To investigate the production and transport mechanism of helicity, we used LES of wall-normal rotating turbulent channel flow to examine the contribution of each term in the budgets of the turbulent helicity and its three parts hxx, hyy, and hzz. hxx and hzz were bigger than hyy and their signs were opposite to each other. We found that the pressure term contributes to hyy and is redistributed to hzz to decrease the absolute value of hzz. Therefore, hxx becomes the biggest term of three and forms positive helicity near the wall and negative helicity away from the wall. In the region away from the wall, the viscous dissipation term reduced the negative helicity of hxx.

A new expression for the scale-space energy density was proposed to better understand and predict inhomogeneous turbulence. Three filtered velocities were used to derive the new expression that consists of homogeneous and inhomogeneous terms; the former is proportional to the velocity variance and is non-negative whereas the latter has the second derivative of the velocity variance. DNS data of turbulent channel flow was used to evaluate the two terms of the turbulent energy and energy density. It was shown that a concave profile of the turbulent energy near the wall accounts for the negative values of the energy density.

By reckoning the plume as the coherent fluctuation, a transport model for the convective turbulence is constructed with the aid of the nonequilibrium effect along plume motions, and applied to a stellar convective flow. One of the prominent characteristics of a surface cooling-driven convection, the enhanced and localized turbulent mass flux below the surface layer, which cannot be reproduced at all by the usual eddy-diffusivity model with mixing length theory (MLT), is well reproduced by the present model with the non-equilibrium effect. Our results show that the incorporation of plume motion into turbulent transport model through the non-equilibrium effect is an important and very relevant extension of mean-field theory beyond the heuristic gradient transport model with MLT.

Numerical calculations for environmental analysis are one of the promising methods to safeguard the sustainable development of marine aquaculture, and it is essential to introduce the effects of fish cage deformation in the model. In this study, the fish cage deformation was reproduced with the CFD method, and its effects on the flow field was investigated. As a result, experimental results have been generally reproduced. The drag force increases in the wake flow area, resulting in a significant decrease in wake flow. However, current parameters are not sufficiently examined, and there is a need for future improvement.

The Reynolds number effects on optimal control of a turbulent channel flow for reducing the skin friction drag are investigated. Although the control performance degrades as the Reynolds number increases, optimal control significantly suppresses the near-wall turbulence structures in a similar manner to those observed at lower Reynolds numbers. To further investigate the effects of increase in the Reynolds number, two control problems are considered; optimal control targeting only a region near or away from the wall and optimal control of an ideal flow where large-scale structures are artificially suppressed. They suggest that control strategies targeting a region away from the wall or only large-scale structures are not effective.

In recent years, reinforcement learning has been attracting attention as a framework for developing a new control strategy in a non-linear system by relating limited sensing information to actuation through deep neural network. So far, although reinforcement learning has been applied to relatively simple flow fields with limited degrees of freedom of a control input, its application to turbulent flows with large degrees of freedom has not been reported. In this study, we apply reinforcement learning to wall turbulence control for drag reduction in order to assess the effectiveness of reinforcement learning in a turbulent flow. It is shown that the current framework successfully yield a new control law which is more effective than the existing opposition control.

The ultimate goal of this research is to develop a deep learning based fast and accurate indoor airflow prediction method that serves as a surrogate for CFD in the coupled simulation. Based on previous investigations, in this study, the prediction target is extended from two-dimensional flow to three-dimensional to explore the feasibility of deep learning neural network (NN) for fast and accurate predictions for more realistic scenarios. The results demonstrate that the NN predicts the non-isothermal airflow distributions with errors less than 12% and up to 80% calculation time is reduced as compared to CFD simulations.

In the urban environment, gas, such as nitrogen dioxide, and particles impose adverse impacts on pedestrians’ health. Conventional computational fluid dynamics (CFD) methods regarding particle as passive scalar cannot reproduce the formation of nitrogen dioxide and secondary aerosols thus leading to uncertain prediction. In this study, SSH-Aerosol, a modular box model that simulates the evolution of gas, primary and secondary aerosols, was coupled with the CFD software OpenFOAM. The transient dispersion of pollutants emitted from traffic in a street canyon was simulated using unsteady Reynolds-averaged Navier–Stokes equations (RANS) model. The simulated pollutant concentrations were validated from field measurements. We compared particle size distributions and chemical compositions between the coupled and the conventional models. The results show that dry deposition played the most important role in aerosol dynamics for particles with diameter larger than 0.4 μm, while coagulation largely influenced the size distribution of small particles. In addition, secondary aerosol formation affected the mass concentration of inorganic and organic matters.

