In fire, any information about the actual condition within the building could be essential for quick and safe response of both fire–fighters and occupants. In most cases, however, the emergency responders will rarely be aware of the actual conditions within a building and they will have to make critical decisions based on limited information. Recent buildings are equipped with numbers of sensors which may potentially contain useful information about the fire; however, most buildings do not have capability of exploiting these sensors to provide any useful information beyond the initial stage of warning about the possible existence of a fire. A sensor–linked modelling tool for live prediction of uncontrolled compartment fires, K– CRISP, has therefore been developed. The modelling strategy is an extension of the Monte– Carlo fire model, CRISP, linking simulations to sensor inputs which controls evolution of the parametric space in which new scenarios are generated, thereby representing real–time “learning” about the fire. CRISP itself is based on a zone model representation of the fire, with linked capabilities for egress modelling and failure prediction for structural members, thus providing a major advantage over more detailed approaches in terms of flexibility and practicality, though with the conventional limitations of zone models. Large numbers of scenarios are required, but computational demands are mitigated to some extent by various procedures to limit the parameters which need to be varied. HPC (high performance computing) resources are exploited in “urgent computing” mode. K–CRISP was demonstrated in conjunction with measurements obtained from two sets of full–scale fire experiments. In one case, model execution was performed live. The thesis further investigates the predictive capability of the model by running it in pseudo real–time. The approach adopted for steering is shown to be effective in directing the evolution of the fire parameters, thereby driving the fire predictions towards the measurements. Moreover, the availability of probabilistic information in the output assists in providing potential end users with an indication of the likelihood of various hazard scenarios. The best forecasts are those for the immediate future, or for relatively simple fires, with progressively less confidence at longer lead times and in more complex scenarios. Given the uncertainties in real fire development the benefits of more detailed model representations may be marginal and the system developed thus far is considered to be an appropriate engineering approach to the problem, providing information of potential benefit in emergency response. Thus, the sensor–linked model proved to be capable of forecasting the fire development super–real– time and it was also able to predict critical events such as flashover and structural collapse. Finally, the prediction results are assessed and the limitations of the model were further discussed. This enabled careful assessment of how the model should be applied, what sensors are required, and how reliable the model can be, etc.

Fire Safety Engineering is required at every stage in the life cycle of modern-day buildings. Fire safety design, detection and suppression, and emergency response are all vital components of Structural Fire Safety but are usually perceived as independent issues. Sensor deployment and exploitation is now common place in modern buildings for means such as temperature, air quality and security management. Despite the potential wealth of information these sensors could afford fire fighters, the design of sensor networks within buildings is entirely detached from procedures associated to emergency management. The experiences of Dalmarnock Fire Test Two showed that streams of raw data emerging from sensors lead to a rapid information overload and do little to improve the understanding of the complex phenomenon and likely future events during a real fire. Despite current sensor technology in other fields being far more advanced than that of fire, there is no justification for more complex and expensive sensors in this context. In isolation therefore, sensors are not sufficient to aid emergency response. Fire modelling follows a similar path. Two studies of Dalmarnock Fire Test One demonstrate clearly the current state of the art of fire modelling. A Priori studies by Rein et al. 2009 showed that blind prediction of the evolution of a compartment fire is currently beyond the state of the art of fire modelling practice. A Posteriori studies by Jahn et al. 2007 demonstrated that even with the provision of large quantities of sensor data, video footage, and prior knowledge of the fire; producing a CFD reconstruction was an incredibly difficult, laborious, intuitive and repetitive task. Fire fighting is therefore left as an isolated activity that does not benefit from sensor data or the potential of modelling the event. In isolation sensors and fire modelling are found lacking. Together though they appear to form the perfect compliment. Sensors provide a plethora of information which lacks interpretation. Models provide a method of interpretation but lack the necessary information to make this output robust. Thus a mechanism to achieve accurate, timely predictions by means of theoretical models steered by continuous calibration against sensor measurements is proposed. Issues of accuracy aside, these models demand heavy resources and computational time periods that are far greater than the time associated with the processes being simulated. To be of use to emergency responders, the output would need to be produced faster than the event itself with lead time to enable planning of an intervention strategy. Therefore in isolation, model output is not robust or fast enough to be implemented in an emergency response scenario. The concept of super-real time predictions steered by measurements is studied in the simple yet meaningful scenario of concurrent flow flame spread. Experiments have been conducted with PMMA slabs to feed sensor data into a simple analytical model. Numerous sensing techniques have been adapted to feed a simple algebraic expression from the literature linking flame spread, flame characteristics and pyrolysis evolution in order to model upward flame spread. The measurements are continuously fed to the computations so that projections of the flame spread velocity and flame characteristics can be established at each instant in time, ahead of the real flame. It was observed that as the input parameters in the analytical models were optimised to the scenario, rapid convergence between the evolving experiment and the predictions was attained.