AAMKS - Integrated Cloud-based Application for Probabilistic Fire Risk Assessment

View the PDF here

AAMKS - Integrated Cloud-based Application for Probabilistic Fire Risk Assessment

By: Adam Krasuski, The Main School of Fire Service, Poland. Simo Hostikka, Aalto University, Finland

Introduction

Available models of fire risk assessment can still hardly be used for practical engineering problems. They are mostly loosely integrated, complex, computationally demanding applications, and require additional pre- or post-hand-calculations. However, the latest achievements in computer science enable establishing an easy-to-use web-application with access to enormous computing power in a cloud requiring minimal management effort.

In this article, we present a new tool for probabilistic fire risk assessment called Aamks. Our goal is to build an easy-to-use, science-based, practical engineering tool to support building design in day-to-day work. To meet the overall goal, we aim at working out an integrated, web-based application with computational efficiency and scalability allowed by cloud computing.

 Aamks performs a stochastic analysis of life safety in building fires using deterministic models for fire and evacuation and stochastic sampling of the uncertain input parameters.

 in day-to-day work. To meet the overall goal, we aim at an integrated, web-based application with computational efficiency and scalability guaranteed by grid computing. The integration proposed in Aamks allows for creating on-demand services feasible to calculate the computationally expensive quantitative fire risk analysis in the convenient web application.

Model Description

Aamks performs a stochastic analysis of life safety in building fires based on deterministic models for fire and evacuation, and stochastic sampling of uncertain or variable input parameters [1]⁠. The main phases of the modeling are:

1. Problem definition. The building configuration and scope details are identified and encoded into the CAD model. Then the set of parameters reflecting occupancy, building and environmental characteristics are defined.
2. Monte Carlo sampling. The input parameters for the model that are uncertain or variable are sampled from defined probability distributions.
3. Fire and evacuation simulations. The input vectors are distributed across nodes of grid architecture. Each node performs deterministic modeling of a single fire scenario defined by the Monte Carlo sampler. The evacuation simulations are performed on top of data obtained from fire modeling; that way the fire and its environment affect the evacuees.
4. Data post-processing. The model output is presented in the forms of individual as well as societal risks. Other available outputs include the distributions of available safe egress time (ASET) and required safe egress time (RSET), failure probability, hot layer height distribution, visibility, as well as maximal temperature.

The general idea of Aamks is based on the approaches proposed in [2,3] -- stochastic simulation on zone models. However, we expanded the approaches by deploying the fully-featured evacuation model. Fire and evacuation are modeled consecutively for each model implementation, allowing quantification of the safety level of people by means of a Fractional Effective Dose (FED) [4]. The tool consists of eight modules presented in Figure 1:

