VOL. 10, NO. 16, SEPTEMBER 2015 ISSN 1819-6608 ARPN Journal of Engineering and Applied Sciences © 2006-2015 Asian Research Publishing Network (ARPN). All rights reserved. www.arpnjournals.com 7190 SYSTEM IDENTIFICATION AND CONTROL OF PRESSURE PROCESS RIG ® SYSTEM USING BACKPROPAGATION NEURAL NETWORKS Benyamin Kusumoputro, Karlisa Priandana and Wahidin Wahab Computational Intelligence and Intelligent System Research Group, Department of Electrical Engineering, Faculty of Engineering, Universitas Indonesia, Kampus Baru Universitas Indonesia, Depok, West Java, Indonesia E-Mail: kusumo@ee.ui.ac.id ABSTRACT A neural networks based direct inverse controller for Pressure Process Rig ® system is presented, including with the performance analysis using an open-loop and a closed loop system. In order to enhance the performance characteristics of this direct inverse controller, a Fine-Tuning method is proposed. Experimental results show that the open-loop system shows lower MSE compare with that of the closed-loop system, and the Fine-Tuned NN-DIC method always performed better with lower MSE compare with that of the normal NN-DIC method. Keywords: direct inverse controller, neural networks, back propagation learning, fine-tuned DIC method. INTRODUCTION The dynamic behavior of a time-dependent nonlinear system is in general could not be accurately modeled by static input-output mapping strategy. In order to deal with this problem, the control system of a time- dependent nonlinear system should design to be adaptive, robust and flexible. The conventional proportional- integral-derivative (PID) controllers are widely used in industry due to their simple control structure, ease of design and low cost (Shin et al. 2012), (Rodriques et al. 2012), (Guzinski et al. 2013), (Holmes et al. 2012), however, the PID controller could not provide a perfect control performance if the controlled system is highly nonlinear. As the consequence, PID controller could not guarantee that the system would work with the same level of accuracy in the entire operating range. Considerable works has been reported, recently, concerning the use of artificial neural networks algorithm as a control system for a time-dependent nonlinear system. Artificial neural networks is a machine that is designed to model the performance of the brain on its ability to solve a particular task of interest based on a pattern recognition scheme. A neural networks is a massive parallel distributed processor made up of a simple processing neuron for memorizing the knowledge and making it available after training. The procedure used to perform the learning process is called the learning algorithm. The main objective of this mechanism is to modify the connection weights between neurons in the networks in an orderly fashion to attain the mapping capability of a set of input patterns onto a corresponding set of output patterns. A simple but powerful neural networks is a multi-layer perceptron (MLP) with one hidden layer, trained by using a back-propagation learning mechanism for updating the neural networks parameters. In recent years, there has been a significant increase in the number of control system methods that are based on nonlinear concepts. The nonlinear inverse model based control is one of such methods, which is dependent on the availability of the inverse of the plant model. As the neural networks have the ability to model any nonlinear system, including their inverse, their use as a controller is promising. In this paper, the design and evaluation of a neural network based inverse controller to a Pressure Process Rig ® system, or PPR ® in short, is presented. Pressure Process Rig ® system is one of the nonlinear systems that the neural network shall be implemented and its control performance could be evaluated for future development of an adaptive and robust controller system based on Neural Networks. Especially, this performance analysis is very important for the development of the error-based direct inverse controller with disturbance rejection capability. This paper is organized as follows. Section II presents description of the Pressure Process Rig ® as a system plant. In addition, the data collection and its processing are also described. Section III discusses the design and the development of the neural networks based controller in detail, including with the system identification of the plant, as the strong system identification capabilities of a neural networks could be extended and utilized to design a better nonlinear controller. The validity of the design procedure and the robustness of the proposed controller are verified by means of a computation simulations and experimental analysis, which is presented in Section IV, follows by the conclusion that is presented in Section V. The Pressure Process Rig ® System Pressure Process Rig ® control that was used in this experiment is a laboratory model developed by Feedback Instrument Ltd. The schematic diagram of the Pressure Process Rig control is borrowed from the manual provided from the manufacture that illustrated in Figure 1. For more specific information regarding the system, please refer to the available manual book (Feedback, 2006). When the system is in the operation mode, a mini compressor supplied a gas through a pipeline into the orifice of the system, and a control valve is put in this line for controlling the Pout outlet pressure. A control signal is