AI-based Inspection Method for Efficiency and Quality Improvement of Construction Work
- On-site trial starts for improving efficiencies in inspection of rebar joints using AI -
NTT COMWARE CORPORATION (Head office: Minato-ku, Tokyo, President: Satoshi Kurishima, hereinafter NTT COMWARE) starts on-site demonstration tests of the image recognition AI "Deeptector(R)" with SHIMIZU CORPORATION (Head office: Chuo-ku, Tokyo, President: Kazuyuki Inoue, hereinafter SHIMIZU) for improvement of efficiencies and quality of inspection concerning gas pressure welded joints*1).
*1) A method of joining the end faces of rebars and heating them while applying pressure to create a bulge in a "red-hot" range without melting the joining end faces.
[Background]The construction industry faces an urgent need to improve the productivity of construction sites and foster young technicians amidst concerns about the high employee turnover because of aging of skilled workers.
While inspections for quality assurance of construction have been performed by visual inspection by workers, it is a challenge to check a large number of inspection items. Therefore, it is necessary to have automation and labor-saving measures in inspection as well as to develop a system to nurture supervisors for construction sites based on the know-how of skilled workers in order to maintain a certain quality of construction work.
SHIMIZU and NTT COMWARE have been jointly conducting feasibility tests for the image recognition AI "Deeptector(R)" for automatic judgement to improve efficiencies in inspection at construction sites by applying image segmentation technology. The study started with the external inspection of gas pressure weld joints.
It has been a challenge to improve the accuracy of AI judgment of images taken in environments with different backgrounds and lighting conditions such as construction sites with conventional methods. Under the circumstances, we decided to start the full-scale demonstration test on-site using "Reinforcement Joint AI Inspection".
[Outline of "AI Inspection for Rebar Joint"]NTT COMWARE's original AI judgment method (*patent pending) is to perform a simple judgment of OK or NG based on the predesigned checklist as its logic. In the pre-study phase, the method is applied with AI learning for the outline image of gas pressure weld joints to improve the accuracy in detecting the image segments.
How the AI works
[Outline of the demonstration test]"Rebar joint AI inspection" method is applied to verify the recognition ratio and usability on-site as a demonstration.
(1) Schedule: January to March 2020
(2) Process: Smartphones are used to take photos of rebar joints at a Shimizu construction site of a building and NTT COMWARE's image recognition AI is used for inspection of the site.
(3) Check items: accuracy of the judgement, length of time for implementation, operability of the screen using "Rebar Joint AI Inspection" in comparison with conventional visual inspection.
[Future scenario]NTT COMWARE and SHIMIZU intend to use "Reinforcement Joint AI Inspection" as a training support tool, aiming to put it into practical use early next fiscal year as well as for human resource development of construction site supervisors.
SHIMIZU continues to strive to improve the efficiency and quality of inspections of construction sites by expanding the application of AI to other items for inspection, starting with this AI inspection of the rebar joint (gas pressure welding).
NTT COMWARE aims to improve customer productivity and collaborative creation of new value as a business partner of its customers based on this initiative, through deployment of this method to the construction and real estate industries as well as to similar businesses.
* "Deeptector" is a registered trademark of NTT COMWARE CORPORATION.
* Company names and product names listed may be trademarks or registered trademarks of each company.