Matsuo Institute Co., Ltd. and Omron Field Engineering Co., Ltd. Systematize Highly Personalized Maintenance Operations with Jointly Developed AI Judgment Model

Matsuo Institute Co., Ltd. (Head office: Bunkyo-ku, Tokyo; Representative Director: Toshifuku Kawakami; hereinafter "Matsuo Institute") has jointly developed a new system with Omron Field Engineering Co., Ltd. (Head office: Chuo-ku, Tokyo; Representative Director and President: Taisuke Tateishi; hereinafter "OFE") that automatically judges inspection photos taken at work sites using AI.

This system is an AI judgment technology jointly developed to realize on-site DX by integrating Matsuo Institute's generative AI and image recognition technologies with OFE's long-cultivated on-site knowledge of maintenance operations. Having confirmed high effectiveness through actual in-house operations, full-scale introduction began in October 2025.

In recent years, against the backdrop of a declining workforce, maintaining work quality and streamlining human resources have become critical issues in the field of maintenance and inspection. OFE has been responsible for maintenance and operation services in social infrastructure areas, including railways and finance, and has provided high-quality services for many years. However, checking inspection photos has been a highly manual task due to its importance as a double-check process.

In this project, we focused specifically on the highly personalized and labor-intensive task of "cross-referencing the setting values (large amounts of textual information) of installed equipment." By combining OFE's accumulated on-site expertise with Matsuo Institute's advanced AI technology, we built an automatic judgment model utilizing generative AI. By systematizing the judgment process that previously relied on human hands, we have established a foundation for high-quality maintenance operations with reduced resources.

Effectiveness Verification in Actual Operations

This system underwent approximately four months of effectiveness verification in internal operations, confirming the following results:

• AI Judgment Count: 8332 cases

• AI Accuracy (*1): 89%

• System Error Rate (*2): 0.2% (achieving high stable operation)

From these results, we confirmed that this system possesses the accuracy and stability to withstand actual business operations.

Starting in October 2025, prioritized introduction has begun, focusing on inspection tasks where AI application is particularly effective. For on-site use, operations are carried out after pre-verification using actual inspection images, contributing not only to improved work efficiency through AI utilization but also to improved quality, such as reducing the risk of human oversight, which is unavoidable in manual work.

Technical Features: Replicating Human Judgment Processes with AI

The greatest feature of this system is its advanced AI processing, which goes beyond conventional image pattern recognition, typified by appearance inspections, to read text information contained in photos and semantically judge whether its content is the correct setting value. Such judgments have been difficult to handle with image recognition alone.

In this project, based on Matsuo Institute's AI technology, we developed an advanced judgment model that replicates the human judgment process—referring to manuals—by combining text extraction through optical character recognition (OCR) and semantic understanding by a large language model (LLM).

• OCR (Optical Character Recognition): Extracts text information such as setting values from inspection photos

• LLM (Large Language Model): Cross-references extracted text against manuals to determine correctness

• Chain-of-Thought (Visualization of Inference Process): Improves transparency and stability of judgment by presenting AI's reasoning step-by-step

FAQ

Which companies jointly developed the new AI judgment system?

The system was jointly developed by Matsuo Institute Co., Ltd. and Omron Field Engineering Co., Ltd.

What is the primary function of the jointly developed system?

It automatically judges inspection photos taken at work sites using AI.

When did the full-scale introduction of the system begin?

The full-scale introduction of the system began in October 2025.

What specific task did the project focus on for automation?

The project focused on cross-referencing the setting values of installed equipment.

How many cases of AI judgment were processed during the effectiveness verification?

There were 8,332 cases of AI judgment processed during the verification period.

Back to Newsroom (40)