In recent years, digital technologies such as ICT and AI have been tried and introduced in the fields of infrastructure construction and maintenance. To promote data-driven infrastructure management through the use of such digital technologies, it is important to develop an infrastructure data platform to manage the information. By utilizing this platform to realize a digital twin, it is expected to improve the productivity in infrastructure management and create added value. This paper describes the digital twin with infrastructure data platforms and examples of IoT and other technologies that support it, as well as the advancement of infrastructure management and future prospects through the use of digital technologies.

The concept of a digital twin is to create a virtual space that is an abstraction of the real space, process information in the virtual space, and reflect it in the real world. Information acquisition function for creating virtual space corresponding to real space, function for sharing and transmitting information in virtual space, information processing function for processing information, and information processed in virtual space to real space. The information utilization function of the action is the skeleton of the digital twin concept. This paper considers the functions required of digital twins, organizes the use cases in which digital twins are currently expected to be used, and organizes them as research issues that need to be resolved in order to implement digital twins in society in the future. do.

In 2014, periodical inspections of major civil infrastructures became mandatory, and a large amount of periodical inspection data was generated. In addition, in 2012, the Electronic Government Open Data Strategy was enforced, and the Japanese government start to promote utilize open data for citizens. However, there is a problem that the format and preservation method of materials are different for each local government, and the level of Open Data is not sufficient. In the study, the authors focus on the utilization of periodical inspection data of social infrastructure facilities, propose a method of distributing inspection data as machine-readable open data on the Internet, and propose efficient generation method and distribution of open data. The prototype system has been developed for this purpose. In addition, as an example of open data utilization, the authors developed a data visualization application that helpful to see the entire view of the condition states of civil infrastructures on a map.

In Japan's social infrastructure, the proportion of buildings over 50 years old is increasing. In addition, it is predicted that the number of workers in the construction industry will decrease in the future, making proper maintenance and management of social infrastructure difficult. To solve this problem, a product model method has been proposed that manages point cloud data by structuring it in units of features and parts. It is considered necessary to develop an efficient update method for the maintenance data update method based on this proposal. Therefore, in this research, we devised a method for correcting the position of the 3D region data of each feature and a method for complementing the omission of identification, after using the technology to automatically identify features. We verified its usefulness in updating maintenance management data of road features using point cloud data acquired by Mobile Mapping System. As a result of the experiment, it was confirmed that this method can update the maintenance management data efficiently.

Current maintenance management methods are based on human work, but new inspection and maintenance management methods utilizing new technologies such as ICT and AI are needed to ensure appropriate maintenance management even as the social structure changes in the future. This paper focuses on dam body’s inspections and examine how to ensure the accuracy of 3D reconstruction using images taken by UAV. Although many ground control points are necessary to ensure accuracy, ground control points can only be placed in limited locations, such as at the top of a dam, in a dam. Therefore, we proposed a method to ensure 3D reconstruction accuracy by combining coordinate measurement results such as point cloud data and photogrammetry, and verified the usefulness of this method.

For training a deep learning model that can estimate the degree of degradation from distress images in road infrastructures, pairs of distress images and their degradation degrees as labels are needed. Although a large number of pairs is desirable for achieving high estimation performance, the total number of such pairs is limited. On the other hand, there is an open dataset composed of distress images in other infrastructures. It is expected to improve the estimation performance of degradation degrees by using distress images of the open dataset in addition to the distress images of the road infrastructure dataset. However, in the open dataset, the distress images are not annotated with labels indicating the degradation degrees. Therefore, we propose a method for estimating degradation degrees across multiple datasets by introducing contrastive learning, regardless independent of the presence or absence of labels. The multi-dataset contrastive learning is performed as a pre-task of supervised learning. The obtained model parameters are used in supervised learning to estimate the degradation degrees of distress images in road infrastructures, and it is possible to achieve the improvement of estimation performance. The effectiveness of the proposed method is verified through experiments using real-world distress images in road infrastructure and an open dataset.

Construction of bridge digital twins using sensors is considered to be expensive for small and mediumspan bridges. This research focuses on drive-by inspection and bridge screening that estimate bridge vibration from vehicle vibration without installing expensive sensors on the bridge. Vehicle vibrations on bridges need to be extracted from continuous data, In this research, the data is extracted from the relative distance between the bridge edge and GPS devices installed on the vehicle, and it is verified that AI leraned them can correcte the position estimation error. In this experiment, the measurement on 4 bridges (3 PC bridges, 1 steel bridge) are carried out and the prediction results of learned AI is validated. For models that were difficult to learn, the improvement of the accuracy rate by post-processing using signal proccesing technology are comfirmed. Therefore, the possibility of constructing and updating bridge digital twins by data accumulation and re-learning was verified using actual data.

