Utilizing a neural network, individual down-axis images can be classified by the number of detonation waves present and their directional behavior in an experimental Rotating Detonation Combustor (RDC).  While the ability to identify the number of waves present might be intuitive, the additional classification of wave direction and multi-wave modal behavior is a result of the wave profile reflecting its angular direction of movement.  In this study, a supervised approach is employed where a series of manually classified images is provided to the network for the purpose of training.  These images, referred to as the training set, are individually labeled as one of the eight modes present in an experimental RDC.  Possible classifications include clockwise and counterclockwise variants of wave quantities ranging from one to three, bi-directional multi-modal (independent vs concurrent), and an unsteady or deflagrative mode.  After training the network, a second set of manually classified images, referred to as the validation set, is used to estimate the model’s performance.  The convergence of the network can be assessed for a variety of image preprocessing and collection methods, including the effects of image filtering and framerate.  Direct high-speed images of the detonation waves were collected down-axis of the RDC using a Photron FASTCAM SA-Z at a framerates of 40,000, 50,000 and 60,000 fps.  Both unfiltered and filtered (to reduce background noise) versions of the images captured at each framerate are used as the training sets for independent neural networks.  The performance of each network is compared to suggest a preferred framerate as well as indicate the value of preprocessing efforts.  The ability to predict detonation wave modes using a trained neural network substantially reduces computational complexity by circumnavigating the need to evaluate the temporal behavior of individual pixels throughout time.  The application of deep learning is highly adaptive and therefore can be trained for a variety of image collection methods across RDC study platforms.  The identification of wave behavior using image classification serves as a stepping stone for further machine learning integration in RDC research.


Application of a Convolutional Neural Network for Wave Mode Identification in a Rotating Detonation Combustor Using High-Speed Imaging

Division / Sponsor:
International Gas Turbine Institute (IGTI)

Paper Proceedings: https://asmedigitalcollection.asme.org/GT/GT2020/volume/84058

Turbo Expo - Virtual Conference

Supporting the need among researchers worldwide for additional online and multimedia content, ASME now offers the ASME Conference Video Collection. This platform will collect and provide long-term access to conference presentation videos, encompassing all technology areas within ASME's technical conference and events program. About 30,000-40,000 engineers across all mechanical engineering and related disciplines participate in ASME conferences each year. This includes authors of approximately 6,000 to 8,000 technical papers published by ASME, the majority of whom also submit video presentations for our virtual conferences. The video collection will be available as an add-on to existing product packages or as a separate subscription.