On Line Estimation of Composition Using Software Sensor in Batch Distillation Operation

This work addresses the design of a software sensor using GRNN model for predictions of product compositions. Product composition in distillation column is a function of operating temperature, heat load, reflux ratio and time duration. Design guidelines using GRNN model has been presented for modeling of batch distillation column. GRNN model is used to correlate input and output data and it has potential to approximate non linear input output relationship efficiently. Model is constructed exclusively from historic process input output data, by undergoing on line training. GRNN formulation and network training being one-step process. The experimental results of compositions found to agree well with model-simulated results.