1. A-GUI: a web application for user input and results visualization. The workflow needs to be commenced using a CAD model of the building. Currently, Aamks allows two methods: i) AutoCAD plugin or ii) a-painter, our own editor. The user’s interaction requires only providing the CAD model of the building, setting a short list of meta parameters (e.g. the occupation type), starting the simulations and later interpreting the results. The administrator needs to install Aamks on the server and the user interacts with Aamks via a web browser, so it requires no configuration on the user’s part.
2. A-GEOM: module for performing geometry processing. The model and parameters defined in A-GUI are to be converted to an internal Aamks format which is needed for topology reasoning and other tasks.
3. A-MC: stochastic producer of input files for A-FIRE and A-EVAC. Aamks has a library of distributions for each of about dozen input parameters that describe various aspects of the scenario. Both Simple Random Sampling (SRS), as well as Latin Hypercube Sampling (LHS) methods can be used.
4. A-GRID: managing computations on the grid/cloud. Obviously, a significant computational power is needed to run simulations. Therefore, Aamks is meant to be run in a cluster or grid environment. The computers that run the simulations are called nodes or workers. 
5. A-FIRE: fire simulations models. Aamks currently uses the CFAST zone fire model for the fire development and smoke spread simulations, and Ozone for the thermal response of steel construction. These are zone models that enable quick (in the meaning of computation costs) exploration of the space of possible fire scenarios. After the fire simulation, the output files, which are generally unsuitable for massive data access queries, are processed for fast tenability information queries in the consequent stage.
6. A-EVAC: simulation of human movement and states. A fully featured evacuation model has been developed [5], which is based on navigation meshesfor global wayfinding and the Optimal Reciprocal Collision Avoidance model for collision avoidance. In addition, evacuees are affected by smoke which restricts visibility and affects their speed. Finally, the fire consequences are measured as the Fractional Effective Dose (FED)⁠⁠. In the case of a traumatic injury, the temperature inside construction elements is calculated using the OZone software. Once the temperature has achieved the critical value, the collapse of the building is assumed.
7. A-RESULTS: post-processing the results and creating the content required for reports. The module summarizes data from the deterministic fire and evacuation simulations, creates output probability distributions, and presents them as collections of charts and tables for risk assessments, such as the FN curves.
8. A-VIS: simulation animation visualization. This module enables the inspection of specific simulations. The results are animated to show the ways the evacuees move and how toxic gases affect their condition.

Figure 1. General overview of the Aamks model and workflow

Risk Model and Metrics

The risk calculated in the model is summarized in an event tree structure. There are currently three primary factors used for the calculation of consequences in Aamks: toxicity, heat and traumatic injury. The toxic injury model is based on the calculation of FED and its further ranking [5]⁠. The thermal injury model is based on the FED of convected heat accumulated per minute as defined in NFPA502⁠. The traumatic injury model is of a rather rough nature and is based on an exceeding of the temperature of construction stability with respect to evacuation time.

In the majority of simulations, the majority of consequences arise from FED. It is based on the predicted concentrations of carbon monoxide, hydrogen cyanide, hydrogen chloride, carbon dioxide, and oxygen. The FED value  equal to one is interpreted as a 50\% of chance of fatality and a high consequence (H). The response to lower or higher values of FED is translated into fatality by using log-normal probability distribution function following ISO 13571. The probability of fatality is calculated independently for all the evacuees. Next the final individual risk of death is calculated as a complement probability that nobody dies. Sublethal effects, also based on the FED, are broken down into: minor (N), low (L) and medium (M).

The risk is calculated as a share of the number of simulations that resulted in a given consequence type to the total number of simulations. It represents the annual risk of death to which specific individuals are exposed. This means the risk to a person in the vicinity of a threat, including the type of consequences and the probability of consequence occurring. The obtained values can be further evaluated using absolute values defined for example in standards PD7974-7, engineering knowledge [6] or set of approaches for relative methods presented for example in [1]⁠.

The individual risk does not, however, carry any information about the number of people affected in the case of a fire. This aspect is addressed by the so-called societal risk⁠⁠. The societal risk is a measure of risk to a group of people. It is most commonly expressed with respect to the frequency distribution of multiple casualty events. The notion of risk in a societal context is expressed as the relation between frequency and the number of people suffering from a specified level of harm as a result of implementation of specified threats. The societal risk may be modeled by the frequency of exceedance curve of the number of deaths, also called the FN curve due to specific threats. Thanks to A-EVAC multi-agent microscopic simulator, it is possible to calculate the FED exposure for each of the evacuees and then rank the severity of the consequences per the number of people affected. This finally results in FN risk curves.

Deterministic Models

The current fire model in Aamks is CFAST version 7.5.1, which provides the fire environment parameters required for the calculation of toxic and thermal FED, as well as movement speed reduction. OZONE is applied for calculating the thermal response of steel construction and its consequences for humans. In practical terms, OZONE calculates whether the critical temperature was reached and the time when this temperature was reached.

