AI Assists in Detecting Anomalies in CFT Column Concrete Filling Work

Kajima Corporation (President: Masafumi Kiryu) and Ridge-i Inc. (President: Naohisa Yanagihara, hereinafter "Ridge-i") have developed an AI-assisted system for detecting anomalies during the concrete filling work within steel pipe columns of CFT structures (hereinafter, "this system"). Kajima has implemented this system into "moni-as®" (monias)¡, which enables remote management of such work.

In CFT structures (structures where concrete is filled inside steel pipe columns), quality control of the concrete filled within the steel pipe columns is crucial for ensuring strength. In the event of an anomaly such as foreign object contamination, prompt and appropriate judgment and response, including immediate work stoppage, are necessary. By implementing this system into moni-as, it is possible to conduct checks with a multi-layered perspective utilizing AI in addition to monitoring by construction personnel and engineers, allowing for faster detection of anomalies. Furthermore, this system utilizes "Kouka VRT," a VM-type GPU cloud service for parallel processing provided by Sakura Internet Inc. (hereinafter, "Sakura Internet"). Combined with moni-as's low-latency delivery, it enables real-time diagnosis of issues like foreign object contamination.

¡ monitoring assist system

[Background and Development Process]

For CFT structures, quality control of the concrete filled inside the steel pipe columns is important. In the event of an anomaly during concrete filling, construction personnel and engineers with specialized knowledge had to be present on-site for extended periods to enable appropriate and prompt judgment and response. Therefore, to reduce this workload, Kajima developed the moni-as system, which allows for real-time visual confirmation of concrete filling progress from remote locations, and has been deploying it at construction sites.

To enhance the performance of moni-as, a new system has been developed where AI assists in detecting anomalies during concrete filling by humans.

[Overview and Features of This System]

This system supports human anomaly detection by having AI monitor the video feed from moni-as during the concrete filling work of CFT structures. The AI model used in this system was built by incorporating Kajima's unique construction know-how into learning data derived from over 2,000 hours of concrete filling footage accumulated by moni-as. Therefore, it detects anomalies based on the same judgment criteria as specialized engineers.

This system checks three items: "Foreign objects inside the steel pipe column," "Concrete properties," and "Diaphragm passage timing"2. Details for each are as follows:

2 Steel plate used to integrate the junction of the steel frame's column and beam.

1. Foreign objects inside the steel pipe column

CFT structure columns achieve their designed strength by having concrete filled uniformly and densely inside the steel pipe columns. This check verifies if any foreign objects that could adversely affect the achievement of the designed strength have been mixed in.

2. Concrete properties

Concrete properties vary depending on design standard strength, manufacturing plant, materials used, etc., and constantly change over time. This check verifies if the properties are in a state suitable for filling.

3. Diaphragm passage timing

When filling concrete into steel pipe columns, managing the filling height speed to be 1m/min or less is done by checking the timing of passage through each diaphragm. To calculate whether concrete is being filled at an appropriate speed, the timing of the concrete passing through each diaphragm is confirmed.

The diagnostic results are displayed on the screen in three levels: "Normal (Green)," "Caution (Yellow)," and "Anomaly (Red)." It is also possible to view the judgment basis as a report. Furthermore, this system is built on Sakura Internet's VM-type GPU cloud service "Kouka VRT" for parallel processing. Combined with moni-as's low-latency delivery, it can diagnose multiple items in real-time.

[Effects of Implementation]

The effectiveness of this system was verified through mock-up tests, reproducing situations that should be detected as anomalies, such as intentionally mixing foreign objects like plastic bottles into the steel pipe columns. As a result, it was confirmed that the system can diagnose and detect "foreign objects inside the steel pipe column," "concrete properties," and "diaphragm passage timing" with high accuracy and in real-time for all three items. Subsequently, this system was introduced to several actual CFT structure sites, and it was confirmed that high quality could be maintained as before.

[Future Development]

Kajima and Ridge-i will continue to promote research and development to improve the performance of this system and work towards ensuring the quality of concrete filling work in CFT structures.

(Reference)

Developed "moni-as" enabling remote management of concrete filling work for CFT columns

(Press Release dated March 14, 2024)

https://www.kajima.co.jp/news/press/202403/14a1-j.htm

VM-type GPU Cloud Service "Kouka VRT" (Sakura Internet Inc.)

https://cloud.sakura.ad.jp/lp/vrt/