# US 2026/0311323 A1 · Automated Defect Recognition and Determinations of Pore Cluster Compliance

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2026/0311323 A1
- Document ID: US20260311323A1
- Status: Published application
- Kind: A1
- Application number: 19/708,659
- Filed: 2026-06-15
- Published: 2026-10-08
- Earliest claimed: 2023-10-27
- Assignees: Relativity Space, Inc.
- Applicants: Relativity Space, Inc.
- Inventors: Christopher Sacco; Viraj Sinha; Kyle Buzza; Salik Syed; Andrew Taormina; Bryson Jones; Cem Tutum; Jeffrey Rossin
- Web page: https://patentgrep.com/patent/US20260311323A1

## CPC

- G06T 7/001: {using an image reference approach}
- G06T 7/11: Region-based segmentation
- G06T 7/62: of area, perimeter, diameter or volume
- G06T 7/64: of convexity or concavity
- G06V 10/26: Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
- G06V 10/32: Normalisation of the pattern dimensions
- G06V 10/762: using clustering, e.g. of similar faces in social networks
- G06V 10/764: using classification, e.g. of video objects
- G06V 10/82: using neural networks
- G06T 2207/10116: X-ray image
- G06T 2207/20081: Training; Learning
- G06T 2207/20084: Artificial neural networks [ANN]
- G06T 2207/20092: Interactive image processing based on input by user
- G06T 2207/30136: Metal
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING

## Abstract

A process can obtain a RT image of an article from the RT imaging system, the article comprising a reference feature, identify a first portion and a second portion of the first RT image, wherein the first portion and the second portion overlap, and classify each pixel in the first portion and the second portion using a pore classification model. A pore map can then be generated and mapped onto the article based on data received from the at least one sensor. A process can identify rejected regions on the pore map, wherein the rejected regions exceed a rejection threshold based on predefined rules. These rejected regions can then be displayed.

## Claims

### Claim 1

A robotic manufacturing inspection system configured to identify defects in manufactured articles, the robotic manufacturing inspection system comprising:
a manufacturing inspection robot, the manufacturing inspection robot comprising:
a robotic arm;
an inspection imaging system mounted to the robotic arm; and
at least one sensor capable of determining a position of the robotic arm; and
an image processing system, the image processing system configured to:
obtain an inspection image from the inspection imaging system and localization information,
wherein the inspection image is of an article created by manufacturing; and
wherein the localization information includes the position of the robotic arm when the inspection image is captured by the inspection imaging system;
adjust values of pixels within the inspection image to normalize the values of pixels within the inspection image against values of pixels in a reference image;
identify a plurality of image portions within the inspection image, wherein each image portion is either adjacent to or at least partially overlaps at least one other image portion;
classify at least one pixel in each image portion using a defect classification model, to obtain classifications of pixels across the plurality of image portions; and
generate a defect map based on applying one or more defect detection filters to the classifications of the pixels in the image portions, wherein the defect map includes identification of defects within the image portions and measurements of the defects based on a calibration of the inspection imaging system.

### Claim 2 (depends on 1)

The robotic manufacturing inspection system of claim 1, wherein the image processing system is further configured to:
display the defect map; and
receive an annotation from a user, wherein the annotation is characterized by at least a location on the defect map.

### Claim 3 (depends on 2)

The robotic manufacturing inspection system of claim 2, wherein the annotation comprises at least one transformation selected from the group consisting of adjusting contrast, adjusting brightness, zooming in, identifying blotches, identifying defects, using rulers to measure distances, notating text, and classifying pores.

### Claim 4 (depends on 1)

The robotic manufacturing inspection system of claim 1, wherein adjusting the values of pixels within the inspection image comprises setting the values so that a histogram of the inspection image approximately matches a histogram of the reference image.

### Claim 5 (depends on 1)

The robotic manufacturing inspection system of claim 1, wherein applying the one or more defect detection filters evaluates characteristics of each defect in the defect map; and wherein the characteristics include convexity, circularity, and a minimum size.

### Claim 6 (depends on 1)

The robotic manufacturing inspection system of claim 1, wherein applying the one or more defect detection filters provides a center point and radius of each detected defect in the defect map.

### Claim 7 (depends on 1)

The robotic manufacturing inspection system of claim 1, wherein the image processing system is further configured to determine whether to identify at least one region within the inspection image as part of a set of rejected regions based upon the defect map, when the at least one region exceeds a rejection threshold.