A-EVAC handles the collision avoidance based on the velocity time-to-collision approaches and linear programming [7]⁠. The wayfinding algorithm is based on the navmesh approach~\cite{navmesh}⁠. For each position of the agent in each step of the simulations, A-EVAC enquires the created data structure for fire parameters and alters agents’ states (FEDs, speed) respectively and behavior correspondingly. When there is smoke on the evacuation route, the agents try to find an alternative route free of smoke. The behavior related to smoke depends on the agent type: conservative, active, herding and followers as defined in [8]⁠. The detailed description of A-EVAC can be found in [5]⁠.

Model Parameters

The simulation process starts with the definition of a model of the building where the fire and evacuation take place. The model reflects the compartmentalization and other type of obstacles that can be present inside a building, as well as openings, deployment of safety measures (fire detectors, sprinklers) and others. These parameters are created using special drawing tools.

The model does not change during the simulation process (at least for a given candidate design), hence it can be considered as a set of a fixed set of parameters defining input. This set is then expanded by other invariants related to the physical properties of the building obstacles, environmental parameters (initial indoor temperature, humidity, and others), and physical parameters of safety measures, such as sprinklers spray density and ventilation flow. The number of parameters as well as their type depends on the problem being addressed, building type and other factors.

The second set of input parameters is uncertain or variable and are drawn from the Monte Carlo sampler. The parameters that are sampled include, but are not limited to: a room of fire origin, heat release rate per unit area (HRRPUA), soot yield, CO yield, locations of people in the building during the pre-evacuation stage, door positions, and the operation of a safety system. The parameters of the probability distributions are mostly based on standards such as BSI 9999:2008 and PD7974-6:2019⁠⁠ or the many scientific records.

The third type of parameter comprises dependent variables. These variables are related to the fixed values or those drawn from the distributions.  The heat release rate (HRR)  may serve as an example of the dependent variable. The HRR is defined as a function of a drawn sample of HRRPUA and the fire area that depends on the room of the fire origin. An effect the peak HRR is defined as the product of the HRRPUA and the area of the room of the fire origin.

Availability and Collaboration

The Aamks software is open source, available at http://github.com/aamks. There is also a demo version available at the project webpage https://aamks.szach.in. The demo version allows preparing a project for simulation, but does not allow launch computation. There are also a number of videos available on the project webpage that present the use of Aamks in the IMO test, compared to other software or just presenting various aspects of software development.

We are open to cooperating both with programmers, fire safety engineers, as well as beta testers. We are particularly expecting assistance in the scope of human behavior in smoke with relation to zone fire modeling, as well as model validation, i.e. the validity of output risks.

Case Study

As a case study, we investigate a building housing a concert hall and surrounding corridors and offices. The main hall is 60 m wide, 34 m deep and 6.5 m high. The main entrance to the hall leads through a cloakroom which is 4.16 m wide, 37.4 m deep, and 3.5 m high. The layout of the model is shown in Figure~\ref{layout}. The people in the concert hall were distributed according to the available seats, and in the remaining rooms placed randomly, considering the expected occupation density BSI 9999:2008. The total number of evacuees was 700.

We applied Aamks to calculate individual and societal risks for this building. Thanks to the zone fire model, the velocity-based model for evacuation, and grid computing, the simulations for fire safety concepts can be performed approximately within 30 min on 160 cores @ 2.4 GHz.

Conclusions and Future Development 

The primary goal of the discussed case study was  to present the main idea of the risk-aware decision making, as well as the easy application of Aamks in the building design process. Therefore, the building layout was rather simple and informative. We also provided the cost of computation of a single scenario for this case study. Taking into account the time required for the computation of a  single scenario, one may project the time of computation of the entire project with respect to the number of nodes being available. For example, having one computer with four cores, the expected time of computation of 1000 simulations will be approximately 14 hours. The accuracy for this number of simulations will be of a magnitude of 1e-5. A better accuracy requires more simulations with error reduction rate proportional to the square root of N. However, the increase in the number of simulations also increases the time of computation (with hardware resources unchanged). The calculation of the simulations on grid computing allows for higher scalability and performance. As was presented in the case study only 30 minutes were needed to calculate 1000 simulations.

