FORWARD PREDICTION OF COLUMN BREAKTHROUGH USING A NEURAL-NETWORK MODEL TO INTERPRET THERMAL SIGNALS

A neural network trained with data sets where a time off-set is introd uced between input and target signals has been used for forward predic tion of breakthrough in an ion-exchange adsorption column. The interva l used for the time scale shift was determined using a linear correlat ion between the relative elution volume and the position of the. Once trained the network proved capable of accurately predicting the forthc oming breakthrough curve using signals derived from a sensor mounted i n the top third of the column.