In recent years, with the development of IoT and ICT technologies, an environment has been created in which digital data related to construction can be easily obtained. This paper describes a basic system that combines such digital data and numerical simulation technology, based on the digital twin concept, to reproduce in a virtual space the "amount of water inflow into the face" and "results of geological survey during construction" obtained during mountain tunnel excavation. The basic system was constructed to numerically simulate the amount of water inflow that will occur in the near future by updating this information. The information predicted by this system could reduce geological risk by disseminating information in a timely manner and in a form that is easy for everyone to understand, thus contributing to safe and secure construction.

The applicability of two methods, SfM and smartphone LiDAR, to the design of countermeasure works for disaster restoration was investigated, with the aim of surveying the post-disaster topography of a slope that collapsed due to an earthquake or rainfall without delaying the construction of emergency countermeasures. The results of the laser survey were assumed to reflect the actual topography, and cross sections were created at the same location from the 3D models created by each method, and their shapes were compared. To confirm whether there are significant differences in the design results of the permanent countermeasures, countermeasures using rock bolt were designed for the cross sections created from each method. The results show that both SfM and Smartphone LiDAR measurement methods can be used for the basic design in terms of measurement accuracy. However, from the viewpoint of measurable range, the application range of smartphone LiDAR is limited in disaster recovery.

In order to realize a digital twin, the 3D model must be a 3D model measured on-site rather than design model data. The model data should also be updated sequentially. On the other hand, it is not desirable from the viewpoint of labor saving and cost reduction to create a new task only for updating model data. In this technology development, 3D scanning by TLS is made parallel work for an autonomous mobile robot that can transport materials linked to BIM / CIM, i-Con Walker that can acquire 3D data by point cloud in addition to realizing the task of material transportation On-site demonstration showed its effect.

In the construction industry, as a countermeasure to the shortage of workers and long working hours, we are promoting DX (digital transformation) of construction to improve productivity and reform work styles. DX is realized not only by digitizing on-site information, but also by making advanced use of that data, supporting decision-making, and simplifying our workflow. We considered that there is a "digital twin" of the construction site as one of the best solutions for realizing DX and developed a new system that we called "CPS construction management system" including a 3D viewer application. This paper introduces several use cases of the digital twins, describes the functions of the "CPS construction management system" for improving the efficiency of construction management work, and discusses future visions based on that system.

Recently in Japan, rapid population decline, low birthrate, and an aging population have caused social problems such as human resource shortages, cutting public transportation systems, lack of transportation for the elderly in underpopulated areas, and deterioration of infrastructures.

The severity of natural disasters has also been intensifying.

To respond to these social problems, Shizuoka prefecture has promoted the "Virtual Shizuoka Project" as a new type of social infrastructure. In this project, laser scanners collect the whole prefecture’s physical data, which will be point cloud data being the copy of the prefecture. It is able to be used in virtual space and provided as open data.

In this paper, we would like to introduce case studies that improved productivity and created new value by using point cloud data as a foundation data of the digital twin.

Recently, it is essential to reduce the risk of disasters by sharing detailed information on rainfall and landslide disaster predictions. Thus, the authors developed a framework (the "iHazard map" project), in which various types of information were aggregated, combined appropriately, and distributed in an easy-to-understand manner. In this study, the authors proposed a disaster prevention hazard map using digital technology (DX) including the Metaverse, followed by effective application. According to a social investigation, it was clarified that, citizens prioritized "ease of viewing’’ and "ease of use’’ over the "amount of information". Moreover, the authors practically applied the proposed method in a briefing session on landslide disasters for citizens. As the result, it was confirmed that the feedback was very positive.

In Japan, where the birthrate is declining and the population is ageing, it is necessary to promote urban planning that suits the times, such as compact city planning. We are promoting the digital twinning of cities, mainly based on the analysis of human flows in Osaka Prefecture, which can be linked to urban planning simulations. In this study, we attempted a geographical similarity serch for H3 hexagonal grid areas in Osaka Prefecture by integrating urban spatial attributes such as the number of stations and bus stops and arrival flow data. Embedded vectors created by distributed representation learning were used for similarity search. Experiments showed that the geographical areas with similar characteristics in terms of human flow + urban spatial attributes to the H3 area including JR Osaka Station were the neighbouring grid areas and the grid area of Kansai International Airport. The zones with intermediate characteristics were indentified from the human flow + urban spatial attributes data of the urban and town/village zones by the addition operation of the embedding vectors.

