Estimation of Heat Release Rate and Fuel Type of Circular Pool Fires Using Inverse Modelling Based on Image Recognition Technique.

A computational model is developed to estimate the heat release rate and fuel type of circular pool fires by analyzing the videos with recognizable flames. The model can be used in both static and mobile platforms. The pool diameter, mean flame height, heat release rate and fuel type are estimated using image recognition technique on flame videos and inverse modelling with traditional fire dynamics theories. A set of experimental videos from different sources are used to validate the model. During image recognition, the model isolates the flame from the non-flame elements in each frame using the "automatic seed placement and region growing" method. The method is found to be effective in non-flame elements removal and improves the flexibility of the model to wide range of flame videos. To automatically determine the pool diameter, fast Fourier transform (FFT) is involved to identify the flame pulsation frequency that is used as the input of inverse modelling. It is found that a sampling duration of 10 s to 20 s gives the most reliable predictions to the flame pulsation frequency for the current set of videos. A shorter duration is not sufficient for FFT to recognize the correct main frequency of the signal while a longer duration increases the low frequency components caused by the unsteady flame. Compared to the traditional fire dynamics theories with power indexes of input variables less than 1, the inverse modelling enlarges the error in the modelling results. Therefore, the main weakness of current model is perhaps the enlarged uncertainty led by the inverse modelling conducted with the traditional theories which are empirical, although the current predictions to the experiments are acceptable. Moreover, the current model divides the pool fires into three categories and based on the predicted pool diameter and heat release rate the fuel type can be estimated, which might benefit the hazard analysis in certain circumstances. Finally, a cross-platform comparison shows that the mobile devices can be considered for fire applications although it is still less powerful than personal computers. [ABSTRACT FROM AUTHOR]