Skip to main navigation Skip to search Skip to main content

Identification and Attitude Estimation of Intelligent Welding Special-shaped Tubes Based on Two-Channel Convolutional Neural Network

  • Harbin Institute of Technology

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Facing toward the development direction of intelligent aerospace manufacturing, the paper, based on deep learning, proposes a model with two-channel convolutional neural network, aiming to solve the bottleneck of flexible robots which is used for intelligent laser welding. One channel is used to classify the special-shaped tubes and the other channel is used to estimate the two-dimensional attitude of the special-shaped tubes. Experiments show that the accuracy of identification for the model on the test set of special-shaped tube can reach 93.6%, while the accuracy of traditional identification method is only 86.3%.Meanwhile, the average error of attitude estimation is 5.49°, while the average error of the traditional method for attitude estimation is 7.69°. According to the data outcome, the model, compared to the traditional method, shows obvious excellence.

Original languageEnglish
Title of host publicationProceedings of the 32nd Chinese Control and Decision Conference, CCDC 2020
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages1457-1462
Number of pages6
ISBN (Electronic)9781728158549
DOIs
StatePublished - Aug 2020
Event32nd Chinese Control and Decision Conference, CCDC 2020 - Hefei, China
Duration: 22 Aug 202024 Aug 2020

Publication series

NameProceedings of the 32nd Chinese Control and Decision Conference, CCDC 2020

Conference

Conference32nd Chinese Control and Decision Conference, CCDC 2020
Country/TerritoryChina
CityHefei
Period22/08/2024/08/20

Keywords

  • Attitude estimation
  • Convolutional neural network
  • Identification
  • Special-shaped tube

Fingerprint

Dive into the research topics of 'Identification and Attitude Estimation of Intelligent Welding Special-shaped Tubes Based on Two-Channel Convolutional Neural Network'. Together they form a unique fingerprint.

Cite this