Convolutional long short-term memory-based feature fusion modeling for reconstructing schlieren images of the supersonic combustor

A multi-input feature fusion model based on the convolutional long short-term memory network is proposed for reconstructing shock wave features in the combustion chamber flow field of a Supersonic Combustion Ramjet. The training dataset includes multiple sets of flow field images and corresponding wall pressure data, all obtained from wind tunnel experiments. An attention-based adaptive loss (ABAL) algorithm was introduced to adaptively balance image and pressure features enhancing the effectiveness of multimodal representation. The model incorporating the ABAL algorithm achieves an average R of 0.9724, peak signal-to-noise ratio of 31.10 dB, and structural similarity index of 0.9447 demonstrating high reconstruction fidelity and strong structural consistency in the predicted flow fields.