Deteksi Anomali Suara Mesin Normal Dan Abnormal Dengan Metode Supervised Dan Unsupervised Learning.

Herlambang, Bagus Adhi (2022) Deteksi Anomali Suara Mesin Normal Dan Abnormal Dengan Metode Supervised Dan Unsupervised Learning. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 02311840000083-undergraduate_thesis.pdf Restricted to Repository staff only Download (3MB)

Abstract

Deteksi anomali suara mesin merupakan salah satu implementasi dari deep learning di industri dalam hal monitoring kondisi mesin. Anomali dalam suara yang dihasilkan oleh mesin dapat menunjukkan adanya indikasi kesalahan atau kerusakan, dengan melakukan deteksi anomali lebih awal dapat menghindari serangkaian masalah terutama dalam hal pemeliharaan prediktif. Anomaly detection umumnya termasuk ke dalam unsupervised learning karena sulitnya mendapatkan informasi data normal atau anomali di industri. Pada penelitian tugas akhir kali ini akan menganalisis metode terbaik antara supervised learning dan unsupervised learning untuk deteksi anomali suara mesin normal dan abnormal. Metode supervised learning menggunakan model Bidirectional Generative Adversarial Network (BiGAN) sedangkan untuk unsupervised learning menggunakan model MobileNet V2 dan Auto Encoder. Hasil disajikan dalam nilai AUC dan pAUC dengan menggunakan rata-rata harmonik pada setiap jenis mesin pada dataset DCASE 2022. Hasil metode supervised learning pada AUC sebesar 57,78% dan pAUC sebesar 56,48%. Sedangkan pada metode unsupervised learning dengan menggunakan model MobileNet V2 menghasilkan nilai AUC sebesar 57,00% dan pAUC sebesar 55,77%, hasil model Auto Encoder pada nilai AUC sebesar 52,21% dan pAUC sebesar 53,75%. Dari hasil yang didapatkan terlihat bahwa metode supervised learning lebih baik dari unsupervised learning.
===================================================================================================================================
Machine sound anomaly detection is one of the implementations of deep learning in the industry in terms of monitoring machine conditions. Anomalies in the sound produced by the machine can indicate an indication of a fault or malfunction, by performing anomaly detection early it can avoid a series of problems especially in terms of predictive maintenance. Anomaly detection is generally included in unsupervised learning because of the difficulty of obtaining normal or anomalous data information in the industry. In this final project, we will analyze the best method between supervised learning and unsupervised learning for the detection of normal and abnormal machine noise anomalies. The supervised learning uses the Bidirectional Generative Adversarial Network (BiGAN) model, while unsupervised learning uses the MobileNet V2 and Auto Encoder models. The results are presented in AUC and pAUC values using the average harmonics for each type of machine in the DCASE 2022 dataset. The results of the supervised learning on AUC are 57.78% and pAUC are 56.48%. While the unsupervised learning using the MobileNet V2 model produces an AUC value of 57.00% and a pAUC of 55.77%, the results of the Auto Encoder model at an AUC value of 52.21% and a pAUC of 53.75%. From the results obtained, it can be seen that the supervised learning is better than unsupervised learning.

Item Type:	Thesis (Other)
Additional Information:	RSF 621.811 Her d-1 2022
Uncontrolled Keywords:	anomaly detection, deep learning, supervised learning, unsupervised learning. anomaly detection, deep learning, supervised learning, unsupervised learning.
Divisions:	Faculty of Industrial Technology and Systems Engineering (INDSYS) > Physics Engineering > 30201-(S1) Undergraduate Thesis
Depositing User:	Mr. Marsudiyana -
Date Deposited:	11 May 2026 07:46
Last Modified:	11 May 2026 07:46
URI:	http://repository.its.ac.id/id/eprint/133130

Actions (login required)

View Item