Določanje lege strukturne spremembe z uporabo podatkovnih pristopov

Knific, Jaka (author), Slavič, Janko (mentor)

Za zaznavanje napak na izdelkih se v industriji pogosto uporablja strojni vid. V primerih, ko to ni mogoče, moramo razviti alternativne metode. V sklopu naloge je bil zasnovan model z avtoenkodersko konvolucijsko nevronsko mrežo, ki na podlagi spektrogramov dinamskega odziva rekonstruira sliko izdelka s prikazano lego napake. Ugotovljeno je bilo, da izdelan model v večini primerov natančno določi lego napake na izdelku in rekonstruira ustrezno sliko.

• Language: Slovenian
• Keywords: konvolucijska nevronska mreža, pytorch, strukturna dinamika, rekonstrukcija slike, avtoenkoder
• Work type: Master's thesis/paper
• Organization: FS - Faculty of Mechanical Engineering
• Year: 2024
• PID: 20.500.12556/RUL-165105
• Publication date in RUL: 23.11.2024

Secondary language

• Language: English
• Title: Data-based location identification of structural change

Abstract

In industry, machine vision is often used for defect detection in products. In cases where this is not feasible, alternative methods must be developed. Within this thesis, a model with an autoencoder convolutional neural network was designed to reconstruct an image of the product, displaying the location of the defect, based on spectrograms of the dynamic response. It was found that in most cases the model accurately determines the defect location on the product and reconstructs the corresponding image.

• Keywords: convolutional neural network, pytorch, structural dynamics, image reconstruction, autoencoder