Analysis of evacuation performance of early stage ship designs using a Markov-Decision-Process model

Abstract

This thesis research is conducted on behalf of the Delft University of Technology and Bureau Veritas Rotterdam BV, and the goal is to improve ship evacuation by analyzing this aspect early on in the design process. First, this summary will provide a background on the difficulties involved with the ship evacuation process . Next, an explanation is given why this thesis proposes a different model. And lastly, how this model is validated is elaborated upon.

A ship evacuation is a highly complex process and it is one of the most important aspects concerning the safety of crew and passengers on-board. This process is affected by various factors, largely by the general arrangement. The layout of a ship varies throughout the design period but the most design freedom occurs during the early stage. An estimated 90% of the decisions which have major impact on final design have been made in this stage. As a result, this stage holds the most potential to improve the ship evacuation performance. This stage is characterized by a lack of detailed layout information available, which results in reduced accuracy in evacuation prediction. As of 1st of July 2020 all passenger ships with more than 36 passengers have to perform an evacuation analysis based on guidelines created by the International Maritime Organization (IMO). The guidelines describe two methods, a simplified method for the early design stage, and an advanced method for the detailed or final design stage. The simplified method translates the ship design to a hydraulic network which in turn results in various method limitations. The goal of an evacuation analysis is, among other things, to provide the designer with information about possible congestion points, identify areas of counter- and cross flows, prove that the escape arrangements are sufficiently flexible, and prove that the evacuation time does not exceed a threshold.

Evacuation models capable of performing an evacuation analysis tend to have four general components that define the various aspects influencing the process. The four components are configuration, environmental, procedural, and behavior component. Including evacuee behavior in an evacuation model tends to be difficult due to the complexity and lack of research available. Evacuation models can be categorized using various descriptors however resolution tends to dominate. High resolution models are called macroscopic models whereas low resolution models are called microscopic models. Microscopic models simulate each individual evacuee as an agent with certain attributes, such as walking speed and destination. Most early stage evacuation analysis are macroscopic models and simulate the population as a whole. A different macroscopic approach is proposed by Kana and Singer, which uses probability theory to incorporate the stochastic nature of an evacuation process. This method consists of a Markov-Decision-Process (MDP) to calculate the most probable evacuation routes. Policies are determined for each small geometric area, called a state, in the layout. These policies affect evacuation routes. The evacuation process can then be simulated by using the MDP policies and an initial population distribution. The initial population distributions are defined by the Fire Safety Systems Code. Design variations can be compared based on the calculated optimal routes, the time it takes for the process to converge, and how the population is distributed among the exits. A lack of literature is noticed that investigates the relationship between early stage ship design and the ship evacuation performance. Two different studies were found which can be used to validate the MDP methodology. The first validation study determines which exit configuration is most optimal from an egress perspective. Kurdi et al. defined four different layouts for two population sizes. Having exits on all sides was determined to be the most efficient layout. The MDP model consisted of the same setup as the validation study. The MDP method is able to identify the same exit configuration as most efficient based on the population distributed among the exits and simulation convergence times. The second validation study evaluates two different frigate layouts with varying passageway configuration. One layout has a single passageway throughout the ship. The other layout has a parallel passageway configuration. Casarosa applied a microscopic model to calculate a range of performance metrics for 7 different scenarios. All evacuees were modeled as singular agents with different behavior and objectives. 3 out of 7 scenarios were emergency scenarios, where only 1 scenario corresponds with the MDP assumptions. Discrepancy between models is considered as a result of limited, and confidential, layout information used by Casarosa. The MDP method was able to differentiate between the two layout configurations and can identify critical areas in the layout. However, due to the assumptions, it cannot be determined to what extend the method inputs were similar. Nevertheless, both validation studies showed the potential of using a Markov-Decision-Process based method to evaluate early stage ship designs on evacuation performance.