Novel Technique for Data Reconciliation and Measurement Bias Detection with Serially Correlated Process Data

Abstract

Process industries implement intricate measurement systems to monitor process variables to newlinefacilitate control, dynamic optimization, online diagnostics, and real-time monitoring. In the newlinemeasured process variables, two types of errors can be observed: random and fixed errors newline(also known as gross errors). Data Reconciliation (DR) techniques address random errors, newlinewhile Gross Error Detection(GED) techniques are utilised to handle gross errors. In many DR newlineand GED techniques, data collected from various sources are assumed to be independently and newlineidentically distributed (i.i.d.). In practice, process loop delays, signal processing elements, newlineand other process phenomena cause serial correlation in measurement data. newlineThis thesis aimed to investigate previous techniques such as Variance Correction (VC) and newlinePre-whitening techniques and to develop a novel approach for effectively handling serial newlinecorrelation in various scenarios, utilising an innovative computational environment. An additional newlinesignificant contribution of our work is the introduction of the Variance Correction newlinePrincipal Component Analysis (VCPCA) based Measurement Test (MT). This novel technique newlinereplaces the conventional WLS-based estimator in the MT with a PCA-based estimator. newlineThe results demonstrated the superior and consistent performance of VCPCA across various newlinescenarios, including different sample sizes, varying biases, variances, and the order of serial newlinecorrelation structures. By employing the VCPCA technique, there is a notable reduction in newlinethe average type I error (AVTI), coupled with a substantial increase in Overall Power (OP) newlineor Overall Performance (OPF). It indicates a decreased frequency of false alarms generated newlineby the control system. The proposed technique applies to steady-state linear systems with newlineknown variances and process constraints. newline newline