Study of the Flow Generated by a Sliding Discharge by Means of a Convolutional Neural Network
A quantitative study has been made of the flow with shock waves generated in air by a sliding surface discharge with a duration of less than one microsecond. The flow was visualized using the shadowgraph method, the process was recorded at a rate of 124,000 frames/s, the exposure time was 1 ?s. The aim of this work is to study the dynamics of a cylindrical blast wave generated during a discharge and a region bounded by a contact surface. Each experiment allowed several hundred images to be taken of a transient gas-dynamic process with a duration of up to 1 ms. A YOLOv8 convolutional neural network was trained and used to determine the positions of the discontinuities. A data set with 984 markups was marked. The model on the mAP50 metric achieved 0.887 and the mAP50-95 was 0.557. The model was used to automatically measure the vertical dimensions of the contact discontinuity. It expands at times of up to 0.4 - 0.8 ms to a vertical size of 5 - 11 mm. The x-t plots and the velocities of the cylindrical shock waves were measured. It is shown that at t < 1 ms the main reason for the development of the flow is the blast wind behind the blast wave. It is shown that the use of computer vision can significantly speed up the analysis of high-speed visualizations and the extraction of quantitative information.