Marbun, Abri Andry Saresa (2022) Deteksi Cacat Pada Permukaan Hasil Pengelasan Friction Stir Welding (FSW) Berbasis Image Processing Dengan Jaringan Fully Convolution Network (FCN) Untuk Material Aa6061-T651. Masters thesis, Institut Teknologi Sepuluh Nopember.
![[thumbnail of 6007201016-Master_Thesis.pdf]](http://repository.its.ac.id/style/images/fileicons/text.png) |
Text
6007201016-Master_Thesis.pdf - Accepted Version Download (2MB) |
Abstract
Monitoring hasil pengelasan perlu dilakukan agar proses pengelasan yang dilakukan telah sesuai dengan ketentuan dan standart yang digunakan selain itu untuk memberikan informasi kepada operator apabila terjadi hambatan dan penyimpangan, serta memberi masukan dalam melakukan evaluasi. Pengujian hasil pengelasan dapat dilakukan dengan metode non desctructive test (NDT) seperti visual inspection yaitu pemeriksaan hasil pengelasan dengan cara pengamatan langsung pada permukaan hasil pengelasan melalui visual manusia. Dengan pengamatan langsung dapat diketahui cacat yang terjadi pada permukaan logam hasil pengelasan. Selain pengamatan langsung inspeksi juga dapat dilakukan dengan mesin berbasis komputer AVIS (automated visual inspection systems). Penelitian ini mengembangkan metode pengolahan citra untuk memonitoring hasil pengelasan Friction Stir Welding (FSW) pada material AA6061-T651 melalui visual inspection yang dilakukan yaitu mendeteksi cacat flash pada permukaan hasil pengelasan FSW. Deteksi dilakukan dengan menggunakan metode deep learning yaitu Fully Convolutional Network (FCN). Jaringan FCN yang dirancang menggunakan 5 convolutional layer, 5 ReLU layer, 5 batchnormalization layer, 5 maxpool layer dan 1 fully connected layer. Sebagai citra pembelajaran, citra validasi dan pengujian digunakan citra hasil pengelasan yang telah dikategorikan menjadi kelas normal dan defect dengan jumlah seluruh citra sebanyak 566 citra. Dari hasil pembelajaran dilakukan pengujian jaringan dengan menggunakan citra yang ada untuk dapat mengetahi performa dari jaringan yang telah dirancang. Berdasarkan hasil pengujian visual inspection oleh jaringan FCN, jaringan mampu mengklasifikasikan citra hasil pengelasan yang digunakan sebagai testing menjadi citra hasil pengelasan dengan kelompok baik dan kelompok cacat (defect). Kinerja jaringan FCN yang dirancang mampu mencapai akurasi deteksi cacat flash sebesar 97.6% dan menghasilkan performa jaringan yaitu precision 96,8% dan sensivity 96,8%. Dari hasil-hasil tersebut diketahui keberhasilan jaringan FCN dalam melakukan visual inspection sangat baik dan dapat digunakan dalam pengamatan hasil pengelasan FSW. Dengan jaringan FCN yang dirancang dapat digunakan dalam penelitian selanjutnya untuk memonitoring proses pengelasan FSW, sehingga memungkinkan untuk mendeteksi cacat flash pada proses pengelasan FSW sedini mungkin.
==============================================================================================================================
Monitoring welding results needs to be carried out so that the welding process carried out is in accordance with the provisions and standards used in addition to providing information to the operator in the event of obstacles and deviations, as well as providing input in conducting evaluations. Testing welding results can be done with non-desctructive test (NDT) methods such as visual inspection , namely the examination of welding results by direct observation on the surface of welding results through human visuals. By direct observation, it can be known the defects that occur on the surface of the metal from welding. In addition to direct observation, inspections can also be carried out with computer-based machines AVIS (automated visual inspection systems). This study developed an image processing method to monitor the results of Friction Stir Welding (FSW) welding on material AA6061-T651 through visual inspection, which was carried out, namely detecting flash defects on the surface of the FSW welding results. Detection is carried out using a deep learning method, namely the Fully Convolutional Network (FCN). The FCN network is designed using 5 convolutional layers, 5 ReLU layers, 5 batchnormalization layers, 5 maxpool layers and 1 fully connected layer. As a learning image, validation and testing images are used welding images that have been categorized into normal and defect classes with a total of 566 images. From the learning results, network testing is carried out using existing images to be able to determine the performance of the network that has been designed. Based on the results of visual inspection testing by the FCN network, the network is able to classify the welding result image used as testing into an image of welding results with good groups and defect groups. The performance of the designed FCN network is able to achieve flash flaw detection accuracy of 97.6% and produce network performance, namely 9 6.8% precision and 96.8% sensivity. From these results, it is known that the success of the FCN network in conducting visual inspection is very good and can be used in observing FSW welding results. With the designed FCN network can be used in subsequent research to monitor the FSW welding process, making it possible to detect flash defects in the FSW welding process as early as possible.
| Item Type: | Thesis (Masters) |
|---|---|
| Additional Information: | RTM 621.367 Mar d-1 2020 |
| Uncontrolled Keywords: | FSW, deep learning, image processing, FCN dan akurasi, accuracy. |
| Subjects: | R Medicine > R Medicine (General) > R858 Deep Learning T Technology > TA Engineering (General). Civil engineering (General) > TA1637 Image processing--Digital techniques. Image analysis--Data processing. |
| Divisions: | Faculty of Industrial Technology and Systems Engineering (INDSYS) > Mechanical Engineering > 21101-(S2) Master Thesis |
| Depositing User: | Anis Wulandari |
| Date Deposited: | 17 Jan 2023 03:45 |
| Last Modified: | 11 Dec 2024 08:11 |
| URI: | http://repository.its.ac.id/id/eprint/95433 |
Actions (login required)
 |
View Item |