FPEExtraIssue23

A Step Towards Real-Time Fire Monitoring

By Paul A. Beata, PhD, Ann E. Jeffers, PhD, and Vineet R. Kamat, PhD

During a fire event, environmental threats to building occupants and first responders include extreme temperatures, toxic gases, disorientation due to poor visibility coupled with unfamiliar surroundings, and a changing indoor environment. In addition to these hazards, firefighters often lack critical information for making decisions on the ground. Such a lack of information, coupled with the dynamics of natural fire events, leads to a number of near-misses, injuries, and deaths each year. These challenges also slow the rescue time for building occupants and prolong the progression of fire.

Integrating real-time measurements from sensors into a fire intervention strategy may provide an opportunity for a new technological advancement to improve the practice of firefighting. Researchers at the University of Michigan have developed a computational framework based on Lightweight Communications and Marshalling (LCM)¹ for connecting real-time fire data to an event detection sub-model to demonstrate the possibility that real-time computing can be used for sensor-assisted firefighting. See Figure 1 for an overview of the proposed system.

Specifically, LCM is a message-passing library for providing communication between applications through consistent data structures (e.g., to send an array of data from a C++ application to a Python one). In the context of fire monitoring, this tool allows multiple sub-models and data sources to be linked through a common interface.

Figure 1: Overview of the components for a real-time fire monitoring system integrating sensor measurements with simulation and visualization.

One primary challenge in this line of research and development, which is also unique when compared with other monitoring applications, is that too much data and information without careful assimilation and presentation will be counterproductive as an aid to firefighting. Cowlard, et al.² have provided, in addition to the benefits, several cautionary warnings for using sensors, real-time data, modeling, and visualization as tools for helping firefighters. As a result, the computing platform presented here aims to minimize interpretation of data and results in the field, opting for simpler visual warnings instead.

In the current stage of development, the fire monitoring tool has two steps: (1) a real-time component for evaluating the fire scenario based on sensor measures and (2) a playback feature to demonstrate the capability for visualization. The real-time monitoring system includes a main node for coordinating data to sub-models in the system; an event detection model was the first sub-model (Figure 2). Building Information Modeling (BIM) software provides visualization, in particular, the AECOsim Building Designer product from Bentley Systems. The visual warnings were represented in the BIM environment using schedule simulation by automatically generating XML-based schedules from the event-detection model.

Figure 2: Component-level view of the LCM-based fire monitoring system.

In the absence of dedicated sensor technology for realizing the implementation of the fire monitoring system, an n-sensor model was developed to simulate the asynchronous measurement and transmission of data from a fire source to the computing system. The fire signatures used in this development are the gas temperature, heat flux, and four species concentrations (CO, CO₂, HCN, and O₂). This used LCM to publish the newly (simulated) measured data on unique LCM channels; receipt of the data in the main program is achieved by subscribing to each unique channel. This constitutes the publish/subscribe model for message-passing between applications.

In the main program, subsets of the measured fire signatures are properly coordinated to the appropriate sub-models in a standardized way via LCM once again, providing flexibility for including future sub-models. The event detection model in Figure 2 was developed to evaluate a fire based on the time series of measurements from sensors. In particular, the model assessed three conditions: (1) smoke toxicity, (2) burn threats, and (3) fire status. Smoke and burn hazards were evaluated on the basis of Fractional Effective Dose (FED) [3] calculations in real time on a per-sensor basis. The fire status hazard aims to provide an early warning for flashover by tracking temperatures and heat fluxes.

Researchers developed a test with four rooms to visualize the results of the event-detection model. The fire-monitoring system operates on measured fire data in real time while the visualization step is used in post-processing as a demonstration for future real-time implementation. The event-detection model writes schedule information in XML format to an output file that is compatible with AECOsim Building Designer.

Scheduling graphical warnings in the BIM environment was based on marking the start and finish times of the various threat levels through the three hazards (smoke, burns, and fire status). Since any of these hazards could exist simultaneously, the program has built-in logic to synchronize combinations of each through colors at the location of the sensor. Figure 3 shows the progression of the hazards, assuming one sensor per room in the four-room model.

Figure 3: Demonstration of the monitoring system visualization in AECOsim Building Designer (right) with the corresponding fire progression shown in Fire Dynamics Simulator (left) for comparison.

From a hardware standpoint, the challenges of sensor measurement in extreme conditions such as fire are known and an expected obstacle to real-time fire monitoring. For example, Silvani, et al.⁴ discuss the issues associated with data losses and time delays using a wireless sensor network (WSN) in forest fire monitoring. However, WSN can be designed and implemented with a resilience property to reroute information around avoid damaged sensors. Such a feature would also be necessary in developing future sensor technology to help mitigate these potential hardware problems in the building setting.

While questions remain about the feasibility of tracking fire signatures in real time for sensor-assisted firefighting purposes, advances in the computing aspects should help motivate the development of additional technology in this area. This work serves as a step towards an intelligent firefighting system based on computing to provide real-time tools for effective decision-making during a fire event.

The authors are with the Department of Civil and Environmental Engineering at the University of Michigan.

Note: This research was part of the Chief Donald J. Burns Memorial Research Grant program through the SFPE Foundation. The purpose of the Burns grant is to use information modeling as a means of improving infrastructure safety and fire service preparedness. The grant is named in memory of FDNY Assistant Chief Donald Burns, who died in the collapse of the World Trade Center Towers on September 11, 2001, while setting up his command post to direct the evacuation. Bentley Systems Incorporated, a global leader in providing architects, engineers, geospatial professionals, constructors, and owner-operators with comprehensive software solutions for sustaining infrastructure, provided funding for the Burns grant.

References
¹ Huang AS, Olson E, Moore DC (2010). LCM: Lightweight communications and marshalling. In: Intelligent Robotic Systems. (IROS), 2010 IEEE/RSJ Int. Conf., pp. 4,057–4,062.

² Cowlard A, Jahn W, Abecassis-Empis C, et al. (2010). Sensor-assisted Firefighting. Fire Technol 46:719–741. doi: 10.1007/s10694-008-0069-1.

³ Purser DA, McAllister JL (2016). Assessment of Hazards to Occupants from Smoke, Toxic Gases, and Heat. In: Hurley MJ, Gottuk DT, Hall Jr. JR, et al. (eds), SFPE Handbook of Fire Protection Engingeering. Springer New York, New York, NY, pp. 2,308–2,428

⁴ Silvani X, Morandini F, Innocenti E, Peres S (2015). Evaluation of a Wireless Sensor Network with Low Cost and Low Energy Consumption for Fire Detection and Monitoring. Fire Technol 51:971–993. doi: 10.1007/s10694-014-0439-9.