WorksConstruction & Infrastructure

key visual

Crack detection on dam walls by Deep Learning

Collaborative work with Dr. Pang-jo Chun, University of Tokyo

Dam picture

We have participated in a project to detect cracks from the photos of the concrete walls of dams in Japan, under the supervision of Project Associate Professor Pang-jo Chun of the University of Tokyo i-Construction System Studies endowed chairs.

There are two types of scratches and cracks that can be observed on the exterior of a dam: the type of cracks that are not very problematic for the long-term maintenance of the dam, and those that can lead to structural problems in the concrete exterior. In this project, we used deep learning to simultaneously detect cracks that were diagnosed by expert civil engineers as likely to lead to problems, and to exclude less critical cracks from detection.

We have built a pipeline that can switch between various image segmentation techniques, and have achieved a level of AI system implementation that can withstand actual operation.

The system is also capable of indicating where cracks exist in a large image, even in orthoimages taken by drones, as the deep learning model performs inference on the entire image.

Since this system can be easily relearned using additional data, it is expected that its performance will be greatly improved sequentially as more data sets are provided in the future.

Automation of rebar reinforcement inspection using image processing

Collaborative work with Mitsui Consultants Co., Ltd

Steel rods

With Mitsui Consultants Co., Ltd., we have tried to automate the inspection of reinforcement used in concrete structures using image processing.

When manufacturing concrete structures, it is necessary to check during the manufacturing process whether the reinforcing bars placed in the concrete are correctly assembled, but until now this inspection has been very time-consuming and labor-intensive. We participated in a demonstration project with Mitsui Consultants Co., Ltd, where we were responsible for automating the inspection process using image processing.

We have built an image processing pipeline that automatically measures the diameter of rebar, the distance between rebar, and the length of overlapping areas, and have succeeded in measuring with relatively small errors, especially for the distance between rebar and the length of overlapping areas.

In the future, this technology is expected to contribute to significant labor savings at construction sites by being used for the inspection of reinforcing bars of various types.