules by a local community college are presented.