Abstract
Heat release rate (HRR) is the principal fire characteristic of materials. There are three known methods for the measurement of HRR (based on oxygen consumption, mass loss rate, and combustion products temperature rise). The method based on oxygen consumption is considered to be the reference. However, this method is expensive and for a large part of laboratories and universities unavailable. The simplest method is based on combustion products’ temperature rise. However, this method has a fundamental problem with the temperature dependence of the heat capacity of combustion products and the thermal inertia of the measurement system. This problem has been solved by training neural networks to predict molar heat capacity and the amount of substance (chemical amount) flow rate of combustion products in the cone calorimeter exhaust duct. Data were obtained for six different wood species: birch (Betula verrucosa Ehrh.), oak (Quercus robur L.) spruce (Picea abies (L.) H. Karst.), locust (Robinia pseudoacacia L.), poplar (Populus nigra × P. maximowiczii L.), and willow (Salix alba L.) woods at heat fluxes from 25 to 50 kW m−2 have been used for neural network training. Data from three other wood species iroko (Milicia excelsa (Welw.) C.C. Berg), pine (Pinus sylvestris L.), and paulownia (Paulownia tomentosa (Thunb.) Steud.) woods have been used for testing of trained neural network. The average percentage ratio of the predicted to the true value of HRR (during the test) has been 103.8%. In addition to that, some key average fire characteristics of wood have been determined: critical heat flux 20.7 kW m−2, effective heat of combustion 14.01 MJ kg−1, and the average value of molar heat capacity of combustion products 0.045 kJ mol−1 K−1.