Automated linking of diagnostic images to specific assets

Patent metadata

Field	Value
Publication number	US-12423947-B2
Application number	US-202318305047-A
Country	US
Kind code	B2
Filing date	Apr 21, 2023
Priority date	Apr 22, 2022
Publication date	Sep 23, 2025
Grant date	Sep 23, 2025

Abstract

Methods and apparatuses that utilize machine learning techniques to identify maintenance assets using sets of machine-health diagnostic images and link individual machine-health diagnostic images to the identified maintenance assets are described. The sets of machine-health diagnostic images may include a set of thermal images, a set of visible-light images, and/or a set of acoustic images. An identified maintenance asset may comprise an individual machine associated with a unique asset identifier. A diagnostic image linking system may acquire machine-health diagnostic images, apply object detection and other computer vision techniques to identify a particular machine within the machine-health diagnostic images, determine machine properties for the particular machine, generate a feature vector using the machine properties, select machine learning models corresponding with maintenance assets, generate predicted answers using the machine learning models, and generate an asset identifier for the particular machine based on the predicted answers.

First claim

The invention claimed is:

1. A system for identifying assets, comprising:
one or more processors in communication with a storage device configured to store a first machine learning model associated with a first machine asset and a second machine learning model associated with the first machine asset, wherein the one or more processors are configured to:
acquire a first set of images using a first imaging modality, and a second set of images using a second imaging modality that is different than the first imaging modality;
determine a first feature vector using the first set of images and a second feature vector using the second set of images;
generate a first predicted answer that the first feature vector corresponds with the first machine asset using the first machine learning model and a second predicted answer that the second feature vector corresponds with the first machine asset using the second machine learning model;
calculate an overall similarity score for the first machine asset using the first predicted answer and the second predicted answer; and
identify an asset identifier for the first machine asset based on the overall similarity score for the first machine asset.

2. The system of claim 1 , wherein:
the first imaging modality is a thermal imaging modality, and the first set of images comprises thermal images acquired using the thermal imaging modality; and
the second imaging modality is a visible light (VL) imaging modality, and the second set of images comprises VL images acquired using the VL imaging modality.

3. The system of claim 1 , wherein:
the one or more processors are configured to detect a machine within the first set of images and identify one or more machine properties for the machine using the first set of images; and
the one or more processors are configured to determine the first feature vector based on the one or more machine properties for the machine.

4. The system of claim 3 , wherein:
the one or more machine properties for the machine include external dimensions of the machine.

5. The system of claim 1 , wherein:
the one or more processors are configured to detect a machine within the first set of images and identify a weighted coefficient for the first predicted answer based on a dimension of the machine; and
the one or more processors are configured to calculate the overall similarity score using the weighted coefficient.

6. The system of claim 1 , wherein:
the one or more processors are configured to determine a weighted coefficient for the first predicted answer based on an estimated distance from a capturing device of the first set of images to the machine.

7. A method for identifying assets, comprising:
acquiring a first set of images using a first imaging modality, and a second set of images using a second imaging modality that is different than the first imaging modality;
identifying a first feature vector using the first set of images and a second feature vector using the second set of images;
generating a first predicted answer that the first feature vector corresponds with a first machine asset using a first machine learning model associated with the first machine asset and a second predicted answer that the second feature vector corresponds with the first machine asset using a second machine learning model associated with the first machine asset;
determining an overall similarity score for the first machine asset using the first predicted answer and the second predicted answer; and
outputting an asset identifier for the first machine asset based on the overall similarity score for the first machine asset.

8. The method of claim 7 , further comprising:
identifying a third machine learning model associated with a second machine asset and a fourth machine learning model associated with the second machine asset;
generating a third predicted answer that the first feature vector corresponds with the second machine asset using the third machine learning model and a fourth predicted answer that the second feature vector corresponds with the second machine asset using the fourth machine learning model;
detecting that the overall similarity score for the first machine asset is greater than a second overall similarity score for the second machine asset; and
outputting the asset identifier for the first machine asset in response to detecting that the overall similarity score for the first machine asset is greater than the second overall similarity score for the second machine asset.

9. The method of claim 7 , wherein:
the first imaging modality is a thermal imaging modality, and the first set of images comprises thermal images acquired using the thermal imaging modality; and
the second imaging modality an acoustic imaging modality, and the second set of images comprises acoustic images acquired using the acoustic imaging modality.

10. The method of claim 7 , further comprising:
detecting a machine captured by the first set of images;
identifying one or more machine properties for the machine using the first set of images; and
generating the first feature vector using the one or more machine properties for the machine.

11. The method of claim 10 , wherein:
the one or more machine properties for the machine include external dimensions of the machine.

12. The method of claim 10 , wherein:
the one or more machine properties for the machine include a set of internal dimensions of an interior space of the machine.

13. The method of claim 10 , wherein:
the one or more machine properties for the machine include a set of components located within an interior space of the machine.

14. The method of claim 7 , further comprising:
extracting metadata information from the first set of images; and
determining the overall similarity score using the metadata information.

15. The method of claim 7 , further comprising:
extracting location information from the first set of images; and
determining the overall similarity score using the location information.

16. The method of claim 7 , wherein:
the determining the overall similarity score for the first machine asset includes identifying a weighted coefficient for the first predicted answer based on an image type of the first set of images.

17. The method of claim 7 , further comprising:
detecting a machine within the first set of images using object detection, wherein the determining the overall similarity score for the first machine asset includes identifying a weighted coefficient for the first predicted answer based on at least one dimension of the machine.

18. The method of claim 7 , further comprising:
detecting a machine within the first set of images, wherein the determining the overall similarity score for the first machine asset includes identifying a weighted coefficient for the first predicted answer based on an estimated volume of the machine.

19. The method of claim 7 , further comprising:
detecting a machine within the first set of images, wherein the determining the overall similarity score for the first machine asset includes identifying a weighted coefficient for the first predicted answer based on an estimated distance from a capturing device of the first set of images to the machine.

20. One or more non-transitory storage devices containing processor readable code for configuring one or more processors to perform a method for identifying maintenance assets,
wherein the processor readable code configures the one or more processors to:
acquire a first set of thermal

Assignees

• Fluke Corp

Inventors

• Schmidt Matthew F
• Stuart Michael D
• Isfahani Seyed Navid Roohani
• Shastry Shreyas Shivaram
• Prabhakar Dileepa
• Ainsworth Ronald

Classifications

• G06T7/73

  using feature-based methods · CPC title

• G06T2207/10048

  Infrared image · CPC title

• G06V2201/10

  Recognition assisted with metadata · CPC title

• G06T7/50

  Depth or shape recovery · CPC title

• G06T7/62

  of area, perimeter, diameter or volume · CPC title