IJCSNS International Journal of Computer Science and Network Security, VOL.9 No.6, June 2009 142 Manuscript received June 5, 2009 Manuscript revised June 20, 2009 Design of Plant Estimator Model Using Neural Network K.SURESH MANIC, R.SIVAKUMAR, V.NERTHIGA, R.AKILA, K.BALU* Research Scholar, Department of Chemical Engineering, A.C. Tech, Anna University, T.N, India. *Professor, Department of Chemical Engineering, A.C. Tech, Anna University, T.N, India. Abstract The construction of a parameter (or state) estimator can be basically considered as a function approximation problem. To design an estimator, it is first necessary, to obtain the training data set ‘G’ such that, this training data set contains as much information as possible about a system ‘g’. Once trained properly, the estimator will adaptively follow the slope of ‘g’ at all times. In this paper, signals are processed in real time and combined with previous monitoring data to estimate, using the neural network, the process variable level in a nonlinear process control plant. Key words: Estimator, Neural Network, Nonlinear control, Sensor validation. 1. Introduction In the area of process engineering, process design and simulation, process supervision, control and estimation, and process fault detection and diagnosis rely on the effective processing of unpredictable and imprecise information. In such situations, the neural network, which can achieve the sophisticated level of information processing the brain is capable of, can excel. The neural networks are generally viewed as process modeling formalism and given the appropriate network topology, they are capable of characterizing nonlinear functional relationships [3]. Furthermore, the structure of the resulting neural network based process model may be considered generic, in the sense that little prior process knowledge is required in its determination. The knowledge about the plant dynamics and mapping characteristics is implicitly stored within the network. 2. Design of Estimator Using Neural Network Training a neural network using input-output data from a nonlinear plant is considered as a nonlinear functional approximation problem. A generic neural network estimator model, used to detect a sensor failure is shown in figure 1. Neural networks have effectively been used in many applications to predict performance degradation of operating systems in real-time. Neural networks are data driven models and data under a variety of conditions need to be obtained. In the present work the experimental setup was used to gather data and the key measurable signals that were collected for training the network consisted of the inflow, outflow rate and the process value level. Different operating conditions were simulated and the change in inflow, outflow and the level were recorded. The data collected from the plant were pre-processed for normalization and fed to the Figure 1 Neural network estimator model