Adaptive nonlinear dynamic data reconciliation and gross error detection

Data reconciliation (DR) is a well-known method in on-line process control engineering aimed at estimating the true values of corrupted measurements under constraints [11]. It is crucial to detect and identify gross errors first or in some cases simultaneously with DR. There have been some new gross error detection (GED) and statistical model identification approaches developed recently and combined with the original nonlinear dynamic DR (NDDR) method in order to remove the negative effects of gross errors. Among these methods there are very few approaches which address the situation where a statistical model is not available. However, they cannot handle either nonlinearity or dynamic behavior of the processes [1].;In the first step, one of the most applicable NDDR methods introduced by Liebman et al. [7], is studied in this thesis. This technique was designed and tested for inputs that undergo step changes and are otherwise constant. Next, an adaptive NDDR (ANDDR) method is proposed that includes the application to processes with an unknown statistical model. A novel GED method is developed as well and combined with the ANDDR algorithm. A new smart tracking system is also combined, to ameliorate the problem of delay seen in both the original and later NDDR methods. Finally, an extension is made to include applications with slowly and smoothly varying inputs. The proposed package has been successfully applied to the simulated continuously stirred tank reactor (CSTR) model cited commonly in the literature. As a more complex case study, this package is also tested and implemented on a simulated jacketed CSTR (JCSTR) model. The proposed package with its smart tracking features is suggested for use in distributed control systems ( DCSs) or chemical process control to improve process monitoring.