AI Case Study
Kawasaki Geological Engineering increases efficiency in subsurface cavity surveying with deep learning
Kawasaki Geological Engineering supported by Fujitsu will apply deep learning technology to subsurface cavity surveying. Radar images captured by the company's underground cavity detection technology are analysed and processed using image recognition through the Zinrai Deep Learning platform service, which was launched by Fujitsu in April 2017. Trained on machine learning and image classification the technology will be able to understand whether changes in radar reflection are displayed as waveforms on images are cavities or sewer pipes. In comparison to the manual procedure by expert technicians the technology was able to accurately determine cavities from image data in one-tenth of the time.
Construction And Engineering
"Currently, using underground radar probe equipment developed by Kawasaki Geological Engineering, underground surveys, which were previously limited to about one meter, can now be done to about five meters deep, which is expected to significantly improve the reliability of cavity explorations beneath road surfaces going forward. Because the massive volume of image data collected by the underground radar probe equipment must be visually evaluated by well-practiced expert technicians, however, there have been issues with preserving objectivity and ever increasing workloads.
Analysis and processing of the massive volume of radar images taken with Kawasaki Geological Engineering's world-leading underground cavity detection technology will be carried out using image recognition through the Zinrai Deep Learning platform service, which was launched by Fujitsu in April 2017. The AI will be trained through machine learning on images where changes in radar reflection are displayed as waveforms, and will then determine whether they are cavities or sewer pipes."
This technology was made available as a service in the summer of 2017.
"In a trial of cavity identification, in comparison with existing visual identification by expert technicians, not only was the cavity identification using AI able to accurately determine cavities from image data, it did the analysis in one-tenth of the time."
"Kawasaki Geological Engineering is using FUJITSU Cloud Service K5 Zinrai Platform Service Zinrai Deep Learning —Fujitsu’s deep learning platform service—to analyze and process the huge volume of radar images collected with underground radar probe equipment. The AI will be trained through machine learning on images where changes in radar reflection are displayed as waveforms, and will then determine whether they are cavities or sewer pipes."
According to Fujitsu's press release, "in Japan, there are about 3,300 incidents of collapse caused by cavities under the surface of roads annually, making an issue in society. The primary cause of these collapses is aging sewer pipes, often located about three meters underground. This creates a growing need for a system that can determine the risk of a collapse without excavating the road to investigate."
Radar images collected with underground radar probe equipment.