A TDLAS Tomographic Forecasting-and-Reconstructing Network for Predictive Temperature Imaging

Tunable Diode Laser Absorption Spectroscopy (TDLAS) tomography is an important optical non-invasive combustion diagnostic technique, which can image two-dimensional cross-sectional distributions of critical flow-field parameters in the combustion field. Most existing TDLAS tomographic temperature imaging methods reconstruct temperature distributions in the current or past time slots, which may indicate delay in the in situ combustion diagnostic. In this work, a forecasting-and-reconstructing network is established to realize predictive TDLAS tomographic temperature imaging, which is applicable for pre-diagnosis in the continuous combustion process. This network consists of two sub-networks. One sub-network forecasts TDLAS measurements in the next time slot based on measurements obtained in the current and past time slots, while the other sub-network reconstructs temperature distribution in the next time slot from the predicted TDLAS measurements. Simulations were conducted with data generated by the flame dynamics simulator, and the prediction accuracy of the proposed network was evaluated quantitatively and visually.