However, the application of grid computing requires high upfront costs related to outlays to be made in the hardware. Moreover, the setup of a grid or cluster is not always easy and may require support from IT experts. Therefore, we also tested Aamks at Microsoft Azure cloud. We applied one of the Azure solutions called ScaleSet, which allows a dynamic increase in the number of nodes depending on CPU usage of those currently existing in ScaleSet. We set up a project of a 3-storey academic building with 1000 simulations. The computation resulted in a dynamic assignment up to 200 cores and total simulation time of several minutes. The monthly cost of such a solution is at the level of magnitude of €5000. The deployment of Aamks on a commercial cloud has a number of advantages, including among others: a) no necessity of maintenance, b) dynamic scalability, c) dynamic fitting of available resources to current computations needs, d) continuous update of software and hardware, e) lack of upfront costs, f) elimination of downtime, g) no necessity of keeping IT experts for the hardware and software management, and finally h)  better energy management and lower CO₂ footprint. Nevertheless there are also certain shortcomings related to keeping calculations in the cloud. The most important are costs of calculations in the cloud and the need of becoming familiar with the cloud-specific software. Therefore, the final cost-benefit result depends on the business model of the company, i.e. the number of project, including also their simultaneity, the complexity of the projects, as well as the currently available computer resources and staff.

In the design of our software, we emphasised the ease-of-use and the feasibility aspects. Given to the fact that the majority of input parameters are drawn from probability distributions, the definition of the simulation process is very easy. We are also preparing our own editor for drawing, because we have found that AutoCAD has proven to be too difficult for a considerable part of the users. Therefore, at the moment we are convinced that our software is crafted for engineering purposes. Moreover, the development of the software as a web application releases the users from problems connected with installation and configuration. All the user needs to do is create an account and then is free to use the application. Moreover, several minutes of calculation time per projects on grid or cloud computing may allow the presumption that the tool can be used in the day-to-day process of building design, in which various fire safety concepts can be considered regarding the safety level.

We have also provided a brief outline of how the calculated risk can be further ranked and reported to parties involved in the project design. In the case study, we have adopted the SFPE risk-ranking matrix and PD 7974-7 standard. We also mentioned a more advanced method of risk-informed decision making.

References

[1] Krasuski A. (2019) Multisimulation: Stochastic simulations for the assessment of building fire safety. The Main School of Fire Service

[2] Hostikka S, Keski-Rahkonen O. (2003) Probabilistic simulation of fire scenarios. Nuclear Engineering and Design ;224(3):301–311.

[3] Wade CA, Baker GB, Frank KM, Robbins AP, Harrison R, Spearpoint M, et al. (2016), B-RISK user guide and technical manual . BRANZ

[4] Purser DA. (2002) Toxicity assessment of combustion products. SFPE handbook of fire protection engineering;3:2–6

[5] Krasuski A, Kreński K. (2019) A-Evac: The Evacuation Simulator for Stochastic Environment. Fire Technology ;55(5):1707–1732.

[6] Hurley MJ, Gottuk D, Jr JMWRH, Harada K, Kuligowski E, Puchovsky M, et al. (2016) SFPE handbook of fire protection engineering, fifth edition. Springer.

[7] Van Den Berg J, Lin MC, Manocha D. (2008) Reciprocal velocity obstacles for real-time multi-agent navigation. In: IEEE International Conference on Robotics and Automation, ICRA 2008. IEEE; p. 1928–1935

[8] Korhonen T, Hostikka S. (2010) Fire Dynamics Simulator with Evacuation: FDS+Evac - Technical Reference and User’s Guide (FDS 5.5.0, Evac 2.2.1)