In recent years, efforts have been made to create interaction and staying spaces in urban areas, and there is an increasing demand for behavioral observation in urban spaces. This paper reports on the development of technology that digitizes pedestrians and cyclists (divided into three categories by vehicle type) in pedestrian spaces from video footage using image recognition technology with AI, building on previous research that automated pedestrian traffic surveys. Based on the digitized results, we converted them into multiple indicators such as speed and stay probability, and considered the possibility of feedback to real-world issues from the characteristics of pedestrian traffic in pedestrian spaces.

The breeding status of raptors is surveyed as part of environmental impact assessments for various infrastructure development projects. While fixed point observation is the conventional method for these surveys, recently, monitoring of raptors in nests using video cameras fixed on nesting trees have become popular. However, the video camera method has challenges including the need to shorten the time required to check the video and securing specialized technicians. Thus, to monitor the breeding status of raptors easily and quickly, the authors developed a behavior detection system for goshawks (Accipiter gentili) using image recognition. First, image data of goshawks were extracted from videos recorded at the nest. Next, the image data were classified into three patterns (nesting, egg-laying, and feeding). Then, deep learning was conducted, and AI modeling was done to automatically detect the behavior of goshawks. The model detected the behavior of the goshawk with satisfactory accuracy. Additionally, to pursue DX further, a dashboard was developed to detect the behavior of goshawks quickly.

In recent years, there has been a growing need to strengthen disaster prevention and mitigation measures for road facilities in order to prepare for increasingly severe heavy rain disasters. It is important to ensure the effective functioning of the extensive road network in the event of a disaster.

In this study, we used GBDT to reproduce a GIS-based search for the shortest route for goods transportation based on the subject of a rain disaster caused by a linear rainfall belt that occurred in August 2022 mainly in the Okitama region of southern Yamagata Prefecture, and improved the prediction accuracy of goods transportation routes when rainfall conditions change by examining explanatory variables using SHAP. In addition, the applicability of the GBDT learning model to future rain disasters was verified by using the model to predict past heavy rain disasters (July 2020). As a result, it was confirmed that the model was able to predict the transportation routes of goods in the past with an accuracy 0.88 (F-value 0.80).

Short-term speed prediction was investigated as a way to utilize the digital twin of traffic conditions. Traffic data for England is publicly available in real time, and its historical data can be downloaded as open data. In this study, we selected a relatively congested point of an arterial road from the downloaded data, compared the traffic condition data with neighboring points of the target, and predicted the speed using machine learning (AI). In addition, we compared the importance of features using SHapley Additive ex-Planations (SHAP), and found that not only the current situation, but also the past speed history had an effect on improving the prediction performance. In the authors’ previous study, only the data of the forecast point were used as input variables for the short-term forecasting, but in this study, the traffic conditions of the neighboring points were input to see if there was any improvement in forecasting performance. The results showed a slight improvement in prediction performance at the points of interest in this study due to the traffic conditions at the neighboring points.

By constructing a digital twin at key locations, real-time traffic flow forecasting and dynamic traffic control can be performed to avoid traffic congestion. In this paper, short-time speed prediction was conducted using open data from England in anticipation of the above applications. Gradient Boosting Decision Tree (GBDT) and Graph Neural Network (GNN) were used for the model, and a comparison was made between the two. The comparison results for the entire 170 target locations showed that the GNN was superior, but the evaluation of individual locations revealed that there were several locations where GBDT was superior. The results also confirmed the GNN was superior at the points where time contributed significantly, and confirmed that the addition of data from other points, which was judged to be valid based on the GNN adjacency matrix, contributed to improving the GBDT accuracy at these points. Finally, the use of GBDT and GNN is discussed.

A method of detecting objects based on images has been utilized to detect vehicles in a parking lot. It is expected to realize optimized operation of parking lots by simulation on the digital twin constructed based on vehicle information obtained from images. However, the accuracy of the vehicle detection method based on images deteriorates significantly under low illuminance conditions. This is due to the decrease in contrast and the noise generated by the image sensor. Low quality images containing these factors are accumulated under low illuminance conditions. In this paper, we propose a method that uses multiple images clarified by applying noise reduction and contrast enhancement as preprocessing for vehicle detection under low illuminance conditions. The proposed method applies object detection using deep learning to these preprocessed images and determines the parking status of the vehicle compartment. Then, the number of detection omissions is reduced by applying the logical OR to the obtained results. Experiments using YOLOv4 for object detection confirmed the effectiveness of the proposed method.