Showing papers by "Anindita Sengupta published in 2010"

Nonparametric modeling of glucose-insulin process in IDDM patient using Hammerstein-Wiener model

Abstract: This paper deals with an identification problem of modeling a nonlinear dynamic system of multivariable glucose-insulin process in an IDDM patient. Out of many model structures that can represent a nonlinear process effectively; the Hammerstein-Wiener model has attracted a lot of attention. The present work proposes a generalized identification method of Hammer stein-Wiener model from the input-output data of multivariable nonlinear glucose-insulin process. The present algorithm consists of a three-block (LNL) realization. For the multivariable system, the first and third blocks are standard impulse response filter (TRF) realization applied to an equivalent linear system using adaptive recursive least square (ARLS) algorithms. In the second block, i.e. the nonlinear part, ARLS algorithms have been used to solve up to second order kernels of Volterra equations with extended input vector consisting of cross components as well. The input-output data taken from the simulated nonlinear process have been used to identify the system with a filter memory length of M=3 and the validation results have shown good fit and in concordance with predicted output.

Identification of servo-driven inverted pendulum system using neural network

Abstract: In the present work, artificial neural network (ANN) has been used to identify a servo-driven inverted pendulum system. The inverted pendulum is a benchmark problem of nonlinear multivariable system with inherent instability. The multi variable system has been considered with servomotor supply voltage as the input and four states of the system being the outputs. An LSVF controller has been used to stabilize the system for identification in closed loop. Here the non linear model of the inverted pendulum has been simulated. The Levenberg-Marquardt back-propagation method has been used for the non linear system identification via Feed-forward Neural Network (FNN). The neural network is trained using the error between the model's outputs and the plant's actual outputs. The results show good match between predicted and actual outputs.

Wavelet transform for determination of state and trajectory sensitivity of a singular control system

Abstract: In this paper wavelet transform is used to determine the state and trajectory sensitivity of homogeneous and non-homogeneous systems. The differential-algebraic equation describing a system is converted via wavelet transform to an algebraic generalized Lyapunov equation which is solved for the coefficients of the state variables in terms of Haar basis. Further, problems of trajectory sensitivity analysis for singular as well as nonsingular systems have also been explored using the same orthogonal basis. Finally, using Kronecker product method, a generalized program is developed to determine the state and sensitivity for any number of basis function in Haar domain.