Use of artificial neural networks in the determination of hydrogen/oxygen laminar diffusion flames at high-pressures

In industrial applications, instant knowledge of flame properties in a flame region is often necessary. This research demonstrates successful use of artificial neural networks in real-time determination of flame properties of high-pressure hydrogen/oxygen laminar diffusion flames. Three back-propagation artificial neural networks (BPNs) are established to compute the distributions of temperature, axial velocity and mixture fraction, respectively, in hydrogen/oxygen diffusion flames at pressures between 25 ∼ 68 atms. The three back-propagation artificial neural networks are trained with calculated results of a theoretical model. Results show that the three BPNs successfully predict the flame properties. The average prediction error of each of the three BPNs is within 5.4%. This study also investigates effects of various network parameters, including number of training sets, number of hidden layers, transfer function of hidden nodes, initial values, etc. on the performance of the neural networks.