(16c) Application-Oriented Review of Data Reconciliation and Gross Error Detection Methods for Heat Exchanger Networks

José Loyola-Fuentes, Carlos Planelles Alemany, Emilio Diaz-Bejarano and Francesco Coletti*

Hexxcell Ltd., Foundry Building, 77 Fulham Palace Rd, London W6 8AF, UK

*corresponding author: f.coletti@hexxcell.com

Extracting useful insights from raw plant data is often a challenging and frustrating experience. On one hand, measurements may be subject to random errors (e.g. arbitrary fluctuations in the environment, signal transmission) and/or systematic errors (measurement bias, instrument degradation, data corruption or complete sensor failure). On the other hand, plant operators rely on these measurements (e.g. flow rates, temperatures, pressures) for process and asset monitoring and make important operating and maintenance decisions based on this information. It is therefore clear that rigorous approaches capable of dealing with the various sources of error are needed.

The field of statistics provides a substantial number of filtering techniques that are able to identify significant disturbances in the measured data, but the physical context where the process states are embedded in is not entirely considered. Therefore, a suitable method that combines the physical and statistical knowledge of a specific system could be beneficial for monitoring purposes. Data reconciliation (DR) and gross error detection (GED) are complementary methods that reduce the effect of random and gross errors by exploiting the relationship between measured variables and a specific process model. The estimations of process variables via DR are expected to be unbiased, more accurate and in accordance with the process model, which in case of heat exchanger networks, could be mass and energy conservation equations. The role of GED is to deliver one or more of the following objectives: to detect the presence of gross errors, to find the sources of single/multiple gross error and to estimate the magnitude of such errors. The challenges related to the completion of these tasks (including DR) depend on the conditions and assumptions associated with the process model, and throughout the past 50 years, a wide variety of methods have been developed to address different data reconciliation problems.

In this presentation, we critically review current practice and state-of-the-art in data reconciliation and gross error detection techniques, applied to heat exchanger networks. The review includes considerations for both steady-state and dynamic systems and an application-oriented summary of the advantages and disadvantages of various techniques (e.g. classic and robust statistics, Kalman filtering, dynamic programming) and presents relevant case studies.