Neural Network based Optimization for a Batch Extractive Distillation Process

In the pharmaceutical industry, batch extractive distillation (BED), which is a combination process between extraction and distillation process, is widely carried out to separate waste solvent mixture of acetone-methanol because of minimum-boiling azeotrope properties of acetone-methanol mixture. In the operation, water has been used as solvent. Semi-continues mode has been proposed to improve purity of acetone. The solvent always has charged into the BED column until it has given purity of desired product. The BED process has operated at total reflux ratio and controlled the reboiler holdup which must not exceed the maximum capacity at any time within the entire operation period to avoid column flooding. A mathematic model has developed to represent process dynamic behavior according to the experimental results obtained in a previous work. The relevant studies have showed that the process behavior of BED is highly nonlinear. Consequently, this work has proposed neural network modeling of BED process to handle nonlinear process dynamic behavior. A single hidden layer with 12 nodes has used in the modeling. Its MSE index of the validate data test is 7.33e-09 which gives a good performance. In addition, the optimization is necessary for an optimal solvent feed rate profile of BED achieving the desired purification of acetone. It is obtained using neural network based optimization strategy. The objective function and the number of intervals of manipulate variable are the factors affecting purification of desired product.