Cosmin Ionete

Neural networks-based adaptive control for a class of nonlinear bioprocesses

The paper studies the design and analysis of a neural adaptive control strategy for a class of square nonlinear bioprocesses with incompletely known and time-varying dynamics. In fact, an adaptive controller based on a dynamical neural network used as a model of the unknown plant is developed. The neural controller design is achieved by using an input–output feedback linearization technique. The adaptation laws of neural network weights are derived from a Lyapunov stability property of the closed-loop system. The convergence of the system tracking error to zero is guaranteed without the need of network weights convergence. The resulted control method is applied in a depollution control problem in the case of a wastewater treatment bioprocess, belonging to the square nonlinear class, for which kinetic dynamics are strongly nonlinear, time varying and not exactly known.

Nonlinear and Neural Networks Based Adaptive Control for a Wastewater Treatment Bioprocess

The paper studies the design and analysis of some nonlinear and neural adaptive control strategies for a wastewater treatment process, which is an activated sludge process with nonlinear, time varying and not exactly known kinetics. In fact, an adaptive controller based on a dynamical neural network used as a model of the unknown plant is developed and then is compared with a classical linearizing controller. The neural controller design is achieved by using an input-output feedback linearization technique.

Remote vs. simulated, virtual or real-time automation laboratory

This paper presents the implementation of an automation laboratory with virtual, remote, real-time and simulated experiments for a workstation (body feed-positioning station, part of a Flexible Manufacturing System FMS-200) and the benefits of the application of Virtual Reality, Telematics and 3D simulation in Control Engineering education. Beyond the theoretical training based on the Internet, practical training based on the remote use of labs tends to hold an important place. The aims of all developed experiments are to improve the training of the students or engineers in the field of automation, but at the same time to offer technology for industrial developments.

Synchronization problem for time-delay recurrent neural networks*

Abstract The subject of the present paper is suggested by the studies reported in the specific literature of the neuroscience domain on the rhythmic activities in the nervous system and on the synchronization of the oscillatory responses with the external time-varying inputs. Regarding synchronization, the aim of the paper is to state conditions ensuring a “good” qualitative behavior for time-delayed recurrent neural networks with time-varying (almost) periodic forcing terms that have to be “reproduced”.

Modeling and estimation strategies for a fed-batch prototype bioprocess

This paper deals with the Bond Graph modeling and the design of estimation strategies for a nonlinear fed-batch prototype bioprocess. The proposed strategies are developed for an aerobic microbial growth process coupled with an enzyme-catalyzed reaction, which is a usual bioprocess that takes place in fed-batch bioreactors. The rules for the design of pseudo Bond Graph model are obtained by using the reaction schemes and the analysis of biochemical phenomena. Two kinds of on-line estimation strategies are approached. First, a general state observer is analyzed and the exponential observability of the bioprocess is tested; two state estimation algorithms are designed: an extended Luenberger observer and an asymptotic observer. Second, an observer-based estimator is derived for the estimation of unknown kinetics. In order to test the behavior of proposed strategies, numerical simulations are included.

