Soft Sensor based on Adaptive Linear Network for Distillation Process

The main objective in refining units is to keep the product quality within specifications in the faces of disturbances. Online measurements of product composition using composition analyser are neither easy nor economically viable. In an effort to overcome these difficulties various soft sensors are designed in the recent years. In this research work, the authors have proposed the design of neural network based soft sensor for two types of chemical processes i.e. reactive distillation process and multicomponent distillation process. The designed soft sensor is based on adaptive linear network, Adaline and is used to infer the product composition from the temperature profile of the respective processes. For a comparative study Levenberg Marquardt based artificial neural network soft sensor is also designed. It is observed from the results that the Adaline based soft sensor is more efficient in comparison to LM based ANN soft sensor in terms of accuracy, time taken for training and memory usage. General Terms Application of Adaptive Linear Network in distillation process.