Anomaly Detection for Manufacturing Applications Using Convolutional Autoencoders

Search for this publication on Google Scholar
Zouar, N. (2021). Anomaly Detection for Manufacturing Applications Using Convolutional Autoencoders. Unc Charlotte Electronic Theses And Dissertations.
Download PDF

Analytics

315 views ◎
290 downloads ⇓


Abstract

ABSTRACTNOURELISLAM ZOUAR. ANOMALY DETECTION FOR MANUFACTURING APPLICATIONS USING CONVOLUTIONAL AUTOENCODERS. (Under the direction of DR. MIN SHIN) Anomaly Detection in manufacturing environments is increasingly gaining popularity among companies and researchers. Computer based visual inspection systems are at the core of this interest. In the last few years, computer vision has made immense advancement thanks to deep learning. More specifically, unsupervised learning has proven its strength in this area, mainly due to its ability to deal with all kinds of anomalies and this is due to the philosophy used in this type of machine learning. This research work was motivated by the availability of what is arguably the most comprehensive public anomaly detection dataset, the MVTec dataset [1]. We implemented a pipeline that starts by preprocessing different categories of the MVTec dataset and ends by calculating the prediction accuracies and the inference times. This pipeline includes a convolutional autoencoder. We started by implementing a CAE that follows the description provided by [1]. Then, we have run many experiments using different CAE architectures found in recently published papers. In order to evaluate the performance of all the experimented CAEs, we used a brute force logic to find the best threshold. We used several portions of the calculated accuracies to find the best threshold. These portions were made using different percentages of the calculated accuracies, ranging from 70% to 100%.

Details
Author

Zouar, Nourelislam

Title

Anomaly Detection for Manufacturing Applications Using Convolutional Autoencoders

Physical Description

1 online resource (51 pages) : PDF

Date

2021

Degree Granting Institution

University of North Carolina at Charlotte

Abstract

ABSTRACTNOURELISLAM ZOUAR. ANOMALY DETECTION FOR MANUFACTURING APPLICATIONS USING CONVOLUTIONAL AUTOENCODERS. (Under the direction of DR. MIN SHIN) Anomaly Detection in manufacturing environments is increasingly gaining popularity among companies and researchers. Computer based visual inspection systems are at the core of this interest. In the last few years, computer vision has made immense advancement thanks to deep learning. More specifically, unsupervised learning has proven its strength in this area, mainly due to its ability to deal with all kinds of anomalies and this is due to the philosophy used in this type of machine learning. This research work was motivated by the availability of what is arguably the most comprehensive public anomaly detection dataset, the MVTec dataset [1]. We implemented a pipeline that starts by preprocessing different categories of the MVTec dataset and ends by calculating the prediction accuracies and the inference times. This pipeline includes a convolutional autoencoder. We started by implementing a CAE that follows the description provided by [1]. Then, we have run many experiments using different CAE architectures found in recently published papers. In order to evaluate the performance of all the experimented CAEs, we used a brute force logic to find the best threshold. We used several portions of the calculated accuracies to find the best threshold. These portions were made using different percentages of the calculated accuracies, ranging from 70% to 100%.

Genre

masters theses

Subjects--Topics

Computer science

Degree

M.S.

Keywords

Anomaly Detection
Computer Vision
Convolutional Autoencoder
MVTEC Dataset
Smart Manufacturing
Unsupervised Learning

Subject Area

Computer Science

Advisor(s)

Shin, Min
Welch, Stephen

Committee Members

Shin, Min
Welch, Stephen
Chen, Chen

Degree Note

Thesis (M.S.)--University of North Carolina at Charlotte, 2021.

Rights Statement

This Item is protected by copyright and/or related rights. You are free to use this Item in any way that is permitted by the copyright and related rights legislation that applies to your use. For other uses you need to obtain permission from the rights-holder(s). For additional information, see http://rightsstatements.org/page/InC/1.0/.

Rights Holder Information

Copyright is held by the author unless otherwise indicated.

Identifier

Zouar_uncc_0694N_12753

Permalink

http://hdl.handle.net/20.500.13093/etd:1866