Dynamic Neural Network Structure Research Articles

A Hybrid Intelligent Active Force Controller for Articulated Robot Arms Using Dynamic Structure Neural Network

The key feature of this paper is the application of a robotic control concept – Active Force Control (AFC). In this type of control, the unknown friction effect of the robotic arm may be compensated by the AFC method. AFC involves the direct measurement of the acceleration and force quantities and therefore, the process of estimating the system ‘disturbance’ due to friction becomes instantaneous and purely algebraic. However, the AFC strategy is very practical provided a good estimation of the inertia matrix of articulated robot arm is acquired. A dynamic structure neural network – Growing Multi-experts Network (GMN) is developed to estimate the robot inertia matrix. The growing and pruning mechanism of GMN ensures the optimum size of the network that results in an excellent generalization capability of the network. Active Force Control (AFC) in conjunction with GMN successfully reduces the velocity and position tracking errors in spite of robot joint friction. The embedded GMN is capable of coupling the inertia matrix estimation on-line that clearly enhances the performance of AFC controller. The robustness and effectiveness of the new hybrid neural network-based AFC scheme are demonstrated clearly with regard to two link articulated robot and a simulated two-degree of freedom Puma 560 robot.

ANN generalised inversion for the linearisation and decoupling control of nonlinear systems

A generalised inverse system (or generalised inversion) method is developed for the linearisation and decoupling control of a general nonlinear dynamic continuous system. The existing condition for generalised inversion of a SISO or MIMO nonlinear system is discussed. Combining the generalised inversion and the original system creates a pseudo-linear composite system whose poles can be placed in the left-half complex plane so as to obtain an ideal open-loop frequency characteristic. To implement the generalised inversion for the control of an unmodelled or only poorly modelled dynamic nonlinear system, a dynamic artificial neural network (ANN) structure formed by connecting a number of linear components to a static ANN is proposed. The static ANN is responsible for approximating the static nonlinear mapping describedby the analytical expression of the generalised inversion and the linear components are used to represent the dynamics of the generalised inversion. Simulation results are also presented to illustrate that the proposed control technique provides a practical approach to realise the linearisation and decoupling control of nonlinear SISO or MIMO systems

SYSTEM IDENTIFICATION USING DYNAMIC NEURAL NETWORKS: TRAINING AND INITIALIZATION ASPECTS

This paper explores training and initialization aspects of dynamic neural networks when applied to the nonlinear system identification problem. A well known dynamic neural network structure contains both output states and hidden states. Output states are related to the outputs of the system represented by the network. Hidden states are particularly important in allowing dynamic neural networks to approximate complex nonlinear dynamics. An optimisation based method is proposed in this paper for properly initialising the hidden states of a dynamic neural network, so as to avoid the introduction of bias in the network parameters as a result of incorrect hidden state initialisation. Furthermore, a simple optimisation based method is proposed to initialise the hidden states once the network has been trained. The methods are illustrated with experimental data taken from a laboratory scale pressure plant.

Neural network controller characteristics with regard to adaptive control

The authors attempt to clarify neural network (NN) controller characteristics by comparison with the adaptive control theory. The authors clarify the classification of the NN controller architecture and the dynamic NN structure. Comparison between the NN controller and the adaptive controller clarifies that the framework of a linear two-layer NN-type controller is the same as that of adaptive control and the nonlinear three-layer NN (PDP type) is a nonlinear extension of the adaptive control. The stability characteristics of the NN control system are discussed. NN controller experiments are demonstrated using a force control servomechanism. Experimental results suggest that the nonlinear sigmoid function of the NN can compensate for the nonlinear plant effect.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>