NTT DOCOMO Uses AWS to Build Cell Tower Inspection Tool That Cuts Costs, Improves Safety
To perform maintenance and preserve network accessibility for its subscribers, Japan’s largest telecommunications company, NTT DOCOMO, needs to inspect about 50,000 cell towers for signs of rust and corrosion every few years. But since sending crew members to climb towers that can be several dozens of meters tall can be risky and inefficient, the company sought to develop a proprietary drone-based system for photographing and inspecting its towers.
To make this system work, NTT DOCOMO needed artificial intelligence (AI) and machine learning (ML) technology plus computing power to develop an image recognition system that would analyze thousands of photographs taken by its drones and detect rust. So it turned to Amazon Web Services (AWS) to quickly deploy a cost-effective solution that would enable the safe and efficient inspection of its cell towers, leading to seamless network availability for end users.
AWS Batch is the most important management service in our system. It saves us a lot.”
Increasing the Accuracy and Safety of Tower Inspection with Drones
Serving some 73 million customers through an LTE network and one of the world’s most progressive LTE-Advanced networks, NTT DOCOMO is no stranger to innovation. It’s a world-leading developer of 5G networks, near-field communication infrastructure, and emerging Internet of Things solutions. The company has long built on AWS, migrating some of its web service systems, its data analytics platform, and its data warehouse to AWS. The company had also developed its own drone-management system called docomo sky utilizing AWS, so when the company sought to use drones to identify rust on its towers, staying on AWS seemed logical. “We knew that AWS was very useful for software and data management, so I didn’t consider on-premises servers or another management system,” notes Issei Nakamura, AI specialist of NTT DOCOMO.
Rust and corrosion are serious problems for cell towers. “If we leave rust for a long time, network availability can decrease,” says Nakamura. Yet physical inspections by human crews are risky and come with inherent danger for those conducting the inspections. “We started to use drones to inspect the towers and to ensure the safety of the facility and our employees, because climbing that tower machinery is quite challenging. We also wanted to ensure the accuracy of the reports.”
NTT DOCOMO began using drones to photograph its cell towers in 2017. Scrutinizing thousands of images was time consuming for its employees, so it turned to AI and ML technology to develop an image recognition system. “We think of the data as a very valuable asset,” says Nakamura. “Since we didn’t want to share the data with other AI vendors, we developed an image recognition system ourselves.”
Quickly Deploying a Fast, Reliable Tool Using AWS
The NTT DOCOMO team developed two prototypes for the rust recognition system in about a month—something it could do because creating prototyping is not expensive and the team had access to a lot of information to understand AWS services—releasing it internally in January 2020. NTT DOCOMO first used about 100 images to train the model to detect rust or damaged hardware but quickly realized it wasn’t enough—so it began collecting data continuously to train the model. The new system is now used to inspect all towers in the company’s network using a two-tiered process: an initial AI inspection with rust recognition comes first; after that, if necessary, there is a precise inspection with an operator. Currently, NTT DOCOMO manually uploads images from the drones’ SD cards to docomo sky using a laptop, though it plans to upload the images or videos directly from the drones to the cloud in the future.
For each tower, the recognition system processes about 400 images, each about 7 MB and 5,472 by 3,648 pixels. NTT DOCOMO originally used the compute power of Amazon Elastic Compute Cloud (Amazon EC2) P2 Instances for inference but in August 2019 quickly switched to Amazon EC2 G4 Instances powered by NVIDIA T4 Tensor Core GPUs, which the company found to be twice as fast and four times more affordable—with a prediction time of 5 seconds compared to 10 seconds per image and a cost of $0.71 versus $1.542 an hour. “Amazon EC2 G4 Instances exceeded our expectations,” Nakamura says.
NTT DOCOMO gains additional cost savings by using AWS Batch, which dynamically provisions the optimal quantity of Amazon EC2 G4 Instances based on the volume and resource requirements of the batch jobs submitted. “AWS Batch is the most important management service in our system,” says Nakamura. “It saves us a lot. We don’t inspect the cell towers every day. So we don’t want Amazon EC2 G4 Instances working all the time.” Amazon API Gateway and AWS Lambda are used to receive HTTP requests and run code to save those HTTP requests to Amazon Simple Storage Service (Amazon S3) and submit jobs to AWS Batch. “We are able to collect images very easily in Amazon S3,” says Nakamura, noting the company previously used an internal data warehouse to share images. “That helps us train models efficiently.”
The image recognition system presently runs 15 concurrent tasks—though NTT DOCOMO can easily increase that number in AWS Batch—and normalizes the results of the inspection, meaning the company can easily compare images from various towers and over time. NTT DOCOMO expects network availability and customer satisfaction for end users to increase in the long run as the system enables the company to more quickly identify and repair damaged towers. Critically, switching from human inspection to drone inspections helps the company improve the safety of its employees: it hasn’t had an accident on its cell towers since switching to drone monitoring. The new system also increases efficiency of tower inspections while drastically reducing costs. It cuts about 100 minutes from each cell tower inspection compared to the manual inspection. “In 2020, we are supposed to inspect about 1,500 towers using drones, which means a decrease of more than 2,000 hours,” Nakamura notes. “That amounts to more than $100,000 in savings.”
Generating New Business and Better Serving Network Users
The successful drone inspection program may even spin off a new business opportunity for NTT DOCOMO. “We regard such inspection data as a valuable asset,” Nakamura explains. “Based on our cell tower inspection experience, we’re thinking of developing other infrastructure inspection services for things such as bridges or buildings.”
Using AWS to migrate its manual cell tower inspection protocol to a drone-based, AI-driven inspection system enabled NTT DOCOMO to improve employee safety, save money, and potentially create a new revenue stream. And fewer rusty and damaged cell towers means Japan’s largest telecommunications company can focus more time and resources on delivering first-rate mobile service to its customers.
About NTT DOCOMO
NTT DOCOMO, the largest telecommunications company in Japan, provides mobile services through advanced wireless networks such as LTE and 5G to more than 73 million customers. It partners with mobile operators to develop smart mobile technologies.
Benefits of AWS
- Developed 2 prototypes in 1 month
- Saves about 100 minutes in each cell tower inspection (saving more than 2,000 hours total in 2020)
- Runs 15 concurrent tasks
- Saved more than $100,000 in 2020
- Reduced reported employee cell tower–related accidents to zero
AWS Services Used
Amazon Elastic Compute Cloud (Amazon EC2) is a web service that provides secure, resizable compute capacity in the cloud. It is designed to make web-scale cloud computing easier for developers.
Amazon EC2 G4 Instances
Amazon EC2 G4 instances deliver the industry’s most cost-effective and versatile GPU instance for deploying machine learning models in production and graphics-intensive applications.
AWS Batch enables developers, scientists, and engineers to easily and efficiently run hundreds of thousands of batch computing jobs on AWS.
Companies of all sizes across all industries are transforming their businesses every day using AWS. Contact our experts and start your own AWS Cloud journey today.