Tuning Linearization Transformation using Back-Propagation Algorithm
S Janakiraman1, Rajagopalan Devanathan2
1S Janakiraman*, Centre for Automation and Robotics, Hindustan Institute of Technology and Science, Chennai, Tamil Nadu, India.
2Rajagopalan Devanathan, Department of Electrical and Electronics, Hindustan Institute of Technology and Science, Chennai, Tamil Nadu, India.
Manuscript received on September 22, 2019. | Revised Manuscript received on October 20, 2019. | Manuscript published on October 30, 2019. | PP: 1471-1476 | Volume-9 Issue-1, October 2019 | Retrieval Number: A1261109119/2019©BEIESP | DOI: 10.35940/ijeat.A1261.109119
Open Access | Ethics and Policies | Cite | Mendeley
© The Authors. Blue Eyes Intelligence Engineering and Sciences Publication (BEIESP). This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Abstract: The objective of linearization of a nonlinear system is to ensure smooth control of the linearized system through well-proven linear control methods. However, residual nonlinearities may still be present in a system after linearization either by design or due to mismatch between the system model and the actual plant. If the residual nonlinearities are not very significant, one can attempt to remove these by tuning the linearizing transformation by comparing the system to a linear canonical form. In this paper, we show how quadratic linearizing transformations of a three-phase horizontal gravity separator (TPS) model derived in an earlier paper by the authors can be tuned as in a neural network using error back-propagation by comparing it to a canonical linear model thus removing the nonlinearities within the tuning error limit.
Keywords: Approximate linearization, back-propagation learning algorithm, control affine model, state space, three-phase horizontal gravity separator, neural network.