Sliding Mode Observers for Rotational Robotics Structures

The problem of controlling uncertain dynamical systems subject to external disturbances has been an issue of significant interest over the past several years. Most systems that we encounter in practice are subjected to various uncertainties such as nonlinearities, actuator faults parameter changes etc. Many of the proposed control strategies suppose that the state variables are available; this fact is not always true in practice, so the state vector must be estimated for use in the control laws. In the past, several types of observers have been designed for the reconstruction of state variables: Kalman filter (Kalman, 1976), adaptive observers (Gevers & Bastin, 1986), high gain observers (Gauthier et al., 1992), sliding mode observers (SMO) (Utkin, 1992; Walcott & Zak, 1986; Edwards & Spurgeon, 1994) and so on see (Thein & Misawa, 1995) for some comparisons. Depending upon the particular application, all these observers can be used with suitable results. Sliding mode observers differ from more traditional observers in that there is a non-linear discontinuous term injected into the observer depending on the output estimation error. These observers are known to be much more robust than Luenberger observers, as the discontinuous term enables the observer to reject disturbances (Tan & Edwards, 2000). The observers based on the variable structure systems theory and sliding mode concept can be classified in two categories (Xiong & Saif, 2000): 1) the equivalent control based methods and 2) sliding mode observers based on the method of Lyapunov. The analysis of these two types of SMO (Edwards & Spurgeon, 1994; Xiong & Saif, 2000) shows that there exist some differences in terms of robustness properties. From practical point of view, the selection of the switched gain for the equivalent control based SMO is difficult (in order to obtain a sliding mode without excessive chattering) (Edwards & Spurgeon, 1994). Also, there exists bounded estimation error for bounded modelling errors (the estimation will not be accurate when uncertainties are presented) (Xiong & Saif, 2000). The Lyapunov based SMO (the so-called Walcott-Zak observer) provides exact estimation for certain class of nonlinear systems under existence of certain class of uncertainties. However, the difficulty in finding the design and gain matrices is the main drawback of this observer. Consider the effect of adding a negative output feedback term to each equation of the Utkin observer. This results in a new error system. The addition of a Luenberger type gain matrix, feeding back the output error, yields the potential to provide robustness against certain classes of uncertainty.

Robotic leg control based on Human motion analysis and neural control methods

This work presents some results obtained from image processing and motion analysis of the human body. The issues raised in motion analysis are of interest for obtaining motion-specific parameters for movements of the human body. The resulting data (spatial coordinates, velocities and accelerations) are used for further processing in humanoid robotics and assistive and recuperative technologies for people with disabilities. The results are implemented on a robotic leg, which was developed in our laboratories. A model based neural control strategy is implemented, too. The performances of the implemented control strategies for trajectory tracking are analyzed by computer simulation.

Neural and Adaptive Control Strategies for a Rigid Link Manipulator

The control of robotic manipulators has become important due to the development of the flexible automation. Requirements such as the high speed and high precision trajectory tracking make the modern control indispensable for versatile applications of manipulators (Middleton & Goodwin, 1998; Ortega & Spong, 1999; Popescu et al., 2008). Rigid robot systems are subjects of the research in both robotic and control fields. The reported research leads to a variety of control methods for such rigid robot systems (Ortega & Spong, 1999; Raimondi et al., 2004; Bobaşu & Popescu, 2006; Dinh et al., 2008). Conventional controllers for robotic structures are based on independent control schemes in which each joint is controlled separately by a simple servo loop. This classical control scheme (for example a PD control) is inadequate for precise trajectory tracking. The imposed performance for industrial applications requires the consideration of the complete dynamics of the manipulator. Moreover, in real-time applications, the ignoring parts of the robot dynamics or errors in the parameters of the robotic manipulator may cause the inefficiency of this classical control. An alternative solution to PD control is the computed torque technique. This classical method is in fact a nonlinear technique that takes account of the dynamic coupling between the robot links. The main disadvantage of this structure is the assumption of an exactly known dynamic model. However, the basic idea of this method remains important and it is the base of the neural and adaptive control structures (Gupta & Rao, 1994; Pham & Oh, 1994; Dumbravă & Olah, 1997; Ortega & Spong, 1999; Aoughellanet et al., 2005; Popescu et al. 2008). Industrial robotic manipulators are exposed to structured and unstructured uncertainties. Structured uncertainties are characterized by having a correct model but with parameter uncertainty (unknown loads and friction coefficients, imprecision of the manipulator link properties, etc.). Unstructured uncertainties are characterized by unmodelled dynamics. Generally speaking, two classes of strategies have been developed to maintain performance in the presence of the parameter uncertainties: robust control and adaptive control. The adaptive controllers can provide good performances in face of very large load variation. Therefore the adaptive approach is intuitively superior to robust approach in this type of application. When the dynamic model of the system is not known a priori (or is not 11