# US 2022/0044378 A1 · Apparatus for predicting wheel performance in vehicle and method therefor

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

- Number: US 2022/0044378 A1
- Document ID: US20220044378A1
- Status: Published application
- Kind: A1
- Application number: 17/235,362
- Filed: 2021-04-20
- Published: 2022-02-10
- Earliest claimed: 2020-08-04
- Assignees: Hyundai Motor Company; Kia Corporation; Sookmyung Women's University Industry-Academic Cooperation Foundation
- Applicants: Hyundai Motor Company; Kia Corporation; Sookmyung Women's university industry-academic cooperation foundation
- Inventors: Jong Ho PARK; Chang Gon KIM; Chul Woo JUNG; Sang Min LEE; Min Kyoo KANG; Ji Un LEE; Kwang Hyeon HWANG; Nam Woo KANG; So Young YOO; Seong Sin KIM; Sung Hee LEE
- Web page: https://patentgrep.com/patent/US20220044378A1

## CPC

- G06T 7/0004: {Industrial image inspection}
- G06T 7/001: {using an image reference approach}
- G06F 18/214: Generating training patterns; Bootstrap methods, e.g. bagging or boosting
- G06T 3/60: Rotation of whole images or parts thereof
- G06T 15/00: Three-dimensional [3D] image rendering
- G06V 10/774: Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
- G06V 10/82: using neural networks
- G06T 2200/08: involving all processing steps from image acquisition to 3D model generation
- G06T 2207/20076: Probabilistic image processing
- G06T 2207/20081: Training; Learning
- G06T 2207/20084: Artificial neural networks [ANN]
- G06T 2207/30164: Workpiece; Machine component
- G06V 2201/06: Recognition of objects for industrial automation
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING

## Abstract

An apparatus for predicting performance of a wheel in a vehicle: includes a learning device that generates a latent space for a plurality of two-dimensional (2D) wheel images based on a convolutional autoencoder (CAE), extracts a predetermined number of the plurality of 2D wheel images from the latent space, and learns a dataset having the plurality of 2D wheel images and performance values corresponding to the plurality of 2D wheel images; and a controller that predicts performance for the plurality of 2D wheel images based on a performance prediction model obtained by the learning device.

## Claims

### Claim 1

An apparatus for predicting performance of a wheel in a vehicle, the apparatus comprising:
a learning device configured to:
generate a latent space for a plurality of two-dimensional (2D) wheel images based on a convolutional autoencoder (CAE),
extract a predetermined number of 2D wheel images from the latent space, and
learn a dataset having the extracted 2D wheel images and performance values corresponding to the extracted 2D wheel images; and
a controller configured to predict performance for a 2D wheel image based on a performance prediction model as learned result of obtained by the learning device.

### Claim 2 (depends on 1)

The apparatus of claim 1, wherein the learning device performs transfer learning based on a learning model generated from the CAE.

### Claim 3 (depends on 1)

The apparatus of claim 1, wherein the learning device generates the plurality of 2D wheel images each of which having a shape different from that of a reference 2D wheel image using generative design.

### Claim 4 (depends on 1)

The apparatus of claim 1, wherein the learning device samples codes in the latent space using Latin hypercube sampling from normal distribution (LHSnorm) algorithm and extracts the predetermined number of 2D wheel images based on the sampled codes.

### Claim 5 (depends on 1)

The apparatus of claim 1, wherein the learning device converts the extracted 2D wheel images into three-dimensional (3D) wheel images and analyzes the 3D wheel images to predict the performance values.

### Claim 6 (depends on 5)

The apparatus of claim 5, wherein the learning device generates dataset in which the predicted performance values and the extracted 2D wheel images are related to each other.

### Claim 7 (depends on 1)

The apparatus of claim 1, wherein the learning device modifies the extracted 2D wheel images in the dataset to increase a number of datasets.

### Claim 8 (depends on 7)

The apparatus of claim 7, wherein the learning device rotates a spoke five times by 72 degrees on a source 2D wheel image to generate five 2D wheel images and flips the five 2D wheel images to generate ten 2D wheel images.

### Claim 9 (depends on 1)

The apparatus of claim 1, wherein the performance includes at least one of natural frequency of the wheel or a mass of the wheel according to a shape of a spoke.

### Claim 10

A method for predicting performance of a wheel in a vehicle, the method comprising:
generating, by a learning device, a latent space for a plurality of 2D wheel images based on a convolutional autoencoder (CAE);
extracting, by the learning device, a predetermined number of 2D wheel images from the latent space;
learning, by the learning device, a dataset having the extracted 2D wheel images and performance values corresponding to the extracted 2D wheel images; and
predicting, by a controller, performance for a 2D wheel image based on a performance prediction model as learned result of obtained by the learning device.

### Claim 11 (depends on 10)

The method of claim 10, wherein the learning of the dataset includes performing transfer learning based on a learning model generated from the CAE.

### Claim 12 (depends on 10)

The method of claim 10, wherein the generating of the latent space for the plurality of 2D wheel images includes generating the plurality of 2D wheel images each of which having a shape different from that of a reference 2D wheel image using generative design.

### Claim 13 (depends on 10)

The method of claim 10, wherein the extracting of the predetermined number of the plurality of 2D wheel images includes:
sampling codes in the latent space using Latin hypercube sampling from normal distribution (LHSnorm) algorithm; and
extracting the predetermined number of 2D wheel images based on the sampled codes.

### Claim 14 (depends on 10)

The method of claim 10, wherein the learning of the dataset includes:
converting the extracted 2D wheel images into three-dimensional (3D) wheel images; and
analyzing the 3D wheel images to predict the performance values.

### Claim 15 (depends on 14)

The method of claim 14, wherein the learning of the dataset further includes:
generating dataset in which the predicted performance values and the extracted 2D wheel images are related to each other.

### Claim 16 (depends on 10)

The method of claim 10, wherein the learning of the dataset includes modifying the extracted 2D wheel images in the dataset to increase a number of datasets.

### Claim 17 (depends on 16)

The method of claim 16, wherein the increasing of the number of the datasets includes:
rotating a spoke five times by 72 degrees on a source 2D wheel image to generate five 2D wheel images; and
flipping the five 2D wheel images to generate ten 2D wheel images.

### Claim 18 (depends on 10)

The method of claim 10, wherein the performance includes at least one of natural frequency of the wheel or a mass of the wheel according to a shape of a spoke.