Daikin tests an AI agent that tells factory technicians why a machine broke

SAKAI, Japan, February 2026 — Daikin has put an AI troubleshooting agent in front of its maintenance crews, and Hitachi has turned it into a product. Here's what launched, what it cost, and what went wrong.

Cost and outcome (as of Feb. 3, 2026)

Speed and accuracy: causes and fixes "within 10 seconds at an accuracy of over 90%" in a preliminary test, the companies say. (Hitachi)
Test size: five failure cases on five equipment types, checked by two experienced technicians. (Nikkei xTECH)
Cost: not disclosed.

What launched

Daikin and Hitachi began a trial at the Sakai plant in April 2025. When a technician inspecting production equipment finds a fault in, say, a pump or valve, the agent suggests the cause and what to do, the companies said. They planned to finish the trial by September 2025, then put it into practical use and roll it out to Daikin plants in Japan and abroad.

Nikkei xTECH reported in November 2025 that technicians reach the agent through a browser on the company intranet, typing questions in a chat. It answers whether the problem is a failure and, if so, its cause and fix. Daikin started further trials at plants in the U.S., India and elsewhere in October 2025, and plans to extend it to more than 100 production sites, the outlet reported.

How it works

Daikin's equipment drawings are turned into knowledge graphs that generative AI can read. The AI also draws on maintenance records and on Hitachi's failure-analysis process, which is based on STAMP, a systems-safety method. Hitachi's R&D feature says the method was first developed for very complex systems such as NASA's, and that the agent searches drawings and standard procedures as it reasons.

Lead researcher Kentaro Yoshimura's team says older approaches that only search past records can't handle failures that have never happened before. For now, people still read the drawings and turn them into structured data for the AI; Hitachi wants to automate that step.

What the companies said

In Daikin's trial, Hitachi says, accuracy in finding the cause rose from 67% with general-purpose AI to over 90%. Answers came in about 10 seconds and named specific suspect parts. Experienced Daikin maintenance staff called it "equal to or better than" a typical maintenance technician and "usable in the field," according to Hitachi.

One aim is passing skills to foreign workers, Nikkei xTECH reported. Daikin wants crews to handle equipment trouble quickly regardless of their experience or the language they speak. The companies' 2025 release said Daikin wants to turn its technicians' tacit knowledge of maintenance into shared knowledge across its plants.

The preliminary test covered machines such as a refrigerant-charging unit and a fin press for heat exchangers, Nikkei xTECH reported.

Who paid, and how

Neither company has disclosed what the project cost or how Daikin pays for the software. On Feb. 3, 2026, Hitachi began selling it as part of its HMAX Industry line, as a cloud API service or a packaged system. Hitachi says the product has also been through a joint test at Mitsubishi Chemical, which began in December 2025, and that it wants to grow it into "physical AI" that collects plant data in real time.

What went wrong

Tiny test. "Over 90%" rests on five failure cases and two reviewers.
Self-graded. The developers ran and reported every test.
What we don't know yet: the full-trial results from Sakai, results from overseas plants, and how often technicians override the agent.

Our follow-up question for Daikin

What did the full Sakai trial measure, and how did the agent do on failures outside the first five cases?

Sources

Hitachi, “Through Collaborative Creation, Daikin and Hitachi Begin Trial Operation of AI Agent Supporting Equipment Failure Diagnostics in Factories,” Apr. 22, 2025. hitachi.com
Daikin, “ダイキンと日立が協創、工場の設備故障診断を支援するAIエージェントの実用化に向けた試験運用を開始,” Apr. 22, 2025. daikin.co.jp
Nikkei xTECH, Seiji Takaichi, “ダイキンが設備故障を診断するAI導入、外国人にも技能・技術を継承,” Nov. 26, 2025. xtech.nikkei.com
Hitachi (via PR Times), “工場の設備故障診断を支援するAIエージェント HMAX Industry「現場サポートAIナビ(Field Support AI Navi)」を提供開始,” Feb. 3, 2026. prtimes.jp
Hitachi R&D, Kazumichi Moriyama, “経験者相当の速度・精度で新規故障の原因も特定できる設備故障診断分析エージェント技術.” rd.hitachi.co.jp

Disclosure: [not in copy] Editor placeholder