日经报道：大金工业联合初创公司 Fairy Devices 开发 AI 智能体培训空调检修技师，计划 2027 年夏季起在日本、印度等市场部署

OSAKA -- Air conditioner maker Daikin Industries has developed an artificial intelligence tool to help train technicians in maintenance inspections as the Japanese company grapples with labor shortages at a time rising higher temperatures are pushing up global demand.
Daikin has developed an autonomous AI agent with startup Fairy Devices -- which has roots in the University of Tokyo and backing from Daikin -- for teaching workers how to perform service inspections.
Inspectors will use a wearable device developed by Fairy Devices while working, pushing a button to start and stop work sessions. The device sends video of the inspection work to the AI agent in real time.
The agent checks the video to make sure that no steps have been missed and sends results to the worker's smartphone within a minute of finishing the inspection. The agent has detailed knowledge of various types of work and can adapt to engineers who might perform the steps in a different order.
Daikin has been using the devices since 2020, initially to allow experienced engineers to remotely supervise newer colleagues. The company's experienced workers had been overextended in summer months -- peak season for air conditioner repairs.
Daikin aims to deploy the wearables in inspections of control boards in home air conditioners starting in the summer of 2027, looking at rolling out AI in such markets as Japan and India.
The company will seek to expand the system to service more complex products including commercial air conditioning equipment in addition to general maintenance and repair services.
As demand for new air conditioning equipment rises, after-sales service is also crucial. The rate of profitability for ongoing maintenance business, including inspections and repairs, is higher than for selling the equipment itself.
The technology is still being tested at a research facility, where the company says it can identify tasks that need to be done with 90%-plus accuracy -- but in actual use, such factors as poor lighting lower this to around 70%. Daikin is working to increase accuracy further with more real-world data.
Daikin has developed a proprietary AI model to detect tasks specific to air conditioners, spurred by the shortcomings of generic models. Development included training the AI on 200 or so hours of video of inspection and repair work gathered by the company, improving accuracy by annotating the video with text to explain the work being done.
"No other company has amassed as much real-world data as Daikin has," said Yoshifumi Seki, chief product officer at Fairy Devices.
"For truly specialized work, we have to ask veteran engineers, but the AI agent can provide probably 80% to 90% of support," said Shohei Hido, chief engineer at Daikin's Technology and Innovation Center.
In the last two years, the AI agent has won a first prize and another special prize in a contest created by Japan's Ministry of Economy, Trade and Industry and the country's New Energy and Industrial Technology Development Organization.
The developers say they are considering going beyond verification after completion to add speech functions that will allow the AI to guide workers through steps during work.
The International Energy Agency estimates that the global stock of installed air conditioners will reach 5.58 billion units by 2050, more than doubling from the 2025 level.
The build-out of data centers and chip foundries in the U.S., Europe and elsewhere is also driving up demand for cooling equipment, and Daikin is expanding its business base in these regions as the market grows.
Daikin has roughly 5,500 engineers responsible for inspecting and repairing air conditioners worldwide, including in the Americas and Asia. With demand set to soar, the company is increasingly concerned that this figure may not be enough, leading to a push to increase staffing and develop talent through mergers and acquisitions.