Dissertation von Deniz Yanar, 1999

Titel

Modellbasierte Risikoanalyse technischer Systeme für Versicherungszwecke

(Dissertation: D-MAVT)

Betreuung

Prof. Dr. W. Kröger, Dr. E. Badreddin

Abgabe, Nummer

1999, Diss. ETH Nr. 15024

Abstract

Background
A part of the risks of companies, mostly of technical nature, is covered by insurance. In order to evaluate risk of companies to be insured, insurance carries out risk analysis, which can become very extensive because of diversity of companies (plants), grade of details and kinds of insurance. Therefore, a demand for an efficient and systematic method of risk analysis is present. State of the Art
Available methods can roughly be divided into groups of check lists, table based and model based methods. Insurance, however, mostly uses only check list methods to evaluate risk.

Shortages, Weakness, Criticism
Most known methods provide scenario based analysis (non parametric model), which makes it structurally difficult to use parts of carried out analysis for similar systems so to re-apply acquired information. Check list methods can be applied in a wide range of fields, which however leads to systematic and reproductive shortages. Tabular methods, e.g. HAZOP, provide a systematic procedure but no possibility to accumulate acquired knowledge. Boolean models, like event-tree and fault-tree analysis permit quantifying but are expendable and useful only for static examination.

Contribution
The present work demonstrates a model based methodology of risk analysis for dynamical systems, which is abstracted in a number of distinct levels. The methodology is based on the following core elements:

1. Parametric, multi-level, object oriented modeling,
2. Analysis through simulation.

The multi-level modeling approach allows to separate different system descriptions into three abstraction levels.

• Process level serves to model single undesired events (faulty processes) of a system, using e.g. differential equations.
• On t he plant level, conditions and their transitions are described through event discrete systems, e.g. by finite state machines or Petri nets (the main focus of the work). ~U
• The evaluation level is intended for the interpretation of plant states in terms of the evaluation focus and expresses corresponding losses, using e.g. neural net or Boolean model.

The separation of modeling and scenario building, as well as parametrising the models enables compact modeling based on system topology. The event trajectories (scenarios), which are necessary for the evaluation, are produced by simulation. Developed parts of the model (modules) can be stored in a library and used for further analysis. The employment of the approach of object oriented modeling contributes to a consistent model building and clear structure of the library. The plant level is implemented with timed Petri-net and provided with a library containing ten modules. The number of modules is sufficient to model two simple process engineering compositions.

The work demonstrates the practicability of the developed concepts. It is illustrated already by modeling small plants that:

1. Needed input data for classical analysis methodologies have a general structure (kind of plant, location, process description, ...) thereby for the presented methodology quantitative and component based data (plant topology and specification of components) are demanded;
2. the modeling of causal connections (presented methodology) leads to a more compact and simple representation in contrast to table based or Boolean methods;
3. the results are reproducible, comprehensive and qualitative compared to conventional table based approaches. By using Petri-net models algebraic loops are impossible;
4. the time needed to perform an analysis is with an available module library moderate low.

Extension
The method is based on an open architecture and can be extended using defined interfaces. The use of stochastic Petri-nets (timed in use) will support modeling of stochastic processes. Another extension can be seen in the development of the evaluation level, e.g. for the evaluation of business interruption.