JP-2026075898-A - Bridge inspection method, bridge inspection device, and program

Abstract

[Problem] To automatically and accurately determine the extent of damage to bridges. [Solution] The bridge inspection device 110 inputs a prompt containing damage criterion information explaining the criteria for the degree of damage to a bridge, and image data to be inspected, into the language model 121, and obtains response information from the language model 121 that contains damage degree information indicating the degree of damage corresponding to the image to be inspected. [Selection Diagram] Figure 1

Inventors

• 櫻井 信彰
• 北 慎一郎

Assignees

• 日鉄エンジニアリング株式会社

Dates

Publication Date

20260511

Application Date

20241023

Claims (20)

1. A prompt input step involves inputting a prompt into a language model that includes damage criteria information describing the criteria for the degree of bridge damage, An inspection target image input step involves inputting image data corresponding to the damage into the language model, A response acquisition step of obtaining response information from the language model, which includes damage degree information indicating the degree of damage corresponding to the image data, A bridge inspection method characterized by comprising the following features.
2. The prompt includes damage definition information that describes the definition of the type of damage, The bridge inspection method according to feature 1.
3. The prompt includes information explaining that, when creating the response information, it is not necessary to consider information regarding the repair of the damage and information regarding the impact of the damage on the structure of the bridge. The bridge inspection method according to feature 1 or 2.
4. The prompt includes at least one of material information describing the material of the bridge and painting specification information describing the painting specifications of the bridge. The bridge inspection method according to feature 1 or 2.
5. The prompt includes a prompt for at least one of the materials of the bridge and the specifications for the painting of the bridge. The bridge inspection method according to feature 1 or 2.
6. The prompt includes member information describing the members of the bridge, The bridge inspection method according to feature 1 or 2.
7. The prompt includes a prompt for each member of the bridge. The bridge inspection method according to feature 1 or 2.
8. The damage includes corrosion. The bridge inspection method according to feature 1 or 2.
9. The damage criteria information includes damage extent information that describes the criteria for the extent of the damage, The bridge inspection method according to feature 8.
10. The damage criteria information includes damage depth information that describes the criteria for the depth of the damage. The bridge inspection method according to feature 8.
11. The aforementioned response information includes findings information that explains the findings of the aforementioned damage. The bridge inspection method according to feature 1 or 2.
12. The damage extent information is associated with the findings information in the response information. The bridge inspection method according to feature 11.
13. The aforementioned response information is structured. The bridge inspection method according to feature 1 or 2.
14. A metadata input step of inputting metadata indicating the specifications of the response information into the language model, The bridge inspection method according to claim 1 or 2, further comprising the following:
15. The image data includes first tag information corresponding to a first region included in the image shown by the image data, and second tag information corresponding to a second region included in the image. The bridge inspection method according to feature 1 or 2.
16. The prompt includes the prompt corresponding to the first tag information and the prompt corresponding to the second tag information. The bridge inspection method according to feature 15.
17. A display control step of displaying an image corresponding to the aforementioned image data on a display unit. Furthermore, In the display control step, information corresponding to the degree of damage is displayed in the first region and the second region, respectively. The bridge inspection method according to feature 15.
18. The prompt includes information instructing the language model to respond using the first tag information and the second tag information, regarding the region where the degree of damage has been assessed. The bridge inspection method according to feature 15.
19. A reference image input step involves inputting reference image data corresponding to each of the multiple degrees of damage into the language model. The bridge inspection method according to claim 1 or 2, further comprising the above.
20. The damage includes damage caused by at least one of the following: water leakage/free lime, peeling/exposed rebar, and floating. The bridge inspection method according to feature 1 or 2.

Description

This disclosure relates to a bridge inspection method, a bridge inspection device, and a program. As a method for inspecting the extent of damage to bridges, Patent Document 1 discloses a method in which an inspector determines the degree of deterioration of the underside and side surfaces of a section of a pier beam. Furthermore, Patent Document 1 discloses a method in which an inspector observes a section of the beam structure and determines its soundness by comparing it against predetermined criteria. Japanese Patent Publication No. 2021-143575 "Guidelines for Periodic Bridge Inspections," [online], March 2019, Ministry of Land, Infrastructure, Transport and Tourism, Road Bureau, National Highway and Technical Division, [Accessed August 30, 2024], Internet <URL: https://www.mlit.go.jp/road/sisaku/yobohozen/tenken/yobo3_1_6.pdf> This is a diagram showing an example of the configuration of a bridge inspection system.This figure shows an example of the functional configuration of a bridge inspection device.This is a diagram illustrating an example of a type of damage.This figure shows an example of the inspection screen before the start of the inspection.This figure shows an example of the inspection screen after the inspection.This figure shows an example of an image to be examined.This figure shows an example of an image to be inspected with tag information.This figure shows an example of information that can be associated with a prompt.This figure shows an example of a prompt set.This figure shows an example of a set of prompts following Figure 8A.This figure shows an example of a set of prompts that follows Figure 8B.This figure shows an example of a set of prompts that follows Figure 8C.This figure shows an example of a set of prompts that follows Figure 8D.This figure shows an example of a set of prompts that follows Figure 8E.This figure shows an example of a set of prompts that follows Figure 8F.This figure shows an example of a set of prompts that follows Figure 8G.This figure shows an example of metadata.This figure shows an example of the response information.This is a flowchart illustrating an example of a bridge inspection method.This figure shows the first concrete example of a prompt.This figure shows the first specific example of the prompt following Figure 12A.This figure shows the first specific example of the prompt that follows Figure 12B.This figure shows a second concrete example of a prompt.This figure shows a second specific example of the prompt that follows Figure 13A.This figure shows a third specific example of a prompt.This figure shows a third specific example of the prompt following Figure 14A.This figure shows a third specific example of the prompt that follows Figure 14B.This figure shows a fourth specific example of a prompt.This figure shows a fifth specific example of a prompt.This figure shows a fifth specific example of the prompt following Figure 16A.This figure shows a sixth specific example of a prompt.This figure shows a sixth specific example of the prompt following Figure 17A. Hereinafter, an embodiment of this disclosure will be described with reference to the drawings. Figure 1 shows an example of the configuration of the bridge inspection system according to this embodiment. Figure 1 illustrates a case where the bridge inspection system comprises a bridge inspection device 110 and a language model server 120. Figure 1 also illustrates a case where the bridge inspection device 110 and the language model server 120 are connected to each other via a network 130, including the Internet. The language model server 120 may reside on the cloud. The language model server 120, in response to requests from external devices such as the bridge inspection device 110, causes a language model 121, trained by machine learning using a large amount of data, to perform natural language processing and transmits the results to the external device. By utilizing the language model 121, for example, machine translation, text generation, and answering questions can be performed automatically. The language model 121 used in this embodiment is a trained model, such as a large language model (LLM) like GPT4-o®, LaMDA®, LLaMA®, and LLaVA®. The bridge inspection device 110 is a device that processes information to determine (inspect) the degree of damage in the inspection target area of a bridge. The inspection target area of the bridge (the part being inspected) may be the entirety of one or more bridge members, a part of one or more bridge members, or the entire bridge. In the following description, the inspection target area of the bridge will be abbreviated as "inspection target area" as needed. In this embodiment, we illustrate a case where the language model 121 outputs a result of determining the degree of damage to the bridge as response information, based on the information input from the bridge inspection device 110. Furthermore, in this embodiment, we illustrate a case where the i