Theory and practice of simultaneous data reconciliation and gross error detection for chemical processes

On-line optimization provides a means for maintaining a process near its optimum operating conditions by providing set points to the process’s distributed control system (DCS). To achieve a plant-model matching for optimization, process measurements are necessary. However, a preprocessing of these measurements is required since they usually contain random and—less frequently—gross errors. These errors should be eliminated and the measurements should satisfy process constraints before any evaluation on the process. In this paper, the importance and effectiveness of simultaneous procedures for data reconciliation and gross error detection is established. These procedures depending on the results from robust statistics reduce the effect of the gross errors. They provide comparable results to those from methods such as modified iterative measurement test method (MIMT) without requiring an iterative procedure. In addition to deriving new robust methods, novel gross error detection criteria are described and their performance is tested. The comparative results of the introduced methods are given for five literature and more importantly, two industrial cases. Methods based on the Cauchy distribution and Hampel’s redescending M-estimator give promising results for data reconciliation and gross error detection with less computation.