predictions be verified in large-scale experiments. For example, CFAST model predictions have been validated with experimental data from the Navy's fire research platform ex-U.S.S. Shadwell (Williams and Carhart, 1992).
There are a number of field models available (Yang et al., 1984; Kou et al., 1986). A two-dimensional finite-difference field model of aircraft fires was developed to predict the movement of hot gases and smoke, as well as temperature and smoke concentration levels in the seating area of an aircraft cabin (Yang et al., 1984; Kou et al., 1986). Additional work included the development of a two-dimensional model of transient cooling by natural convection (Nicolette et al., 1985). This model utilized a fully transient semi-implicit upwind-differencing scheme with a global pressure correction. Baum and Rehm (1978, 1982a, b, 1984) have developed several field models for prediction of fires using time-dependent inviscid Boussinesq equations to simulate three-dimensional buoyant convection and smoke aerosol coagulation.
There have been several field modeling projects to develop a computer model as a low-cost alternative to predict the spread of fire and smoke in enclosed spaces on naval vessels (Nies, 1986; Raycraft, 1987; Hauck, 1988). The similarity of the enclosed spaces of naval vessels to an aircraft interior makes these types of models valuable in evaluating the effectiveness of suppression systems and new designs in the prevention and control of fires in aircraft.
Field modeling requires a large, fast computer with significantly more memory than is required in zone modeling. The accuracy of the solution depends on reducing the size of the control volumes, thus increasing the number of individual cells and the computing expense.
Fire-hazard assessment models include both zone and field models for compartment fires, with submodels for fire endurance, activation of thermal detectors or sprinkler systems, generation of toxic gases, evacuation, and survival models.
One of the early room fire models, HARVARD V, was developed in the early 1980s (Emmons, 1981; Mitler and Emmons, 1981). With this model, the user specifies room characteristics, technical information on objects contained in the room, and where the fire starts. The program calculates the fire growth, fire plume, the accumulated hot layer at the ceiling, and the outflow of hot gases and inflow of air after the smoke layer reaches the soffit. The program also calculates the radiation from the flame and hot layer to all of the objects. As each object reaches ignition temperature, the program ignites a new fire and new plume and models the additional hot gas going to the upper layer. As the flames and hot layer grow, radiation to the burning objects controls the rate of fire growth over their surface. When all the objects ignite, flashover has occurred. As the hot layer descends and envelops a burning object, the program calculates the reduction of airflow and the fire slows down. The program keeps track of the total mass and indicates when the fire of each object goes out. It also calculates the time at which a smoke detector in the room will sound its alarm.
Two fire-hazard assessment models currently in use are HAZARD 1 (Bukowski et al., 1991) and FPEtool (Nelson, 1990). According to the authors, HAZARD 1 will calculate:
the production of energy and mass (smoke and gases) by one or more burning objects in one room, based on small-or large-scale measurements;
the buoyancy-driven ventilation, as well as forced flow, of this energy and mass through a series of user-specified rooms and connections (doors, windows, cracks, holes in ceiling or floor);
the resulting temperatures, smoke optical densities, and gas concentrations after accounting for heat transfer to surfaces and dilution by mixing with clean air;
the evacuation of a user-specified set of occupants accounting for delays in notification, decision making, behavioral interactions, and inherent capabilities; and
the impact of the exposure of these occupants to the predicted room environments as they move through the building; and the time, location, and cause of each incapacitation or fatality.
This model requires detailed knowledge of the fire scenario, including the geometry of the room(s), the location of the items that are burning, the combustion properties of those items, and also information on the exposed occupants (i.e., their initial location and characteristics such as age, and whether they have any disabilities or small children to assist). Therefore, the users of this model need to be familiar with fire physics and understand the limitations of the model.
FPEtool is also a hazard assessment computer model (Nelson, 1990), but is less mathematically rigorous than HAZARD 1 and therefore takes less time to run. It will estimate ignition of exposed objects, smoke flows, gas concentrations and toxicity, pressures on a door from the fire and wind, actuation of detectors and sprinklers, and egress time of occupants.
Table C-1 describes some limits inherent to HAZARD 1, FPEtool, and HARVARD V and which probably bracket the majority of limitations inherent within two-zone models. However, there is an incomplete understanding of the physical phenomenon involved in fires. Each limitation results in a computer code that may deviate from the correct representation of the fire physics that could introduce errors into