A review of techniques for instrumentation design and upgrade in process plants

Abstract

nstrumentation is needed in process plants to obtain data that are essential to perform several activities: control of plants, assessment of I the quality of products, production accounting (sometimes called yield accounting), detection of failures related to safety, parameter estimation. This review concentrates on the techniques for the optimal allocation of instruments in grassroots as well as in retrofit scenarios. The article is a condensed version of a book by the author (Bagajewicz, 2000). Model-Based MonitoringModel-Based Monitoring consists of the use of a combination of models and on-line measurements to: a) produce estimates of measured and unmeasured variables; b) identify malfunctioning instrumentation; c) identify unsafe and/or faulty operating conditions and their origin; and d) identify events that can impact efficiency and quality of products.Kramer and Mah (1993), in an excellent overview of this issue, discussed several scenarios in which good estimates of data can be obtained. Data reconciliation and gross error detection are techniques that help accomplishing tasks (a) and (b) and constitute a particular case of the concept of data rectification. While data reconciliation r#elies on analytical constraints and mostly on least square estimation, data rectification can obtain these estimates using techniques like Kalman filtering, pattern recognition, neural networks, principal component analysis, partial least squares, among others. Likewise, the detection of faults relies on techniques based on statistics, but several other techniques can be used. Tasks (c) and (d) are direct consequences of using a model.The field of sensor network design and upgrading has traditionally relied on model-based concepts. Almost all work addressing monitoring goals includes:the ability of the sensor network to provide estimates of variables of interest; the ability to guarantee certain accuracy through data reconciliation; reliability; capability of identifying gross errors; and capability of identifying process faults. Quality of DataQuality of data is a term that is often used to refer to several properties that a set of data should have. These properties can be condensed in the following three broad attributes: accuracy; precision or reproducibility; E-mail address: bagajewicz@ ou.eduThe Canadi an Journal of Chemical Engineering, Volume 80, February 2002This article is a review of the area of optimal instrumentation location in process plants for both grassroots designs and upgrades. Emphasis is put in methods for monitoring purposes and based on the cost minimization model. These methods are mostly related to data reconciliation paradigms and include especial consideration to the detection of gross errors. Connections between this model and other models based on other objective functions and goals are analyzed. Finally, the emerging field of sensor design for efficient process fault detection is briefly covered.On passe en revue dans cet article la problematique du positionnement optimal de I'instrumentation ...

Optimal sensor network design for multirate systems

“…The measurement of variable 14 is not part of any optimal sensor network scheme. Without this optimization, a process designer might have chosen these sensors (for variables [12][13][14] due to their low variances (Table 6). By optimizing the sensor network design, we are able to find other sensor network combinations which provide comparable quality of estimation of states at lower cost compared to measurements 12-14.…”

“…Additionally, Bagajewicz and Cabrera [11] have discussed visualization schemes for displaying solutions for multiobjective sensor network design and upgrade problems. These approaches are fundamentally different in scope from the problem considered in the current article, and the interested reader can refer to the survey paper by Bagajewicz [12] for further information.…”

Boolean Network-Based Sensor Selection with Application to the Fault Diagnosis of a Nuclear Plant

“…Sensor selection was treated as different DG-based optimization problems in most early studies. Bagajewicz et al summarized the sensor selection in a process as mix integer linear programming (MILP) problems focusing on optimizing cost or (and) reliability [1][2][3][4]. Bhushan, Narasimhan, and Rengaswamy added the criteria of robustness to the MILP problems [5].…”

“…Nodes Description f 1 abnormal reactivity injection to the reactor f 2 malfunction of the primary helium blower f 3 heat transfer degradation of OTSG two sides…”

Robust sensor network design for fault diagnosis

“…The aim of sensor location problem as posed and solved in the literature is to identify variables to be measured such that various criteria, viz: observability (ability to estimate values of all variables), precision (estimation of certain key variables with minimum variance), residual precision (estimation of certain key variables with required variance even if some sensor measurements are deleted), reliability of estimation of variables (in presence of sensor failure probabilities), gross error detectability (ability to detect gross errors of a certain size), and error resilience (ability to limit the smearing effect of undetected gross errors), are satisfied. An interested reader is referred to Bagajewicz's (2002) review article for further detailed information. As opposed to the above approaches which are mainly variable-centric, the aim of sensor location from fault diagnostic perspective is to identify variables to be measured such that various faults can be detected and diagnosed.…”