Risk Management Methods for Industrial Systems

Risk management (RM) is today a commonly used term in many different disciplines (e.g. economy, industry, human resources, IT). Its goal is firstly to identify risks (e.g. of a machine operation, human behaviour or of a whole project) which can cause harm to properties, persons or environment. Secondly the RM should evaluate probabilities and severities of these unwanted events and propose measures for their elimination or a reduction of their impact. This whole process should be periodically repeated to assess influences of the measures of a risk control on detected risks. If the risks are mitigated insufficiently, then there should be additional measures proposed. This general interdisciplinary approach could be split into three parts or phases. This Ph.D. thesis is focused on the first phase: the risk analysis (RA). There are few commonly used traditional RA methods in the industry area e.g. Fault Tree Analysis (FTA) or Event Tree Analysis (ETA). These methods were developed many years ago, so their original definitions do not meet today’s requirements for analysis of large and complex systems or accidental scenarios with different types of dependencies, dynamic changes and other pitfalls. Different industrial areas have developed their own narrowly focused methods during last years, even though there could be, for selected tasks, used some of the mentioned common, but slightly modified methods. Based on the stated facts, this Ph.D. thesis is focused on an analysis of traditional RA methods, on a detection of their specific problems and mainly on a proposal of new alternative universal methods which are able to solve the mentioned problems. These new methods should integrate more than two phases of the RA together and they have to be practically usable. As a framework for a design of new methods, Petri nets (PN) were chosen. They are often used for a modelling and analysis of discrete event systems (DES), but they are still not common in the area of the RA. The first part of the thesis shows how advantageous it is to use general (in this case traditional) methods instead of the narrowly focused ones. As an example, a risk analysis of the Strahov city road tunnel is chosen. The goal of the analysis is to select which option of a reconstruction of the tunnel is the best from the point of view of the risk/cost ratio. A slightly modified general analysis called Probabilistic Risk Assessment, which is known from the nuclear and aerospace industry, provides superb results and supports the idea of using non-specialized methods. The second part of the thesis introduces a concept of the PN as a tool for a solution of selected problems of traditional RA methods. The non-marked PN are used for a theoretical solution of dependencies of pivotal events in the ETA. The final part of the thesis presents a proposal of a new alternative complex RA method based on stochastic PN, which is able to model the whole accidental scenario without pitfalls of the traditional FTA and ETA. It extends a classical approach with an ability to easily model dependencies and dynamic changes of an event sequence in the scenario. This method is fully usable and is supported by existing commercial software tools. This doctoral thesis demonstrates that the PN are still a little bit overlooked but powerful framework for risk analysis and management and offer new possibilities for modelling, simulation and analysis.