IDENTIFICATION OF GROSS ERRORS IN MATERIAL BALANCE MEASUREMENTS BY MEANS OF NEURAL NETS

Reliable sets of steady-state component and total flow rate data form tbe cornerstone for the monitoring of plant performance. The detection and isolation of gross errors in these data constitute an essential p art of the process of reconciliation of the measurement data, which ar e generally inconsistent with process constraints. By using a neural n et to classify measurement or constraint residuals, gross errors in th e data can be identified accurately and efficiently. Gross error detec tion and isolation with artificial neural nets do not require explicit knowledge of the distribution of random errors in measurement values and can be applied to processes with arbitrary constraints.