This study aims to develop a fast and accurate indoor airflow prediction method using deep neural network (DNN). Here, the effect of training data selection on transient prediction is investigated. Five data selection scenarios were considered to find out a method selecting training data, which is benefit to training DNN efficiently, from two aspects: (1) time selected for each case, (2) selection of cases with different setting conditions. As a result, it is better to take time points into consideration as many as possible for each case. At the same time, it is better to take cases in the interior of condition space (a coordinate space formed by setting conditions of condition change experiment case) as training cases.

A method combining two-dimensional Fourier transform (2DFT) and proper orthogonal decomposition (POD) was proposed and applied to analyze the turbulent structures in a two-dimensional urban canyon flow. In this method, 2DFT was used to decompose the turbulence along the homogeneous spanwise direction and time, and POD was used to decompose the turbulence along the non- homogeneous streamwise and vertical directions. It was found that the large- and small- scale turbulent structures affect the flow field around the canyon in different forms, which were visualized by the POD modes. However, due to the self-similarity of turbulence, the time scales of the main turbulent structures were always proportional to the length scales in the span direction, regardless of the scales of the structures. Furthermore, the roles of turbulent structure on pollutant removal at the roof-level were quantitatively elucidated using the SPOD co-spectrum defined in this study. Roof-level Kelvin-Helmholtz instability was found to contribute most to pollutant removal.

This research proposed a sensor configuration optimization method for statistical source term estimation by designing an objective function and applying a simulated annealing algorithm. The objective function is based on the entropy of adjoint concentration distribution to evaluate the information that a configuration can measure. The performance of the proposed method was evaluated by producing an optimal sensor configuration for a regular block-arrayed building group model. The optimal configuration was applied in the estimations for 25 unknown sources. Its performance was compared to that of uniform and random configurations. According to the results, the estimation accuracy of configurations is positively related to their values of objective function. The optimal configuration outperformed other two configurations because it can measure more information of unknown sources.

This research verified the effectiveness of the sensor configuration optimization (SCO) method based on the entropy of adjoint concentration for the indoor model space. By using this SCO method, the optimal configuration consisting of a finite number of sensors was determined from the candidate positions near the ceiling. To evaluate the estimation accuracy of sensor configurations, we estimated 10 unknown sources in three-dimensional space using the optimal configuration obtained and compared the results with those by other two random configurations. As a result, the optimal arrangement showed the highest estimation accuracy. In addition, it was found that the sensors in optimal arrangement tended to be distributed near the inlet and outlet and at the four corners of the indoor model.

Research of automated driving was grown from a small topic on individual technologies in sensing, control and intelligent system. This research expands to various related topics now, and collaborating in automated driving research fields is important. However, there are limited numbers of research visualizing the current situation of automated driving. In SIP-adus 2nd phase, the Japanese national project for automated driving, the research body to collaborate with overseas researchers also domestic research groups is expected to be established, and the Alliance for Promoting Mobility Innovation was organized under this national project. Overview of the study about automated driving researches with the Alliance is introduced in this paper.

Communication with citizens about actual situation of automated driving taking an important role for implementation of automated driving technology. Data collection campaign in FoT bring basic and valuable data for those communications also functional development of automated driving system. In Kashiwa-no-ha area in Kashiwa city, the long-term filed operational testing by automated driving bus with driver is carrying out from November 2019. This FoT seems to be relatively similar situation of that with automated driving system implemented compared with other FoT. In this research the data collection campaign is started to study about factor when safety driver taking over control from automated driving systems. In this paper, the overview of this campaign is introduced.

