Dissertation von Stefan Fabricius, 2003

Titel

Modeling and Simulation for Plant Performability Assessment with Application to Maintenance in the
Process Industry
(Dissertation: D-MAVT)

Betreuung

Prof. Dr. W. Kröger, Dr. E. Badreddin

Abgabe, Nummer

2003, Diss. ETH Nr. 15024

Abstract

Component failures in process plants can trigger incidents or cause costly production interruptions. Minimizing the failure count or their adverse effects potentially lead to increased safety, reduced downtime and consequently higher production efficiency. This work explores options for assessing and optimizing plant performability, i.e., the degree of performance (e.g., productivity) while being dependable (safe, reliable, available). Collaboration with a partner from the process industry assures that the economic interests and pragmatic needs that arise in practise are addressed and offers an area to apply system analysis and modeling methods research results.

Plant maintenance strategy and execution play a crucial role in avoiding failures. Preventive-periodic and condition-based strategies are well known, but not suited in all circumstances. In practice, maintenance still often relies on unplanned repairs or exchange of previously failed components (a breakdown strategy). To improve maintenance strategy allocation, a pragmatic, "quick-and-sound" nine-step procedure for maintenance strategy selection was developed. It integrates already available analysis methods (such as Failure Mode and Effects Analysis), jointly considers technical, organizational and economic aspects, treats component as well as plant levels and provides decision aids in the form of flow-charts and checklists. The procedure leads to a process-oriented maintenance strategy choice — i.e., the plant structure and simple dynamic behavior are accounted for — and to a targeted use of preventive schemes where appropriate. The transparent, simple-to-use and systematic approach serves as a first attempt towards enhanced, cost efficient maintenance with facilitated plant operation and maintenance work planning as side effects. In addition, a concept for real-time, online plant fault monitoring was developed, employing available fault detection techniques, extended by parameterizable modular and distributed schemes for fault prognosis and diagnosis. Further experimental research and development work will be necessary to realize working monitoring solutions though. A laboratory test-bed was designed and installed for this purpose.

Two aspects were seen to complicate the determination of cost-effcient maintenance migration and process plant performability improvement options. First, in practice the economic impact of foreseen maintenance innovations is often hard to estimate prior to introduction as well as to quantify thereafter (or only possible after long time delays), but nevertheless is of decisive importance. Second, plant operation — involving control actions, product changes and dynamic characteristics of the plant — strongly affects component and plant dependability. Hence, performability improvement indeed constitutes a multiaspect, non-trivial optimization problem. To tackle it, available system analysis and modeling methods — e.g., as used in safety analysis — could be employed. However, most classical safety analysis methods are limited to static or scenario-oriented views, and tend to stand isolated in determining individual dependability attributes. Therefore, an approach to extend the capabilities of the classical analysis methods was taken, based on: i) integration of isolated methods, ii) a move from safety to performability analysis (which inherently leads to consideration of dynamics), iii) multi-formalism modeling and iv) use of computers and the simulative solution approach for quantitative analysis.

Starting with discrete-event modeling formalism, new high-level, timed, generalized stochastic Petri nets (PNs) were developed for component and system functional state representation. It is shown how PNs can be modularly implemented using an available stand-alone computer tool and be deployed for availability assessment by means of simulation. Result accuracy is proved with model versions of analytically tractable systems. Moreover, it is demonstrated how plant maintenance strategy, costs and resource scarcity can be conveniently included with further extended PNs. This was not previously possible with formalisms as e.g., Markov chains, which are limited to memoryless probability distributions and potentially suffer from combinatory state space explosion.

Discrete-event formalism is well suited to model distinct states and complex logic, but not to represent continuous variables, e.g., tank levels or flow rates. Hence, PNs are not suffcient alone to adequately represent process plants if continuous-type variables are of importance. Therefore, a concept and implementation of hybrid (discrete-event and continuous-time) dynamic performability assessment (HDPA) was developed. It comprises modeling of component failure and repair characteristics, plant topology and flow dynamics, control aspects and performance evaluation in holistic unified models. Leverage of classical analysis methods is featured by appropriate transformation of Fault Trees, enabling their inclusion in HDPA. To facilitate investigation of dynamic behavior, the principles of System Dynamics (J. Forrester) were transferred in a novel approach to problems of performability evaluation. Readily applicable, effcient computer tools for hybrid modeling not suffering from narrow focus on special-purpose applications or from proprietary inflexibility are still rare. Due to this lack, considerable effort was necessary to work out and customize new models and libraries for HDPA. Using the general-purpose, open Modelica modeling language specification and related tools proved most suitable in this respect, since they support many of the modeling principles thought necessary for handling of complex systems, e.g., object-orientation, scalability and multi-formalism. Several new Modelica libraries — among them a Petri net, a System Dynamics and a fluid flow library — were developed, which when combined allow implementation of HDPA and support re-usability of modeling know-how. With a special focus on mass-flow in process plants — and interruptions due to component failure — it is shown in analysis studies how changes to the plant configuration (e.g., increasing intermediate storage tank capacity, introducing more reliable or redundant components, relocating the bottle neck) affect overall production output. The feasibility of the modeling concept is demonstrated in particular for chemical plants with continuous, discontinuous and combined processes. It is proved that the results of process plant analysis using traditional static methods can be misleading and differ significantly from those yielded by the new dynamic ones.

Application of performability analysis and simulation can save money in a production company (not only of the process industry), avoiding potentially costly "trial-and error" experiments with the real plant. In particular, it can help to quantify return-on investment of maintenance or monitoring innovations, generating results that support informed, founded decision-making. The new performability models offer progress over available static models and methods, help to better understand dynamic system failure and recovery behavior, assist in locating weaknesses and identify potential for cost efficient improvements. Classical safety analysis (e.g., as carried out for a certain critical chemical reaction in a plant segment) can be supplemented by HDPA in order to combine safety considerations with economic and dynamic operational issues.