Applicability of Convolutional Neural Network for Estimation of Turbulent Diffusion Distance from Source Point

Abstract

For locating the source of leaking gas in various engineering fields, several issues remain in the immediate estimation of the location of diffusion sources from limited observation data, because of the nonlinearity of turbulence. This study investigated the practical applicability of diffusion source-location prediction using a convolutional neural network (CNN) from leaking gas instantaneous distribution images captured by infrared cameras. We performed direct numerical simulation of a turbulent flow past a cylinder to provide training and test images, which are scalar concentration distribution fields integrated along the view direction, mimicking actual camera images. We discussed the effects of the direction in which the leaking gas flows into the camera’s view and the distance between the camera and the leaking gas on the accuracy of inference. A single learner created by all images provided an inference accuracy exceeding 85%, regardless of the inflow direction or the distance between the camera and the leaking gas within the trained range. This indicated that, with sufficient training images, a high-inference accuracy can be achieved, regardless of the direction of gas leakage or the distance between the camera and the leaking gas.

Deep learning estimation of scalar source distance for different turbulent and molecular diffusion environments

“…The distance between the turbulence grid and the cylinder was set to 40, which was denoted as "Case R", while the case without a turbulence grid was labeled as a less turbulent condition (Case LT). This condition of Case LT is same with that of our previous qDNS (Ishigami et al, 2022b). In contrast, the case with a close distance to the turbulence grid was labeled as a highly turbulent condition (Case HT).…”

“…The time period for training data acquisition and that for test data was separated by a large time interval ∆T , which was much We evaluated individual learners on unknown test images of the same size as the training phase. As previously reported (Ishigami et al, 2022a(Ishigami et al, , 2022b, high-performance inference with a high accuracy of over 90% was achieved by a CNN learner of Inception-ResNet-v2 (Szegedy et al, 2017), which was also used in this study. The network employs residual inception blocks based on ResNet (He et al, 2015) and consists of 164 layers: see Fig.…”

“…This is a common situation for external flows. Ishigami et al (2022b) studied the wake turbulence in a transverse cylinder as in the present study, but did not consider freestream turbulence created by turbulence grids. The Schmidt-number dependence was also not discussed.…”

“…They demonstrated that the source-to-observation distance can be estimated with at least 90% accuracy for scalar (dye) diffusion from a point source in water channel. Ishigami et al (2022b) applied CNN to a distribution image of concentrations integrated in the line-of-sight direction, assuming an infrared camera image, and demonstrated the estimation in a gas outflow from a single pipe. In our previous studies, the training images used are limited to single Reynolds number and Schmidt numbers for a fully developed turbulent channel flow and a cylinder wake.…”

Special Issue on “Machine-Learning-Assisted Intelligent Processing and Optimization of Complex Systems”

“…In order to locate the source of gas leaks in several engineering areas, Ishigami et al [5] investigated the practical applicability of adopting a convolutional neural network (CNN) to predict the location of gas leaks based on captured infrared images. The study found that a single learner trained with a sufficient number of images achieved an inference precision higher than 85%.…”