Measurement and modelling of the fire behaviour of composite laminates

Abstract

This thesis describes the measurement and modelling of the fire reaction and resistance properties of fibre glass polyester and vinyl ester composite laminates. A new small-scale loading frame was designed and fabricated to apply tensile and compressive stresses in a cone calorimeter under a constant heat flux of 75 kW/m2. It was found, for the first time, that stress has a small but significant effect on the fire reaction properties. Increasing tensile stress increased heat release rate and smoke production, while shortening the time-to-ignition. Compressive stress had the reverse effect. This was attributed to the fact that tensile stress promotes the formation of matrix microcracks, facilitating the evolution of flammable volatiles. This hypothesis is further supported by the observation that stress has the greatest effect on the early heat and smoke release peaks, with a lower effect on the final `run-out' values. Another new small scale testing rig was developed along with a calibrated conical radiant heater, which resembles the cone calorimeter's heater, to allow testing composite laminates at high tensile and compressive stresses. Stress rupture (time-to-failure) curves were produced. It was shown that, in tension, the behaviour was fibre dominated, with failure times roughly ten times those in compression. Compressive failure involved resin dominated local fibre kinking initiated near the cold face and was controlled by the matrix glass transition temperature. The existing thermal model was upgraded to include the mass flux effect into the net total energy on the hot face of the composite laminate at flashover. This novel approach has resulted in very good predictions of the most important fire reaction properties obtained from the cone calorimeter such as time-to-ignition, mass loss rate, heat release rate and flame-out time. In addition, the predicted temperature and resin content profiles combined with the temperature dependent properties values were used to model the failure times by applying a suitable empirical relationship. Good agreement was found.