Neural network based level control in two tank conical interacting system

The control of liquid level in tanks form an essential part in many process industries. A level that is too high may upset the reaction equilibrium, cause damage to the equipment or result in spillage of valuable or hazardous material. If the level is too low, it may have bad consequences for the sequential operation. In conical tanks the control of liquid level presents a challenging problem due to its non-linearity and constantly changing cross section. In a two tank conical interacting system it is difficult to control the level in both tanks with a single PID controllers because it can be used only in linear systems. Non-linearity of the system can be controlled by piece wise linearization of the process input-output response curve and dividing the curve into various regions. Separate PID controllers are designed for each linearized regions. The system is tuned and the control parameters are obtained. Though PID controllers are very efficient, it is time consuming and produces damped oscillation. To overcome this problem neural network based controllers are designed.