### Claim 8 (depends on 7)

The robotic manufacturing inspection system of claim 7, wherein the rejection threshold is based on a set of predetermined criteria including spacing, size, and frequency between defects in the defect map.

### Claim 9 (depends on 1)

The robotic manufacturing inspection system of claim 1, wherein, when classification of at least one pixel in a first image portion of the plurality of image portions differs relative to classification of at least one pixel in at least one second overlapping image portion of the plurality of image portions, generating the defect map further comprises following the classification with a highest confidence value.

### Claim 10 (depends on 1)

The robotic manufacturing inspection system of claim 1, wherein the article comprises a reference feature and the image processing system is further configured to obtain the calibration of the inspection imaging system based on the reference feature.

### Claim 11 (depends on 1)

The robotic manufacturing inspection system of claim 1, wherein the one or more defect detection filters comprise a plurality of defect detection filters, and each defect detection filter includes a different combination of values for parameters of convexity, circularity, and minimum size.

### Claim 12

A robotic additive manufacturing inspection system configured to identify porosity defects in manufactured articles, the robotic additive manufacturing inspection system comprising:
an additive manufacturing robot, the additive manufacturing robot comprising:
a robotic arm;
a radiographic testing (RT) imaging system mounted to the robotic arm; and
at least one sensor capable of determining a position of the robotic arm; and
an image processing system, the image processing system configured to:
obtain an RT image from the RT imaging system and localization information,
wherein the RT image is of an article created by additive manufacturing; and
wherein the localization information includes the position of the robotic arm when the RT image is captured by the RT imaging system;
identify a plurality of image portions within the RT image;
classify at least some pixels in each image portion as “pore” or “not pore,” using a pore classification model, to obtain classifications of pixels in the plurality of image portions;
generate a pore map based on applying one or more porosity detection filters to the classifications of the pixels in the image portions, wherein the pore map includes identification of pores within the image portions; and
determine whether to identify at least one region within the RT image as a rejected region based upon the pore map, when the at least one region exceeds a rejection threshold.

### Claim 13 (depends on 12)

The robotic additive manufacturing inspection system of claim 12, wherein the image processing system is further configured to adjust brightness values of pixels within the RT image to normalize the brightness values against brightness values of pixels in a reference image.

### Claim 14 (depends on 12)

The robotic additive manufacturing inspection system of claim 12, wherein the article created by additive manufacturing comprises at least one weld produced from a three-dimensional (3D) print.

### Claim 15 (depends on 12)

The robotic additive manufacturing inspection system of claim 12, wherein the pore classification model comprises a U-Net convolutional neural network; and wherein the pore map is a segmentation map.

### Claim 16

A robotic additive manufacturing inspection system configured to identify defects in manufactured articles, the robotic additive manufacturing inspection system comprising:
an additive inspection robot, the additive inspection robot comprising:
a robotic arm;
a radiographic testing (RT) imaging system mounted to the robotic arm; and
at least one sensor capable of determining a position of the robotic arm; and
an image processing system, the image processing system configured to:
obtain an RT image from the RT imaging system and localization information,
wherein the RT image is of an article created by additive manufacturing; and
wherein the localization information includes the position of the robotic arm when the RT image is captured by the RT imaging system;
adjust brightness values of pixels within the RT image to normalize the brightness values;
classify a plurality of pixels in the RT image as “pore” or “not pore,” using a pore classification model, to obtain classifications of the plurality of pixels in the RT image;
generate a pore map based on applying one or more porosity detection filters to the classifications of the pixels in the RT image, wherein the pore map includes identification of pores within the RT image and measurements of the pores based on a measurement calibration of the RT imaging system; and
determine whether to identify at least one region within the RT image as a rejected region based upon the pore map, when the at least one region exceeds a rejection threshold.

### Claim 17 (depends on 16)

The robotic additive manufacturing inspection system of claim 16, wherein the image processing system is further configured to identify a plurality of image portions within the RT image, wherein each image portion at least partially overlaps at least one other image portion.

### Claim 18 (depends on 17)

The robotic additive manufacturing inspection system of claim 17, wherein identifying the plurality of image portions within the RT image is performed using a sliding-window sub-process.

### Claim 19 (depends on 16)

The robotic additive manufacturing inspection system of claim 16, wherein the article created by additive manufacturing comprises at least one weld produced from a three-dimensional (3D) print.

### Claim 20 (depends on 16)

The robotic additive manufacturing inspection system of claim 16, wherein the pore classification model comprises a U-Net convolutional neural network; and wherein the pore map is a segmentation map.
