Hitachi, Ltd.
3D digital twins can teach trainees where experts look first
Hitachi has developed a method that extracts from operational logs how experts move their viewpoints and choose documents when inspecting equipment on a 3D digital twin, allowing newcomers to follow the same sequence for finding information. In a simulated inspection of railway rolling stock, the median time needed to reach the required information was reduced by 60%, suggesting that “where to look first”—something difficult to convey except verbally—could become a training asset.
Newcomers in maintenance and inspection may be given 3D models and manuals but still not know which components to look at or in what order. They have no choice but to search for the necessary materials or ask an expert. Experts’ inspection procedures have traditionally been explained verbally on a case-by-case basis, using the limited time of a small number of people each time training is provided.
The “order of looking” that is hard to write into manuals
Maintenance and inspection require workers to connect the equipment’s structure, inspection points, past records, and related documents when making decisions. Manuals can show standard procedures, but it is difficult to spell out which part of the equipment to look at first in order to narrow down the information needed.
A 3D digital twin is a recreation of equipment in a digital space. While it helps users understand the equipment’s position and structure, the model alone cannot convey why an expert checked a particular component first or which documents the expert chose.
Turning operational logs into training assets
The method developed by Hitachi is not a way to automate experts’ “intuition” as it is. Instead, it records what experts looked at on a 3D digital twin and preserves it as the sequence they used to find information.
The recorded operations include how the camera was moved, whether components were shown or hidden, whether the equipment was disassembled, which documents were opened, and what was searched for. The method extracts and links the viewpoint, displayed components, and reference documents at the point when the expert judged them useful.
A less-experienced worker selects the problem they are having. Supporting information is then displayed in a list, and the scene the expert inspected is recreated on the 3D digital twin. In a verification test simulating part of the maintenance and inspection work for railway rolling stock, the median time needed to reach the required information was 60% shorter than when the experts’ knowledge was not shown. Hitachi says the results demonstrate the possibility that even less-experienced workers can check information at a level close to that of experts.
Toward inspection preparation without searching everywhere
If this method can be put into practice, a newly appointed worker at a railway rolling-stock maintenance depot could select “how to inspect this part” and prepare by looking at a vehicle model reproduced on a device in a disassembled state, along with related documents. Experts would no longer need to repeatedly explain the same basic search procedures. For operators with many pieces of equipment, experts’ operations could become reusable training assets.
However, simply retaining operational logs is not enough. The content must be updated as equipment changes and documents are revised, and it is necessary to select the situations in which an expert’s procedure was appropriate. An operating framework that verifies procedures and safety judgments on actual equipment in the field is also a prerequisite. Hitachi plans to continue verification in the field and aims to combine the accumulated OT knowledge with AI agents. Some of the results are scheduled to be presented at IEEE ITSC 2026.