Vehicle localization is one of the important technical factors in autonomous driving vehicles. High accuracy, high preciseness, and robustness towards various road conditions are required for localization of autonomous vehicles. A method using magnets as road markers has proven to achieve above requirements. In such method, it is unable to localize the vehicle when it veers off the course of magnetic marker system. This is problematic especially in places where vehicles can take various paths, such as intersections. In this paper, we propose a concept where road markers are placed in grid-like patterns to enable localization for various paths. Maximum likelihood estimation is used to distinguish individual markers.

The applicability is discussed for the traffic signal, which controls the right-of-way dynamically and externally to traffic flow and it is located at the at-grade intersection where hierarchical traffic functions of streets/avenues. The announcement function of traffic signal to multiple users at the real physical field, which provides the condition of rightof-way dynamically, is examined. The significance of autonomous distributed traffic signal control, which strictly follows the principle, is also examined. These examinations are based on the signal lights locations, control restriction of traffic flow, drivers' behavior, and relationship between real-world traffic phenomena and legal rules. Based on the development of advanced technology, such as automated vehicles, a future direction of the advanced traffic management for urban road network is proposed.

Origin-Destination (OD) traffic volume data are expected to improve the performance of traffic simulators by providing more precise demand input, which can support efficient traffic management and operation. This study analyzed the temporal fluctuation of OD traffic volume in the whole Tokyo metropolitan expressway network using electronic toll collection (ETC) data in 7 months. Specifically, by applying the state space model, traffic volume hourly fluctuation in a day and daily fluctuation in a week were extracted for each OD pair. A prediction model of OD traffic volume based on these fluctuations was proposed. The temporal fluctuation patterns and their proportions in specific OD pair groups were further investigated using hierarchical clustering.

This study investigates the geometric impacts on the gap distributions of human-driven vehicles (HDVs) to consider the locations of exits for dedicated connected-and-automated vehicle (CAV) lanes, where CAVs must merge into HDVs. The gap distributions were modelled by the combined gamma distributions for different locations and traffic conditions using Hanshin and Chuo Expressway data. Based on the estimated model, the required extra time for merging CAV was hypothetically computed by assuming the random arrival of HDVs in the normal lane at different locations. The results indicate that the waiting time tends to increase in a long-stretch uphill, suggesting that such locations should be avoided when locating an exit.

In signal control, which is one of the main functions of traffic control systems, appropriate signal control parameters are calculated based on the measurement data from vehicle detectors installed on the road. However, the installation and maintenance of vehicle detectors is costly, so realization of a signal control system that can reduce the number of vehicle detectors used while maintaining control level is required. In this paper, we propose a method to calculate signal control parameters without using measurement data from vehicle detectors using traffic information obtained from probe vehicle data.

In order to ensure safe driving at unsignalized intersections with poor visibility for automated driving and driving assistance, we use on-board camera images to understand traffic conditions in a blind spot beyond the traffic mirror, and predict risk. The system uses deep learning to detect/track the position of the traffic mirror and the position of on-road objects including vehicles reflected in the mirror. Then, whether the blind spot is risky or not is determined by the approach of the on-road object. As a result of experiment in a real environment, it was confirmed that the accuracy of the approach discrimination reached 75-92% under the environment where the viewpoint conversion worked stably.

Detecting and predicting abnormal driving behaviors such as sudden braking counts for preventing traffic accidents. Keeping distances from other vehicles is considered one of the most common situations where sudden braking occurs. However, to detect distances, state-of-the-art 3D object detection usually requires advanced sensors such as LiDAR. In this work, we use a monocular visual odometry system to estimate distances between vehicles with only a common drive recorder. As a preparation for predicting sudden braking events, we train a mining model to detect them with the estimated distance information along with probe data. The experiments show that with the estimated distances, the model gains an improved performance on braking detection.

The causes of environmental value loss are different for each tenant. In response to this, it is useful to formalize the relationships among “Building Elements” to understand the semantic structure, and then monitor the situation and implement.

In this paper,since we need an information infrastructure that can be integrated and expanded, we proposed to use a “Semantic Data Model”.

I verified the description of the semantic structure of the “Building Element” involved in the CO₂ concentration monitoring system, and the integration with the other models. And it was found that the graph is integrated at common nodes of locations.

An overview of the need for a general-purpose spatial description format called Common Ground, its ideal format, directions for collaboration and extension among different formats, and possible industrial implementations.